id
stringlengths 27
33
| source
stringclasses 1
value | format
stringclasses 1
value | text
stringlengths 13
1.81M
|
|---|---|---|---|
warning/0506/astro-ph0506552.html
|
ar5iv
|
text
|
# The infrared Hourglass cluster in M8Based in part on observations with NASA/ESA Hubble Space Telescope obtained from the archive at the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS 5-26555. Based in part on observations obtained at European Southern Observatory, La Silla, Chile.
## 1 Introduction
The Lagoon Nebula (Messier 8 = NGC 6523-NGC 6530) is an extended H ii region mainly ionized by two O-type stars, 9 Sagitarii \[O4 V((f))\] and HD 165052 (O6.5 V + O7.5 V). It is embedded within a molecular cloud which extends to the star cluster NGC 6530. This very young open cluster is believed to be the starting point for a sequential star formation process (Lightfoot et al. 1984) which is still active in the region. Within M8’s core lies a distinctive bipolar nebula called the Hourglass, a blister-type H ii region which has been produced by the O7.5 V star Herschel 36 (Her 36). The Hourglass, which is about 15” EW $`\times `$ 30” NS in size, is believed to be an ionized cavity in an inhomogeneous clumpy molecular cloud. The Hourglass region also harbours the ultracompact H ii region G5.97-1.17 as well as a number of infrared sources, first observed by Allen (1986), which may form a cluster of very young hot stars, analogous to the Orion Trapezium (Allen 1986). Narrow band optical imaging with HST reveals that G5.97-1.17 could be a young star surrounded by a circumstellar disk that is being photoevaporated by Her 36, similar to the so-called proplyds seen in the Orion Nebula (Stecklum et al. 1998). Furthermore, strong H<sub>2</sub> line emission is produced from around the Hourglass, showing a morphology very similar to that of the CO J=3-2 distribution in the region (White et al. 1997; Burton 2002). This fact and the detection of a jet-like feature extending from Her 36 in HST images (Stecklum et al. 1995) suggest there might be a molecular outflow in the core of M8 (Burton 2002). Recently, Rauw et al. (2002) reported X-ray emission from Her 36 as well as probably diffuse X-ray emission from the Hourglass region that might reveal a bubble of hot gas produced by the interaction of the stellar wind of Her 36 with the denser part of the molecular cloud.
In this paper, new high-resolution near-infrared images and photometry of the field surrounding the Hourglass are presented and compared with archival HST emission-line images. All these data provide strong evidence that the core of M8 is an important region of active star formation and pre-main sequence stellar evolution.
## 2 Observations and Data Reduction
### 2.1 Infrared Images
$`J`$, $`H`$ and $`Kshort`$ ($`K_s`$) images of the Hourglass region were obtained on September 26 1999, using the near-infrared camera IRCAM, attached to the 2.5-m Du Pont Telescope at Las Campanas Observatory (LCO), Chile. IRCAM was equipped with a NICMOS III $`256\times 256`$ array (Persson et al. 1992) and has a pixel scale of 0$`\stackrel{}{.}`$35 $`px^1`$. The seeing during the observations was typically of 0$`\stackrel{}{.}`$8-0$`\stackrel{}{.}`$9 giving an optimum sampling. In each band, five partially overlapping frames were taken, covering a total area of 135”$`\times `$139” centered approximately at the position of Her 36. The total on-source times were 360 s in $`K_s`$, 500 s in $`H`$ and 600 s in $`J`$. The frames were combined (median-averaged) after being linearised, dark- and sky-substracted, flat-fielded and cleaned of bad pixels, using SQIID processing routines layered in IRAF<sup>1</sup><sup>1</sup>1IRAF is distributed by NOAO, operated by AURA, Inc., under agreement with NSF.. Several standard stars selected from Persson et al. (1998) were observed during the night; these fields were reduced following a similar procedure. The combined (final) images of the Hourglass were then shifted to align stars in each frame.
### 2.2 HST/WFPC2 Imaging
A partial HST/WFPC2 coverage of the Hourglass Nebula is available from the public database. These images include the brightest part of the nebula and the star Her 36, both centred in the Planetary Camera chip. The orientation of the images (P.A.= 56$`\stackrel{}{.}`$0919) allow to cover the west part of the nebula and they correspond to about $`2/3`$ of the field observed in our infrared images. The WFPC2 data set (Proposal ID. 6227) was obtained in 1995 July 13, 14, and September 16, using several narrow band and broad band filters. These observations were retrieved from the HST Archive and processed using the “on-the-fly” pipeline. In this work, we use images obtained in F487N (H$`\beta `$), F656N (H$`\alpha `$), F658N (\[N ii\]), F673N (\[S ii\]) narrow band filters, and F547M ($`v`$), F814W (Johnson $`I`$) broad band filters. The filter characteristics are described in the WFPC2 Instrument Handbook. The first steps of data reduction were straightforward, using IRAF and STSDAS tasks to do cosmic ray cleaning, and integrating the four individual WFPC2 CCD images into a single mosaic. In order to obtain flux calibrated ”pure nebular” H$`\beta `$ and H$`\alpha `$ images, we followed the procedure given by O’Dell & Doi (1999). A ”pure” H$`\beta `$ calibrated image was derived from the F487N image, using the F547M images as continuum. A ”pure” H$`\alpha `$ calibrated image was obtained from F656N ”line” image, considering the contamination of \[N ii\] emission lines through the F658N filter, and again using the F547M as continuum image. In the case of F673N filter used to derive ”pure” \[S ii\] calibrated emission line image, the continuum was interpolated between the F547M and F814W images and the zero-points determined by synphot (STScI 1999).
### 2.3 Long-slit Echelle Spectroscopy
Two long-slit spectra obtained with ESO Multi-Mode Instrument (EMMI) on NTT were retrieved from the ESO Archive Facility. The spectra were obtained on July 31st 2000 in the REMD (red medium dispersion) configuration with echelle (#10) grating. The order containing H$`\alpha `$ and \[N ii\] emission lines was isolated using the filter Ha#596. The detector was a $`2048\times 2048`$ Tek CCD with 24-$`\mu `$m pixels, given a pixel scale of 0$`\stackrel{}{.}`$265 $`\times `$ 0.041 Å and a spatial extent along the slit of $`330^{\prime \prime }`$. The spectra were obtained with a $`P.A.=0^{}`$. The total exposure time was 1800 s.
## 3 Results
Figure 1 (left-hand panel) shows the $`K_s`$ frame of the field surrounding the Hourglass Nebula. Figure 1 (right-hand panel) is a $`JHK_s`$ colour composite image, produced combining the $`J`$ (1.25 $`\mu `$), $`H`$ (1.65 $`\mu `$) and $`K_s`$ (2.2 $`\mu `$) images as blue, green, and red channels, respectively. This false-colour picture shows the effects of extinction, the differences in colour of the stars mainly being due to differences in reddening. This figures confirm the existence of an infrared star cluster around Her 36, as was recently reported by Bica et al. (2003).
### 3.1 Distance
Most previous investigators estimated the distance to NGC 6530 as 1.8 Kpc ($`V_0M_V=11.2511.3`$)(van Altena & Jones 1972; Sagar & Joshi 1978; van den Ancker et al. 1997). A smaller value was however obtained by Kilambi (1977) ($`V_0M_V=10.7`$). More recently, Sung et al. (2000) calculated the distance modulus of several individual stars, finding that some of them gave values near to $`V_0M_V=11.1511.35`$, while other early-type stars gave somewhat smaller values ($`V_0M_V=10.511`$). A more robust determination by Prisinzano et al. (2005), considering a sample that reaches down to $`V23`$, lead to the value of $`V_0M_V=10.48`$, which means a significantly lower cluster distance of 1.25 Kpc. This distance determination is based on more complete colour-magnitude diagrams, for which it is possible to define the blue envelope of the star distribution with $`V>14`$ because of the obscuration effect of the giant molecular cloud which strongly reduces the detection of background stars at optical wavelengths.
Although this indetermination in the distance to M8 does not change the main conclusions of this work, we explore the possibility of deriving our own estimate by considering a group of stars around Her 36 with known spectral types. We gathered optical photometry from Sung et al. (2000) and from the Galactic O Star Catalog (Maíz-Apellániz et al. 2004a), as well as near-infrared photometry from this work or from the 2MASS All-Sky Point Source Catalog (PSC) (Second Incremental Data Release, Cutri et al. 2000) for Her 36, 9 Sgr (HD 164794), W 9 (HD 164816), SCB 182, SCB 325 and SCB 354 (hereafter we use “SCB number” to refer to the sources from Sung et al. 2000). For the O-type stars, the adopted spectral types were either obtained from Maíz-Apellániz et al. (2004a) (Her 36, 9 Sgr) or derived from our spectroscopic data (W 9). The stars 9 Sgr and W 9 are both suspected to be spectroscopic binaries (cf. Mason et al. 1998). Arias et al. (2005) have detected both components in W 9, resulting a spectral classification of O 9.5V + B0 V, and have also confirmed the binary nature of 9 Sgr and Her 36. For B-type stars (SCB 182, SCB 325, SCB 354) we adopted the spectral types from van den Ancker et al. (1997).
We used the code CHORIZOS developed by Maíz-Apellániz (2004) to derive the individual distance modulii of these stars. CHORIZOS is a code that uses $`\chi ^2`$ minimization to find all models compatible with the observed data set in the model ($`N`$-dimensional) parameter space, in our case, broadband photometry and spectral types. For a complete description of how this method works see Maíz-Apellániz (2004). We considered TLUSTY (Lanz & Hubeny 2003) atmosphere models and Kurucz (Kurucz 2004) models for the spectral energy distribution (SED) of O-type and B-type stars, respectively. Figure 2 shows the SED corresponding to the best CHORIZOS fit (spectrum) and the synthetic photometry associated with the best SED (stars) for Her 36. The photometric data used for the CHORIZOS fit are also shown in the figure. $`H`$ and $`K_s`$ filters were not considered due to the infrared excess of Her 36, which is readily apparent in the plot. Table 1 lists all the compiled quantities and the derived parameters, i.e., the colour excess $`E(44055495)`$ and the ratio of total to selective absorption $`R_{5495}`$, which are the monocromatic equivalents to the usual $`E(BV)`$ and $`R_V`$, respectively (for a detailed explanation see Maíz-Apellániz 2004), and the distance modulus. The $`\chi ^2`$ values corresponding to the best fit in each case have also been included in the last column. The main source of error comes from the adopted values of the absolute magnitudes. The $`R_{5495}`$ value is close to the canonical one of 3.1 only for W9; for the rest of the stars it shows ”anomalous” values. In particular, the value of $`R_{5495}=5.36`$ derived for Her 36, which represent one of the highest $`R_V`$ values known (cf. Cardelli, Clayton, & Mathis 1989), is in excellent agreement with earlier estimations. Besides, as mentioned before, the analysis of its SED using CHORIZOS shows that this star presents an evident infrared excess, which is also apparent in the near-infrared colour-colour diagram (see Section 3.3).
As also shown in Table 1, the four earliest stars lead to an average distance modulus of nearly 10.5 (1.25 Kpc), clearly lower than the previously adopted by other authors and similar to the value derived by Kilambi et al. (1977). On the other hand, the distance modulii derived for the two B7-type stars is somewhat larger ($`11.3`$). Such problem is illustrated in the Hertzprung-Russell diagram compiled from the CHORIZOS output which is shown in Figure 3. Ellipses indicate the 68% likelihood contour for each star. The main sequence for $`5\mathrm{log}d5=10.5`$ has been marked with a solid line. As seen in the plot, such a distance is compatible with the CHORIZOS output for the four earliest stars but not for the latest two.
We note here that the association of 9 Sgr, located $`3^{}`$ to the NE of Her 36, with the Hourglass region is based not only in a common distance but also in the observed morphology of the gas and dust in the region. Besides a dust “finger-shape” structure that will be discussed later, the northern part of the nebula seems to be illuminated by this massive star. This is especially evident in the photodissociation regions (PDRs) facing to 9 Sgr observed in HST images (see Figures 11 and 12 in Section 3.8).
Based on the values derived for the earliest stars, which are systematically smaller than the previously published, we adopt a distance modulus of $`V_0M_V=10.5`$ for the Hourglass region. A posible explanation for the disagreement with the values derived by other authors is considered here.
As pointed out by van den Ancker et al. (1997), the study of a very young open cluster like NGC 6530 can not be performed using average extinction laws, since each star has its own individual extinction characteristics. In fact, anomalous extinction laws have been found for several stars embedded in the cluster. This may be a serious obstacle when deriving distance modulii from the observed optical colours, as done in most previous studies. In the infrared however, the greatly reduced extinction and the independence of the extinction law with the ratio of total to selective absorption $`R_V`$ (Jones & Hyland 1980; Cardelli et al. 1989; Martin & Whittet 1990; Whittet et al. 1993), make distance determinations more reliable.
An alternative interpretation for the disagreement between the distance values derived by different authors may be that, when looking at the direction of NGC 6530, we are actually seeing distinct groups of stars placed at different depths. Figure 10 in van den Ancker et al (1997) shows the distribution of the distances of individual stars in the cluster, from which they estimated an average distance of 1.8 kpc. However, the distribution has a remarkable secondary peak around 1.5 kpc. In this way, the two B7-type stars analyzed with CHORIZOS (SCB 325 and SCB 354) lead to distance modulii of 11.04 and 11.55, respectively. Both stars are located to the east of the Hourglass Nebula, in a region with lower reddening, as inferred from their colour excesses $`E(44055495)0.26`$. So they presumably locate deeper in the spiral arm but are being observed through a window relatively free of dust. Finally, as mentioned above, Sung et al. (2000) also obtained two values for the distance modulus, but they left out the shorter distance estimate with the argument that many of the B-type stars were probably binaries.
Distance estimations to such very young open clusters are extremely complicated since they require the knowledge of the extinction, which is really due to circumstellar rather than intracluster material (van den Ancker et al. 1997). Each star have its own extinction law, and hence a very detailed study is needed. We conclude that a definitive distance to the Hourglass Nebula is not yet established, but the CHORIZOS analysis of the earliest stars in the region is in favour of a distance modulus of 10.5 (1.25 kpc).
### 3.2 Photometry of infrared sources
Point-spread function (PSF) photometry was performed using IRAF/DAOPHOT-II software running in a Linux workstation. Stellar-like sources with fluxes significantly above the mean background were extracted using a threshold flux of $`3\sigma `$. A set of PSF star candidates was selected in each image avoiding nebular and crowded regions. The PSF was calculated using a lorentz function and one look-up table. The final PSF photometry was performed on all the extracted sources using an aperture 3 pixels in radius. This corresponds to an aperture diameter which is roughly that of the measured full width at half-power of a typical stellar image. Aperture corrections were computed between apertures of 3 and 15 pixels in radius using the same PSF stars. The final number of detected sources was 762 in $`K_s`$, 748 in $`H`$ and 653 in $`J`$ with sensitivity limits of 17.5, 18.5 and 19.2, respectively. The photometric limits were assumed to be the magnitude corresponding to the peak of the star distribution. This maximum occurred at about 1.5 mag brighter than the magnitude where the distribution fell to zero, i.e., 16.0 in $`K_s`$, 17.0 in $`H`$ and 17.7 in $`J`$. The average photometric errors in all colours are 0.03 for sources with mag brighter than these limits and up to 0.2 for the faintest sources. Table 4 (in separated file) gives the position and photometry of all the infrared sources detected in the field. Running number sources are in column 1, column 2 and 3 are right ascension and declination (J2000), columns 4, 5, 6, 7, 8, 9 are $`J`$, $`H`$, $`K_s`$ magnitudes and their errors, columns 10 and 11 are $`JH`$ and $`HKs`$ colours, and column 12 contains comments related to the identification of optical counterparts.
In order to detect systematic photometric errors produced by aperture corrections and standard zero-points, we compared the magnitudes derived from our photometry for some relatively bright stars with those from the 2MASS All-Sky Point Source Catalog (PSC) (Second Incremental Data Release, Cutri et al. 2000). Considering only the sources in 2MASS with $`photometric`$ quality “$`AAA`$”, we found an average difference in $`J`$, $`H`$ and $`K_s`$ magnitudes of $`0.005\pm 0.085`$, $`0.016\pm 0.052`$ and $`0.008\pm 0.067`$, respectively, so no further correction was performed.
Positions in equatorial coordinates for the individual sources were established based on the 2MASS Point Source Catalogue (All-Sky 2003). The rms residuals between the positional tables from this work and from 2MASS database are found to be 0$`\stackrel{}{.}`$11 and 0$`\stackrel{}{.}`$12, in $`\alpha `$ and $`\delta `$ respectively.
### 3.3 Colour-Colour and Colour-Magnitude Diagrams
A near-infrared colour-colour (CC) diagram of the Hourglass region is illustrated in Figure 4 for the 647 sources in the field detected in all three wavelength bands. Also plotted in Figure 4 is the locus of points corresponding to the position of the unreddened main sequence and the position of the red giants. The two parallel dashed lines represent the reddening vectors for early- (O3 V) and late-type (M0 III) stars, determined using the values of extinction from Rieke & Lebofsky (1985); their length corresponds to $`A_V=20`$ mag. Figure 5 shows the $`K_s`$, $`(HK_s)`$ colour-magnitude (CM) diagram for the same stars. The position of the main sequence has been plotted, corrected to an apparent distance modulus of 10.5, which, as was previously discussed, is appropiated for the Hourglass region. The reddening vector for an O7 V is also plotted for a visual extinction of 20 mag. The vertical solid line denotes the location of giant stars (Koornneef 1983; Zombeck 1990) for the distance of the Galactic Bulge (8 kpc) and reddened by $`E(HK_s)=0.64`$ due to the interstellar component. The position of Her 36, using near-infrared data from 2MASS, has been marked for reference. We note here that, if we compare the data with the zero-age main sequence (ZAMS) instead of the main sequence, both Her 36 and SCB 182 (source #694) fall about 1 mag above the expected location according to their spectral types (O7.5 V and B2.5 V, respectively). This shift is even larger ($``$ 2 mag) considering the distance modulus of 11.3 suggested by other authors.
As M8 is located not far from the direction to the Galactic Centre, many field interlopers can be expected. In particular, it is very probable to be intercepting part of the giant population of the galactic inner disk and Bulge. In this way, the CC and CM diagrams show a superposition of well-distinguished stellar components. Unfortunately, no control fields were taken for these observations, but in order to account for the distribution of the foreground/background field stars, $`JHK_s`$ photometry corresponding to a 5 arcmin radius circular field located off the nebula at $`\alpha (2000)=270.654`$ and $`\delta (2000)=24.618`$ ($``$ 18’ in $`l`$ at the same $`b`$) was retrieved from the 2MASS catalog. We considered only the sources with photometric errors lower than 0.1 mag. This field was chosen to be free of significant interstellar material by examination of IRAS surveys in the region. Based on these data we studied the $`(HK_s)`$ distribution of the field stars, which shows a significant concentration of sources around $`(HK_s)=0.8`$, probably indicating the average colour of the Bulge red giants which must be highly affected by interestellar reddening (see Figure 6). A similar analysis for the sources with $`K_s<14`$ of the Hourglass field revealed a peak around $`(HK_s)=1.0`$, with a considerable spread to larger values, due to the differential extincion originated in the molecular cloud. In addition, both histograms show a secondary peak around $`(HK_s)=0.3`$, probably associated with main sequence stars moderately affected by reddening. We note here that the number of infrared sources detected in the Hourglass field almost doubles the number expected from the ratio of the areas of the on and off field observations. Two facts can account for this: the existence of an infrared cluster around Her 36 and the much higher resolution of our observations.
Using 2MASS data we also constructed the CC and CM diagrams for the control field (Figure 7). The CC diagrams for the control field and the Hourglass region clearly differ. The stars in the control field appear confined in two regions: a very crowded region corresponding to the highly reddened giants of the Bulge and disk, and a less populated region associated with field main sequence stars with little reddening. On the other hand, the stars in the Hourglass region spread over a much larger area of the CC diagram. Besides the two components observed in the control field, which confirm that a significant fraction of the stars are background field stars unrelated to the cloud, there is a significant number of sources located along and to the right of the reddening vector of an O dwarf star. These stars with intrinsic colours indicative of near-infrared excess emission must be associated with young stellar objects (YSOs), such as Class I “protostars”, Herbig Ae/Be objects and T Tauri stars.
In Figure 4 two kinds of symbols have been used in order to discriminate between objects of different nature. In this way, filled triangles represent sources with evident infrared excess, which are prime candidates to be included among the YSOs previously mentioned. On the other hand, open circles may denote either classical T Tauri stars affected by moderate to large amounts of reddening, or extremely reddened early-type main-sequence stars. For example, according to its position in the CC and CM diagrams, source #349 could be either a classical T Tauri star with $`A_V\mathrm{\hspace{0.17em}4}`$ mag, or an early B-type star affected by more than 11 mag of extinction. Spectroscopic observations are needed for a definitive conclusion.
Also evident from Figure 4 is the lack of stars within the lower part of the reddening band $`[0<(HK_s)<0.3]`$, reflecting a threshold in the extinction caused by the presence of the molecular cloud. Just beside this gap, there is a group of sources vertically distributed around $`(HK_s)=0.3`$, which presumably represents the main sequence at the Hourglass distance. This idea is supported by the presence of a similar distribution $`(HK_s)=0.3`$ to the right and parallel to the main sequence of the CM diagram. If so, the mean extinction toward the region can be estimated from a simple approximation (Rieke & Lebobfky 1985), $`A_K=1.78\times E(HK)`$, adopting for the main sequence stars an average intrinsic colour excess of 0.2 mag. This leads to $`A_K=0.36`$ mag ($`A_V=A_K/0.112=3.2`$ mag).
The CM diagrams for the control and the Hourglass fields are also clearly different. The bulk of the stars in the control field distribute in a band around $`(HK_s)=0.8`$ (which approximately coincides with the vertical line that represents the location of red giants for the distance of the Galactic Bulge), in contrast with the Hourglass field, where the sources appear in a wider range of $`(HK_s)`$ and a great number of them show large values of this colour. The large $`(HK)`$ colour for these sources is likely intrinsic and due to excess infrared emission, as deduced from their locations in the infrared excess region of the CC diagram.
Before concluding this section, we want to stress that the comparison of the infrared photometric diagrams suggests that an important fraction of the stars observed toward the Hourglass Nebula are background field stars unrelated to the cloud, presumably red giants of the galactic inner disk and Bulge. These objects are significantly reddened as a result of interestellar extinction through the line-of-sight, plus a diferential contribution of an inhomogeneus molecular cloud, which, as will be discussed in next section, posseses clumps dense enough to block background radiation.
As previously mentioned, when we observe in the direction of M8 we are actually seeing a superposition of distinct stellar populations placed at different distances from us, which significantly complicates the interpretation of the results. Trying to clarify this situation, we divided the infrared sources observed toward the Hourglass in six groups according to their infrared colours. Groups I, II, III and IV include sources located above the track ($`JH=1.7(HK_s)+0.2`$), with $`HK_s<0.4`$, $`0.4<HK_s<0.8`$, $`0.8<HK_s<1.2`$ and $`HK_s>1.2`$, respectively. This track has been chosen as the straight line parallel to the reddening vectors, passing approximately through the blue end of the classical T Tauri loci of Meyer et al. (1997). In some way it may be considered as a dividing line between early- and late-type stars, as it coincides with the reddening track of a mid-F type dwarf. Since these sources are expected to be members of the stellar population of the inner disk and Bulge, we call them “giants”, even though group I may have a contribution of late-type main sequence dwarfs with lower reddening. Groups V and VI include sources below the former track. Whereas group V is expected to contain both classical T Tauri stars with lower reddening and/or highly reddened early-type dwarfs, group VI will include only objects with genuine infrared excess. In Figure 8 we plotted the spatial distribution of each stellar group using different symbols and colours. A colour CC diagram has been included for reference.
Furthermore, in order to quantify the complex distribution of sources in the region, we subdivided the observed field in nine areas of about 45”$`\times `$45” (which we call from upper left to lower right, NE, N, NW, E, C, W, SE, S and SW, respectively), and counted the number of stars of each group in every subregion. The resulting star counts are presented in Table 2. The interpretation of Figure 8 and Table 2 will be discussed in the following subsections.
### 3.4 The extinction map
We note here that, in contrast to other studies of dust extinction in molecular clouds (for example: Lada et al. 1994), it is not easy to infer the mean intrinsic colour $`(HK)`$ of background stars from the control field stars, since these objects themselves suffer significant random extinction due to a large depth in the line-of-sight. Anyway, we choose $`(HK)=0.15`$ as a representative value, which is equivalent to the colour of an early K star, and follow the procedure described in Lada et al. (1994), in order to map the distribution of colour excesses (and hence, extinctions) through the cloud. We derived the mean colour excess for all the sources located along the reddenning vector of a giant star in the Hourglass CC diagram, i.e. all the candidate late-type stars of the galactic inner disk and Bulge. In a rough approximation, these can be considered as the objects in the groups I, II, III and IV introduced in section 3.3. These measurements were then converted to a mean extinction in the $`K`$ band, $`A_K`$, using the reddening law from Rieke & Lebosfky (1985):
$`A_K=1.78E(HK)`$
We consider $`A_K`$ instead of the visual extinction $`A_V=15.9E(HK)`$, since the latter expression is valid only to the extent that the reddening law in this cloud is a normal one. Disregarding the objects in groups V and VI (plotted with magenta and red, respectively), which can be roughly adopted as belonging to the M8 region itself, Figure 8 (bottom) represents a randomly sampled map of the distribution of extinctions across the Hourglass nebula, obtained by plotting the position of each field star with a circle whose size and colour depend on the total extinction derived from its infrared colours (from the largest to the smallest extinction, we have big blue, cyan, green and open circles). In spite of the depth of our infrared observations, the centre of the cloud remains devoid of stars, indicating the presence of extremely high extinctions. Also evident is the existence of an extinction gradient extending from the position of Her 36. In Figure 8 we overlay a contour map of the integrated CO $`J=32`$ emission from White et al. (1997). The correspondence between our extinction map and the distribution of molecular material is impressive, demostrating that the extinctions derived for the individual candidates for background giants are good tracers of the molecular cloud.
### 3.5 The infrared cluster around Herschel 36
By photometry, 763 sources have been extracted from a 135”$`\times `$139” area around the massive star Her 36, the vast majority of which had not been detected by previous observations. 652 of these sources have been detected in all three wavelength bands and are plotted in the CC and CM diagrams in Figures 4, 5 and 8. The analysis of these infrared diagrams lead us to suggest that an important fraction of the stars observed toward the Hourglass are background field stars, probably red giants of the inner disk and Bulge. Adjusting for this background contamination is however extremely difficult. Although somewhat rough, a direct alternative is to consider the extreme case in which all the sources in groups I, II, III and IV are foreground/background field stars, while all the sources in groups V and VI belong to the cluster (see section 3.3 and Figure 8 for the definition of the stellar groups). Then the number of infrared sources expected to be related to the region may be about 214. 95 of them show genuine moderate and large infrared excesses, yielding a fraction of near-infrared excess sources of about 45 per cent. We should keep in mind that we have presumably overestimated the number of objects associated to the molecular cloud, and therefore the resulting fraction of infrared excess sources may be only a lower limit. Anyway, the estimated value is close to the fraction found within the Taurus (50 %) (Kenyon & Hartmann 1995), NGC 1333 (60 %) (Lada et al. 1996) and S87E (40 %) (Chen et al. 2003) star forming regions. A strikingly important fact is what happens in the central part of the field (subregion C in Table 2), where the infrared excess objects reach a total of 27 or over 70 % of all the point sources assumed to belong to the cluster. From the analysis of molecular hydrogen and CO emission, Burton (2002) suggested the existence of a molecular outflow from Her 36. This fact, along with the relative large fraction of infrared excess sources and the morphological evidence of jets from some of them (see section 3.8), indicate the cluster population is extremely young. From the comparison with similar star forming regions for which ages have been estimated (see Section 5 in Lada et al. 1996), one could say that this cluster is probably between $`12\times 10^6`$ years, if not younger.
CO line emission toward the Hourglass Nebula was studied by White et al. (1997), who detected an elongated molecular core, 30” $`\times `$ 20” in size and orientated SE-NW. The superposition of CO $`J=32`$ contours on the spatial distribution of sources in the field is shown in Figure 8. The CO emission peaks at Her 36 and 40” to the NW. Note that the first peak is the second most intense CO source observed with a single dish antenna.
As clearly seen in Figure 8, the infrared excess sources do not distribute uniformly in the region. Instead, they extend along the molecular core. While Her 36 and the infrared sources KS 1 to K S4 designated by Woodward et al. (1990) lie close to the most intense peak, another group of candidate YSOs coincides with the secondary peak. Additionally, a number of sources with infrared excess appear grouped to the South of the Hourglass, within the cavity open by the strong wind of Her 36. It is remarkable that almost no source is detected to the NE of Her 36, either with or whitout infrared excess, indicating an extremely high densiy of the molecular cloud.
The infrared excess sources distribution can be quantitatively described using the star counts in Table 2. Three representative ratios between distinct stellar populations have been computed. The ratio of genuine infrared excess sources to candidate late-type objects, ex/”giants”, has a very low value ($``$ 0.15 in average) in 7 of the 9 regions in which the Hourglass field was subdivided (see section 3.3). Nevertheless, it greatly exceeds its average value in the NW region and reaches 1.17 in the central (C) region. These two areas aproximately coincide with the secondary and primary CO emission peaks respectively, confirming that the youngest objects locate along the molecular core.
### 3.6 Individual stars
Among the most peculiar objects in the field we find the infrared sources designated by Woodward et al. (1990) as KS 1 to KS 4, as well as the ultracompact H ii region G5.97-1.17 studied by Stecklum et al. (1998). What follows is a brief description of the infrared properties of these objects.
a) KS 1 : This source (#312) is $``$ 3$`\stackrel{}{.}`$3 north and $``$ 0$`\stackrel{}{.}`$3 east of Her 36. Allen (1986) suggested that this star and Her 36 may be part of a Trapezium-like stellar cluster, based on the similar separation distances. KS 1 is actually a binary star with a very red northern component, as pointed out first by Stecklum et al. (1995). The binarity of this object compelled us to perform aperture photometry on it. According to its near-infrared colours ($`K_s=9.53`$, $`JH=1.63`$, $`HK_s=1.08`$), it is very probably a Herbig Ae/Be object affected by a few magnitudes of visual extinction.
b) KS 2: Located $``$ 1$`\stackrel{}{.}`$5 north and $``$ 11$`\stackrel{}{.}`$1 east of Her 36, this star is also a binary system. In our PSF photometry we identified two well-separated (angular separation $``$1$`\stackrel{}{.}`$3) sources: #385 (north) and #382 (south). While source #385 (KS 2-N) is presumably a late B-type dwarf attenuated by $``$ 5 mag of visual extinction, the infrared colours of source #382 (KS 2-S) are typical of a classical T Tauri star ($`K_s=12.55`$, $`JH=1.03`$, $`HK_s=0.77`$).
c) KS 3 and KS 4: These two sources are located $``$16$`\stackrel{}{.}`$8 east of Her 36, very near the apex of the optical bicone (the “waist”) of the Hourglass. The observed infrared colours of KS 3 (#410) can be replicated either by a relatively hot $``$ B9/A0 dwarf affected by $``$ 8 mag of visual extinction, or by a cooler T Tauri star ($`K_s=12.01`$, $`JH=0.78`$, $`HK_s=0.47`$). On the other hand, KS 4 (#414) is much redder than KS 3 and shows significant near-infrared excess emission. Its position on the CC diagram is typical of very young stellar objects such as Class I protostars ($`K_s=11.53`$, $`JH=2.17`$, $`HK_s=1.46`$).
d) UC H ii region G5.97-1.17: High resolution optical, infrared and radio observations revealed that this object, 2$`\stackrel{}{.}`$7 distant from Her 36, may be a proplyd, i.e., a young star surrounded by a circumstellar disk that is being photoevaporated by the nearby hot star (Stecklum et al. 1998). Because of its close proximity to Her 36, we had to perform aperture photometry on this source (#330). The near-infrared colours obtained indicate that G5.97-1.17 likely belong to the class of Herbig Ae/Be objects ($`K_s=10.62`$, $`JH=1.40`$, $`HK_s=1.20`$).
### 3.7 Distribution of dust and gas
Using $`HST`$ images, we have produced the maps of the following line ratios: H$`\alpha `$/H$`\beta `$ (Balmer ratio) and (\[S ii\]6717+\[S ii\]6731)/H$`\alpha `$ (S2H ratio). Figure 9 shows the Balmer ratio map. The H$`\alpha `$ map has been included in Figure 10 for reference. The Balmer ratio can be used as a tracer of extinction by dust. For temperatures of $``$10<sup>4</sup> K and electron densities of $``$100 cm<sup>-3</sup>, the theoretical H$`\alpha `$/H$`\beta `$ value is 2.86 (Osterbrock 1989). The presence of extinction increase the observed ratio, since it affects H$`\beta `$ wavelenghts more than H$`\alpha `$ wavelengths. The measured ratio will depend on the geometry of the dust and gas clouds, as well as on the extinction law.
As seen in Figure 9, the Balmer ratio goes clearly above the expected value of 2.86 over the whole region of the Hourglass. In fact, it has a minimum value close to 4.0. We interpret this effect as consequence of the interstellar extinction toward the Hourglass Nebula. A colour excess $`E(BV)`$ can be associated to a measured ratio $`r_B=I`$(H$`\alpha `$)/$`I`$(H$`\beta `$) according to the expression:
$`E(BV)=E_0log_{10}(r_B/2.86)`$,
where $`E_0`$ is a constant which depends on the reddening law. If we adopt the extinction law parametrized by Cardelli et al. (1989), $`E_0`$ = 2.307, and then $`E(BV)`$ = 0.34 when $`r_B`$ = 4.0. This is in excellent agreement with the mean value of the reddening derived from the optical colours of the stars in NGC 6530 ($`<E(BV)>`$ = 0.35; Sung et al. 2000).
The distribution of dust in the central part of the Nebula is quite complex, showing strong variations in a scale of arcseconds, which is clearly appreciable in the enlargement shown at the upper right corner of the panel. The whole central area seems to be highly reddened and the maximum values of $`r_B`$ are found around Her 36 ($`r_B>7`$ and $`E(BV)>0.9`$). The region immediately to the E of Her 36 shows an interesting spatial correlation between $`r_B`$ (Fig. 7) and H$`\alpha `$ (Fig. 8): brighter (in H$`\alpha `$) regions present higher values of $`r_B`$, rather than the opposite, which is what would be expected if optically-thin clouds where partially occulting the ionized gas behind. Furthermore, the correlation is also apparent when comparing the spatial distribution of H$`\alpha `$ and the S2H ratio in the sense that low values of the S2H ratio correspond to bright areas. Similar correlations have been detected in NGC 604 (Maíz-Apellániz et al. 2004b) and in 30 Doradus (Walborn et al. 2002, Maíz Apellániz et al. 2005 in preparation) and can be explained if the variations in extinction are caused by optically-thick clouds instead of by optically-thin ones. In such a model, a strong variable foreground screen is present and located between the main part of the H ii region and the observer. In those areas where the screen is thinner, we can see into the bright part of the H ii region but at the cost of measuring large values of $`r_B`$ (note that values of $`r_B`$ larger than about 4.0 imply that dust and gas cannot be uniformly mixed but rather that the dust is located in a foreground screen, Maíz-Apellániz et al. 2004b). There, the gas we are seeing is directly exposed to the ionizing radiation of Her 36, located a short distance away, and therefore shows a high excitation, implying low values of the S2H ratio. In those areas where the screen is thicker, dust completely blocks the bright areas of the H ii region and all we can see is the backside (as seen from Her 36) of the cloud, which only receives low-intensity ionizing-radiation (scattered from Her 36 or originating from other ionizing sources at large distances, see Fig. 8 of Maíz-Apellániz 2005), thus producing H ii gas with a low ionization parameter. There, the S2H ratio is much higher, as expected, and $`r_B`$ is lower because there is little dust between the source and the observer since the cloud is located at larger distances than the source.
Extinction seems to be fairly uniform over the rest of the field. The cavity toward the South open by the winds of Her 36 (marked with a letter C in Figure 9) suffers almost no obscuration, thus showing the lowest values of $`r_B`$, i.e., $`4.0<r_B<4.5`$, or equivalently, $`0.34<E(BV)<0.45`$. A kind of thin flap of dust seems to cover all the region toward the West of Her 36. This fact implies that we are not actually penetrating the molecular cloud but only detecting the front gas H$`\alpha `$ emission, which can explain the relatively low extinction measured ($`r_B<5`$ and $`E(BV)<0.56`$).
By comparing the Balmer ratio and the H$`\alpha `$ emission, we find a number of dark regions with very low H$`\alpha `$ intensity (I(H$`\alpha `$)$`<0.5`$) which also show small values of $`r_B`$ ($`r_B4.0`$). An example of such region has been marked with a letter A in Figure 9. Here, a low $`r_B`$ does not mean low extinction. On the contrary the optical depth is so high that in case of existing gas emission hidden by dust, it can not be detected only with optical data. Such possible emitting regions should be unveiled using near-infrared Br$`\gamma `$ imaging.
To conclude we want to stress that extinctions derived from the Balmer emission lines are only representative of the outer “shell” of the nebula, up to about $`A_V\mathrm{\hspace{0.17em}3}`$ mag. Optical observations do not probe the gas and dust distribution more deeply into the cloud, and thus estimating ionizing fluxes and other properties only from optical data will lead to uncertain results. Near-infrared imaging are inevitably needed to analyze what processes are really taking place behind the dust.
### 3.8 Signatures of ongoing star formation in HST images
Archival HST emission-line images in H$`\alpha `$, \[O iii\] and \[S ii\] were searched for proplyds, jets and other features that might be associated with the newly detected candidate YSOs. The HST observations reveal a rich variety of structures related to star formation, similar to those observed in M16 and M42.
Herbig-Haro (HH) objects are shock-excited nebulae powered from YSOs (see Reipurth & Bally 2001). The morphological analysis of \[S ii\] images is a commonly used technique for surveying star forming regions for the location of excited gas arising from HH flows (e.g. Wang et al. 2003). This is helped by the analysis of \[S ii\]/H$`\alpha `$ ratio maps in which regions with high values indicative of shock-excited gas are searched.
Figures 11 and 12 show the HST \[S ii\] continuum-subtracted emission map and the corresponding \[S ii\]/H$`\alpha `$ (S2H) ratio map of the Hourglass Nebula, respectively. The strongest line emission arises from the inner part of the Hourglass and is mainly concentrated on the east wall of the cavity. Both, the \[S ii\] and the H$`\alpha `$ emissions, have a clumpy structure close to Her 36. When compared to H$`\alpha `$ emission, the \[S ii\] emission shows a more complex and filamentary pattern to the north and west of the nebula. On the other hand, both images look smoothed in the southern part of the nebula, where the H ii cavity is an opened blister. These differences in the emission appears as large fluctuations in the S2H map, in a scale of few tenths of arcsecond, especially close to Her 36 and to the west part of the image.
It is very interesting to note that the S2H flux ratio goes below 0.10 over almost the whole region of the Hourglass, as expected for a photoionized plasma. We are particularly interested in detecting emission features produced in shock-excited gas, for which the expected S2H flux ratio values are typically of 0.3–0.5. Nevertheless, the highest S2H flux ratio hardly reaches 0.2 (over an arc located at 75” to the NW of Her 36). This could be explained by considering normal photoionized gas filling the volume in front of the nebula. This gas shows a S2H flux ratio close to 0.05, and then reduces the contrast between the shock-excited features with higher values and the molecular cloud. As mentioned in Section 3.7, the dust associated with this foreground material could be responsible for most of the foreground extinction in the nebula.
Perhaps the most stricking features in the \[S ii\] and S2H images are three nebular knots located 40” to the SE of Her 36 that appear to form one large structured bow shock for which the HH number HH 870 has been assigned in the HH catalogue<sup>2</sup><sup>2</sup>2HH catalogue numbers are assigned by B. Reipurth in order to correspond with the list of Herbig-Haro objects that he maintains.. Equatorial positions for this and the other new HH objects discovered in the Hourglass region are shown in Table 3. The morphological appearance of HH 870 is pretty similar to the HH 203 and HH 204 objects in the Orion Nebula (O’Dell et al. 1997). Its three flow components have been labelled as A, B and C in Figure 13. The right-hand panel of this figure shows a long-slit spectrum in the \[N ii\] 6584 Å emission line obtained across the Hourglass Nebula with NTT. The placement of the long-slit aperture has been plotted on the \[S ii\] WFPC2 mosaic in the left-hand panel of the same figure. Two noticeable kinematic structures (labeled as ”1” and ”2”) are clearly observed. These velocity components are associated with nebular structures located at 25” (0.15 pc) and 37” (0.22 pc) from the middle point of the slit. Their receding velocities respect to the H ii region reach 80 km s<sup>-1</sup> for feature “1” and 45 km s<sup>-1</sup> for feature “2”. Feature ”1” seems to origin in the gas located inmediately to the south of a dusty finger-like structure whose head points directly to Her 36. Feature ”2” is clearly associated with the flow component A of HH 870. It must be taken in mind that the long-slit images were originally obtained with other scientific purposes, and so the position on the sky was chosen. It was a fortunate coincidence that the slit aperture crossed the western bow shock feature, but it is presumably not aligned along the jet axis. This could explain the undetection of the northern counterjet, although it also might be due to the fact that the driven source is buried in the molecular cloud core. Both infrared sources #414 (KS 4) and #410 (KS 3) are located along the simmetry axis of the main bow shock component B of HH 870. As was remarked in Section 3.6, KS 4 is probably a Class I protostar and therefore a good candidate to be the driving source of this structure.
One-sided outflows are commonly observed emerging from dense cores in the photoionized medium (cf. HH 616 and HH 617 in S140, Bally et al. 2002; HH 777 in IC 1396N, Reipurth et al. 2003). Additional evidence in favour that kinematic feature “2” (spatially associated with the feature HH 870 A) is a bona fide bow shock is found in the very high \[N ii\]/H$`\alpha `$ (N2H) ratio derived from the long-slit spectra. Figure 14 shows the superposition of the H$`\alpha `$ and \[N ii\] 6584 Å (scaled by 2) profiles along the position of feature ”2”. The N2H ratio value is about 0.15 for the main emission component, but increases to about 0.7 for the radial velocity component at +45 km s<sup>-1</sup> , reinforcing the idea that this feature is in fact shock-excited.
About 50” (0.3 pc) to the West of Her 36, we find another set of nebular structures which are also identified as new HH objects. Figure 15 shows a portion of the WFPC2 H$`\alpha `$ and \[S ii\] mosaics and the S2H ratio map, where these interesting features have been labelled as HH 867, HH 868 and HH 869 according to the numbers assigned in the HH catalogue.
The object HH 867 is a very intriguing elliptical ring, approximately $`4^{\prime \prime }\times 2\stackrel{}{.}5`$ in size ($`5000\times 3100`$ AU). The ring is uncomplete and presents strong emission in \[S ii\] and much weaker emission in H$`\alpha `$, leading to a S2H ratio of about 0.11. A jet-like feature seems to cross axially the center of the ring with a certain inclination angle respect to the ring plane. This candidate jet shows two main condensations with high S2H ratio: the first one (S2H$``$0.18) is placed toward the west of the ring, whereas the other (S2H$``$0.11) locates on its eastern border. Furthermore, the jet seems to extend beyond the west end of the ring in a diffuse path of about 28” (0.17 pc), which can be mainly seen in the S2H ratio map. Near the apparent jet axis, there are two infrared excess sources, #32 ($`K_s=13.04`$, $`JH=1.50`$, $`HK=1.00`$) and #11 ($`K_s=15.73`$, $`JH=1.14`$,$`HK=1.25`$), which seem to be connected by a thin H$`\alpha `$ filament. Source #449 coincides with the eastern jet condensation. Since both of these sources are candidates to be embedded YSOs, any of them might be the driving source of the jet feature.
HH 868, located about 20” to the SE of HH 867, is an ever larger structure (6” in size = 7500 AU) composed by three nebular emission arcs. These bow shock features show high S2H ratio values ($`0.13`$) and they are undetectable in the H$`\alpha `$/H$`\beta `$ map, which means that the extinction has foreground origin. The northern arc shows a bright \[S ii\] condensation inside the bow show, suggesting a reverse shock (”Mach disk”), indicative of the action of a faster jet overtaking cooled gas behind the preceding bow shock. Note that it seems very likely based on the morphology of the flow components that HH 867 and HH 868 form independent bow shocks in a single large flow coming from the upper left of the image.
A third HH object is found in Fig. 15. HH 869 is a bright \[S ii\] condensation located 15” (19000 AU) to the west of the HH 867, almost perpendicularly to the previously mentioned ring plane. This condensation is not appreciable in H$`\alpha `$ emission, showing a S2H ratio value of $`0.12`$. Our infrared survey does not reach this peculiar emission feature and the 2MASS catalog does not show any infrared source close to this location. A diffuse filament seems to develop from HH 869, grow in the southwest direction and intersect the filament coming from HH 867. Bally et al. (2002) describe the breakout in an ionized medium of a jet arised inside the molecular cloud. They found that the wall of the molecular cloud shows arcs with enhanced \[S ii\] emission where the jet appears. In this way, we should also take in mind the hypothesis that the ring of HH 867 represents the point in the molecular cloud where a jet arised in (or around) Her 36 emerges, since the normal direction to the ring points directly to that star. Furthermore, the HH 869 feature is placed along the line joining the ring and Her 36.
Chakraborty & Anandarao (1997, 1999) performed Fabry-Perot observations of the Hourglass region in the \[N ii\] 6583Å and \[O iii\] 5007Å emission lines. They found high expansion velocities up to 50 km s<sup>-1</sup>, which they interpreted as indicative of Champagne flows. Unfortunately, the area observed by these authors does not reach the location of the outflow HH 870. Nevertheless a careful inspection of their plots suggests the possibility that some of the high velocity features observed in their spectral line profiles may represent the contribution of collimated outflows to the general turbulent motion. For example, among the line profiles shown in Figure 2 of Chakraborty & Anandarao (1997), there is a high velocity component of -50 km s<sup>-1</sup>, which appears very prominent at P.A. 190, 20” from Her 36 (Fig. 2h) but it is not observed at P.A. 175, 12” from this star (Fig. 2f). This kind of small-scale variations in the structure of the velocity field may be indicating the existence of collimated outflows.
Located 30” (0.18 pc) to the NW of Her 36, source #156 (SCB 1040) is a remarkable object which has already been suggested as pre-main sequence candidate from its optical colours in HST data by Sung et al. (2000). Its H$`\alpha `$ image shows extended emission ($`0\stackrel{}{.}5`$ in size) and reveals a chain of nebular emission features extending 10” to the south, which clearly resembles the one-sided jets detected in the Orion Nebula (Bally et al. 2000).
About $`7\stackrel{}{.}5`$ (9400 AU) to the South of source #156 is source #144, which shows an intense and very elongated H$`\alpha `$ emission. This star has infrared colours similar to T Tauri stars ($`JH=1.36`$, $`HK_s=0.82`$). Additionally, its $`K_s`$ magnitude of 10.14 corresponds to the same brightness as T Tauri itself ($`JH=1.02`$, $`HK_s=0.91`$ and $`K_s=5.33`$, according to 2MASS Second Incremental Data Release, Cutri et al. 2000), if moved from its location in the Taurus T association (140 pc) to the distance of M8 (1300 pc).
A rich clustering of faint infrared sources (#369, 374, 497, 320, 370) is found 35” (0.2 pc) to the North of Her 36, close to bright filaments facing 9 Sgr. These filaments are part of the PDR of the molecular core detected in CO (see Figure 8).
The central part of the Hourglass Nebula shows an incredible complex structure which will be better described in a subsequent paper. We remark here a large dust pillar orientated to Her 36 in the SE-NW direction, which is observed as an extinction feature against the nebular background. Its very bright rimmed head indicates that it is presumably illuminated on its far side, and so is somewhat in front of Her 36. The projected distance between the head of the pillar and the star is 11” (0.07 pc). Our $`K_s`$-band image reveals a bright stellar infrared source (#347) located in the tip of this finger, resembling those found in the dense knots of 30 Doradus by Walborn et al. (1999). Source #347 has no optical counterpart and its near-infrared colours are typical of a highly reddened early-type star ($`JH=1.72`$, $`HK_s=1.05`$).
In the southern region of the field, the WFPC2 images reveal another interesting dust structure, which might probably be an externally-ionized molecular globule similar to that found in the core of the Carina Nebula by Smith et al. (2004). This molecular globule, located $`105^{\prime \prime }`$ to the SW of the Hourglass, has also a shape resembling a human finger that points toward its likely source of ionizing photons. Curiously, the finger does not point to Her 36 but to the O4 V((f))-type star 9 Sgr. While the finger is seen only as a silhouette against a brighter background in H$`\alpha `$, its surface is very bright in \[S ii\], which is especially clear from the S2H ratio map in Figure 12. Although most of this structure is out of the FOV of our infrared images, a very bright infrared stellar source is noticeable in the summit of the pillar. 2MASS archive data reveal the presence of a highly reddened object at this location ($`J=14.603`$, $`JH=3.018`$, $`HK_s=1.502`$).
Finally we have identified two large bow shocks surrounding young stars whose near-infrared colours are typical of reddened T Tauri stars (sources #349, $`JH=1.25`$, $`HK_s=0.64`$, and #334, $`JH=1.44`$, $`HK_s=0.84`$; see Fig. 13). These objects are marked with stars in Figure 13. It is interesting to note that, while source #334 is a single star in HST images, it is observed as a double star with components separated by 0$`\stackrel{}{.}`$5 in our infrared images. The eastern component is not detected in the optical images, but it is brighter than its companion in the infrared. This object seems to be a resolved binary with components affected by a different degree of extinction.
The bow shocks detected in sources #349 and #334 are seen as arcs consisting in bright H$`\alpha `$ emission and facing Her 36. In both cases, they are concave toward the corresponding infrared source and convex toward Her 36. The bow shock around star #349 is brighter than the one around star #334. Their morphologies resemble strongly the hyperbolic bow shock around the T Tauri star LL Ori (Bally et al. 2000, Bally & Reipurth 2001). Thus, these bow shocks are candidates to be wind-wind collition fronts.
In summary, the morphological analysis of the HST images of the Hourglass region reveals that the nebula has a complex structure, showing dozens of features that closely resemble those observed in other star forming regions (for example, M42, M16, NGC 3372). However, almost nothing is known yet about the proper motion and radial velocity of these objects, so it is premature to draw any conclusions about these objects here. In a forthcoming paper, we will discuss optical spectroscopy recently obtained using the Boller and Chivens Spectrograph on the 6.5 m Magellan I telescope at Las Campanas Observatory, which will undoubtedly contribute to unveil their true nature.
## 4 Summary and Conclusions
We have obtained near-infrared $`JHK_s`$ images of the Hourglass Nebula in M8. These data were complemented with HST images and longslit spectroscopy retrieved from on-line databases. The main results can be summarized as follows:
We used the recently developed numerical code CHORIZOS to obtain a new estimate for the distance to the Hourglass region. Using optical and infrared photometry from this work and from literature for a group of early-type stars around Herschel 36 (including this star), we derived a distance modulus of 10.5 (1.25 kpc). This value is sensibly smaller than the 11.25 previously accepted for the region, but it is in excellent agreement with the last determination for NGC 6530 by Prisinzano et al. (2005).
Our infrared images confirm the existence of a very young stellar cluster around the massive O-type star Herschel 36. We have detected almost 100 sources with colours indicative of intrinsic infrared excess emission, which appear as prime candidates to be T Tauri and Herbig Ae/Be stars. These objects are not uniformly distributed in the region, but they extend along the molecular core, mostly concetrated near the two most intense peaks of the CO distribution (White et al. 1997). The large fraction of infrared excess sources (over 70 % in the central part of the field) along with the evidence of outflow activity from Her 36 (Burton 2002), suggest the stellar population within the cluster must be extremely young ($`10^6`$ years).
By comparing the $`JHK_s`$ photometric diagrams of the Hourglass and a 2MASS control field, we have found that a significant fraction of the stars observed toward M8 may be red giants belonging to the galactic inner disk and Bulge. Based on the method described by Lada et al. (1994), we have used the positions and near-infrared colour excess estimates of over 400 background field giants to directly measure the extinction through the molecular cloud, finding a strong correlation with the distribution of molecular material.
Archival HST emission-line images reveal a variety of ongoing star formation features in the core of M8. Several of the candidate YSOs detected here seem to be associated to structures similar to those observed in other star forming regions like M16 and M42, namely, proplyds, jets, dense knots, molecular globules and bow shocks. Furthermore, four Herbig-Haro objects have been detected for the first time in the region (HH 867, HH 868, HH 869 and HH 870). A longslit spectrum in the \[N ii\] 6584 Å emission line obtained across the Hourglass Nebula with NTT and retrieved from the ESO Archive Facility confirmed the identification of HH 870, providing the first direct evidence of active star formation by accretion in M8. In order to understand better the nature of some of the rest of the objects we have obtained optical spectroscopy with the 6.5 m Magellan Telescope, which will be analyzed in a future paper.
## acknowledgements
This publication makes use of data products from the Two Micron All Sky Survey, which is a joint project of the University of Massachusetts and the Infrared Processing and Analysis Center/California Institute of Technology, funded by the National Aeronautics and Space Administration and the National Science Foundation. This research has made use of Aladin and the Simbad Database, operated at CDS, Strasbourg, France.
Financial support from “Fondo de Publicaciones de la Dirección de Investigación de la Universidad de La Serena (Chile)” and from the Center for Astrophysics (FONDAP No. 1501003, Chile) are acknowledged by RHB and MR, respectively. JIA thanks the Departamento de Física of Universidad de La Serena for the use of their facilities and the warmest hospitality. Also, the authors gratefully thank the staff at LCO for kind hospitality during the observing run.
|
warning/0506/cond-mat0506425.html
|
ar5iv
|
text
|
# Magnetic field-induced quantum critical point in YbPtIn and YbPt0.98In single crystals
## I Introduction
In recent years, stoichiometric Yb-based heavy fermion compounds have raised a lot of interest, particularly due to the limited number of such systems known to date. YbPtIn is one of the few examples of such heavy fermion systemstro09 ; kac10 ; yos11 , as indicated by measurements performed on single crystal samples extracted from on-line melts of the polycrystalline material. Based on the existing datatro09 ; kac10 , YbPtIn has a relatively low magnetic ordering temperature (T$`{}_{ord}{}^{}3`$ K) and an enhanced electronic specific heat coefficient ($`\gamma >400`$ mJ$`/`$mol K<sup>2</sup>), whereas the magnetic entropy at T<sub>ord</sub> amounts to only about 60$`\%`$ of the $`R\mathrm{ln}2`$ value expected for a doublet ground state. In light of these observations, this system appears to be qualitatively similar to the other recently studied Yb-based heavy fermion antiferromagnets, YbRh<sub>2</sub>Si<sub>2</sub>tro05 ; geg06 ; ish07 ; pas08 ; pas09 , and YbAgGebey27 ; kat28 ; mor05 ; bud11 ; bud12 , the latter compound being isostructural to YbPtIn. In YbRh<sub>2</sub>Si<sub>2</sub>, magnetic order occurring at very low temperature (below 70 mK)tro05 was associated with a low entropy release (around 0.01 $`R\mathrm{ln}2`$)geg06 . For the YbAgGemor05 ; bud11 compound, the temperature associated with the magnetically ordered state, whereas still fairly low ($`1.0`$ K), is enhanced compared to that of YbRh<sub>2</sub>Si<sub>2</sub>; also, the magnetic entropy at the ordering temperature is larger than in YbRh<sub>2</sub>Si<sub>2</sub>, but still less than 10$`\%`$ of $`R\mathrm{ln}2`$. Thus these two compounds can be regarded as systems with small moment ordering. Having an even higher ordering temperature and magnetic entropy at T<sub>ord</sub>, YbPtIn seemed a good candidate to further study a progression from small moment to reduced moment ordering, in stoichiometric Yb-based heavy fermion compounds, with field-induced quantum critical point.
In contrast to a classical phase transition at finite temperatures, driven by temperature as a control parameter with thermal fluctuations, a quantum phase transition is driven by a control parameter C other than temperature (e.g., C = pressure, doping or magnetic field) at T = 0, with quantum mechanical fluctuations. Such a control parameter tunes the system from a magnetically ordered state towards a state without magnetic order, at zero temperature, crossing a quantum critical point. Due to the hybridization of the 4f electrons and the conduction electrons in heavy fermion (HF) systems, which can be modified by any one of the aforementioned C control parameters, the (HF) compounds are very suitable to study quantum critical behavior. Moreover, close to the critical value C<sub>crit</sub> which drives the ordering temperature close to zero, pronounced deviations from the Fermi liquid-like FL behavior can occur. This has been observed in a large number of HF systems where C = doping or pressure, and only a few doped systems have been field-tuned through a QCPste2001 . To date, YbRh<sub>2</sub>Si<sub>2</sub>tro05 ; geg06 ; ish07 ; pas09 and YbAgGekat28 ; bud11 ; bud12 are the only stoichiometric Yb-based HF compounds in which a field-induced quantum critical point QCP has been observed. The heavy fermion character of the YbPtIn system has already been reportedtro09 ; kac10 ; in this paper we will discuss the existence of a field-induced QCP in YbPtIn which renders it very similar to the other two stoichiometric, Yb-based heavy fermions.
Recently we presented anisotropic low-field susceptibility measurements, as well as specific heat data in zero applied field, on solution-grown single crystals of the RPtIn series, including YbPtInmor10 . No features indicative of magnetic order could be identified in our magnetization measurements on single crystals of the R = Yb compound, down to T $``$ 2 K. However, a well-defined peak at $`T=2.1`$ K in the zero-field specific heat data suggests that this compound orders magnetically below 2.1 K, just above the low temperature limit of our magnetization measurements. As we will present in this paper, detailed transport and thermodynamic measurements down to 0.4 K not only confirm the magnetic ordering in this compound below 2.1 K, but also suggest that another phase transition might exist in this system around 1 K.
However, our measurements were in part inconsistent with previous reports on YbPtIn single crystals: whereas Kaczorowski et al.kac10 have also presented magnetization data showing no features associated with magnetic ordering above 1.7 K, in their specific heat measurements three anomalies can be observed (at 3.1, 2.3, and 1.2 K), all at slightly different temperatures than in our data. Furthermore, the low-field susceptibility data presented by Trovarelli et al.tro09 are consistent with an ordering temperature around 3.4 K, also confirmed by their resistivity and specific heat measurements.
In order to address these apparent discrepancies between our data on solution grown single crystals and the two previous reports on on-line grown single crystalstro09 ; kac10 , we reproduced the growth as described by Trovarelli et al.tro09 . The anisotropic specific heat and transport measurements on our single crystals extracted from the melt confirmed the existence of magnetic phase transitions at 3.4 K and 1.4 K, as observed previously.
In this paper we will try to examine the differences between solution and on-line grown single crystals, given that a small Pt-deficiency occurred in the former types of crystals, leading to a stoichiometry closest to YbPt<sub>0.98</sub>In, and no disorder could be detected in the on-line grown crystals. Also, given that heavy fermion compounds with small moment ordering are likely to be driven to a quantum critical point QCP by disorder or applied field, we will study the evolution of both YbPt<sub>0.98</sub>In and YbPtIn towards a field-induced QCP. Detailed anisotropic measurements of specific heat and resistivity on the two types of crystals, for fields up to 140 kG and temperatures down to 0.4 K, will be used for the comparison between the two types of crystals. Additional Hall effect measurements were performed on the solution grown single crystals for the same temperature and field ranges, allowing us to further explore the effects of the QCP in this compound.
After briefly describing the methods used for synthesizing the two types of YbPtIn single crystals, and the measurement techniques used for the sample characterization, we will present the results of single crystal x-ray diffraction. Next we will compare the low field data for the two types of compounds, pointing out the similarities, as well as the significant differences in their physical properties. Next we will present the higher field data for both types of crystals, for $`Hab`$ and $`Hc`$. This will allow us to follow, in both systems, the progression from reduced magnetic moment order to the quantum critical point QCP, as driven by the application of increasing magnetic field. Also, we will extend the zero-field comparison between the two YbPtIn systems to their field- dependent properties, and will try to identify common features as well as possible effects of the site deficiency.
## II Experimental methods
YbPtIn is one of the RPtIn compounds, reported to crystallize in ZrNiAl hexagonal structurekac10 ; mor10 , space group P$`\overline{6}2m`$. Single crystals of this compound were obtained by two methods: from high-temperature ternary solution, as described in Ref. , and from stoichiometric on-line melttro09 . The initial concentration used for the solution growth was Yb<sub>0.4</sub>Pt<sub>0.1</sub>In<sub>0.5</sub>, and the ternary solution, sealed in a 3-cap Ta cruciblecan17 under partial argon pressure, was slow-cooled from 1200<sup>0</sup> C to 1000<sup>0</sup> C over approximately 100 hours. Subsequently, the excess liquid solution was decanted, and the resulting hexagonal rods were briefly etched in HCl to remove residual flux from the surface. For the on-line melt, 1:1:1 atomic ratios of Yb, Pt and In were sealed in a Ta crucible under partial argon atmosphere. After briefly heating the sealed crucible up to approximately 1650<sup>0</sup> C in an induction furnace, the crucible was sealed in a quartz ampoulle and annealed at 700<sup>0</sup> C for 120 hours. Small rods were subsequently extracted from the resulting crystalline conglomerate.
Anisotropic magnetization measurements as a function of temperature and applied field M(H,T) were performed in a Quantum Design MPMS SQUID magnetometer (T $`=1.8350`$ K, H$`{}_{max}{}^{}=55`$ kG). Temperature- and field-dependent specific heat C<sub>P</sub>(H,T), resistivity $`\rho `$(H,T) and Hall resistivity $`\rho _H`$(H,T) measurements were taken in a Quantum Design PPMS-14 instrument with He-3 option, for temperatures down to 0.4 K, and applied magnetic fields between 0 and 140 kG. For the anisotropic (i.e., $`Hab`$ or $`Hc`$) specific heat measurements a relaxation technique with fitting of the whole calorimeter (sample with sample platform and puck) was used. The sample platform and grease background, measured separately for all necessary (H,T) values, was subtracted from the sample response. Also, the non-magnetic contribution to the total specific heat was estimated based on measurements on flux-grown LuPtIn, used as the non-magnetic analogue of YbPtIn. The single crystals of LuPtIn were grown using similar initial composition (R<sub>0.05</sub>Pt<sub>0.05</sub>In<sub>0.90</sub>) and temperature profile (slow cooling from 1190<sup>0</sup>C to 800<sup>0</sup>C over $``$100 hours) that yielded fully-occupied RPtIn single crystals for the lighter members of the seriesmor10 . We therefore anticipate that the LuPtIn crystals are also fully-occupied, and we used this system as the non-magnetic analogue for our both Yb compounds. It should be noted that we made the assumption that the non-magnetic background is insensitive to the 2$`\%`$ Pt-deficiency in the solution-grown compound. The specific heat for LuPtIn was measured in the same temperature range as for YbPtIn and YbPt<sub>0.98</sub>In, for 0 and 140 kG applied field, and was found to be virtually field-independent over the measured temperature range.
A standard AC four-probe resistance technique ($`f=16`$ Hz, $`I=10.3`$ mA) was used for the field- and temperature-dependent resistivity measurements. Given the rod-like geometry of the samples, the current was always flowing along the crystallographic c-axis, whereas the field was either parallel (for longitudinal magnetoresistance), or perpendicular (for transverse magnetoresistance) to the direction of current flow. The Hall resistivity measurements were performed on a sample that was polished down to a plate-like shape. This sample geometry restricted the Hall resistivity measurements to the field applied within the hexagonal crystallographic plane ($`Hab`$). The Hall resistivity $`\rho _H`$(H,T) was measured for field perpendicular to both the current and the Hall voltage directions. In order to minimize the inherent (small) misalignment of the voltage contacts, these measurements were taken for two opposite directions of the applied field, H and -H, and the odd component, ($`\rho _H`$(H) - $`\rho _H`$(-H))$`/2`$ was taken as the Hall resistivity.
For each of the two systems (YbPtIn and YbPt<sub>0.98</sub>In), we will compare the similar data for the the two orientations of the applied field (i.e., for $`Hab`$ and $`Hc`$), as well as the resulting T - H phase diagrams. The criteria used for determining the points on all presented phase diagrams will be discussed in more detail in the appropriate sections, but in general they were maxima in the C<sub>P</sub>(H,T) data, and derivative maxima or onset values for the transport data. For $`Hab`$, the maximum temperature for which the $`\rho (T^2)`$ could be fit to a straight line was the criterium used in determining the upper limit of this Fermi liquid-like region. Additional Hall resistivity measurements were performed for the field applied within the basal plane, for the YbPt<sub>0.98</sub>In (flux-grown) compound. Various criteria were used for determining critical points similar to the Hall line for the YbRh<sub>2</sub>Si<sub>2</sub>pas09 and YbAgGebud12 systems.
## III Results
### III.1 Crystal structure
For both solution and on-line grown crystals, the crystal structure was confirmed by powder x-ray diffraction, with no detectable impurity peaks. However, additional single crystal x-ray measurements were performed on the two types of YbPtIn compounds. Crystals with dimensions $`2\times 7\times 11\mu m^3`$ and $`2\times 6\times 13\mu m^3`$ were extracted from the flux and on-line grown samples respectively. Room-temperature X-ray diffraction data were collected on a STOE IPDSII image plate diffractometer with Mo K$`\alpha `$ radiation, and were recorded by taking $`1^0`$ scans in $`\theta `$ in the full reciprocal sphere. The range of 2 $`\theta `$ extended from $`6^0`$ to $`63^0`$. Numerical absorption corrections for both crystals were based on crystal face indexing, followed by a crystal shape optimization. Structure solution and refinement were done using the SHELXTL program. The crystallographic and structural data are summarized in Tables 1 \- 2.
A high temperature factor was inferred for Pt(1) of the flux-grown crystal during final stages of the refinement, which is usually indicative of possible atomic deficiencies, symmetry reduction or superstructure formation. No superstructure reflections were observed, and symmetry reduction did not resolve the issue. However, relaxing Pt(1) site occupancy resulted in a statistically significant deficiency (0.06(1), Table 1), and led to improvements in the Pt(1) temperature factor and overall residual R value. We also tested for possible deficiencies on the other atomic sites in this crystal, but refined occupancies for Yb (1.00(6)), Pt(2) (1.00(6)) and In (1.02(2)) did not suggest presence of atomic deficiencies. Thus the composition of the flux-grown crystal can be written as YbPt<sub>0.98</sub>In, noting that the Pt<sub>0.98</sub> value reflects the Pt stoichiometry of the whole unit cell, not just of the Pt(1) site. In contrast to the flux-grown sample, the Pt(1) temperature factor of the on-line grown crystal had a reasonable value. Relaxing the Pt(1) occupancy yielded only a slight and statistically insignificant deficiency of 0.014(9) (Table 2). Occupancy refinement for other atomic sites showed no deviations from unity. Although small deficiencies on the Pt(1) site could not be excluded, the YbPtIn formula is a good presentation of the composition of the on-line crystal in terms of sensitivity of our X-ray diffraction experiments. This difference in the stoichiometry of the solution and on-line grown samples is consistent with YbPtIn having a small width of formation extending towards the Pt-deficient side; given that the initial melt composition is very Pt-poor (i.e., Yb<sub>0.4</sub>Pt<sub>0.1</sub>In<sub>0.5</sub>), it is expected to be sensitive to such a small width of formation.
A closer look at the atoms’ positions given in Tables 1-2 suggests that two compounds might be mirror images of each other; whereas racemic twinning could not be excluded for either the solution or the on-line grown compounds, no evidence of the existence of both ”left” and ”right” structures in each system could be found.
The lattice parameters and unit cell volume for the solution-grown, YbPt<sub>0.98</sub>In crystals were $`a=(7.55\pm 0.01)`$ Å, $`c=(3.76\pm 0.01)`$ Å and $`Vol=(186.28\pm 0.51)`$ Å<sup>3</sup>. The analogous unit cell dimensions on the on-line grown crystals were slightly smaller: $`a=(7.54\pm 0.01)`$ Å, $`c=(3.75\pm 0.01)`$ Å and $`Vol=(185.61\pm 0.11)`$ Å<sup>3</sup>.
### III.2 Low magnetic field comparison
Anisotropic magnetization measurements are presented in Fig.1 for both the YbPt<sub>0.98</sub>In (full symbols) and the YbPtIn (open symbols) compounds. As can be observed in Fig.1a, the paramagnetic susceptibility indicates moderate anisotropy for both systems (with $`\chi _{ab}/\chi _c6`$ at the lowest temperature), and no clear sign of magnetic ordering down to $`T=1.8`$ K. The anisotropic M(H) isotherms show a continuous increase of the magnetization, with a trend towards saturation above 40 kG (Fig.1b) for H applied within the ab-plane; the axial magnetization remains linear and significantly smaller than M<sub>ab</sub> up to our maximum field available for these measurements (55 kG). Whereas qualitatively there is an overall similarity between the corresponding data of the two compounds, the absolute values of both susceptibility and field-dependent magnetization are slightly larger for YbPt<sub>0.98</sub>In than for YbPtIn. We believe the $``$ 10 to 20$`\%`$ difference to be too large to have been caused by weighing errors alone, and thus we conclude that it may reflect the different Kondo temperatures and exchange coupling due to the change of stoichiometry in the two compounds.
The zero-field specific heat and resistivity data shown in Fig.2 are consistent with magnetic ordering in both compounds, however at different temperatures: two well defined peaks at T$`3.4`$ K and 1.4 K are visible in the YbPtIn C<sub>P</sub>(T) data (open symbols, Fig.2a), whereas only one peak can be distinguished, around 2 K, in the YbPt<sub>0.98</sub>In data (full symbols). These transition temperatures are marked by the vertical dotted lines for the former compound, and by one dashed line for the latter. It can be seen that the corresponding resistivity measurements (Fig.2b) show changes in slope around the respective transition temperatures. Another noticeable difference between YbPtIn and YbPt<sub>0.98</sub>In manifests in the resistivity values (Fig.2b and 3) with the ones for the former compound being approximately three times smaller in the latter one. The error bars shown for the lowest temperature $`\rho `$ values give a caliper of the uncertainty in estimating the resistivity values for the two compounds, further confirming the aforementioned difference. The larger residual resistivity in YbPt<sub>0.98</sub>In is consistent with the additional disorder (i.e., site disorder) or presence of additional vacancies in this type of crystals.
Thus the zero-field measurements indicate a dramatic effect of the small Pt-deficiency on the ordered state in YbPtIn: the upper transition is shifted down in temperature in the Pt-deficient compound, whereas a second one is clearly identifiable only in YbPtIn. In order to explore the differences between these two samples more thoroughly, a systematic study of the field-dependence of $`\rho (T)`$ and C$`{}_{p}{}^{}(T)`$ was undertaken.
### III.3 High magnetic field measurements: YbPtIn
$`Hab`$
The low-temperature specific heat data for the on-line grown compound YbPtIn is shown in Fig.4a, for various values of the applied field. As already seen, two sharp peaks present in the H = 0 data can be associated with magnetic phase transitions at T $`=3.4`$ K and 1.4 K. As the applied field is increased, these transitions (indicated by small arrows in Fig.4a) move to lower temperatures, and eventually drop below 0.4 K around 20 kG, and 60 kG respectively. Trovarelli et al.tro09 have reported similar measurements up to 80 kG, which are consistent with our data; however, their study did not include a systematic analysis of the H - T phase diagram or the potential quantum critical behavior in this compound. From the linear extrapolation of the zero-field C$`{}_{P}{}^{}/T`$ vs. T<sup>2</sup> data from T $``$ 5 K down to T = 0 (dotted line in the inset, Fig.4a), the electronic specific heat coefficient $`\gamma `$ can be roughly estimated as $`\gamma 500`$ mJ / mol K<sup>2</sup>. The magnetic component of the specific heat is defined as $`C_m=C_P`$(YbPtIn)$`C_P`$(LuPtIn), and is shown in Fig.4b as $`C_m/T`$ vs. T<sup>2</sup>, for the same field values as before.
When the magnetic specific heat is plotted in C$`{}_{m}{}^{}/T`$ vs. $`\mathrm{ln}`$ T coordinates (Fig.5), a reduced region of logarithmic divergency (non-Fermi liquid NFL behavior) is apparent; however this linear region in C$`{}_{m}{}^{}/T(lnT)`$ is more ambiguous than in other heavy fermion compounds displaying NFL behavior (e.g., YbRh<sub>2</sub>Si<sub>2</sub>tro05 and YbAgGebud11 ). Because of a downturn in the high field data (H $``$ 50 kG \[Fig.5b\]) around 5 K for H = 50 kG, the largest logarithmic divergency which occurs for H $``$ 60 kG, is limited to only a fraction of a decade in temperature (1.5 K $`<`$ T $`<`$ 6.5 K). The above observations suggests that a QCP may exist around a critical field value H$`{}_{c}{}^{}{}_{}{}^{ab}`$ just above 60 kG, but the presence of a NFL region at intermediate field values is less clearly defined than in the previously studied Yb-based heavy fermion compounds.
One of the expressions considered in the scaling analysis at a QCPtsv18 is the cross-over function $`[C(H)C(H=0)]/T`$ vs. $`H/T^\beta `$. In the case of YbRh<sub>2</sub>Si<sub>2</sub>pas09 and YbAgGebud12 , the $`H/T^\beta `$ range over which universal scaling was observed in high fields corresponded to 1$`/`$T $`<`$ 3 K<sup>-1</sup>, and 1.2 K<sup>-1</sup> respectively (with $`\beta >1`$). Due to the slightly enhanced magnetic ordering temperature $`T_{ord}=3.4`$ K in YbPtIn, the analogous $`1/`$T range is drastically reduced (1$`/`$T $`<`$ 0.3 K<sup>-1</sup>), making the unambiguous determination of the critical exponent $`\beta `$ essentially impossible.
Low-temperature resistivity curves for different values of the applied magnetic field are shown in Fig.6. The $`\rho (T)_H`$ data are consistent with the presence of two magnetic phase transitions at low fields; the small arrows indicate these transition temperature values, as determined from maxima in the d$`\rho /dT`$. Both these transitions are suppressed by increasing applied field. Whereas the upper transition can still be detected for H $`<60`$ kG, a field of about 20 kG is sufficient to drive the lower one below our base temperature of 0.4 K. It is worth noting that the critical field H$`{}_{c}{}^{ab}60`$ kG, determined from the $`\rho `$(T,H) data as the field required to suppress the magnetic ordering, is close to the position of the QCP as inferred based on the C<sub>P</sub> data (Fig.4-5).
As Fig.7 indicates, when the magnetic field is increased beyond H$`{}_{c}{}^{}{}_{}{}^{ab}`$, the temperature dependence of the resistivity is ambiguous; particularly at fields just above H$`{}_{c}{}^{}{}_{}{}^{ab}`$ (H $``$ 100 kG), it is difficult to distinguish between linear or quadratic behavior of the resistivity as a function of temperature. Consequently we leave the complete analysis of the temperature-dependent resistivity outside the ordered state for the discussion section.
When magnetoresistance measurements ($`\rho (H)_T`$) are performed (Fig.8) three features are apparent at very low temperatures, as indicated for T = 0.4 K by the small arrows. The inset shows $`\rho `$(H) at T = 0.8 K, to exemplify how these transition temperatures were inferred from these data. For T $``$ 1 K, the two lower transitions merge and the resulting one is still distinguishable up to approximately 1.4 K. The upper transition moves down towards zero field as the temperature approaches T$`{}_{N}{}^{}=3.4`$ K.
Based on the above field- and temperature-dependent thermodynamic and transport measurements, a $`TH`$ phase diagram for $`Hab`$ can be constructed (Fig.9): in zero magnetic field, two magnetic phase transitions can be observed, around T$`{}_{N}{}^{}=3.4`$ K and T$`{}_{m}{}^{}=1.4`$ K. Increasing magnetic field splits the lower transition into two separate ones around H $`10`$ kG; one of these phase lines drops towards our lowest temperature at almost constant field, whereas the second one has a slower decrease with field, such that it approaches T = 0.4 K around H = 20 kG. In a similar manner, the upper transition is driven down towards 0 around $`H_c=60`$ kG.
$`Hc`$
Fig.10 presents specific heat data for YbPtIn, $`Hc`$, for fields up to 80 kG. High torques on this sample for this orientation of the field prevented us from completing these measurements up to 140 kG. Moreover, as will be shown below, there are significant discrepancies between the transition temperatures determined even from the intermediate-field specific heat data and transport measurements (i.e., for H $``$ 40 kG). This observation prompts us to suspect that significant torques may have changed the sample orientation for the specific heat measurements, even for fields significantly lower than 80 kG.
In zero field, we can confirm the two magnetic transitions observed before, at T$`{}_{N}{}^{}=3.4`$ K, and T$`{}_{m}{}^{}=1.4`$ K respectively; as the small arrows indicate, the lower-T transition is driven down in field, and falls below 0.4 K for H $`>40`$ kG, whereas the upper transition persists above 80 kG.
The temperature and field dependent resistivity data (Fig.11-12) indicate a much slower suppression of the magnetic order with the applied field. In Fig.11a, $`\rho (T)`$ curves are shown, with the arrows indicating the transition temperatures as determined from $`d\rho /dT`$. Fig.11b presents a subset of these derivatives, to illustrate the criteria for determining the temperatures: for the lower transition, a peak in $`d\rho /dT`$ broadens as field is increased, and disappears for H $`>`$ 80 kG; the upper transition is marked by a step in these derivatives, which also broadens as H increases. At the highest measured field (i.e., 140 kG) we are unable to distinguish between a very broad step (with a possible transition temperature marked by the small arrow) or a cross-over in corresponding $`d\rho /dT`$. The field-dependent resistivity data (Fig.12) are indicative of a low temperature transition consistent with that seen in $`\rho (T)`$, with the critical fields determined from onsets.
Given the above C<sub>P</sub>(T,H) and $`\rho `$(T,H) data, we suspect that magnetic fields H $`>`$ 20 kG deform the four wires supporting the He-3 specific heat platform used for the C<sub>P</sub>(T,H) measurements, whereas the resistivity sample appears to be well held in place by grease on the rigid platform. Consequently, at high fields, the two sets of data (C<sub>P</sub>(T,H) and $`\rho `$(T,H)) may not correspond to the same orientation of the field ($`Hc`$), yielding different transition temperature values for the corresponding applied fields.
As a result, in constructing the T - H phase diagram for H$`c`$ (Fig.13), we will consider the T<sub>c</sub> values as determined from the $`\rho `$(T,H) data up to H = 140 kG, and only the H $``$ 20 kG ones based on specific heat measurements. Also shown are error bars for points determined from $`\rho `$(T) data at several field values (i.e., for H = 20, 80, 100, 120 and 140 kG), and for the point obtained from $`\rho `$(H) at our minimum temperature (T = 0.4 K); these give a caliper of the errors bars in determining the points on this phase diagram for the whole field and temperature range. Two transitions can be observed in Fig.13, at low fields, around 3.4 K, and 1.4 K respectively. As H is being increased, the low temperature line slowly approaches T = 0 around H $`85`$ kG. The step in $`d\rho /dT`$ associated with the upper transition (Fig.11b) broadens as the field increases, resulting in increasingly large error bars in determining these transition temperatures. As already mentioned, it is uncertain if the transition persists up to H = 140 kG, or if cross-over occurs between 120 and 140 kG.
### III.4 High magnetic field measurements: YbPt<sub>0.98</sub>In
$`Hab`$
Given the differences between YbPtIn and YbPt<sub>0.98</sub>In evidenced by both thermodynamic and transport data (Figs.1-3), it is desirable to compare similar measurements on the two compounds, and to study the effect of the small stoichiometry change on the field-induced QCP.
Consequently, in Fig.14 we present the low-temperature specific heat data of YbPt<sub>0.98</sub>In, for various values of the applied field H $`ab`$. A well-defined peak at T $`=2.1`$ K in the H $`=0`$ data may be associated with the magnetic ordering of this compound. As the applied magnetic field is increased, this transition (indicated by small arrows) moves to lower temperatures, and eventually drops below 0.4 K around 35 kG. From the linear extrapolation of the H = 0 data from T $`5`$ K down to T = 0 (inset, Fig.14a), we can estimate the electronic specific heat coefficient $`\gamma `$ as $`\gamma 500`$ mJ / mol K<sup>2</sup>. Fig.14b the magnetic specific heat data for the same field values, $`C_m=C_P`$(YbPt<sub>0.98</sub>In)$`C_P`$(LuPtIn), plotted as C$`{}_{m}{}^{}/T`$ vs. T<sup>2</sup>.
Fig.15 shows the magnetic specific heat as $`C_m/T`$ vs. $`\mathrm{ln}T`$, for the same field values as before. A logarithmic divergence can be observed in these data, with the largest temperature region where $`C_m/T(lnT)`$ occurring around H = 35 kG (dotted line). However, on the next measured curve (i. e., for H = 40 kG) the linear region extends over a comparable temperature interval, at slightly higher temperatures than in the H = 35 kG case. It thus appears that the largest temperature region (close to a decade) for the logarithmic divergency of the C$`{}_{m}{}^{}/T`$ data may occur for some intermediate field value (35 kG $`<`$ H $`<`$ 40 kG). These data could be described as $`C_m/T=\gamma _0^{}\mathrm{ln}(T_0/T)`$, with the ranges for $`\gamma _0^{}`$ and T<sub>0</sub> determined from the linear fits on the H = 35 and 40 kG curves: 420 mJ$`/`$mol K$`{}_{}{}^{2}<\gamma _0^{}<`$ 430 mJ$`/`$mol K<sup>2</sup>, and 14.7 K $`>`$ T$`{}_{0}{}^{}>`$ 13.5 K. The above observations seem consistent with a QCP in this compound with critical field just above 35 kG.
Fig.16 shows the low-temperature resistivity data for various values of the applied magnetic field. A maximum in d$`\rho /dT`$, associated with the magnetic ordering can be identified at T $`2.2`$ K in the H = 0 data, and is marked by small arrow in Fig.16a; as the field is increased above 35 kG, small arrows indicate that this transition temperature drops below 0.4 K , consistent with the specific heat data.
As was the case of YbPtIn, in YbPt<sub>0.98</sub>In the temperature dependence of the resistivity outside the ordered state is ambiguous. From figs.16b-17 it is unclear whether the resistivity is linear or quadratic in field for H $`>H_c^{ab}`$, particularly for intermediate field values (H $``$ 55 kG). A detailed analysis of the resistivity data above the critical field will be performed in the discussion section, for both compounds. This should allow us to better clarify the position of the QCP and the existence of a Fermi liquid-like (FL) regime in these systems.
Transverse magnetoresistance measurements were taken at constant temperatures ranging from 0.4 K to 10 K. As shown in Fig.18, two different temperature regimes can be identified in these data: for T $`=0.41.8`$ K (Fig.18a), two transitions can be distinguished. The small arrows mark the positions for these two transitions for T = 0.4 K, whereas the inset illustrates how these critical field values were determined. As the temperature increases up to about 1.8 K, the upper transition moves down in field and broadens, whereas the position of the lower one seems almost unaffected by the change in temperature. For temperatures higher than 2 K (Fig.18b), the magnetoresistance isotherms display only a broad feature that looks more like a cross-over rather than a transition.
Using the detailed C<sub>P</sub>(T,H) and $`\rho `$(T,H) measurements discussed above, the YbPt<sub>0.98</sub>In T - H phase diagram for H$`ab`$ can be constructed. As can be seen in Fig.19, it is qualitatively similar to the corresponding T - H phase diagram for the stoichiometric compound (Fig.9): in YbPt<sub>0.98</sub>In magnetic ordering occurs at around 2.2 K. An almost field independent phase line is apparent in the $`\rho (H)_T`$ data around 8 kG, and it appears to persist close to the magnetic ordering temperature. Increasing applied field drives the higher transition towards T = 0 at a critical field values around 35 kG.
$`Hc`$
The similarities observed previously between YbPtIn and YbPt<sub>0.98</sub>In are also present for the $`Hc`$ direction, as the C<sub>P</sub>(T,H) and $`\rho `$(T,H) measurements indicate. As in the case of the stoichiometric compound, significant torques on the specific heat platform and sample for H $``$ 50 kG may be the cause of the different transition temperature values, as determined by the two data sets mentioned above. Therefore we will only take into consideration C<sub>P</sub>(T,H) data for H $`<`$ 50 kG (Fig.20). Similar to the $`Hab`$ measurements (Fig.14), the $`Hc`$ C<sub>P</sub> curves reveal a magnetic transition around 2.1 K for H = 0 (Fig.20), which drops to $``$ 2 K for H = 40 kG, before the sample torques significantly; small arrows indicate the position of the transition temperature for the three curves shown in Fig.20.
The temperature and field dependent resistivity data (Fig.21) indicate a slow suppression of the magnetic order with the applied field. Fig.21a shows the $`\rho `$(T) curves in various applied fields, with the large circles marking the phase transition as determined from d$`\rho /dT`$; the inset illustrates how the transition temperature was determined for H = 0. The critical field required to suppress this transition below our base temperature appears to be around 120 kG. Given the limited temperature range at these high fields, for H = 130 and 140 kG we were unable to distinguish a linear or quadratic temperature dependence of the resistivity. The magnetoresistance isotherms are presented in Fig.21b, and the large squares on this plot also indicate the high-T magnetic phase transition; in the inset, a few d$`\rho /dH`$ curves are shown to illustrate how the critical field values for the transition were determined.
Based on the C<sub>P</sub>(T,H) and $`\rho `$(T,H) presented above, the $`Hc`$ T - H phase diagram for YbPt<sub>0.98</sub>In can be obtained, as shown in Fig.22. At low fields, this T - H phase diagram is consistent with the in-plane one for this compound: a magnetic transition is apparent around 2.2 K, but the possible second one around 1.0 K is not visible in the H$``$c measurements; the T $``$ 2.2 K transition is driven down in temperature by increasing applied fields, and it approaches our base temperature (i.e., 0.4 K) around 120 kG. Lack of measurements below 0.4 K or above 140 kG limits our ability to probe the existence of a QCP in this orientation, similar to the one at H$`{}_{c}{}^{ab}=35`$ kG for $`Hab`$.
## IV discussion
For both the YbPtIn and YbPt<sub>0.98</sub>In compounds a number of similar properties, as well as systematic differences, could be distinguished. First, the small stoichiometry difference was apparent from single crystal x-ray measurements, which also resulted in slightly reduced lattice parameters and unit cell volumes in the YbPtIn system. Furthermore, the resistivity values were shifted towards higher values in the Pt-deficient compound (Figs.2b-3) for the whole temperature (T = 0.4-300 K) and field ranges (H = 0-140 kG) of our measurements. High enough fields suppressed the magnetic order in both systems, at least for field applied within the basal plane; in a similar manner to the T-scale in these compounds, the critical field value was reduced in YbPt<sub>0.98</sub>In by comparison to the analogous one in YbPtIn. For $`Hc`$, the field values required to suppress the magnetic order in YbPt<sub>0.98</sub>In and YbPtIn were close to, or respectively higher than our maximum available field (i.e., 140 kG); this precluded us from studying the low temperature properties of the two compounds outside the ordered state, for this orientation of the field.
For $`Hab`$ NFL-like behavior seems to occur in YbPt<sub>0.98</sub>In for intermediate field values above the ordering temperatures, as indicated by the logarithmic divergence of the C$`{}_{m}{}^{}/T`$ data (Fig.15a). In YbPtIn the NFL region is less clearly defined in the specific heat, as its logarithmic divergence was limited to a small temperature range by a downturn in the C$`{}_{m}{}^{}/T(\mathrm{ln}T)`$ data (Fig.5) towards high T. Even more ambiguous was the temperature dependence of the resistivity data in both compounds, for H $`>H_c^{ab}`$. Below we are presenting a detailed analysis of these data, aimed at distinguishing between, if existent, the NFL and FL regimes, and providing a more accurate estimate of the QCP position in each of the two compounds.
### IV.1 H$``$ab resistivity data above the magnetically ordered state
Given the ambiguity already mentioned in the power law that best describes $`\rho (T)`$ data above the ordered state, we attempt to determine the exponent $`\beta `$ when the temperature dependence of the resistivity is given by: $`\mathrm{\Delta }\rho (T)=AT^\beta `$, where $`\mathrm{\Delta }\rho (T)=\rho (T)\rho _0`$. This is best done by determining the slope of the $`\mathrm{\Delta }\rho (T)`$ data on log-log plots at various field values. However, the accuracy of the resulting $`\beta (H)`$ values is highly dependent on accurate estimates of the residual resistivity values $`\rho _0(H)`$. In turn, the estimates of the $`\rho _0(H)`$ values are made difficult by their enhanced values compared to the over-all resistivity values (fig.3).
Consequently, we determine the residual resistivity values using two criteria: firstly, from fits of the low-T resistivity data to a $`\rho _0+AT^\beta `$ function; secondly, given that figs.16b-17 (for YbPt<sub>0.98</sub>In), or fig.7 (for YbPtIn) suggests that low-T $`\rho (T)`$ data are close to being either linear or quadratic in T, we also choose (for mathematical rather than physical reasons) a second-order polynomial function to determine $`\rho _0`$ as a function of field: $`\rho (T)=\rho _0+AT+BT^2`$. For each field value, the average $`\rho _0`$ resulting from the two aforementioned criteria is used to calculate $`\mathrm{\Delta }\rho (T)=\rho (T)\rho _0`$ plotted in Fig.23a and 24a respectively, on a log-log scale. The large data scattering towards the lowest temperatures, particularly at high fields, is due to the fact that $`\mathrm{\Delta }\rho `$ is calculated as the difference between two comparably large values, $`\rho (T)`$ and $`\rho _0`$. As a consequence, the values of the exponent $`\beta `$ are being determined from (i) linear fits shown as solid lines, for $`\mathrm{\Delta }\rho (T)=\rho (T)\rho _0`$ larger than $``$ 0.1 $`\mu \mathrm{\Omega }cm`$ (marked by the dashed horizontal line), and (ii) from similar linear fits down to the lowest measured temperature (dotted lines). Together with the exponent values originally determined from fitting the low-T data to $`\rho _0+AT^\beta `$ functions, we can now determine the $`\beta `$ values shown in Fig.23b and 24b, as full symbols, with their corresponding error bars.
In the case of YbPt<sub>0.98</sub>In (Fig.23a), the resistivity is linear on the log-log scale, up to some maximum temperatures T<sub>max</sub> marked with large triangles, with a large change in slope for 60 kG $`<`$ H $`<`$ 70 kG. This is equivalent to the exponent $`\beta `$ having a discontinuous change from $`\beta 1`$ below 60 kG (i.e., for H $`>`$ H$`{}_{c}{}^{}{}_{}{}^{ab}`$), to $`\beta 2`$ in higher fields (H $``$ H$`{}_{c}{}^{}{}_{}{}^{ab}`$), as can be seen in Fig.23b. Also shown in Fig.23b are the temperatures T<sub>max</sub> below which the low-temperature $`\mathrm{\Delta }\rho `$ manifests a Fermi liquid-like exponent of $`\beta `$ 2, analogous to the cross-over previously reported for YbRh<sub>2</sub>Si<sub>2</sub>tro05 and YbAgGebud11 . The departure from linearity on the resistivity plots can be determined up to a temperature range, and not a unique temperature value, which results in finite error bars on these cross-over temperatures. However, a monotonic increase of the cross-over temperatures with field is observed, with the corresponding error bars also increasing. When adding this cross-over line to the H-T phase diagram (see below Fig.25, full triangles), it appears to converge with the magnetic order upper boundary (solid line) around the QCP between 45 and 70 kG.
When we describe the YbPtIn low-T resistivity data as $`\rho \rho _0=AT^\beta `$ for H $`>`$ H$`{}_{c}{}^{}{}_{}{}^{ab}`$, we encounter a situation very similar to that seen in YbPt<sub>0.98</sub>In: the logarithmic plot in Fig.24a, linear at low T, is indicative of a jump in slope for an applied field between 70 kG and 80 kG, where the slope gives the values of the exponent $`\beta `$. It appears that the exponent $`\beta `$ has a discontinuous variation with field (Fig.24b (left)) from $`\beta 1`$ for H $`>`$ H$`{}_{c}{}^{}{}_{}{}^{ab}`$, to $`\beta 2`$ when H $``$ H$`{}_{c}{}^{}{}_{}{}^{ab}`$. This suggests a NFL-like regime just above the magnetically ordered state, with a cross-over to a FL state above 100 kG, making this compound fairly similar to the Pt-deficient one, but also to the previously reported YbRh<sub>2</sub>Si<sub>2</sub>tro05 and YbAgGebud11 .
As observed before for YbPt<sub>0.98</sub>In, due to the data scattering, the departure of the low temperature resistivity plots from the FL-like $`\beta `$ 2 behavior at the highest field allows us to determine a cross-over temperature range, rather than a temperature value. This results in finite error bars on the cross-over temperatures, as seen in Fig.24b (right). However, an increase of the cross-over temperatures with field is observed, both in Fig.24b, and also in the revised H-T phase diagram (Fig.26) (full symbols).
### IV.2 Hall resistivity measurements in YbPt<sub>0.98</sub>In: H$``$ab
Based on the low temperature thermodynamic and transport measurements, YbPtIn and YbPt<sub>0.98</sub>In can be regarded as Yb-based heavy fermion compounds with long range, possibly reduced moment ordering that can be driven through a field-induced quantum critical point, similar to the previously studied Yb-based HF systems, YbRh<sub>2</sub>Si<sub>2</sub>pas09 and YbAgGebud12 . In the latter two compounds, the Hall effect served as an additional tool for characterizing the QCP and its effects on the finite-temperature properties of these materials.
In order to further explore the field-induced QCP our Yb-based compounds, field-dependent Hall resistivity measurements were performed for temperatures up to 300 K. However, the data on the stoichiometric compound was noisy, given the limited (small) crystal size, and meaningful Hall resistivity data could only be collected for the solution-grown compound for H $`ab`$. These measurements are shown in Fig.27.
We determined the Hall coefficient R<sub>H</sub> for YbPt<sub>0.98</sub>In as $`\rho _H(H)/H`$, in the low-field (H $`<30`$ kG) and high-field (H $`>60`$ kG) regimes, shown in Fig.27 for various T. (This alternative definition for the R$`{}_{H}{}^{}(T)`$ was preferred to $`d\rho _H/dH`$ due to the scattering of the data at low fields as seen later in Fig.29b). The results are shown in Fig.28 on a semi-log scale. The high-H points (squares) show the expected leveling off of R$`{}_{H}{}^{}(T)`$ as $`T0`$, whereas the low-H data appear to have a stronger temperature dependence. At temperatures higher than 25 K, the two data sets merge (as the $`\rho _H(H)`$ data become roughly linear for the whole field range \[Fig.27a\]). These data indicate that the field-dependent Hall resistivity will be non-trivial and potentially of interest at low temperatures.
In the field-dependent Hall resistivity measurements shown in Fig.27, two possible regimes can be distinguished: a high-T regime (T $`>25`$ K), where a monotonic decrease with field can be observed, despite the scattering of the data (Fig.27a), and a low-T regime (T $`25`$ K), for which a minimum in the $`\rho _H(`$H) data appears and sharpens as the temperature decreases (Fig.27b). For 5 K $``$ T $``$ 25 K, the Hall resistivity curves show a fairly broad minimum (Fig.27b), marked by the large gray dots, which moves down in field with decreasing temperature. Below 5 K, this minimum is more and more pronounced, and is almost unaffected by the change in temperature.
Coleman et al.col20 have indicated that R<sub>H</sub>(C) data (where C is a control parameter, i.e., H in our case) can be used to distinguish between two possible QCP scenarios: diffraction off of a critical spin density wave SDW (manifesting as a change in the slope of R<sub>H</sub>(C) at C<sub>crit</sub>) or a breakdown of the composite nature of the heavy electron (signaled by the divergence of the slope of R<sub>H</sub>(C) at C<sub>crit</sub>). In our case, it is not clear what definition of the Hall coefficient should be used for comparison with the theory (i.e., either R$`{}_{H}{}^{}(C)=\rho _H/H`$ or R$`{}_{H}{}^{}(C)=d\rho _H/dH`$) since the magnetic field H is itself the control parameter C. Therefore in Fig.29 we are presenting the R<sub>H</sub>(H) curves determined using both of the aforementioned definitions. When R<sub>H</sub>(H)$`=\rho _H/H`$ (Fig.29a), the Hall coefficient is linear up to a field value which varies non-monotonically with T, as indicated by the dashed line. This line is shifted to higher field values with respect to the Hall resistivity line (Fig.29a), and is marked here by the large diamonds. When using the R<sub>H</sub>(H)$`=d\rho _H/dH`$ definition for the Hall coefficient (Fig.29b), the maximum-H points on the low-field linear fits could be used as the criterium for defining the Hall line (indicated by the dotted line).
Regardless of what criterium is being used, the Hall measurements define a new phase line, distinct from any of the ones inferred from either the C<sub>P</sub>(T,H) or the $`\rho `$(H,T) data. Fig.30 shows the $`TH`$ phase diagram with the phase lines determined from the $`\rho `$(H,T) (H $`H_c^{ab}`$) or the C<sub>P</sub>(T,H) data (lines), together with the cross-over line and the Hall line inferred from the various definitions (symbols). The line resulting from the Hall resistivity $`\rho _H`$ data (diamonds) seems to persist even below T<sub>ord</sub>, down to our lowest T. However, the other two criteria used for the definition of the Hall line (R$`{}_{H}{}^{}(H)=\rho _H/H`$ or H<sub>max</sub> for the linear fit of R$`{}_{H}{}^{}(H)=d\rho _H/dH`$) delineate lines, which more closely resembles the Hall line observed for either YbRh<sub>2</sub>Si<sub>2</sub>pas09 and YbAgGebud12 as it appears to converge with the ”coherence” line (i.e., the line defining the cross-over between the non-Fermi liquid and Fermi-liquid like regimes), and the high-T magnetic ordering phase boundary, at the QCP, around H$`{}_{c}{}^{}{}_{}{}^{ab}`$. It should be noted, though that, unlike YbAgGe, for which the Hall line is clearly not associated with the saturation of the Yb momentstak05 , for YbPt<sub>0.98</sub>In the Hall line appears to be close to the saturation fields, at least at the lowest temperatures.
### IV.3 Summary
Based on the presented data, we can conclude that, in both YbPt<sub>0.98</sub>In and YbPtIn, when magnetic field is increased above $`H_c^{ab}`$, the system first enters a NFL state, followed, at higher fields, by a FL-like state: the specific heat has a logarithmic divergence (Figs.15 and 5) and the resistivity shows a $`\mathrm{\Delta }\rho AT^\beta `$ functional dependence (Fig.23a,b and 24a,b), with $`\beta `$ having a discontinuous change at intermediate field values, from $`\beta 1`$ to $`\beta `$ 2. Moreover, from Hall resistivity measurements on the Pt-deficient compound, we were able to determine another totally distinct line in the T - H phase diagram (Fig.25), convergent, towards the lowest temperature, with the cross-over line and the high-T magnetic order line. All these observations lead us to believe that a QCP exists in both compounds, around $`H_c^{ab}3555`$ kG for YbPt<sub>0.98</sub>In, and 60 kG for YbPtIn.
To further study the nature of the field-induced QCP in the two systems, we analyze the field dependence of the electronic specific heat coefficient $`\gamma `$. Also, in order to extend the comparison of the disordered state in our systems and in previously studied compounds (YbRh<sub>2</sub>Si<sub>2</sub>pas09 and YbAgGebud12 ), we analyze the field-dependence of the coefficients A in the $`\mathrm{\Delta }\rho AT^\beta `$ ($`\beta `$ = 2) resistivity, even though in both YbPtIn and YbPt<sub>0.98</sub>In the resistivity was not exactly quadratic for H $``$ H$`{}_{c}{}^{}{}_{}{}^{ab}`$, but $`\beta `$ 2.
The field-dependent electronic specific heat coefficient $`\gamma `$ could be estimated at low T, outside the ordered state, and these values are shown in Fig.31. For both compounds, $`\gamma (H)`$ was taken as the corresponding C$`{}_{m}{}^{}/T`$ value at T $`1.3`$ K, such as to avoid the low-T upturn in the highest field data. A drastic decrease of $`\gamma `$ (almost an order of magnitude) can be observed in Fig.31 for both compounds, for fields above their respective $`H_c^{ab}`$.
The field dependence of the coefficient A of the quadratic resistivity is shown in Fig.32a,b for YbPtIn and YbPt<sub>0.98</sub>In respectively. A $`1/(HH_{c0}^{calc})`$ divergence can be observed for both compounds, with $`H_{c0}^{calc}`$ estimated from the fit as 64.4 kG for YbPtIn, and 56.0 kG for YbPt<sub>0.98</sub>In. This critical field value for the former system is consistent with the $`Hab`$ specific heat and resistivity data (Figs.4-9), which suggested that $`H_c^{ab}`$ 60 kG. In the case of YbPt<sub>0.98</sub>In, where NFL behavior was observed between 35 kG and 55 kG, the critical field value can only be determined to a range within the above two field values; the $`H_{c0}^{calc}56`$ kG determined from the A vs. $`1/(HH_{c0}^{calc})`$ fit (Fig.32b) is close to the above field range, thus consistent with the specific heat and resistivity data (Figs.14-19,23-25).
The proportionality $`A\gamma ^2`$ between the resistance coefficient A and the electronic specific heat coefficient $`\gamma `$ is emphasized by the logarithmic plots in the insets in Fig.32. In the case of the YbPtIn compound (Fig.32a, inset), the solid line represents the Kadowaki-Woods ratio $`A/\gamma ^22.610^5\mu \mathrm{\Omega }`$ cm $`/`$ (mJ / mol K)<sup>2</sup>. Such a value is close to that observed for many heavy fermion systemskad21 , but slightly higher than the $`A/\gamma ^2`$ ratio reported for YbRh<sub>2</sub>Si<sub>2</sub>geg06 ; this, in turn, was larger than the values observed for most Yb-based intermetallic compoundstsu22 . The corresponding value was larger still in YbAgGebud11 , and it would appear that this is a common feature of Yb-based materials with field-induced NFL-like behavior.
When we turn to the Pt-deficient compound (Fig.32b, inset), a smaller $`A/\gamma ^2`$ ratio is indicated by the solid line: $`A/\gamma ^20.410^5\mu \mathrm{\Omega }`$ cm $`/`$ (mJ / mol K)<sup>2</sup>. This value is again one order of magnitude larger than that expected for many Yb-based compoundstsu22 , and several times smaller than the ratio observed for the stoichiometric YbPtIn compound (Fig.32a). In light of observations of suppression of the A coefficient due to site disordertsu22 ; law23 , we may have to consider the $`A/\gamma ^2`$ ratio for YbPt<sub>0.98</sub>In as a reduced value from the enhanced ratio observed for the stoichiometric compound.
## V conclusions
The detailed field- and temperature-dependent measurements presented here allowed us to confirm that YbPtIn is a heavy fermion compound, as has been presented by Trovarelli et al.tro09 . In addition, we showed that a field-induced QCP exists in this material, at least for $`Hab`$, with $`H_c^{ab}60`$ kG. In addition to the magnetic field used as a control parameter, we showed that a small Pt-deficiency introduced in this system had effects consistent with positive pressure applied to Yb-based heavy fermion compounds. Thus, in the YbPt<sub>0.98</sub>In compound we also see a suppression of the magnetic ordering by applied magnetic field; in addition, the small disorder also suppresses the ordered state (both T<sub>ord</sub> and H<sub>c</sub> have smaller values in this compound than the similar ones in YbPtIn).
NFL regions can be observed in both our compounds for H $`>`$ H$`{}_{c}{}^{}{}_{}{}^{ab}`$, characterized by linear $`\rho (T^\beta )`$ ($`\beta `$ 1) and logarithmic divergency of the specific heat. For H $``$ H$`{}_{c}{}^{}{}_{}{}^{ab}`$, the exponent $`\beta `$ approaches 2, suggesting a cross-over to a low-T FL state at these field values, similar to YbRh<sub>2</sub>Si<sub>2</sub>geg06 and YbAgGebud11 .
As the critical field required to suppress the magnetic order in YbPt<sub>0.98</sub>In and YbPtIn in the c direction appeared to exceed our maximum field, experiments to higher fields would be desirable in order to extend the comparison with YbRh<sub>2</sub>Si<sub>2</sub> and YbAgGe to the effects of anisotropy on the field-induced QCP.
## VI acknowledgments
Ames Laboratory is operated for the U.S. Department of Energy by Iowa State University under Contract No. W-7405-Eng.-82. This work was supported by the Director for Energy Research, Office of Basic Energy Sciences.
|
warning/0506/hep-th0506208.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
Models with $`N=(2,2)`$ worldsheet supersymmetry play a major role in various applications of string theory, most notably in the context of Calabi-Yau compactifications, mirror symmetry and the construction of four-dimensional string vacua. The simplest non-trivial theories with superconformal symmetry are the $`N=2`$ minimal models which have an ADE classification. These models play a central role in the construction of Gepner models that describe certain Calabi-Yau compactifications at specific points in their moduli space. In addition to their realisation as abstract conformal field theories, the $`N=2`$ minimal models also have a description as Landau Ginzburg theories. For the case of the D-model that shall mainly concern us in this paper, the relevant superpotentials are
$$W_D=x^{\frac{n+1}{2}}xy^2,W_D^{}=x^{\frac{n+1}{2}}xy^2z^2.$$
(1.1)
The two different choices correspond to the two possible GSO-projections that can be imposed in conformal field theory.
In this paper, we perform a complete analysis of the B-type branes in both of these theories. These can be studied from two points of view: on the one hand, they have a description in terms of boundary states in rational conformal field theory. On the other hand, according to a proposal in unpublished work of Kontsevich, B-type branes in Landau Ginzburg models can be characterised in terms of matrix factorisations of the superpotential. This proposal was discussed in the physics literature in . One would thus expect that the boundary state analysis in conformal field theory should agree with that of the D-branes in the Landau Ginzburg theory. This was for example seen to be the case for the A-type models in .
In the following we shall compare in detail the boundary states of the D-type minimal models with the matrix factorisations of the corresponding Landau Ginzburg models. Furthermore, we shall relate the matrix factorisations of the Landau Ginzburg superpotential $`W`$ to matrix factorisations that appear in the study of singularity theory. Surface singularities can formally be related to Landau Ginzburg theories by studying the locus $`W=0`$ as a hypersurface in $`^3`$. The matrix factorisations encode then the geometrical details of the blow-ups necessary to resolve the surface singularity. Interpreted on the singular variety, they also correspond to the elements of Orlov’s category $`D_{Sg}`$ . Our analysis suggests that the Landau-Ginzburg theory, despite having smaller central charge, captures some of the physical properties of the D-branes on these surface singularities that become massless at the singular point.
The paper is organised as follows. In section 2 we review the relation between matrix factorisations and boundary states for the A-type minimal models. As explained in , adding a square to the Landau Ginzburg superpotential corresponds to a change in the GSO projection in conformal field theory. In section 3 we construct all B-type boundary states in the D-model for both possible GSO projections. Since the D-model is a rational conformal field theory (with respect to the $`N=2`$ superconformal algebra), the boundary states we construct generate all superconformal boundary states. In section 4 we discuss the matrix factorisations of the corresponding Landau Ginzburg models, completing the list of factorisations provided in . Once these additional factorisations are considered, we obtain complete agreement with the conformal field theory description. On the other hand, since the latter description is complete (in the above sense), we can conclude that the matrix factorisations we have found are all the fundamental matrix factorisations for these superpotentials. This is to say, any matrix factorisation must be the direct sum of these fundamental factorisations. Our analysis also confirms the conjecture of regarding the relation between the different GSO-projections in conformal field theory and the addition of a square to the Landau Ginzburg model. Finally, in section 5, we investigate the relation between D-branes on singular surfaces, the corresponding Landau Ginzburg theories and conformal field theory. We also comment on the description of the surface singularity in terms of the singular quotient $`^2/\mathrm{\Gamma }`$, where $`\mathrm{\Gamma }`$ is a subgroup of $`SU(2)`$. This allows us to use the methods of to describe D-branes on those orbifolds in terms of representations of $`\mathrm{\Gamma }`$. We find complete agreement with the results from the matrix factorisation point of view, confirming yet again that we have found all D-branes.
## 2 Matrix factorisations
Let us briefly review the description of D-branes in terms of matrix factorisations. This approach was proposed in unpublished form by Kontsevich, and the physical interpretation of it was given in ; for a good review of this material see for example .
According to Kontsevich’s proposal, D-branes in Landau Ginzburg models correspond to matrix factorisations of the superpotential $`W(x_i)`$,
$$d_0d_1=d_1d_0=W\mathbf{\hspace{0.17em}1},$$
(2.1)
where $`d_0`$ and $`d_1`$ are $`r\times r`$ matrices with polynomial entries in the $`x_i`$. To a matrix factorisation of this form, one can then associate the boundary BRST operator $`Q`$ of the form
$$Q=\left(\begin{array}{cc}0& d_1\\ d_0& 0\end{array}\right).$$
(2.2)
The spectrum of open string operators consists of polynomial matrices of the same size. It is naturally $`_2`$ graded, where the bosons $`\varphi `$ correspond to block-diagonal matrices, while the fermions $`t`$ are off-diagonal:
$$\varphi =\left(\begin{array}{cc}\varphi _0& 0\\ 0& \varphi _1\end{array}\right),t=\left(\begin{array}{cc}0& t_1\\ t_0& 0\end{array}\right).$$
(2.3)
To obtain the physical spectrum between two such branes, one restricts to the degree $`0`$ operators satisfying $`D\varphi =[Q,\varphi ]=0`$, and identifies operators that differ only by BRST-exact terms. To calculate this cohomology, we can follow the strategy of . A BRST invariant boson has to satisfy
$$d_1\varphi _1=\varphi _0d_1,\text{and}d_0\varphi _0=\varphi _1d_0,$$
(2.4)
where $`\varphi _0`$ and $`\varphi _1`$ are $`r\times r`$ matrices. This equation can be solved for $`\varphi _0`$ as
$$\varphi _0=\frac{1}{W}d_1\varphi _1d_0,$$
(2.5)
provided that the matrix on the right hand side is divisible by $`W`$. After moding out $`Q`$-exact terms, the bosonic cohomology can therefore be described using $`\varphi _1`$ only. The BRST trivial $`\stackrel{~}{\varphi }_1`$ are derivatives of fermions, and hence take the form
$$\stackrel{~}{\varphi }_1=(Dt)_1=t_0d_1+d_0t_1.$$
(2.6)
For the fermions, the condition for BRST invariance is
$$t_0d_1+d_0t_1=0=d_1t_0+t_1d_0,$$
(2.7)
which can be solved for $`t_1`$
$$t_1=\frac{1}{W}d_1t_0d_1,$$
(2.8)
resulting again in a divisibility condition. An invariant fermion $`(\stackrel{~}{t}_0,\stackrel{~}{t}_1)`$ is BRST trivial if $`\stackrel{~}{t}_0`$ can be written as
$$\stackrel{~}{t}_0=\varphi _1d_0+d_0\varphi _0,$$
(2.9)
for a boson $`(\varphi _0,\varphi _1)`$.
It is easy to see that the spectrum of two factorisations $`Q`$ and $`\widehat{Q}`$ is the same, if
$$Q=U\widehat{Q}U^1,U=\left(\begin{array}{cc}A& 0\\ 0& B\end{array}\right),$$
(2.10)
provided that $`U`$ and its inverse, $`U^1`$, are polynomial matrices. We therefore identify two such factorisations .
For a given factorisation $`Q`$, we call the factorisation where the roles of $`d_0`$ and $`d_1`$ are reversed the reverse factorisation $`Q^r`$. It is clear from the above discussion that the bosonic spectrum between two factorisations $`Q_1`$ and $`Q_2`$ and the fermionic spectrum between $`Q_1`$ and $`Q_2^r`$ coincide.
To make contact with conformal field theory, one has to make sure that the $`U(1)`$ $`R`$-symmetry that becomes the $`U(1)`$ current symmetry of the $`N=2`$ superconformal algebra in the IR is preserved . This means that one has to restrict to homogeneous superpotentials and specify a consistent assignment of R-charge in the boundary theory. By (2.1) and (2.2) this implies that $`Q`$ should have charge one
$$e^{i\lambda R}Q(e^{i\lambda q_i}x_i)e^{i\lambda R}=e^{i\lambda }Q.$$
(2.11)
Then one can assign R-charge to the boundary operators by
$$e^{i\lambda R}\varphi (e^{i\lambda q_i}x_i)e^{i\lambda R}=e^{i\lambda q}\varphi .$$
(2.12)
This description of D-branes should then have a direct correspondence in conformal field theory. In particular, the different factorisations (up to the aforementioned equivalence) should correspond to the different B-type boundary states in conformal field theory. The physical boson spectrum described above corresponds then to the topological open string spectrum from the point of view of conformal field theory. The reverse factorisation of a given factorisation corresponds to the anti-brane of the given brane; thus the physical fermion spectrum in the above description corresponds to the topological open string spectrum between a brane and an anti-brane.
In order to illustrate these concepts and set up notation for the rest of the paper, we briefly review the case of the A-type minimal models.
### 2.1 A-type minimal models
In A-type minimal models were studied from the matrix point of view. For the potential
$$W_A=x^n$$
(2.13)
it can be shown that a class of inequivalent matrix factorisations that generate all factorisations are described by the simple rank $`1`$ factorisations $`W=x^lx^{nl}`$, where $`l=1,\mathrm{},n1`$. The corresponding BRST operator is
$$Q_l=\left(\begin{array}{cc}0& x^l\\ x^{nl}& 0\end{array}\right).$$
(2.14)
The factorisations $`Q_l`$ and $`Q_{nl}`$ are related to one another by the exchange of $`d_0`$ and $`d_1`$; they are therefore reverse factorisations of one another.
The corresponding conformal field theory is described by a single $`N=2`$ minimal model with $`n=k+2`$. (Our conventions are chosen as in .) The spectrum of this theory is (after GSO-projection)
$$_A=\underset{[l,m,s]}{}\left(_{[l,m,s]}\overline{}_{[l,m,s]}\right).$$
(2.15)
The GSO projection chosen here is the analogue of the Type 0A projection. B-type branes are characterised by the gluing conditions
$`(L_n\overline{L}_n)||B`$ $`=`$ $`0`$
$`(J_n+\overline{J}_n)||B`$ $`=`$ $`0`$ (2.16)
$`(G_r^\pm +i\eta \overline{G}_r^\pm )||B`$ $`=`$ $`0.`$
The corresponding B-type boundary states were constructed some time ago (see for example ), and are explicitly given as
$$||L,S=\sqrt{k+2}\underset{l+s2}{}\frac{S_{L0S,l0s}}{\sqrt{S_{l0s,000}}}|[l,0,s].$$
(2.17)
Here $`L=0,1,\mathrm{},k`$ and $`S=0,1,2,3`$. The boundary states with $`S`$ even (odd) satisfy the gluing conditions with $`\eta =+1`$ ($`\eta =1`$); in the following we shall restrict ourselves to the case $`\eta =+1`$, and thus to even $`S`$.The D-branes corresponding to $`\eta =1`$ or $`S`$ odd preserve a different supercharge at the boundary. The branes that are described by the different matrix factorisations however always preserve the same supercharge at the boundary. We also note that
$$||L,S=||kL,S+2$$
(2.18)
and thus there are only $`k+1`$ different boundary states with $`\eta =+1`$ (and $`k+1`$ different boundary states with $`\eta =1`$). These boundary states therefore account for all the $`N=2`$ Ishibashi states. Finally we note that $`||L,S`$ and $`||L,S+2`$ are anti-branes of one another (since they differ by a sign in the coupling to the RR sector states).
It is then suggestive to identify the matrix factorisation corresponding to $`Q_l`$ (2.14) with the boundary state with label $`L=l1`$ and $`S=0`$. The equivalence (2.18) mirrors the factorisation reversal map $`lnl`$ on the matrix side. Restricting without loss of generality to the range $`ln/2`$ one can then easily check, both in conformal field theory and the matrix language, that there are $`l`$ topological states in the open string spectrum on each of these branes (they correspond to the ‘bosons’ from the matrix point of view), and $`l`$ topological states in the open string spectrum between brane and anti-brane (the ‘fermions’ in the matrix description).
From the conformal field theory perspective it is immediately clear that there is another closely related theory, where one uses a type 0B like GSO projection instead of the GSO projection discussed above. The relevant Hilbert space is
$$_A^{}=\underset{[l,m,s]}{}\left(_{[l,m,s]}\overline{}_{[l,m,s]}\right).$$
(2.19)
It was shown in that from the matrix point of view this theory is described by the superpotential
$$W_A^{}=x^ny^2.$$
(2.20)
Since the D-models that we shall discuss in this paper are closely related to this theory, let us briefly summarise the results of on factorisations and their relation to boundary states. The superpotential $`W_A^{}`$ has two classes of factorisations: the first class is given by
$$d_0=\left(\begin{array}{cc}x^l& y\\ y& x^{nl}\end{array}\right),d_1=\left(\begin{array}{cc}x^{nl}& y\\ y& x^l\end{array}\right).$$
(2.21)
One can easily see that the factorisation corresponding to $`l`$ and $`nl`$ are equivalent in the sense of (2.10); in the following we shall therefore always take $`ln/2`$. Since these are reverse factorisations of one another, the corresponding branes have to be their own anti-branes.
These factorisations are just a special class of the tensor product factorisations discussed in . In fact, they consist of one factor coming from the potential $`x^n`$, and another one from the ‘trivial’ factor $`y^2`$ whose physical significance has been discussed in . Accordingly, the open string spectrum inherits the tensor product structure: there are $`2l`$ bosons of the type
$$a_i=\left(\begin{array}{cc}x^i& 0\\ 0& x^i\end{array}\right),a_{l+i}=\left(\begin{array}{cc}0& x^i\\ x^{n2l+i}& 0\end{array}\right),i=0,\mathrm{},l1,$$
(2.22)
where the $`2\times 2`$ matrices $`a_j`$ describe $`\varphi _1`$, and $`\varphi _0`$ is then determined by (2.5). Likewise, there are $`2l`$ fermions
$$\eta _i=\left(\begin{array}{cc}0& x^i\\ x^i& 0\end{array}\right),\eta _{l+i}=\left(\begin{array}{cc}x^i& 0\\ 0& x^{n2l+i}\end{array}\right),i=0,\mathrm{},l1,$$
(2.23)
where the $`2\times 2`$ matrices $`\eta _j`$ describe $`t_0`$, and $`t_1`$ is then determined by (2.8). In the case that $`n`$ is even, the factorisation $`l=n/2`$ is not irreducible. In fact, it is equivalent (in the sense of (2.10)) to the direct sum of the rank $`1`$ factorisation
$$d_0=(x^{\frac{n}{2}}+y),d_1=(x^{\frac{n}{2}}y),$$
(2.24)
and its reverse. The physical open string spectrum of these rank $`1`$ factorisations consists of $`n/2`$ bosons (but no fermions); correspondingly there are $`n/2`$ fermions (but no bosons) between the factorisation (2.24) and its reverse.
In conformal field theory, the rank $`2`$ factorisations reviewed above correspond to the boundary states
$$||L,S=(2k+4)^{1/4}\underset{l\mathrm{even}}{}\underset{\nu =0,1}{}\frac{S_{Ll}}{\sqrt{S_{0l}}}e^{i\pi \nu S}|[l,0,2\nu ].$$
(2.25)
Here $`S`$ is only defined modulo $`2`$, and $`||L,S=||kL,S`$. As before, the identification relates the factorisation (2.21) to the boundary state with $`L=l1`$ and $`S=0`$. (Recall that $`n=k+2`$.) These boundary states do not couple to RR states and are therefore equivalent to their own anti-branes. (This is simply the statement that $`S`$ is only defined modulo $`2`$.)
If $`k`$ is even, then the $`L=k/2`$ boundary state contains two vacua in its open string spectrum and thus can be further resolved. The explicit form of the two resolved boundary states is
$$||k/2,S_{\mathrm{res}}=\frac{1}{2}(||k/2,S+\sqrt{k+2}\underset{s=\pm 1}{}e^{\frac{\pi iSs}{2}}|[k/2,(k+2)/2,s]),$$
(2.26)
where $`S`$ is now defined mod $`4`$, and $`||k/2,S_{\mathrm{res}}`$ and $`||k/2,S+2_{\mathrm{res}}`$ are anti-branes of one another. The two branes with $`S=0`$ and $`S=2`$ then correspond to the matrix factorisation (2.24) and its reverse, respectively.
We shall now construct the boundary states of the various GSO-projections of the D-model; in section 4 we shall discuss the corresponding matrix factorisations and identify the two descriptions.
## 3 Boundary states for the D-model
The D-model is only defined if the level $`k`$ is even. As before for the case of the A-type minimal model there are two possible GSO-projections one can consider. We begin by discussing the theory that is analogous to type 0B. Its spectrum depends on whether $`k/2`$ is even or odd, i.e. on whether $`k`$ is divisible by $`4`$ or not. In the former case, the spectrum is
$$=\underset{[l,m,s],l\mathrm{even}}{}\left(\left(_{[l,m,s]}\overline{}_{[l,m,s]}\right)\left(_{[l,m,s]}\overline{}_{[kl,m,s]}\right)\right).$$
(3.1)
On the other hand, if $`k/2`$ is odd the spectrum is
$$=\underset{[l,m,s],l\mathrm{even}}{}\left(\left(_{[l,m,s]}\overline{}_{[l,m,s]}\right)\right)\underset{[l,m,s],l\mathrm{odd}}{}\left(\left(_{[l,m,s]}\overline{}_{[kl,m,s]}\right)\right).$$
(3.2)
We are interested in the B-type boundary states of this theory. In our conventions, the B-type Ishibashi states are characterised by the gluing conditions (2.16). The structure of these Ishibashi states (and thus the corresponding boundary states) depends on the two cases above, and we shall therefore discuss them in turn.
### 3.1 The case $`k/2`$ odd
For $`k/2`$ odd, the theory (3.2) has the Ishibashi states
$$|[l,0,s]_{[l,0,s]}\overline{}_{[l,0,s]},$$
(3.3)
where $`l`$ and $`s`$ are both even, and the Ishibashi states
$$|[l,(k+2)/2,s]_{[l,\frac{k+2}{2},s]}\overline{}_{[kl,\frac{k+2}{2},s]}$$
(3.4)
where $`l`$ and $`s`$ are both odd. In total there are therefore $`2(k+1)`$ Ishibashi states of the coset theory; they give rise to $`(k+1)`$ Ishibashi states of the $`N=2`$ theory with one spin structure $`\eta =+1`$, and $`(k+1)`$ Ishibashi states of the $`N=2`$ algebra with the other $`\eta =1`$.
The corresponding boundary states are described by
$`||L,M,S`$ $`=`$ $`\sqrt{{\displaystyle \frac{2k+4}{2}}}({\displaystyle \underset{l,s\mathrm{even}}{}}{\displaystyle \frac{S_{LMS,l0s}}{\sqrt{S_{000,l0s}}}}|[l,0,s]`$ (3.5)
$`+{\displaystyle \underset{l,s\mathrm{odd}}{}}{\displaystyle \frac{S_{LMS,l\frac{k+2}{2}s}}{\sqrt{S_{000,l\frac{k+2}{2}s}}}}|[l,(k+2)/2,s]).`$
Here $`L+M+S`$ is even, and we have the identifications
$`||L,M,S`$ $`=`$ $`||kL,M+k+2,S+2=||kL,M,S+2`$ (3.6)
$`=`$ $`||L,M+4,S=||L,M+2,S+2.`$
For fixed spin structure $`\eta =+1`$ (corresponding to $`S=0,2`$), we have for every value of $`L`$ only two possible values of $`M`$, namely $`M=0,2`$ or $`M=1,3`$. Furthermore, because of the last identification, of the four choices ($`S=0,2`$, $`M=0,2`$) or ($`S=0,2`$, $`M=1,3`$) only two describe different boundary states. For $`Lk/2`$, the pair of boundary states with $`L`$ and $`kL`$ account therefore each for two different boundary states giving in total $`k`$ different boundary states. For $`L=k/2`$ on the other hand, it is easy to see that the second sum in (3.5) vanishes, and thus $`||k/2,+1,0=||k/2,1,0`$, thus giving only one additional boundary state. In total the above construction therefore gives $`k+1`$ different boundary states of fixed spin structure, and thus accounts for all the Ishibashi states.
These boundary states satisfy the Cardy condition since their overlap equals
$`L_1,M_1,S_1||q^{L_0+\overline{L}_0\frac{c}{12}}||L_2,M_2,S_2={\displaystyle \underset{[l,m,s]}{}}\chi _{(l,m,s)}(\stackrel{~}{q})\delta ^{(2)}(S_2+sS_1)\times `$
$`\times (N_{L_2l}{}_{}{}^{L_1}\delta _{}^{(4)}(ms+M_2M_1S_2+S_1)`$
$`+N_{L_2kl}{}_{}{}^{L_1}\delta _{}^{(4)}(ms+M_2M_1S_2+S_1+2)).`$ (3.7)
Here $`N_{Ll}^{\widehat{L}}`$ denotes the level $`k`$ fusion rules of $`su(2)`$, and $`\chi _{(l,m,s)}`$ is the character of the coset representation. We note that the right hand side is invariant under the field identification, $`(l,m,s)(kl,m+k+2,s+2)`$, as it must be.
#### 3.1.1 Topological spectrum
It is now straightforward to read off from this formula the topological spectrum between two such branes. If we consider only D-branes of a fixed spin structure, we may set, without loss of generality, $`S=0`$. Then we can restrict the open string sum without loss of generality to the states with $`s=0`$; then the topological states arise for $`l=m`$.
For example, for the case $`L=L_1=L_2`$ (and $`S_1=S_2=0`$), the number of topological states $`T`$ is
$$M_1=M_2(\mathrm{bosons}):|T|=\{\begin{array}{cc}L+2& \text{if }L\text{ is even}\hfill \\ L+1& \text{if }L\text{ is odd,}\hfill \end{array}$$
(3.8)
and
$$M_1=M_2+2(\mathrm{fermions}):|T|=\{\begin{array}{cc}L& \text{if }L\text{ is even}\hfill \\ L+1& \text{if }L\text{ is odd.}\hfill \end{array}$$
(3.9)
Note that the boundary state corresponding to $`L=k/2`$ (which is an odd number) has the same number of topological fermions and bosons in its open string spectrum.
### 3.2 The case $`k/2`$ even
For $`k/2`$ even, the theory (3.1) has the $`(k+2)`$ Ishibashi states
$$|[l,0,s]_{[l,0,s]}\overline{}_{[l,0,s]},$$
(3.10)
where $`l`$ and $`s`$ are both even, and the $`(k+2)`$ Ishibashi states
$$|[l,(k+2)/2,s]_{[l,\frac{k+2}{2},s]}\overline{}_{[kl,\frac{k+2}{2},s]},$$
(3.11)
where $`l`$ is even and $`s`$ is odd. In addition, there are the two Ishibashi states from the first sum in (3.1)
$$|[k/2,(k+2)/2,s_{[k/2,\frac{k+2}{2},s]}\overline{}_{[k/2,\frac{k+2}{2},s]},s=\pm 1$$
(3.12)
and the two Ishibashi states from the second sum in (3.1)
$$|[k/2,0,s_{[k/2,0,s]}\overline{}_{[kk/2,0,s]},s=0,2.$$
(3.13)
In total there are therefore $`2(k+4)`$ Ishibashi states of the coset theory; they give rise to $`(k+4)`$ Ishibashi states of the $`N=2`$ theory with one spin structure $`\eta =+1`$, and $`(k+4)`$ Ishibashi states of the $`N=2`$ algebra with the other $`\eta =1`$.
The corresponding boundary states are described by
$`||L,M,S`$ $`=`$ $`\sqrt{{\displaystyle \frac{2k+4}{2}}}{\displaystyle \underset{l\mathrm{even}}{}}({\displaystyle \underset{s\mathrm{even}}{}}{\displaystyle \frac{S_{LMS,l0s}}{\sqrt{S_{000,l0s}}}}|[l,0,s]`$ (3.14)
$`+{\displaystyle \underset{s\mathrm{odd}}{}}{\displaystyle \frac{S_{LMS,l\frac{k+2}{2}s}}{\sqrt{S_{000,l\frac{k+2}{2}s}}}}|[l,(k+2)/2,s]).`$
Here $`L+M+S`$ is even, and $`Lk/2`$. We have the identifications
$`||L,M,S`$ $`=`$ $`||kL,M+k+2,S+2=||kL,M,S`$ (3.15)
$`=`$ $`||L,M+4,S=||L,M+2,S+2.`$
For fixed spin structure $`\eta =+1`$ (corresponding to $`S=0,2`$), we have for every value of $`L`$ only two possible values of $`M`$, namely $`M=0,2`$ or $`M=1,3`$. Furthermore, because of the last identification, of the four choices ($`S=0,2`$, $`M=0,2`$) or ($`S=0,2`$, $`M=1,3`$) only two describe different boundary states. For $`Lk/2`$, the pair of boundary states with $`L`$ and $`kL`$ account therefore each for two different boundary states giving in total $`k`$ different boundary states. The remaining four boundary states correspond to the ‘resolved’ boundary states for $`L=k/2`$, and will be described shortly.
These boundary states satisfy the Cardy condition since their overlap equals
$`L_1,M_1,S_1||q^{L_0+\overline{L}_0\frac{c}{12}}||L_2,M_2,S_2={\displaystyle \underset{[l,m,s]}{}}\chi _{(l,m,s)}(\stackrel{~}{q})\delta ^{(2)}(S_2+sS_1)\times `$
$`\times (N_{L_2l}{}_{}{}^{L_1}+N_{L_2kl}{}_{}{}^{L_1})\delta ^{(4)}(ms+M_2M_1S_2+S_1).`$ (3.16)
We observe again that the right hand side is invariant under the field identification, $`(l,m,s)(kl,m+k+2,s+2)`$, as it must be.
It is easy to see that the topological open string spectrum of these branes is described by the same formulae as (3.8) and (3.9) above.
#### 3.2.1 The resolved branes
The remaining boundary states are of the form
$`||k/2,M,S,\pm `$ $`=`$ $`{\displaystyle \frac{1}{2}}||k/2,M,S_{(\text{3.14})}`$ (3.17)
$`\pm {\displaystyle \frac{\sqrt{2k+4}}{4}}({\displaystyle \underset{s\mathrm{odd}}{}}e^{i\pi M/2}e^{i\pi sS/2}|[k/2,(k+2)/2,s]`$
$`+{\displaystyle \underset{s\mathrm{even}}{}}e^{i\pi sS/2}|[k/2,0,s]).`$
Here $`M+S`$ is even, and we have the same identifications as in the second line of (3.15). Thus these boundary states account for four more boundary states (of a fixed spin structure).
One easily checks that the overlaps of these boundary states with the previously constructed boundary states is simply
$`L_1,M_1,S_1||q^{L_0+\overline{L}_0\frac{c}{12}}||k/2,M_2,S_2,\pm ={\displaystyle \underset{[l,m,s]}{}}\chi _{(l,m,s)}(\stackrel{~}{q})\delta ^{(2)}(S_2+sS_1)\times `$
$`\times N_{k/2l}{}_{}{}^{L_1}\delta _{}^{(4)}(ms+M_2M_1S_2+S_1).`$ (3.18)
On the other hand, the overlaps involving two resolved branes is
$`k/2,M_1,S_1,\pm q^{L_0+\overline{L}_0\frac{c}{12}}k/2,M_2,S_2,\pm `$ (3.19)
$`={\displaystyle \underset{[l,m,s]l=4n}{}}\chi _{(l,m,s)}(\stackrel{~}{q})\delta ^{(2)}(S_2+sS_1)\delta ^{(4)}(ms+M_2M_1S_2+S_1),`$
and
$`k/2,M_1,S_1,\pm q^{L_0+\overline{L}_0\frac{c}{12}}k/2,M_2,S_2,`$ (3.20)
$`={\displaystyle \underset{[l,m,s]l=4n+2}{}}\chi _{(l,m,s)}(\stackrel{~}{q})\delta ^{(2)}(S_2+sS_1)\delta ^{(4)}(ms+M_2M_1S_2+S_1).`$
It is now straightforward to determine the topological open string spectrum of these branes. As before, we may set, without loss of generality (if we restrict ourselves to a specific spin structure) $`S_1=S_2=0`$. Then the number of topological states on the $`+`$brane is
$`M_1=M_2(\mathrm{bosons}):`$ $`|T|={\displaystyle \frac{k}{4}}+1,`$ (3.21)
$`M_1=M_2+2(\mathrm{fermions}):`$ $`|T|=0.`$ (3.22)
Obviously the result for the $``$brane is identical. On the other hand, the topological spectrum between the $`+`$brane and the $``$brane is
$`M_1=M_2:`$ $`|T|=0,`$ (3.23)
$`M_1=M_2+2:`$ $`|T|={\displaystyle \frac{k}{4}}.`$ (3.24)
### 3.3 The other GSO-projection
As we mentioned before, we can also consider the 0A-like GSO-projection. Then the spectrum of the D-model is for $`k/2`$ even
$$=\underset{[l,m,s],l\mathrm{even}}{}\left(\left(_{[l,m,s]}\overline{}_{[l,m,s]}\right)\left(_{[l,m,s]}\overline{}_{[kl,m,s]}\right)\right).$$
(3.25)
On the other hand, if $`k/2`$ is odd, the spectrum is
$$=\underset{[l,m,s],l\mathrm{even}}{}\left(\left(_{[l,m,s]}\overline{}_{[l,m,s]}\right)\right)\underset{[l,m,s],l\mathrm{odd}}{}\left(\left(_{[l,m,s]}\overline{}_{[kl,m,s]}\right)\right).$$
(3.26)
In both cases we have the $`k+2`$ Ishibashi states
$$|[l,0,s]_{[l,0,s]}\overline{}_{[l,0,s]},$$
(3.27)
where $`l`$ and $`s`$ are both even, as well as the two Ishibashi states
$$|[k/2,0,s]_{[k/2,0,s]}\overline{}_{[k/2,0,s]},k/2+s=\mathrm{even}.$$
(3.28)
We therefore expect to have $`k/2+2`$ boundary states of each spin structure. Some of these boundary states are given by
$$||L,S=\sqrt{2k+4}\underset{l,s\mathrm{even}}{}\frac{S_{L0S,l0s}}{\sqrt{S_{000,l0s}}}|[l,0,s],$$
(3.29)
where $`L=0,1,\mathrm{},k`$ with $`Lk/2`$ and $`S`$ is defined modulo $`4`$. We observe that
$$||L,S=||kL,S=||L,S+2.$$
(3.30)
For fixed spin structure (say $`S`$ even) there are thus $`k/2`$ such boundary states. Their overlap equals
$`L_1,S_1q^{L_0+\overline{L}_0\frac{c}{12}}L_2,S_2`$
$`={\displaystyle \underset{[l,m,s]}{}}\chi _{(l,m,s)}(\stackrel{~}{q})\delta ^{(2)}(S_2+sS_1)(N_{L_2l}{}_{}{}^{L_1}+N_{L_2kl}{}_{}{}^{L_1}).`$
Thus the open string spectrum on the brane $`||L,S`$ has $`2(L+1)`$ topological states; the same is true for the open string between the $`||L,S`$ brane and its anti-brane.
The remaining two boundary states are then
$$||k/2,S,\pm =\frac{1}{2}||k/2,S_{(\text{3.29})}\pm \sqrt{\frac{k+2}{4}}\underset{s}{}e^{i\pi Ss/2}|[k/2,0,s],$$
(3.31)
where the sum over $`s`$ runs over $`0,2`$ if $`k/2`$ is even, and over $`\pm 1`$ if $`k/2`$ is odd. For $`k/2`$ odd, these two branes are anti-branes of one another, i.e. $`||k/2,S,+=||k/2,S+2,`$, and their overlaps equal
$`k/2,S_1,\pm q^{L_0+\overline{L}_0\frac{c}{12}}k/2,S_2,\pm `$
$`={\displaystyle \underset{[l,m,s],l\mathrm{even}}{}}\chi _{(l,m,s)}(\stackrel{~}{q})\delta ^{(2)}(S_2+sS_1)\delta ^{(4)}(l+S_2+sS_1).`$
There are then $`(k+2)/4`$ topological states in the open string spectrum of each of these branes, and $`(k+2)/4`$ topological states in the open string spectrum between the brane and the anti-brane.
For $`k/2`$ even, on the other hand, each of the two branes $`||k/2,S,\pm =||k/2,S+2,\pm `$ is its own anti-brane. In this case their overlaps equal
$`k/2,S_1,\pm q^{L_0+\overline{L}_0\frac{c}{12}}k/2,S_2,\pm `$
$`={\displaystyle \underset{[l,m,s]}{}}\chi _{(l,m,s)}(\stackrel{~}{q})\delta ^{(2)}(S_2+sS_1)\delta ^{(4)}(l),`$
$`k/2,S_1,\pm q^{L_0+\overline{L}_0\frac{c}{12}}k/2,S_2,`$
$`={\displaystyle \underset{[l,m,s]}{}}\chi _{(l,m,s)}(\stackrel{~}{q})\delta ^{(2)}(S_2+sS_1)\delta ^{(4)}(l+2).`$
In this case there are $`k/4+1`$ topological states in the open string spectrum of either of these two branes, and $`k/4`$ topological states in the open string between the two different branes.
## 4 The matrix factorisation description
We now want to describe the matrix factorisations that correspond to the above boundary states. As for the case of the A-type minimal model, the two different GSO-projections correspond to two different superpotentials. We shall now discuss them in turn.
### 4.1 The first D-model
The theory whose boundary states we discussed in section 3.1 and 3.2 corresponds to the superpotential
$$W_D=x^{n+1}xy^2,$$
(4.1)
where $`n=k/2`$. Its matrix factorisations have partially been studied in , but, as we shall see, the list of factorisations given there is incomplete. If we include another class of factorisations, we obtain perfect agreement with the conformal field theory results described above. Given that these boundary states generate all boundary states of the superconformal field theory, we can turn the argument around and conclude that we have identified all fundamental factorisations for this potential, i.e. that any factorisation is equivalent to a direct sum of these fundamental factorisations.
#### 4.1.1 Rank 1 factorisations
We start with a discussion of the rank 1 factorisations, all of which have been discussed in . The most obvious factorisation is
$$_0:d_0=x^ny^2,d_1=x,$$
(4.2)
as well as its reverse. The bosonic spectrum of either factorisation consists of two states, whereas the fermionic spectrum is empty. The only boundary states with two bosons and no fermions in their open string spectrum are the $`L=0`$ and $`L=k`$ boundary states of (3.5) and (3.14), respectively. They must therefore correspond to the above factorisation (and its reverse).
Generically this is the only rank 1 factorisation, but if $`n`$ is even, there are in addition the two resolved rank 1 factorisations, given by
$`_+`$ $`:`$ $`d_0=(x^{n/2}+y),d_1=x(x^{n/2}y),`$
$`_{}`$ $`:`$ $`d_0=x(x^{n/2}+y),d_1=(x^{n/2}y),`$
as well as their reverses. Either of these factorisations has a purely bosonic spectrum consisting of $`n/2+1`$ states, and there are $`n/2`$ fermionic operators propagating between $`_+`$ and $`_{}`$. Given that $`n/2=k/4`$, this matches precisely with the spectrum of the resolved branes described in section 3.2.1 that also only exist provided that $`k/2`$ is even.
#### 4.1.2 Rank 2 factorisations
Next we turn to the rank $`2`$ factorisations. In the factorisations of the form
$$d_0=\left(\begin{array}{cc}x^l& \alpha \\ \beta & x^{n+1l}\end{array}\right),d_1=\left(\begin{array}{cc}x^{n+1l}& \alpha \\ \beta & x^l\end{array}\right),\alpha \beta =xy^2$$
(4.3)
were considered. Exchanging $`l`$ with $`n+1l`$ amounts to exchanging $`d_0`$ and $`d_1`$, and hence to considering the reverse factorisation. The same holds for the exchange of $`\alpha `$ and $`\beta `$. The only two inequivalent choices for $`\alpha ,\beta `$ are therefore $`\beta =x`$ or $`\beta =y`$. As was shown in the choice $`\beta =x`$ leads to a class of factorisations that is equivalent to the direct sum of the factorisation (4.2) and its reverse. This leaves us with the factorisations
$$𝒮_l:d_0=\left(\begin{array}{cc}x^l& xy\\ y& x^{n+1l}\end{array}\right)d_1=\left(\begin{array}{cc}x^{n+1l}& xy\\ y& x^l\end{array}\right),$$
(4.4)
where $`l=1,\mathrm{},n`$. One easily finds that the factorisation $`𝒮_l`$ has $`2l`$ bosons and $`2l`$ fermions. Comparing with the conformal field theory analysis, one would then like to identify the factorisation $`𝒮_l`$ with the boundary state (3.5) or (3.14) with $`L=2l1`$. Note that $`ln+1l`$, which maps the factorisation to the reverse factorisation, then corresponds to $`L=2l1kL`$, which maps the brane to the anti-brane. \[If $`k/2`$ is even, the anti-brane of $`||L,M,0`$ is $`||kL,M,2`$, while for $`k/2`$ odd it is $`||kL,M,0`$.\]
As is clear from the conformal field theory analysis, the above factorisations do not yet account for all branes since there are also the boundary states (3.5) or (3.14) with $`L`$ even. The corresponding class of factorisations can be directly obtained from the factorisations (2.21) of the A-type minimal model with potential $`W_A^{}=x^ny^2`$: since $`W_D=x(x^ny^2)=xW_A^{}`$, one obtains a factorisation for the D-model by multiplying the matrix $`d_1`$ of any A-model factorisation by $`x`$. This leads to the following factorisations
$$𝒯_l:d_0^D=\left(\begin{array}{cc}x^l& y\\ y& x^{nl}\end{array}\right)=d_0^A^{},d_1^D=\left(\begin{array}{cc}x^{nl+1}& xy\\ xy& x^{l+1}\end{array}\right)=xd_1^A^{}.$$
(4.5)
Here $`l=0,\mathrm{},n`$, and the factorisation $`𝒯_l`$ is equivalent to $`𝒯_{nl}`$, but neither is equivalent to $`𝒯_l^r`$. We now propose that the factorisation $`𝒯_l`$ corresponds to the boundary state (3.5) or (3.14) with $`L=2l`$. To confirm this proposal, we have to determine the spectrum of these factorisations.
The calculation of the fermionic part of the spectrum proceeds exactly as in the case of the A-model. For this, consider a fermion described by its two components $`(t_0,t_1)`$. BRST-invariance means that
$$t_0d_1^D=d_0^Dt_1,$$
(4.6)
which can be solved by
$$t_1=\frac{1}{W_D}d_1^Dt_0d_1^D.$$
(4.7)
This requires, of course, that the entries of the matrix appearing on the right hand side are all divisible by $`W_D`$. In terms of the A-model data, this equation can be rewritten as
$$t_1=x\frac{x}{W_D}d_1^A^{}t_0d_1^A^{}=x\frac{1}{W_A^{}}d_1^A^{}t_0d_1^A^{},$$
(4.8)
where $`W_D/x`$ is the A-model potential $`W_A^{}`$. Without the additional factor of $`x`$ on the right hand side, this condition is simply the A-model result, implying a divisibility condition for the entries of the matrix $`d_1^A^{}t_0d_1^A^{}`$. Since $`x`$ does not divide $`W_A^{}`$ the divisibility conditions of the A- and D-model are therefore equivalent.
Any solution for $`t_0`$ is BRST-trivial if
$$t_0=\varphi _1d_0^D+d_0\varphi _0^D.$$
(4.9)
Since $`d_0^D=d_0^A^{}`$ it is obvious that the fermionic part of the spectrum for the A- and D-model is the same for this class of factorisations. Thus there are $`2l`$ fermions for the factorisation $`𝒯_l`$.
Let us now turn to the bosonic spectrum. The condition for an operator $`(\varphi _0,\varphi _1)`$ to be BRST invariant is
$$d_1^D\varphi _1=\varphi _0d_1^D,$$
(4.10)
which is the same condition as in the A-model. Solving for $`\varphi _0`$, one obtains the divisibility condition
$$\varphi _0^D=\frac{1}{W_D}d_1^D\varphi _1d_0^D=\frac{1}{W_A^{}}d_1^A^{}\varphi _1d_0^A^{},$$
(4.11)
which is explicitly (since $`𝒯_l𝒯_{nl}`$ we may assume that $`ln/2`$)
$$y(\varphi _1^{11}\varphi _1^{22})+x^l\varphi _1^{21}x^{nl}\varphi _1^{12}=0\mathrm{mod}W_A^{},$$
(4.12)
where $`\varphi _1^{ij}`$ denote the matrix elements of the $`2\times 2`$ matrix $`\varphi _1`$. The possible solutions for $`\varphi _1`$ are further restricted by the requirement that the operator is BRST non-trivial. BRST trivial operators can be written in the form (see (2.6))
$`\varphi _1^{11}`$ $`=`$ $`t_0^{11}x^{nl+1}t_0^{12}xy+x^lt_1^{11}+yt_1^{21}`$ (4.13)
$`\varphi _1^{12}`$ $`=`$ $`t_0^{11}xyt_0^{12}x^{l+1}+x^lt_1^{12}+yt_1^{22}`$ (4.14)
$`\varphi _1^{21}`$ $`=`$ $`t_0^{21}x^{nl+1}t_0^{22}xyyt_1^{11}x^{nl}t_1^{21}`$ (4.15)
$`\varphi _1^{22}`$ $`=`$ $`t_0^{21}xyt_0^{22}x^{l+1}yt_1^{12}x^{nl}t_1^{22}.`$ (4.16)
One can now use the freedom in $`t_1`$ to restrict the powers of $`x`$ appearing in the representatives of the BRST cohomology classes to lie in the range $`0,\mathrm{},l1`$ (for $`\varphi _1^{11}`$ and $`\varphi _1^{12}`$)) and $`0,\mathrm{},nl1`$ (for $`\varphi _1^{21}`$ and $`\varphi _1^{22}`$)). On the other hand, this then uses up the freedom described by $`t_1`$, and in particular, the $`y`$-dependence of the solution cannot be eliminated any longer. Indeed, in addition to the obvious $`x`$-dependent solutions (that are as for the A-model, see (2.22))
$$a_i=\left(\begin{array}{cc}x^i& 0\\ 0& x^i\end{array}\right),a_{l+i}=\left(\begin{array}{cc}0& x^i\\ x^{n2l+i}& 0\end{array}\right),i=0,\mathrm{},l1,$$
(4.17)
there are now the following two non-trivial solutions (for $`\varphi _1`$)
$$b_1=\left(\begin{array}{cc}y& 0\\ 0& y\end{array}\right),\text{and}b_2=\left(\begin{array}{cc}0& y\\ yx^{n2l}& 0\end{array}\right).$$
(4.18)
This means, in particular, that there are two more bosons than fermions for each factorisation $`𝒯_l`$. This is then in perfect agreement with the conformal field theory spectra (3.8) and (3.9) for the branes with $`L=2l`$.
In summary, we therefore propose the identifications:
$$\begin{array}{ccc}& & \\ \mathrm{Boundary}\mathrm{state}\hfill & \mathrm{Matrix}\mathrm{factorisation}& \\ & & \\ & & \\ ||0,M,S\hfill & _0,_0^r& \\ & & \\ ||2l1,M,S\hfill & 𝒮_l,𝒮_l^r& l=1,\mathrm{},[(n+1)/2]\hfill \\ & & \\ ||2l,M,S\hfill & 𝒯_l,𝒯_l^r& l=0,\mathrm{},[n/2]\hfill \\ & & \\ ||k/2,M,S,\pm \hfill & _\pm ,_\pm ^r& n\mathrm{even}\hfill \end{array}$$
As a consistency check we note that we have now identified the boundary state with $`L=0`$ with two factorisations, namely with $`_0`$ and with $`𝒯_0`$. Thus we need to have that these two factorisations are actually equivalent (in the sense of (2.10)), and this is indeed easily checked. Furthermore, for $`k/2=n`$ even, the boundary state $`||k/2,M,S`$ is not fundamental, but can be resolved as explained in section 3.2.1. We have identified these resolved branes with the factorisations $`_\pm `$; thus we expect that for $`n`$ even the factorisation $`𝒯_{n/2}`$ is equivalent to the direct sum of $`_+`$ and $`_{}`$, and this is also straightforwardly confirmed.
Another consistency check concerns the various flows of D-brane configurations. By switching on suitable tachyons, one can show that there are flows from the point of view of the matrix factorisation (such flows were first discussed, for the case of the A-model, in )
$`_0_0^r`$ $``$ $`𝒮_1`$
$`𝒮_l_0`$ $``$ $`𝒯_l`$
$`𝒯_l_0^r`$ $``$ $`𝒮_{l+1}.`$
These flows are easily seen to be compatible with the RR-charges of the corresponding boundary states. They are also in agreement with what one expects based on the analysis of . Finally, we note that the $`L=0`$-brane (together with the two resolved branes if $`n`$ is even) generates all D-brane charges since all D-branes (except for the resolved branes) can be obtained from it by the above flows. The corresponding matrix factorisations of these three D-branes all have rank $`1`$. This is similar to the situation for the tensor product of two A-models where the permutation branes (whose factorisations also have rank $`1`$) generate all the charges .
### 4.2 The other D-model
Finally we consider the superpotential
$$W_D^{}=x^{n+1}xy^2z^2$$
(4.19)
that should correspond, following the logic of , to the D-model with the other GSO-projection that was discussed in section 3.3. The easiest class of factorisations of this model are of tensor product type: for any factorisation of $`W_D`$ one obtains a factorisation of $`W_D^{}`$, where one factorises $`z^2`$ as $`z^2=zz`$. Since the spectrum on a brane with superpotential $`W=z^2`$ consists of the identity $`\varphi _0=\varphi _1=1`$ and a fermion $`t_0=t_1=1`$, it is very simple to calculate the spectrum of these factorisations. The number of bosons is simply given by the sum of the number of bosons and fermions of the corresponding factorisation of the D-model $`W_D`$. The same holds for the number of fermions. In particular, the number of bosons always equals the number of fermions for any factorisation of this type.
We now propose that we can identify the boundary states (3.29) with these tensor product factorisations. As is explained in section 3.3, each of these branes has $`2(L+1)`$ topological states in its open string spectrum (irrespective of whether $`L`$ is even or odd), and the same is true for the open string between the brane and its anti-brane. This then matches precisely with the above spectrum of the corresponding tensor product factorisations, given our previous identifications of the boundary states with the matrix factorisations for the $`W_D`$ model: in particular, the sum of the number of bosons and fermions in (3.8) and (3.9) is precisely equal to $`2(L+1)`$ for all $`L`$.
The analysis works similarly for the two resolved branes $`||k/2,S,\pm `$ for $`k/2`$ even. This leaves us with identifying the resolved brane $`||k/2,S,\pm `$ for $`n=k/2`$ odd, which does not come from such a tensor factorisation. For $`n`$ odd there are however additional factorisations of the form
$$d_0=\left(\begin{array}{cc}x^{\frac{n+1}{2}}z& \alpha \\ \beta & (x^{\frac{n+1}{2}}+z)\end{array}\right),d_1=\left(\begin{array}{cc}x^{\frac{n+1}{2}}+z& \alpha \\ \beta & (x^{\frac{n+1}{2}}z)\end{array}\right),\alpha \beta =xy^2.$$
(4.20)
Exchanging $`\alpha `$ and $`\beta `$ maps the factorisation to its reverse, and thus effectively the only inequivalent choices are $`\beta =x`$ and $`\beta =y`$. As before for the case of the D-model, one may expect that $`\beta =x`$ is in some sense trivial, and indeed one can show that this factorisation is equivalent to the tensor product factorisation corresponding to the D-model factorisation $`_0`$,
$$d_0=\left(\begin{array}{cc}x^ny^2& z\\ z& x\end{array}\right),d_1=\left(\begin{array}{cc}x& z\\ z& (x^ny^2)\end{array}\right).$$
(4.21)
On the other hand, the topological spectrum of the factorisation (4.20) with $`\beta =y`$ has $`(n+1)/2`$ bosonic and fermionic states, which thus agrees with the result of section 3.3.
We have therefore also managed to identify the boundary states of this D-model with the factorisations of the superpotential $`W_D^{}`$. Since we know that the boundary states we have described generate all possible boundary states, the same must be true for the above factorisations.
## 5 Matrix factorisations and singularity theory
Matrix factorisations are also a well-known tool in the theory of singularities in mathematics. In the case of complex dimension two, the simple singularities are known to have an ADE classification. The relevant singularities are described by hypersurfaces in $`^3`$ characterised by the equations<sup>§</sup><sup>§</sup>§By a change of variables one can also rewrite the first equation as $`W=x^ny^2z^2=0`$.
$$\begin{array}{ccc}\hfill A_{n1}:& \hfill W=x^nyz=& 0\hfill \\ \hfill D_{n+2}:& \hfill W=x^{n+1}xy^2z^2=& 0\hfill \\ \hfill E_6:& \hfill W=x^4+y^3+z^2=& 0\hfill \\ \hfill E_7:& \hfill W=yx^3+y^3+z^2=& 0\hfill \\ \hfill E_8:& \hfill W=x^5+y^3+z^2=& 0.\hfill \end{array}$$
(5.1)
The resolution of the singularities is obtained by blow-ups, where each node of the associated ADE Dynkin diagram corresponds to an exceptional divisor. For the case of these simple surface singularities, the explicit description of the blow-up can be encoded in a matrix factorisation of the polynomial $`W`$ defining the hypersurface .
D-branes on such singular geometries have recently been discussed in . There it was proposed to define a category $`D_{Sg}`$ that is supposed to capture the topological sector of B-type branes on singular spaces. On smooth manifolds, any coherent sheaf has a finite resolution of locally free sheaves of finite type. On singular spaces this is no longer the case and Orlov defined $`D_{Sg}`$ to be the quotient of the bounded derived category of coherent sheaves modulo those sheaves that have such a finite resolution. It is shown in that this category is equivalent to the category of matrix factorisations; in particular, any object in $`D_{Sg}`$ corresponds to a matrix factorisation, and the open string spectrum that is described in $`D_{Sg}`$ in terms of morphisms of modules describing the branes, corresponds exactly to the BRST invariant spectrum of the Landau Ginzburg theory. More precisely, consider a Landau Ginzburg potential $`W:^n`$ with an isolated critical point at the origin. \[In our case the relevant LG potentials are $`W_{ADE}:^3`$, and thus describe precisely the singular hupersurfaces of (5.1).\] Denoting the fiber of $`W`$ over $`0`$ by $`S_0`$, allows us to establish a relation between $`D_{Sg}(S_0)`$ and the Landau-Ginzburg category. For this, we associate to any factorisation $`W=d_0d_1`$
$$\left(\text{}\right)$$
the short exact sequence
$$0P_1\stackrel{d_1}{}P_0\mathrm{Coker}d_10.$$
(5.2)
The geometrical object associated to the factorisation is then the sheaf $`\mathrm{Coker}d_1`$, which, since it is annihilated by $`W`$, is a sheaf on $`S_0`$. For further details on this functor and mathematical proofs, we refer to .
### 5.1 Singular geometry vs. Landau Ginzburg model
As we have just seen, the very same matrix factorisations that describe B-type D-branes in the Landau Ginzburg theory with superpotential $`W`$ also characterise the B-type D-branes of the singular hypersurface $`W=0`$ and its possible resolutions. Despite their formal similarities, these two description are however rather different: string theory on the singular geometry is, in sigma-model language, a singular limit of a theory with $`c=6`$ that has no well-behaved conformal field theory description. On the other hand, the Landau Ginzburg model flows to a perfectly well defined conformal field theory, namely the $`N=2`$ minimal model with $`c=3k/(k+2)`$.
From the closed string point of view the relation between singular geometries and $`N=2`$ conformal field theories has been studied before in . In particular, characteristic properties of the Landau Ginzburg model, such as the central charge and the chiral ring, have been compared with concepts appearing in singularity theory (singularity index, local ring of $`W`$). The agreement of the matrix factorisations thus provides a natural extension of this correspondence to the open string sector. Since the matrix factorisations determine (the topological part of the) open string spectrum, this may even suggest that the B-type D-branes of the Landau-Ginzburg model (or of the $`N=2`$ minimal model) provide a worldsheet description of the D-branes that become massless in the singular geometry.
### 5.2 Singular geometry vs. the orbifold description
String theory compactified on a singular surface does not have a well-defined perturbation theory since at the singular point non-perturbative (D-brane) states become massless. However, it is possible to give a well-defined conformal field theory description for a closely related background where a non-zero $`B`$-field has been switched on for all singular cycles. (This $`B`$-field prevents the D-branes from becoming massless in the singular limit, and therefore avoids the breakdown of string perturbation theory .) The relevant conformal field theory is the orbifold theory $`^2/\mathrm{\Gamma }`$, where $`\mathrm{\Gamma }`$ is a finite subgroup of $`SU(2)`$. Orbifold theories are well behaved, and it is known how to describe their D-branes. In particular, following , the different D-branes are in essence characterised by the representation of the orbifold group $`\mathrm{\Gamma }`$ that acts on the corresponding Chan-Paton indices. The charges are then generated by the branes that are associated to the non-trivial irreducible representations of $`\mathrm{\Gamma }`$. If we associate to each such representation a node of a (Dynkin) diagram, and connect nodes if the open string between the corresponding D-branes has a (massless) hypermultiplet in its spectrum, we recover precisely again the corresponding ADE Dynkin diagrams. Furthermore, the dimension of the irreducible representation of $`\mathrm{\Gamma }`$ equals the Kac-label of the corresponding node. This suggests, in particular, that the orbifold description captures at least some of the structure of the geometry described by $`W=0`$.
Since the fundamental D-branes of the orbifold theory also give rise to the same ADE Dynkin diagram, they should be in natural one-to-one correspondence with the matrix factorisations of the LG potential.However, the open string spectra should not (and do not) agree: because of the resolution certain massless open string states of the Landau Ginzburg theory become massive in the orbifold theory. In fact this relation can be understood fairly directly. As described in following the ideas of , for each irreducible representation of $`\mathrm{\Gamma }`$ one can find a module of the $`\mathrm{\Gamma }`$-invariant part of $`[X,Y]`$. Unless the representation of $`\mathrm{\Gamma }`$ is trivial, these modules are not free, and the relations among the generators of the module can be described by a matrix. This is then precisely the matrix that appears as one factor of the matrix factorisation. For the two cases of interest in this paper, this can be done explicitly as follows.
#### 5.2.1 A-type singularity
In the case of A-type singularities, the generator of the orbifold group acts on $`^2`$ as
$$gX=\xi X,gY=\xi ^1Y,\xi =e^{\frac{2\pi i}{n}}.$$
(5.3)
The singular surface can be described by the polynomial subring in the two variables $`X,Y`$ that is invariant under the orbifold action. This invariant subring is generated by $`x=XY`$, $`y=Y^n`$ and $`z=X^n`$. These polynomials are however not independent, but satisfy the equation $`x^nyz=0`$, which is just the singular hypersurface equation of the $`A`$-type singularity.
The orbifold group $`\mathrm{\Gamma }`$ is in this case $`_n`$, which has $`n1`$ irreducible non-trivial representations. All of these representations are one-dimensional: for $`l=1,\mathrm{},n1`$ the corresponding representation associates to the generating element $`g\mathrm{\Gamma }`$ the phase $`e^{\frac{2\pi il}{n}}`$.
These $`n1`$ representations are then in one-to-one correspondence to the matrix factorisations of the Landau-Ginzburg potential (2.13), or equivalently, the boundary states of the minimal model (2.15), whose label take values $`L=l1=0,\mathrm{},k=n2`$.The superpotential $`W=x^nyz`$ (that is equivalent by a change of variables to $`W=x^ny^2z^2`$) differs from $`W_A=x^n`$ by changing the GSO-projection twice. One would therefore expect that these two superpotentials are equivalent. From the point of view of singularity theory this equivalence is known as Knörrer periodicity . As explained in we can associate to each representation of $`\mathrm{\Gamma }`$ a module for the invariant part of the polynomial ring in two variables. For example, for the representation labelled by $`l`$ the corresponding module is generated by $`X^l`$ and $`Y^{nl}`$. This is not a free module since we have the relations between the generators $`s_1=Y^{nl}`$, $`s_2=X^l`$:
$$x^ls_1ys_2=0,zs_1+x^{nl}s_2=0.$$
(5.4)
The matrix of relations
$$d_1=\left(\begin{array}{cc}x^l& y\\ z& x^{nl}\end{array}\right)$$
(5.5)
is then one of the matrices appearing in the matrix factorisation. Conversely, we can apply the functor of Orlov and recover the module (and thus the representation of $`\mathrm{\Gamma }`$) from the factorisation.
#### 5.2.2 D-type singularity
The D-type minimal model corresponds to the case where the orbifold is a binary dihedral group. Its two generators act on $`^2`$ as
$$g=\left(\begin{array}{cc}\beta & 0\\ 0& \beta ^1\end{array}\right),h=\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right),\mathrm{where}\beta =e^{\frac{\pi i}{n}}$$
(5.6)
and satisfy the relations
$$g^{2n}=\mathrm{𝟏},h^2=g^n,hgh^1=g^1.$$
(5.7)
The ring of invariant polynomials under the group is generated by
$$x=(XY)^2,y=\frac{1}{2}(X^{2n}+Y^{2n}),z=\frac{i}{2}(XY)(X^{2n}Y^{2n}),$$
(5.8)
which satisfy the D-type hypersurface equation $`x^{n+1}xy^2z^2=0`$. Generically, the irreducible representations of this group are two-dimensional, with
$$\rho ^{(l)}(g)=\left(\begin{array}{cc}\beta ^l& 0\\ 0& \beta ^l\end{array}\right)$$
(5.9)
and
$$\rho ^{(l)}(h)=\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right)(l\mathrm{odd}),\rho ^{(l)}(h)=\left(\begin{array}{cc}0& i\\ i& 0\end{array}\right)(l\mathrm{even}).$$
(5.10)
Here $`l=0,1,\mathrm{},2n1`$, and the representations $`\rho ^{(l)}`$ and $`\rho ^{(2nl)}`$ are equivalent. Furthermore, the representations $`\rho ^{(0)}`$ and $`\rho ^{(n)}`$ are not irreducible but can be further decomposed into one-dimensional representations: $`\rho ^{(0)}`$ contains the trivial representation and the one where $`g1,h1`$. Likewise, $`\rho ^{(n)}`$ can be decomposed into the two one-dimensional irreducible representations $`g1`$ and $`h\pm 1`$ ($`l`$ even) or $`h\pm i`$ ($`l`$ odd).
These representations are exactly in one-to-one correspondence with the boundary states of section 3.3, or equivalently the matrix factorisations of section 4.2. More explicitly, the representation $`\rho ^{(l)}`$ corresponds precisely to the boundary state with $`L=l`$. Since $`n=k/2`$, the reducible representation $`\rho ^{(n)}`$ corresponds then to the non-fundamental brane with $`L=k/2`$ which can be resolved into two boundary states (that correspond in turn to the two one-dimensional representations with $`g1`$). Also the identification $`\rho ^{(l)}\rho ^{(2nl)}`$ mirrors the equivalence of boundary states (3.30). As in the A-case, one can also relate these representations to certain modules of the invariant algebra $`[X,Y]^\mathrm{\Gamma }`$, and obtain the corresponding matrix factorisations in this manner (see ). It is maybe remarkable that the blow-ups that correspond to the one-dimensional representations of $`\mathrm{\Gamma }`$ are quite different to those that are associated to the two-dimensional representations.
For completeness we should also mention yet another class of models with $`c=6`$ describing the motion of strings on ALE spaces. In this approach (which is due to ), the minimal models are tensored with another theory that contributes the missing central charge to get $`c=6`$. To be more precise, consider the tensor product of two coset theories $`SL(2,)/U(1)\times SU(2)/U(1)`$, and orbifold by an appropriate discrete group to impose the charge integrality condition. In terms of Landau-Ginzburg potentials, these models can be written as
$$\begin{array}{ccc}\hfill A_{n1}:& \hfill W=\mu t^n+x^nyz=& 0\hfill \\ \hfill D_{n/2+1}:& \hfill W=\mu t^n+x^{n/2}xy^2z^2=& 0,\hfill \end{array}$$
(5.11)
which are again subject to an integer charge projection. As argued in , the value of the $`B`$-field in these models is $`0`$, but now $`\mu 0`$, and we are therefore discussing the resolved geometry. The D-branes of this model have been investigated in . If one wants to consider B-type D-branes in these models, one can effectively reduce the discussion to A-type D-branes in the minimal model part. The reason is that the orbifold procedure (followed by the mirror map) maps A-type to B-type branes, and, as noticed in , the contribution of the non-compact coset can be captured by universal factors. Since this description involves effectively the A-type D-branes of the minimal model, the relation to matrix factorisations and branes in the singular geometry is however less evident.
Acknowledgements
This research has been partially supported by a TH-grant from ETH Zurich, the Swiss National Science Foundation and the Marie Curie network ‘Constituents, Fundamental Forces and Symmetries of the Universe’ (MRTN-CT-2004-005104). We thank Stefan Fredenhagen, Wolfgang Lerche and Sebastiano Rossi for useful discussions.
|
warning/0506/cond-mat0506655.html
|
ar5iv
|
text
|
# Lattice-Boltzmmann simulations of the sedimentation of charged disks
## I Introduction
Suspensions of charged disks are of great practical importance. Examples range from clay suspensions to blood. In the present paper, we present calculations of the electrokinetic behavior of charged disks. As disks are not spherically symmetric, they also provide an ideal model system to study the effect of shape on the coupling between electrostatic and hydrodynamic response of a macroscopic particle.
There exists an extensive experimental literature on the transport properties of suspensions of charged disks. Yet, in spite of the importance of these systems, there is surprisingly little theoretical knowledge about the effect of the charge of the disks on their transport properties. One reason may be that the non-spherical geometry greatly complicates the use of the analytical approach that is used to describe charged, spherical particles. Whilst there are papers that consider the hydrodynamical properties of uncharged disks or the electrostatic properties of charged disks , we are not aware of any theoretical publications that treat the interplay between electrostatics and hydrodynamics for charged disks.
Several theoretical studies suggest that, in general, there may be non-trivial coupling effects due to shape asymmetries . The need for a numerical (rather than an analytical) approach is related to the fact that the analytical approaches are usually only tractable in certain limits. For example, to arrive at tractable analytical expressions it is often necessary to assume that the Poisson-Boltzmann equation can be linearized or that the Debye screening length is small compared to the linear dimensions of the charged colloid. Yet these conditions are rarely satisfied for charged biocolloids and biopolymers. These (non-spherical) particles usually have high charges under physiological conditions making the use of the linearized Poisson-Boltzmann equation questionable.
The total number of ionizable groups of colloids can be determined experimentally by titration. However, the experimental determination of the effective charge of colloids usually exploits non-equilibrium techniques, such as electrophoresis or sedimentation. The interpretation of many of these experiments is based on theoretical expressions that have been derived for weakly charged, spherical colloids. Yet, it is not at all obvious that expressions valid for spheres can be extrapolated to disks. Hence, it is important to understand the effect of shape and charge on the transport behavior of non-spherical charged particles.
In the present paper we report a numerical study of the electrokinetics of charged disks. In order to perform these simulations, we extended a Lattice-Boltzmann scheme that we had developed to study the electrokinetic behavior of spherical colloids. With this simulation method, we can study the dynamics of charged particles at the Poisson-Boltzmann level. Hence our simulations ignore charge fluctuations and static charge correlation effects. In the absence of polyvalent ions, the Poisson-Boltzmann description is expected to work quite well.
Our simulations extend from the regime where the electrohydrodynamic coupling is small (low charge, thin double layer) to the case where the charge is large and the double layer is extended. By performing simulations of many different conditions, we disentangle the effects of charge, double layer width and shape.
The paper is organized as follows. In Sec. II we present the basic electrokinetic equations and the simulation method that we use to solve them. In order to study the effect of charge on the sedimentation of disks, we must first know the sedimentation behavior of uncharged disks. This is done in Sec. III, where we use our model to compute the friction coefficients of a neutral sedimenting disk for motions parallel to and perpendicular to its axis of symmetry (we refer to these motions as “longitudinal” and “transverse”, respectively). In Sec. IV we discuss the effect of the charge of the disk on the sedimentation velocity. In Sec. V we consider the effect of the concentration of the disks on the sedimentation velocity. Sec. VI focuses on the variation of the sedimentation velocity with Debye-screening length. Sec. VII contains a comparison of the sedimentation behavior of disks and spheres. Concluding remarks are contained in Sec. VIII.
## II Electrokinetic model
We analyze a simple geometry in which one disk-like colloidal particle with radius $`a`$ and height $`h`$ sediments due to the action of a uniform external field. The disk has an overall charge $`Q=Ze`$, where $`Z`$ is the valency and $`e`$ is the elementary charge unit. The aspect ratio $`p`$ is defined as $`p=2a/h.`$ The disk is suspended in a symmetric electrolyte and, for the sake of simplicity, we assume that coions and counterions have the same mobility. The fluid mixture is characterized locally by the solvent density, $`\rho _s`$, and by the microion electrolyte densities, $`\rho _\pm .`$ The latter also determine the local charge density of the fluid, $`q(\text{r})=ze[\rho _+(\text{r})\rho _{}(\text{r})]`$. We restrict ourselves to monovalent electrolytes, i.e. $`z=1`$.
On a macroscopic length scale, the dynamics of the system is governed by the standard electrokinetic equations that specify the interplay between the electrical potential, local charge density, electrical currents and fluid flow:
$`{\displaystyle \frac{}{t}}\rho _k`$ $`=`$ $`𝐣_kk=+,`$ (1)
$`{\displaystyle \frac{d}{dt}}\left(\rho \text{v}\right)`$ $`=`$ $`\eta ^2\left(\rho \text{v}\right)P+{\displaystyle \frac{k_B\text{T}}{e}}q\mathrm{\Phi }.`$ (2)
$$𝐣_k=\rho _k\text{v}+D_k\left[\rho _k+z_k\rho _k\mathrm{\Phi }\right]$$
(3)
where $`\eta `$ is the shear viscosity, $`P`$ is the pressure, $`𝐯`$ is the fluid velocity, $`k_B\text{T}\beta ^1`$ measures the temperature, and $`D_k`$ stands for the diffusivity of each electrolyte species (which reduce to a single constant for symmetric electrolytes). $`\widehat{\mathrm{\Phi }}`$ is the electrostatic potential, while $`\mathrm{\Phi }\widehat{\mathrm{\Phi }}(k_BT/e)`$ is an appropriate dimensionless potential which satisfies the Poisson equation
$$^2\mathrm{\Phi }=4\pi l_B\left[\underset{k=\pm }{}z_k\rho _k+\rho _w\right],$$
(4)
where $`l_B=\beta e^2/(4\pi ϵ)`$ is the Bjerrum length, $`ϵ=ϵ_0ϵ_r`$ denotes the dielectric constant of the medium, whilst $`\rho _w`$ refers to the charge density due to embedded solid objects, either colloids of solid walls.
Equation (1) simply expresses a conservation law and Eqn. (3) is the constitutive equation. Together with the incompressibility condition $`𝐯=0`$, Eqn. (2) corresponds to the Navier-Stokes equations for an incompressible, isothermal electrolyte. In the presence of external forces (such as the gravitational field), the corresponding force must be added to the right-hand side of Eqn. (2).
### II.1 Simulation method
To simulate the sedimentation of charged disks, we used the lattice-Boltzmann scheme reported in Ref. . We showed therein that lattice-Boltzmann can be used to compute transport coefficients of charged spherical colloids. Below, we briefly summarize the main features of the method and refer the reader to that reference for further details.
The Lattice-Boltzmann (LB) method is the lattice counterpart of the Boltzmann equation. It prescribes a dynamical evolution rule for the distribution function $`n_i(𝐫,t)`$, which represents the density of particles at the lattice node $`𝐫,`$ at the discrete time $`t`$ and with the discrete velocity $`𝐜_i`$. The density-weighted moments of the local velocity distribution correspond to the hydrodynamic fields. In particular, $`_i\text{c}_in_i(\text{r},t)=\rho 𝐯`$ is the fluid’s momentum; it satisfies the Navier-Stokes equation on length and time scales that are large compared to the lattice spacing and the LB time step, respectively.
The electrolyte species are simulated by following the diffusion and convection of the local densities of coions and counterions described by Eqns. (1) and (3). This equation is based on the flux of each species along the links that connect neighboring nodes, and ensures strict local charge conservation. This local charge, combined with the corresponding electrostatic potential—computed by a numerical solution of the Poisson equation (4)—provides the local force that accelerates the fluid. With this technique, colloidal particles are simply introduced as surfaces where the collision rules of the populations of the neighboring nodes are modified to ensure non-slip boundary conditions . The link-based definition of the flux of the electrolyte species leads to a straightforward implementation of the no-flux boundary condition for each of the ionic species at solid surfaces. This suppresses possible charge leakage through the solid walls.
For reasons of computational convenience, we choose the value of the kinematic viscosity $`\nu =1/6`$ (in lattice units) and the $`\rho _s=1`$, as the density unity. The external (gravitional) field that induces sedimentation was chosen to be $`10^6,`$ a value that is well inside the linear-response regime. This gravitational field generates fluid velocities of the order of $`10^8`$ (again, all expressed in lattice units). The diffusivity of the electrolyte is set to $`D=0.19`$, a value for which spurious diffusion due to lattice advection (see Ref. ) is negligible. The values of the diffusivity and the flow velocity correspond to Péclet numbers smaller than $`10^1.`$ In the simulations described in the subsequent sections we vary the salt concentration between $`7\times 10^4`$ and $`5\times 10^3`$ as a way to control the electrical double layer thickness.
## III Sedimentation of neutral disks
Before assessing the role of electrostatics on the sedimentation of non-spherical particles, we performed LB simulations to compute the sedimentation velocity of uncharged hard disks. Such reference calculations are needed because, in contrast to the case of hard spheres, analytic expressions for the sedimentation velocity of an isolated hard disk only exist in the limit of infinitely thin disks . We are not aware of analytical results for disks with finite aspect ratios.
We have simulated the sedimentation of disks with two different nominal aspect ratios $`p=10`$ and $`p=5`$, corresponding to disks of lateral dimension $`h=2,`$ and radii $`a=10`$ and $`5`$, respectively. However, these aspect ratios are only approximate: in the LB approach, the hydrodynamic boundary of a solid particle is usually located close to the midpoint of links joining fluid and solid nodes. In practice, the hydrodynamic shape of an object may differ slightly from the nominal one. For this reason, we need to calibrate the shapes of the disks. An unambiguous way to determine these effective sizes (the hydrodynamic radius and height) would be to measure the friction coefficients of the particles and compare the results with the appropriate analytical expression of an infinitely thin disk. Great care must be taken when comparing the LB numerical results for the friction coefficient with results obtained for an isolated disk. Since we use periodic boundary conditions, the simulations measure the friction coefficient of a regular array of particles at volume fraction $`\phi =\pi a^2h/L^3,`$ where $`L`$ is the diameter of the simulation box. Hence, in order to extrapolate to infinite dilution, we must perform a series of simulations with increasing box size. The same procedure will also be used to compute the sedimentation velocity of a charged disk in the dilute limit.
### III.1 Friction coefficients
In the following calculations, the reference frame is centered on the disk and the disk is fixed on the lattice. Hence, the gravitational force acting on the disk becomes a body force, with opposite direction, that acts to the fluid. We then determined the total fluid velocity at steady state, $`v_F`$. As this velocity is equal and opposite to the sedimentation velocity of the particle, $`U_d=v_F`$, we can relate the sedimentation velocity to the friction coefficient through $`U_d=F_g/\xi `$, where $`U_d`$ is the sedimentation velocity of the disk, $`\xi `$ is its friction coefficient, and $`F_g`$ is the applied gravitational force. We computed both the friction coefficient for motion along the symmetry axis of the disk and the one for perpendicular (transverse) motion.
Hashimoto has shown that the friction coefficient of an array of hard spheres depends, at low volume fractions $`\phi `$, as $`(\xi /\xi _0)^1=11.76\phi ^{1/3}+\phi +O(\phi ^2),`$ which shows that the initial decrease in the sedimentation velocity of an array of hard spheres is controlled by the colloid volume fraction.
In Figure 1 we show the transverse and longitudinal friction coefficients for a neutral disk with $`p=5`$ at various volume fractions, normalized by the Stokes friction coefficient of a sphere with the same area, i.e. by $`\xi _A6\pi \eta R`$, $`R=\sqrt{a(a+h)/2}.`$ The figure shows that, just as in the case of neutral spheres, the friction varies linearly with $`\phi ^{1/3}`$. Knowledge of this concentration dependence allows us to extrapolate the numerical results to estimate the friction coefficients ($`\xi _{}`$ and $`\xi _{}`$) at infinite dilution. One limiting case is known: the friction coefficients of an infinitely thin disks ($`p\mathrm{}`$) is identical to that of an oblate spheroid with the same aspect ratios between its main axis .
We summarize the results for the normalized friction coefficients for disks of two aspect ratios in Table 1. For the disk with $`p=5`$, we conclude that, to a good approximation, the hydrodynamic radius and height correspond to the nominal ones. The numerical values obtained for the larger disk ($`p=10`$) are less satisfactory than for the shorter one. While one could conclude that, for a larger aspect ratio the hydrodynamic radius and height are different with respect to the nominal ones, and hence recompute the effective values for ”$`a`$” and $`h`$ so as to adjust $`\xi _A`$, this does not seem satisfactory, since for a larger object the disagreement between the nominal and hydrodynamic size is expected to decrease. Moreover, because we have always used cubic simulation boxes, we can attribute such a difference to the stronger coupling between image disks for longitudinal sedimentation. On the other hand, the ratio between the perpendicular and parallel frictions $`\xi _{}/\xi _{}`$ do agree with the value estimated on the basis of the nominal size, suggesting that the deviations come mostly from uncertainties related to the extrapolation from finite volume fraction values. Hence, we conclude that, also for this shape, the disagreement between the two sizes is negligible, and we ascribe the deviations to interactions with the periodic images.
## IV Sedimentation velocities of charged disks: charge dependence
Before discussing the computed sedimentation velocities of charged disks, we briefly recall the theoretical results concerning the sedimentation velocity of weakly charged spheres, since this theory serves as a reference point for the discussion of the results for disks. Booth (and Ohshima *et al.* ) predicted that the sedimentation velocity $`U_s(Z)`$, where Z is the valence, of an isolated sphere is a quadratic function of the sphere valence, which can be expressed as
$$\frac{U_s(Z)}{U_s(0)}=1c_2(\kappa R)Z^2,$$
(5)
where $`U_s(0)`$ corresponds to the sedimentation velocity of a hard sphere in the dilute limit. In the regime where the Debye-Hückel theory is valid, the pre-factor $`c_2`$ for a symmetric 1-1 electrolyte is given by
$$c_2(\kappa R)=\frac{k_BTl_B}{72\pi R^2\eta D}f(\kappa R),$$
(6)
where $`R`$ is the radius of the sphere and, with $`\rho _k^0`$ denoting the bulk value of the charge density of species $`k`$, $`\kappa =\sqrt{4\pi l_B_kz_k^2\rho _k^0}`$ stands for the inverse Debye length that characterizes the size of the electrical double layer. The function $`f(\kappa R)`$ is defined as
$`f(\kappa R)`$ $`=`$ $`{\displaystyle \frac{1}{1+(\kappa R)^2}}[\text{e}^{2\kappa R}(3E_4(\kappa R)5E_6(\kappa R))^2+8\text{e}^{\kappa R}(E_3(\kappa R)E_5(\kappa R))`$ (7)
$`\text{e}^{2\kappa R}(4E_3(2\kappa R)+3E_4(2\kappa R)7E_8(2\kappa R))],`$
expressed as a linear combination of the integral function, $`\text{E}_n(x)x^{n1}_x^{\mathrm{}}𝑑tt^n\mathrm{exp}(t)`$.
It is reasonable to assume that, in the case of disks, the dependence of the sedimentation velocity on colloidal charge in the Debye-Hückel limit has the same functional form as Eqn. (5), where all the shape dependence and hydrodynamic coupling enters through the factor $`c_2(\kappa R).`$ To test this, and to analyze the role of charge on the sedimentation velocity of disks, we performed a series of simulations at constant Debye screening length and volume fraction ($`\phi =7.2\times 10^4`$ for $`p=5`$, and $`\phi =2.9\times 10^3`$ for $`p=10`$). Since in many cases of practical interest colloidal particles are highly charged (see e.g for the case of disk-like clay particles) we performed numerical simulations covering a wide range of disk charges.
The results obtained are displayed in Fig. 2, where the velocity, normalized by the sedimentation velocity of a hard disk at the same volume fraction, is depicted as a function of the surface charge $`\sigma =eZ/\left[2\pi a(a+h)\right]`$. One can identify the quadratic dependence at low charge, consistent with Booth theory for spheres. Such a dependence can be clearly appreciated in the figure’s inset. After this quadratic growth, a crossover region is identified, for surfaces charge densities between $`0.1`$ and $`0.4,`$ before entering the asymptotic regime at even larger surface charge densities, where the sedimentation velocity increases much more slowly. This behavior is consistent with numerical results on the sedimentation velocity of charged colloidal spheres which also show a deviation from Booth’s predictions for surface charge densities around $`0.1.`$
Comparing the deviation of the sedimentation velocity for transverse and longitudinal motion, as displayed in Fig. 2, one can see that the sedimentation velocity decreases faster with charge for transverse motion, regardless of the aspect ratio. This is indicative of a stronger electrokinetic coupling for transverse motion. Interestingly, the sedimentation velocities (expressed by $`1U_d(Z)/U_d(0)`$) become almost independent of the aspect ratio of the disks for large surface-charge densities. In the same figure, we also plot the decrease in sedimentation velocity for a sphere of radius $`R=4.2`$, which shows the same dependence on surface charge as that of the disks.
Figure 2 allows us to draw some qualitative conclusions concerning the nature of the errors that are made when estimating the charge of disk-like particles by assuming the validity of Booth’s theory . As mentioned above, Booth’s theory is valid only for *weakly* charged sedimenting *spheres*. As we now have numerical results for the sedimentation velocity of disks, we can identify two sources of errors. At low charge, where the quadratic dependence of the sedimentation velocity on colloidal charge holds, there will still be some discrepancy in the coefficient $`c_2`$. We address this issue in more detail in Section VII. At high charge, where even the quadratic surface-charge dependence does not apply, the use the Booth theory leads to serious errors in the estimation of the zeta potential. To give an idea of the magnitude of this error, one should compare the curves shown in Fig. 2 with a parabolic extrapolation of the curves up to $`\sigma 0.1.`$ Our calculations also suggest that if the sedimentation velocity of a charged colloid is plotted as a function of the surface-charge density (and not as a function of the total charge as one might be tempted to do when comparing with Booth’s theory) the curves corresponding to the sedimentation velocity of a variety of charged disks show the same functional dependence.
## V Sedimentation velocities of charged disks: volume fraction dependence
As discussed above, the inverse friction coefficient of a dilute, ordered array of hard spheres scales as $`\phi ^{1/3}.`$ In Ref. , we have verified that this functional dependence also holds for charged spheres, provided that the system is dilute enough to guarantee that there is no significant overlap of the double layers. For charged disks, we expect the $`\phi ^{1/3}`$ dependence to hold under the same circumstances. In order to test whether there is a detectable effect of the overlap of electric double layers of different disks, we have computed the normalized friction coefficients for disks of aspect ratio $`p=5`$ as a function of the volume fraction, for volume fractions up to $`10\%,`$ for different widths of the diffuse layer.
In Fig. 3 we show the results for a weakly charged disk, both for transverse and longitudinal motion. Note that, in the dilute limit, the friction coefficients will depend on $`\kappa a`$ due to the electrohydrodynamic interaction. For transverse motion the convergence to the dilute limit is slower, indicating a stronger coupling between disks; we attribute this to the fact that the distance of closest approach coincides with the external field direction. Although for $`\kappa a=1/2`$ and high volume fractions the diffuse layers overlap, the effect of the diffuse layer is much weaker than the volume-fraction dependence or than the effect of the particle shape. The shape effect is reflected in the substantial difference between the friction coefficients for transverse and longitudinal sedimentation. Although barely visible in Figure 3, there is a small but significant dependence of the friction coefficients on $`\kappa `$. This we discuss in more detail in the next section.
The dependence on $`\kappa `$ is better visible in Fig. 4, where we display the friction coefficients of highly charged disks Although here the dependence on $`\kappa `$ is clearly visible, it becomes less important at higher volume fractions. This suggests that with increasing volume fraction, the effect of the overlap of diffuse layers becomes less important than the direct effect of hydrodynamic interaction between disks. At small volume fractions, we always recover the $`\phi ^{1/3}`$-dependence of the inverse friction coefficients. However, deviations form this scaling are already noticeable at small $`\phi `$. This fact indicates that the dilute regime is confined to smaller $`\phi `$ values for highly charged disks. In the limit $`\kappa a\mathrm{}`$ (vanishing Debye length), the sedimentation friction coefficient of a charged and a neutral disk should coincide. The fact that, for finite $`\kappa a`$, the ratio of the friction coefficients extrapolates at $`\phi =0`$ to a value different from one indicates the importance of electrokinetic coupling for the sedimentation of charged disks.
We stress that the values of $`\kappa `$ are computed on the basis of the electrolyte densities in the bulk. For concentrated suspensions of highly charged disks, the density of counterions added to the system to ensure charge neutrality ($`\rho _{}=Z/V_\text{f}`$, where $`Z`$ is the valency of the sphere and $`V_\text{f}`$ the volume occupied by the electrolyte) may exceed the concentration of added salt. In that case, $`\kappa ^1`$, the screening length in the salt “reservoir” is not simply related to the apparent screening length in the dense suspension. For our simulations this phenomenon becomes important only for the highest volume fractions (typically $`\phi >0.08`$). Hence, our extrapolation at low $`\phi `$ is not affected by this complication.
## VI Sedimentation velocity of charged disks: Debye length dependence
Having analyzed the role of charge and volume fraction on the sedimentation velocity, we now consider in more detail the effect of the double layer width on the sedimentation of the disks. We follow the same procedure as in Section III.1 and study the sedimentation velocity $`U_d(Z)`$ at different volume fractions, normalized by the corresponding velocity of isolated charged-neutral disks $`U_d(0)`$. In this way, the ratio $`U_d(Z)/U_d(0)`$ measures the reduction is the sedimentation velocity of one charged disk due to its electrokinetic interaction with the electrolyte. The reduction in sedimentation velocity in the dilute limit is interesting theoretically, because we can compare with analytic results for weakly charged spheres, although in experiments the reduced sedimentation velocity $`U_d(Z,\phi )/U_d(0,\phi )`$ at finite $`\phi `$ is the relevant quantity. For simplicity, in the remaining part of this section, we will be writing $`U_d(Z)`$ instead of $`U_d(Z,\phi )`$ but, unless explicitly stated, the volume fraction dependence is always assumed.
In Figure 5(a), we show the normalized sedimentation velocity as a function of the double layer width for a disk with aspect ratio $`p=5`$ in transverse motion, for different volume fractions. For infinitely thin and infinitely broad diffuse layers, the sedimentation velocity should coincide with that of a charged-neutral disk, and hence the curve should approach one for both small and large $`\kappa a`$, as is indeed observed. The decrease at intermediate values of $`\kappa a`$ is the result of the interplay between hydrodynamic dissipation and electrolyte diffusion. The largest effect is observed when the size of the double layer is of the order of the largest dimension of the disk, i.e. $`\kappa a1`$. The effect increases with decreasing volume fraction, consistent with the discussion in the previous section, and above volume fractions around $`1\%,`$ the changes in normalized sedimentation become negligible. The minimum velocity also depends on volume fraction, an effect which is consistent with previous findings for spheres . A similar behavior is observed in Fig. 5(b), where longitudinal sedimentation for a weakly charged disk is depicted. It is interesting to note that the decrease in sedimentation velocity is slightly smaller. We can ascribe this effect to the fact that the distorted double layer is not isotropic and has a smaller contribution to the friction when the wider side of the disk is exposed to a region where the velocity gradients are smaller.
In Figures 6(a) and 6(b), we show the sedimentation velocity for a weakly charged disk with a somewhat larger aspect ratio: $`p=10`$. The trends are the same as for the disk with $`p=5`$, although the minimum velocity seems to depend on aspect ratio, and is achieved now for slightly narrower double layers. The reduction in absolute terms is now smaller, but this is simply due to the lower surface charge density as compared with the smaller ($`p=5`$) disk. It is worth mentioning that for disks with $`p=5`$ we could effectively reach the dilute limit for $`U_d(Z)/U_d(0)`$. In contrast, for disks with $`p=10`$ we had to perform the dilute-limit extrapolation.
In Figure 7 we show the sedimentation velocity for a highly charged disk with $`p=5`$, normalized by the sedimentation velocity of uncharged disks at the same volume fraction. Although, again, the relevance of the electrokinetic coupling in the sedimentation velocity diminishes upon increasing volume fraction, the coupling between electric friction and velocity dissipation becomes much more dominant now. The sedimentation velocities decrease by almost $`50\%`$, and the range of the values of $`\kappa a`$ where appreciable deviations from the uncharged disk behavior is observed is wider than in the case of weakly charged disks. Hence, the electrokinetic coupling for disks is much larger than that observed for spheres, and is in fact consistent with mobility reductions observed in laponites. For higher aspect ratios the same trends are observed, as shown in Fig. 8 for the particular aspect ratio $`p=10`$.
In all figures we observe that the reduction in sedimentation velocity for transverse motion is larger than for longitudinal motion. For wide double layers the differences may amount to $`20\%`$. This effect can be intuitively understood in terms of the different forces felt by the electric double layer in the two configurations. For transverse sedimentation, most of the diffuse layer is exposed to the flow induced by the sedimenting array of disks. On the contrary, for longitudinal motion most of the electric double layer is located in a region where the fluid velocity is small and is not subject to large gradients. One would then naively expect that this difference will be enhanced by an increase of the surface charge. But, in fact, the relative difference decreases with the charge of the disk. Hence, as qualitatively illustrated in Figs. 9 and 10 there are non-trivial couplings between the electrostatic restoring force and the flow field. From the figures it is clear that for longitudinal sedimentation most of the diffuse layer is in a region of smoothly varying velocity, whilst for transverse sedimentation the double-layer is exposed to large velocity gradients. More interestingly, by comparing the flow fields past the weakly and the highly charged disk, we observe that the effective hydrodynamic shape of the particle becomes more isotropic. Moreover, close to the surface of the particle in Fig. 9b, the direction of the flow is reversed with respect to the direction of the bulk flow. This effect can only be caused by the different behavior of the electrostatic and the hydrodynamic fields at the edge of the disks, an effect we have not analyzed in detail, because much more expensive simulations are required to gain a more quantitative understanding of the flow patterns in Fig. 9b. Such simulations fall out of the scope of the present paper.
## VII Sedimentation velocity of charged disks: shape effects
It is not straightforward to quantify the effect of shape variations on the sedimentation of (charged) disks because one cannot change the shape of without modifying either the surface charge density or the overall particle charge. Then, because the electrostatic field next to a particle is proportional to the surface charge $`\sigma `$, a change in the surface area will change the electric field surrounding the particle, making it impossible to isolate the effect of shape change. On the other hand, keeping $`\sigma `$ constant by varying the overall particle charge is not a solution either since the reduction in sedimentation velocity does depend also on $`Z`$ \[see Eqn. (5)\]. As a result, we will have to modify both valency and volume (to keep the surface area constant) to disentangle charge effects from effects arising from shape changes. However, even if we take care of this problem, we can only compare each disk with the corresponding sphere, because the two disks we study have different areas.
In order to focus on shape effects as much as possible, we computed the normalized sedimentation velocity $`U_d(Z)/U_d(0)`$ (with $`U_d(Z)`$ the sedimentation velocity of an isolated particle with valency $`Z`$, and $`U_d(0)`$ the velocity of the same object with $`Z=0`$) with the corresponding normalized sedimentation velocity of a sphere with the same valency $`Z`$ and surface area, $`U_s(Z)/U_s(0)`$.
For weakly charged particles, we can make use of Booth’s prediction to analyze the results. To this end, rather than studying the scaled velocity directly, we have found fruitful to consider $`[1U_d(Z)/U_d(0)]/Z^2`$, which is the coefficient $`c_2`$ \[see Eqn. (6)\] in the case of a sphere. This is a direct measure of the electrokinetic reaction induced by the electric double layer. Since we have argued (see Section IV) that the charge dependence of disks is the same as the one observed for spheres in the Debye-Hückel limit, the previous ratio is a quantitative way of assessing the role of shape on the sedimentation velocity.
In Figures 11(a) and 11(b), we show $`[1U_d(Z)/U_d(0)]/Z^2c_2^d`$ for disks with two different aspect ratios and with a small charge, $`Z=10`$, both for transverse and longitudinal sedimentation. $`c_s^d`$ is expressed in units of $`A_2^sk_BTl_B/\left(72\pi D\eta a^2\right)`$, in such a way that for spheres it reduces to $`f(\kappa a)`$ as predicted by Booth. For weakly charged disks and thin double layers, the decrease in velocity does not depend strongly on shape. This is consistent with Smoluchowski’s theory for electrophoresis , which predicts that the electrophoretic velocity of particles with the same zeta potential (the electrostatic potential at contact) is independent of the particle shape if $`\kappa a\mathrm{}`$. However, the deviation from this Smoluchowski limit appears confined at narrower double layers for longitudinal motion; hence, shape affects significantly the sedimentation velocity of suspended particles. Moreover, in the case of asymmetric objects, the orientation of the particle also affects the velocity. For both longitudinal and transverse sedimentation, the electrokinetic coupling of a disk is always smaller than the decrease for an equivalent sphere. One can clearly see that the decrease in velocity for longitudinal motion is smaller than for transverse motion.
In the high-charge regime we use the same quantity, $`c_2^d,`$ to assess the role of shape, although we know that the Booth theory fails in this case. In Figures 12(a) and 12(b) we show $`c_2^d,`$ again for two aspect ratios. In the thin diffuse-layer limit, our data are consistent with Smoluchowski’s theory, and we observe again a departure from the results for a sphere upon increasing the width of the electric double layer. The maximum effect is observed for electric-double-layer widths of the order of the largest linear dimension of the object, and, again, the decrease for longitudinal motion is smaller than for transverse motion. This is consistent with the intuitive picture that the hydrodynamic shape of a disk becomes more isotropic upon increasing its charge.
By comparing Figs. 12(a) and 12(b), the reader might conclude that the reduction in sedimentation velocity is higher for the disk with a smaller aspect ratio. However, one should bear in mind that the two disks have the same valency and therefore very different surface charges $`\sigma _{p=5}3.4\times \sigma _{p=10}.`$ To show how much the surface charge affects $`c_2`$, we show $`c_2`$ for the disk with $`p=10`$ also at $`Z=300.`$ Even though the surface charge of this disk is still lower than the surface charge for the other disk, the electrokinetic effect is already more pronounced.
An illuminating way of displaying the relevance of shape for sedimentation is to consider what the effective Stokes radius of a sedimenting disk is. In Fig. 13 we show the effective Stokes radius \[$`R_{\text{eff}}F/\left(6\pi \eta U\right)`$\], where $`F`$ is the magnitude of the external force acting on the disk. For small charges, the effective radius depends weakly on the width of the double layer, and is larger for the transverse motion, as can be expected. At high charges the behavior is qualitatively different since $`R_{\mathrm{eff}}`$ depends on the width of the double layer for $`\lambda _D/a<2.`$ For larger $`\lambda _D/a`$ it tends to level off. As the diffuse layer broadens, the effective size that characterizes the sphere and the disk in longitudinal motion tend to converge, leading to a same effective shape for wide layers.
The physical origin of this effect is already implicit in Figs. 9 and 10. These figures show the velocity fields around the disk for both orientations. Different flow fields develop around the sedimenting disk for low and high surface charge.The flow profiles look more isotropic for high $`Z,`$ therefore one might expect that for high $`Z`$, the friction coefficients of a disk approaches that of a sphere with the same $`Z.`$
## VIII Discussion
In this paper we have presented simulations of the sedimentation of an array of charged disks. We have treated the electrolyte at the Poisson–Boltzmann level, while we have incorporated the relevant hydrodynamic couplings between the solvent and the dissolved electrolyte. Using the lattice-Boltzmann method we have modelled highly charged colloids and arbitrary $`\kappa a`$ values, which greatly expands the parameter range that can be covered.
Since no exact analytical expressions exist for the sedimentation velocity of isolated neutral disks and finite thickness, we have first checked the performance of our method by validating the sedimentation of neutral disks using approximate expressions that become exact for infinitesimally thin disks. Such a computation has provided us with values for the sedimentation velocities of uncharged disks, which are needed for the subsequent analysis.
In order to clarify the role of electrohydrodynamic coupling and the relevance of shape, we have performed a systematic study to assess the role of shape, volume fraction, charge, and ionic strength on the sedimentation velocity. We find that in the linearized Debye-Hückel regime, the sedimentation velocity has the same functional dependence on volume fraction and surface charge as that for spheres, although with different amplitudes. This deviation should be accounted for when using diffusivity measurements of disks to infer the effective charge of colloids and this work represent, to our knowledge, the first were the sedimentation velocity is computed systematically. So far experimental findings could only be compared with the theory for weakly charged spheres , which can lead to numerical errors in the estimates of their effective charges . At fixed $`\kappa a,`$ we have studied the surface charge dependence of the disk sedimentation velocity, from which we have observed that in the high-charge regime, the accumulation of charge near the disk surface layer decreases the effect of electrokinetic coupling on the sedimentation velocity, and also shows that such accumulation becomes more relevant as the disk becomes more anisotropic.
We have shown that the geometrical anisotropies of neutral disks are reduced by the presence of the electric double layer, especially for highly charged disks. In fact, we have seen that when the double layer is exposed to larger velocities, the reduction in sedimentation velocity is larger. Hence, this mechanism tends to generate a more symmetric disk response, as can be effectively characterized in terms of an effective disk radius which becomes less sensitive to shape details as charge increases.
## Acknowledgments
The work of the FOM Institute is part of the research program of FOM and is made possible by financial support from the Netherlands organization for Scientific Research (NWO).
I.P. acknowledges financial supprot from DGICYT of the Spanish Government and from DURSI, Generalitat de catalunya (Spain), and thanks the FOM institute for its hospitality.
|
warning/0506/astro-ph0506254.html
|
ar5iv
|
text
|
# Deuterated H⁺₃ in proto-planetary disks
## 1 Introduction
Proto-Planetary disks are the sites of planet formation. Their physical, dynamical and chemical structure and evolution determine if, when, how, where and what planets form. Two very important parameters that are difficult to determine are the gas mass of the disk, and the ionization degree. Critical questions that need to be answered by observations are: what is the evolution of the gaseous component of the disk, in particular with respect to the dusty component? Is it dispersed before or after dust coagulates into planetesimals and/or rocky planets? At what radius? By what process? At the same time, theory predicts that the accretion in the disk is regulated by its ionization degree (Balbus & Hawley 1998, Gammie 1996). So the questions here are: what is actually the measured ionization degree across the disk? What ionizes the gas? Cosmic rays and/or X-rays? Where do the two effects balance each other, if they do?
Observationally answering to those questions is all but an easy task, especially in solar type systems. This is because the bulk of the disk mass resides in the outer midplane, which is cold and dense. As a consequence, all heavy-bearing molecules freeze-out onto the grain mantles, and disappear from the gas phase where they could be observed. So the first difficulty in the study of the outer disk midplane is to find probes of it. Last year, Ceccarelli et al. (2004) detected abundant H<sub>2</sub>D<sup>+</sup> in the proto-planetary disk which surrounds the solar type protostar DM Tau. They claimed that H<sub>2</sub>D<sup>+</sup> probes the cold outer midplane and, in addition, its abundance is a direct measure of the ionization degree. The reason behind this claim is the peculiar chemistry of the H<sub>2</sub>D<sup>+</sup> ion. The basic idea is that in cold and dense gas, where CO and all heavy-bearing molecules freeze-out into the grain mantles, two things happen: first, only H<sub>2</sub> and the ions from this molecule, namely H$`{}_{}{}^{+}{}_{3}{}^{}`$ and its isotopomers, remain in the gas phase; second, the molecular deuteration is dramatically enhanced, up to having H<sub>2</sub>D<sup>+</sup>/H$`{}_{}{}^{+}{}_{3}{}^{}`$ larger than unity (Caselli et al. 2003). Therefore, Ceccarelli et al. concluded that a) H<sub>2</sub>D<sup>+</sup> line emission probes the gas disk midplane, and b) the H<sub>2</sub>D<sup>+</sup> abundance measures the ionization degree there.
In the present article, we examine the chemistry of the deuterated forms of H$`{}_{}{}^{+}{}_{3}{}^{}`$ in proto-planetary disks, with the goal of exploring, on a solid theoretical basis, the exact diagnostic value of the H<sub>2</sub>D<sup>+</sup> observations. Besides, the present study concerns also the other H$`{}_{}{}^{+}{}_{3}{}^{}`$ isotopomers, HD$`{}_{}{}^{+}{}_{2}{}^{}`$ and D$`{}_{}{}^{+}{}_{3}{}^{}`$. To accomplish the goal, we develop a chemical model of the outer disk midplane, focused in particular on the H$`{}_{}{}^{+}{}_{3}{}^{}`$ deuteration chemistry. The physical model that describes the disk computes self-consistently the temperature and density profiles for a given disk mass and star luminosity (Dullemond et al. 2001; Dullemond & Dominik 2004). The chemical model is based on what has been understood from the studies of molecular deuteration in pre-stellar-cores and protostars, both observationally (see e.g. the review by Ceccarelli 2004) and theoretically (e.g. Roberts, Millar & Herbst 2003, 2004; Walmsley, Flower & Pineaut des Forets 2004, Flower, Pineaut des Forets & Walmsley 2004). Particular emphasis is devoted to the role of dust grains, and the effect of dust coagulation/fragmentation on the disk midplane chemical structure.
Aiming to give observable predictions, besides to provide the abundances of the H$`{}_{}{}^{+}{}_{3}{}^{}`$ isotopomers across the disk, and the average column density of each species, we compute the intensities of the four ground state lines from the ortho and para forms of H<sub>2</sub>D<sup>+</sup> and HD$`{}_{}{}^{+}{}_{2}{}^{}`$ respectively. We carry out a wide parameter study, and explore the dependence of our results on three major parameters: the dust-to-gas ratio, the cosmic rays ionization rate, and the dust grain average sizes. Besides, we also discuss how the results depend on the basic properties of the star-disk system, namely the star luminosity, disk age, mass and radius. Finally, we discuss the case in which N<sub>2</sub> disappears from the gas phase simultaneously with CO, which is not what has been observed so far, but what it would be expected based on the laboratory measurements of the N<sub>2</sub> and CO binding energies (e.g. Oberg et al. 2005).
The article is organized as follows. In §2 we develop our model. In §3 we report the model predictions of a standard case, and as function of the parameters of the model. In §4 we discuss the diagnostic values of the observable quantities (line intensities and column densities). Finally, §5 summarizes the content of the article.
## 2 The model
In this section we describe the model that we developed to calculate the abundance profiles and the average column densities of H$`{}_{}{}^{+}{}_{3}{}^{}`$ and its isotopomers across the disk, and the line emission of the two deuterated forms of H$`{}_{}{}^{+}{}_{3}{}^{}`$ which have observable ground state rotational transitions, H<sub>2</sub>D<sup>+</sup> and HD$`{}_{}{}^{+}{}_{2}{}^{}`$.
### 2.1 Disk structure
For the model calculations we use a model of passively irradiated hydrostatic flaring circumstellar disks (Dullemond et al 2001; Dullemond & Dominik 2004). This model computes the structure (i.e. the density and temperature distribution) in a selfconsistent way. In the center of the disk, a low-mass star is located with a mass of $`M_{}`$ and a luminosity of $`L_{}`$. Around the star we distribute $`M_{\mathrm{disk}}`$ of material in a disk ranging from 0.1 AU to $`R_{\mathrm{disk}}`$, with a surface density powerlaw $`\mathrm{\Sigma }r^1`$, implying that the disk mass per unit radius is constant, i.e. both inner and outer disk contain significant amounts of mass. The disk contains dust at a mass fraction of $`f_{\mathrm{d}/\mathrm{g}}`$ which we assume to be fully mixed with the gas. While in reality, there probably exists a distribution of grain sizes in the disk, we choose a single grains size for the present calculation. This allows us in a simple way to study the effects of grain size. The structure of the disk is then computed by iterating between a 1+1D continuum radiative transfer which computes the dust temperature in the entire disk, and a hydrostatic equilibrium code which computes the vertical density and pressure distribution under the assumption that the gas and dust temperatures are equal. For the technical details of the modeling procedure we refer to Dullemond et al (2001). A typical temperature and density profile across the disk (our standard case described in §3.1) is shown in Fig.1. The disk is flaring as can be seen from the upwards-curved temperature contours. We anticipate that heavy-elements bearing molecules will freeze-out, and therefore H$`{}_{}{}^{+}{}_{3}{}^{}`$ deuterium fractionation will be significant at about 25 K (approximatively the CO condensation temperature for the involved densities) and below. These conditions are only fulfilled in the outer disk, approximately outside of 20 AU. The low temperature region becomes geometrically thick at large distances from the star. For example, at a distance of 300 AU, the 15 K contour reaches a height of 60 AU. Typical densities in this region are $`n_{\mathrm{H}_2}=10^7`$ cm<sup>-3</sup> and above, up to $`n_{\mathrm{H}_2}=10^{8.5}`$ cm<sup>-3</sup> in the innermost parts of the outer disk, near 100 AU.
### 2.2 Chemistry of deuterated H$`{}_{}{}^{+}{}_{3}{}^{}`$
Complex chemical networks are often necessary to compute the abundances of molecules in interstellar environments. In particular in regions with a rich chemistry, it is difficult to predict which are the important reactions leading to and from a certain molecule (e.g. Semenov, Wiebe & Henning 2004). However, in the cold and metal-depleted regions in protostellar cores and circumstellar disks, the chemical network is limited and a much simpler treatment is possible. Also, if the chemical timescales involved are short, a steady state solution of the chemistry is often appropriate. In the following we address the most important processes leading to the formation of deuterated H$`{}_{}{}^{+}{}_{3}{}^{}`$ and develop a set of equations for the steady state solution of this network.
In gas under standard molecular cloud conditions, H$`{}_{}{}^{+}{}_{3}{}^{}`$ ions are formed with a rate $`\zeta `$ by the ionization of H<sub>2</sub> (and He) due to cosmic rays, and destroyed by the reactions with all neutral molecules and atoms in the gas, and by recombination reactions with electrons and grains. The reaction H$`{}_{}{}^{+}{}_{3}{}^{}`$ with H<sub>2</sub> – the most abundant species in molecular gas – returns H$`{}_{}{}^{+}{}_{3}{}^{}`$ and therefore has no net effect on the H$`{}_{}{}^{+}{}_{3}{}^{}`$ abundance. However, the reaction with HD, the most abundant H<sub>2</sub> isotopomer, forms H<sub>2</sub>D<sup>+</sup>:
$$\mathrm{H}_3^++\mathrm{HD}\mathrm{H}_2\mathrm{D}^++\mathrm{H}_2$$
(1)
with a rate coefficient $`k_1`$ (see Table 1). The backward reaction is endothermic with an energy barrier of about 220 K. At low temperatures, it is inefficient, making the reaction 1 the effective route to H<sub>2</sub>D<sup>+</sup> formation. When considering the abundances in cold molecular gas, the most important molecules which cause destruction of H$`{}_{}{}^{+}{}_{3}{}^{}`$ are: HD, CO (the second most abundant molecule after H<sub>2</sub>), and N<sub>2</sub> (see §2.3). Also recombination with grains (see § 2.5) is an important process. Equating H$`{}_{}{}^{+}{}_{3}{}^{}`$ formation and destruction rates, and ignoring the destruction by molecules/atoms less abundant than CO and N<sub>2</sub>, the equilibrium abundance ratio between the H$`{}_{}{}^{+}{}_{3}{}^{}`$ and H<sub>2</sub>D<sup>+</sup> is given by the following equation:
$$\frac{\mathrm{H}_2\mathrm{D}^+}{\mathrm{H}_3^+}=\frac{2[\mathrm{D}]k_1}{k_{\mathrm{rec1}}x_\mathrm{e}+k_{\mathrm{CO}}x_{\mathrm{CO}}+k_{\mathrm{N}_2}x_{\mathrm{N}_2}+k_{\mathrm{gr}}x_{\mathrm{gr}}+2k_1[\mathrm{D}]+2k_2[\mathrm{D}]}$$
(2)
where the rate coefficients are defined in Table 1, $`x_e`$ is the electronic abundance, and \[D\] is the elemental abundance of deuterium relative to H nuclei, equal to $`1.5\times 10^5`$ (Lynsky 2003).
As noticed by other authors (e.g. Caselli et al. 2003), the H<sub>2</sub>D<sup>+</sup>/H$`{}_{}{}^{+}{}_{3}{}^{}`$ ratio, given by Equation 2, can be larger than unity if the gas is cold and very depleted of CO and N<sub>2</sub>. In the limit of very cold and completely depleted gas, it reaches $`k_1/k_22`$ (Walmsley et al. 2004).
Just like H$`{}_{}{}^{+}{}_{3}{}^{}`$, H<sub>2</sub>D<sup>+</sup> is destroyed by reacting with molecules, atoms, electrons, and grains. In analogy with H$`{}_{}{}^{+}{}_{3}{}^{}`$, the interaction of H<sub>2</sub>D<sup>+</sup> with HD leads to the formation of HD$`{}_{}{}^{+}{}_{2}{}^{}`$, whereas the reaction with H<sub>2</sub> is endothermic and therefore suppressed at low temperatures.
$$\mathrm{H}_2\mathrm{D}^++\mathrm{HD}\mathrm{HD}_2^++\mathrm{H}_2.$$
(3)
As in the case of reaction (1), the reverse reaction of (3) is endothermic, and therefore inhibited at low temperatures. H<sub>2</sub>D<sup>+</sup> is therefore destroyed by reactions with HD, CO, N<sub>2</sub>, electrons, and grains, and forms HD$`{}_{}{}^{+}{}_{2}{}^{}`$. In the same way, the triply deuterated form of H$`{}_{}{}^{+}{}_{3}{}^{}`$ is formed by the successive reaction of HD$`{}_{}{}^{+}{}_{2}{}^{}`$ with HD:
$$\mathrm{HD}_2^++\mathrm{HD}\mathrm{D}_3^++\mathrm{H}_2$$
(4)
and destroyed by the reaction with CO, N<sub>2</sub>, grains and electrons. Now the reaction D$`{}_{}{}^{+}{}_{3}{}^{}`$ \+ HD just exchange the deuterium atoms, and D$`{}_{}{}^{+}{}_{3}{}^{}`$ is therefore the dead-end of the chain. As in the case of Equation 2, the abundance ratios of HD$`{}_{}{}^{+}{}_{2}{}^{}`$ and D$`{}_{}{}^{+}{}_{3}{}^{}`$ with respect to H<sub>2</sub>D<sup>+</sup> and HD$`{}_{}{}^{+}{}_{2}{}^{}`$ respectively can be derived by equating formation and destruction rates:
$`{\displaystyle \frac{\mathrm{HD}_2^+}{\mathrm{H}_2\mathrm{D}^+}}`$ $`={\displaystyle \frac{2[\mathrm{D}]k_2}{k_{\mathrm{rec2}}x_\mathrm{e}+k_{\mathrm{CO}}x_{\mathrm{CO}}+k_{\mathrm{N}_2}x_{\mathrm{N}_2}+k_{\mathrm{gr}}x_{\mathrm{gr}}+2k_2[\mathrm{D}]+2k_3[\mathrm{D}]}}`$ (5)
$`{\displaystyle \frac{\mathrm{D}_3^+}{\mathrm{HD}_2^+}}`$ $`={\displaystyle \frac{2[\mathrm{D}]k_3}{k_{\mathrm{rec3}}x_\mathrm{e}+k_{\mathrm{CO}}x_{\mathrm{CO}}+k_{\mathrm{N}_2}x_{\mathrm{N}_2}+k_{\mathrm{gr}}x_{\mathrm{gr}}+2k_3[\mathrm{D}]}}`$ (6)
One can easily see that the HD$`{}_{}{}^{+}{}_{2}{}^{}`$/H<sub>2</sub>D<sup>+</sup> ratio can reach unity. In the limiting case of very low temperatures, extreme CO and N<sub>2</sub> depletion and low ionization, it reaches a value of $`k_2/k_31.3`$. On the other hand, the abundance of D$`{}_{}{}^{+}{}_{3}{}^{}`$ is only limited by the electron abundance. In extreme conditions – cold and heavily CO and N<sub>2</sub> depleted gas – D$`{}_{}{}^{+}{}_{3}{}^{}`$ can be the dominant charge carrier.
Table 1 summarizes the reactions involving H$`{}_{}{}^{+}{}_{3}{}^{}`$ and its isotopomers, along with the rates used in this study. The reaction rates have been taken from Roberts et al. (2003, 2004), except for recombination reactions on grains which we discuss in §2.5.
In practice, the adopted chemical network is a small subset of the extensive chemical networks implemented by Roberts et al. (2003, 2004) or Walmsley et al. (2004) or Flower et al. (2004). Our goal here is to study the behavior of the H$`{}_{}{}^{+}{}_{3}{}^{}`$ and its isotopomers, which can indeed be described by the few equations discussed above. This sets the limit of applicability of the present model to regions where CO and N<sub>2</sub> are the only molecules affecting the H$`{}_{}{}^{+}{}_{3}{}^{}`$ chemistry. Furthermore, we did not treat the ortho and para forms independently, but used, when necessary, the results by Walmsley et al. (2004) and Flower et al. (2004).
With this reduced chemical network, we are able to reproduce the results by Roberts et al. (2003, 2004), Walmsley et al. (2004) and Flower et al. (2004) accurately as far as the abundances of the deuterated forms of H$`{}_{}{}^{+}{}_{3}{}^{}`$ are concerned.
### 2.3 CO and N<sub>2</sub> freeze-out
The CO and N<sub>2</sub> abundances are obviously critical parameters in the deuteration of H$`{}_{}{}^{+}{}_{3}{}^{}`$. CO is known to freeze-out onto the grains mantles at large enough densities ($`10^5`$ cm<sup>-3</sup>) and low temperatures ($`25`$ K) (e.g. Caselli et al. 1999; Bergin et al. 2001; Bacmann et al. 2002; Tafalla et al. 2002). Recent laboratory experiments show that the CO binding energy depends on the matrix where CO is embedded (e.g. Collings et al. 2003; Fraser et al. 2004; Oberg et al. 2005). It varies from $``$ 840 K in CO-CO ices (average of the values measured by Collings et al. 2003 and Oberg et al. 2005), to 885 K in a CO–N<sub>2</sub> ice (Oberg et al. 2005), and 1180 K in a CO–H<sub>2</sub>O ice (Collings et al. 2003; Fraser et al. 2004). In this study, we will adopt the value 885 K.
There is overwhelming observational evidence, mainly towards pre-stellar cores, that N<sub>2</sub>H<sup>+</sup> remains in the gas phase at larger densities than CO (e.g. Caselli et al. 1999; Bergin et al. 2001; Tafalla et al. 2004; Pagani et al. 2004; Crapsi et al. 2005). Since N<sub>2</sub>H<sup>+</sup> is believed to be formed from N<sub>2</sub>, these observations suggest that N<sub>2</sub>, which is the major nitrogen reservoir, freezes-out onto grains at higher densities than CO. The reason for that is not fully understood, because the binding energy of N<sub>2</sub>, measured in the laboratories, is only slightly lower that the CO binding energy: 790 K in a pure N<sub>2</sub> ice and 855 K in a N<sub>2</sub>-CO ice (Oberg et al. 2005). However, the effect of having N<sub>2</sub> in the gas phase where CO molecules disappear greatly influences the abundances of H$`{}_{}{}^{+}{}_{3}{}^{}`$ and its isotopomers, because it causes an additional term of H$`{}_{}{}^{+}{}_{3}{}^{}`$ (H<sub>2</sub>D<sup>+</sup> and HD$`{}_{}{}^{+}{}_{2}{}^{}`$) destruction in Equation 2 (and Eqs. 5 and 6). Therefore, we decided to adopt a semi-empirical approach, and to assume that the binding energy of N<sub>2</sub> is 0.65 times that of CO, following the theoretical studies of other authors (e.g. Bergin et al. 1995, 1997). However, we will discuss the effect of a larger N<sub>2</sub> binding energy, as measured in laboratory, in §3.4.
We treat the freeze-out of CO and N<sub>2</sub> in a time-dependent way. CO molecules freeze out onto the grain mantles at a rate:
$$k_{\mathrm{dep}}=S\pi a_{gr}^2n_\mathrm{g}v_{\mathrm{CO}}$$
(7)
where we adopted a sticking coefficient $`S=0.3`$ (Burke & Hollenbach 1983), and a mean grain radius of $`a_{\mathrm{gr}}`$. The grain number density $`n_g`$ is given by the (mass) dust-to-gas ratio $`f_{\mathrm{d}/\mathrm{g}}`$ multiplied by the gas density $`n`$, and divided by the grain mass (see Eq.(15)). A similar equation can be written for N<sub>2</sub>.
Frozen CO and N<sub>2</sub> molecules can be released back into the gas phase by thermal evaporation, and we followed the first order desorption kinetics approach to describe it (Hasegawa & Herbst 1993):
$$k_{\mathrm{ev}}=\nu _0\mathrm{exp}[E_\mathrm{b}/kT]$$
(8)
where $`\nu _0=10^{12}`$s<sup>-1</sup> is the lattice frequency, $`E_b`$ is the binding energy per molecule of CO and N<sub>2</sub> ice respectively, and $`k`$ is the Boltzman constant.
Cosmic rays also contribute to the release of CO and N<sub>2</sub> from the ice, with a rate of (Hasegawa & Herbst 1993):
$$k_{\mathrm{cr}}=9.8\times 10^{15}\frac{\zeta }{3\times 10^{17}s^1}\mathrm{s}^1.$$
(9)
The time dependent number density of gaseous CO, $`n_{\mathrm{CO}}`$, is therefore given by the solution of the following equations:
$`{\displaystyle \frac{\mathrm{d}n_{\mathrm{CO}}}{\mathrm{d}t}}=k_{\mathrm{dep}}n_{\mathrm{CO}}+(k_{\mathrm{ev}}+k_{\mathrm{cr}})n_{\mathrm{CO}}^{\mathrm{ice}}`$ (10)
$`n_{\mathrm{CO}}^{\mathrm{ice}}+n_{\mathrm{CO}}=n_{\mathrm{H}_2}A_{\mathrm{CO}}`$ (11)
where we adopted the CO abundance $`A_{\mathrm{CO}}`$ observed in molecular clouds, $`9.5\times 10^5`$ (Frerking et al. 1982). $`n_{\mathrm{CO}}^{\mathrm{ice}}`$ is the number density of frozen CO molecules. Similar equations can be written for N<sub>2</sub>, where the N<sub>2</sub> abundance in molecular clouds is assumed to be $`4\times 10^5`$ (Bergin et al. 1995, 1997). In our model, the CO and N<sub>2</sub> abundances in the gas phase across the disk depend therefore on the time. They start at time=0 equal to their relative molecular cloud abundances. At large times, they are given by the equilibrium between thermal evaporation, cosmic ray desorption and freeze-out onto the mantles (Leger et al. 1986).
Note that we do not treat in any way the CO and N<sub>2</sub> photo dissociation by the UV and/or X-rays photons emitted by the central star. Therefore, our model just describes the regions where those photons are fully shielded, i.e. the warm molecular layer and the cold midplane of sufficiently massive disks. As our model focuses on the H$`{}_{}{}^{+}{}_{3}{}^{}`$ and its isotopomers only, it does not have any vocation in treating the CO and/or N<sub>2</sub> chemistry, other than computing their disappearance from the gas phase because of the freezing onto the dust grains. Besides, our model does not consider any possible grain surface chemistry involving CO and N<sub>2</sub>, which may transform (part of) condensed CO (and N<sub>2</sub>) into different, more complex molecules.
### 2.4 Charge balance
Ionization in disks can be due to cosmic rays, X rays and UV rays. For the current study we restrict ourselves to regions of the disk which are shielded from UV radiation and X rays, i.e. to the disk midplane. In this case, cosmic rays are the dominating source of ionization (Semenov, Wiebe & Henning 2004).
In regions where the heavy-elements bearing molecules are depleted, the positive charge is carried by the H-bearing species, namely H$`{}_{}{}^{+}{}_{3}{}^{}`$, its isotopomers, and H<sup>+</sup>. The negative charge, on the other hand, is mostly carried by free electrons. The fraction of electrons attached to grains remains small, in particular if the smallest grains have been removed by coagulation (Walmsley et al. 2004). We will therefore ignore negatively charged grains when computing the abundances of H<sup>+</sup>, H$`{}_{}{}^{+}{}_{3}{}^{}`$ (with isotopomers), and $`e^{}`$.
Similar to H$`{}_{}{}^{+}{}_{3}{}^{}`$ (see §2.2) H<sup>+</sup> is formed by the interaction of cosmic rays with H<sub>2</sub>, with a rate $`0.08\zeta `$. Recombination of H<sup>+</sup> occurs mainly on grains because the direct recombination with electrons is too slow, and the proton exchange with HD. The abundance of H<sup>+</sup> is therefore independent of the electron density. Equating the formation and destruction rates, we have:
$$x_{\mathrm{H}^+}=\frac{0.08\zeta }{2k_{\mathrm{HD}}[\mathrm{D}]+k_{\mathrm{gr}}x_{\mathrm{gr}}}$$
(12)
where $`\zeta `$ is the cosmic ray ionization rate. The recombination rate on grains k<sub>gr</sub> is discussed in § 2.5.
The electron abundance may be derived by equating the cosmic ray ionization rates with the processes removing electrons from the gas. As source term of free electrons we consider the reactions $`\mathrm{H}_2+\mathrm{cr}\mathrm{H}_3^++\mathrm{e}^{}`$ and $`\mathrm{H}_2+\mathrm{cr}\mathrm{H}^++\mathrm{e}^{}`$. As electron sink term we use the recombination of H$`{}_{}{}^{+}{}_{3}{}^{}`$ and its isotopes with electrons, with a dissociative recombination rate $`\beta `$, measured by McCall et al. (2003) to be $`4\times 10^7`$ at 10 K. In addition, we assume that every recombination of H$`{}_{}{}^{+}{}_{3}{}^{}`$ (and its isotopomers) and H<sup>+</sup> on grains remove an electron from the gas. This is a reasonable assumption because positively charged grains quickly recombine with free electrons, consistent with the very low abundance of positively charged grains found by (Walmsley et al. 2004). This leads to the following balance equation for the electron density:
$$\zeta (1.08)n_{\mathrm{H}_2}=\left(\beta x_{\{\mathrm{H},\mathrm{D}\}_3^+}+k_{\mathrm{gr}}x_{\mathrm{H}^+}+\frac{k_{\mathrm{gr}}}{\sqrt{m/m_\mathrm{H}}}x_{\{\mathrm{H},\mathrm{D}\}_3^+}\right)n_{H_2}n_e^{}$$
(13)
where $`x_{\{\mathrm{H},\mathrm{D}\}_3^+}`$ is the total abundance of H$`{}_{}{}^{+}{}_{3}{}^{}`$ and its isotopomers, and $`m`$ is the average mass of these molecules which we set equal to $`4.5m_\mathrm{H}`$. Together with charge conservation $`x_\mathrm{e}^{}=x_{\mathrm{H}^+}+x_{H_3^+}+x_{\mathrm{H}_2\mathrm{D}^+}+x_{\mathrm{D}_2\mathrm{H}^+}+x_{\mathrm{D}_3^+}`$, a 2nd degree equation for the electron density results which can be easily solved.
We can consider two limiting cases. If the electron abundance is set by the recombination of electrons with H$`{}_{}{}^{+}{}_{3}{}^{}`$ and its isotopes, we can expect the canonical dependence $`x_e^{}\sqrt{\zeta /n_{H_2}}`$. If on the other hand the electron abundance is set by recombination of H<sup>+</sup> on grains, the electron abundance will be $`x_e^{}\zeta /n_{\mathrm{gr}}`$ where $`n_{\mathrm{gr}}`$ is the number density of grains. We will use this difference below to identify the main recombination process in the model calculations.
### 2.5 Recombination on grains
At low electron densities, collisions of ions with grains become an important contributer to the recombination rates. This is true in particular for protons, and, depending on the rate coefficients for dissociative recombination, also for molecular ions. At high densities, the charge distribution is dominated by neutral and singly negatively charged grains. Draine and Sutin (1987) argue that the sticking coefficients for positive ions on both neutral and negatively charged grains should be 1. The different recombination rates then only depend on the Coulomb focusing factor in the collision cross section. The Coulomb focusing factor $`\stackrel{~}{J}`$ is given by Draine & Sutin (1987) as a function of the grain charge, ion charge and the gas temperature. For a full treatment, the relative abundances of neutral and negatively charged grains would have to be calculated, as e.g. done by (Walmsley et al. 2004). However, in the case of really dense environments, in which the abundance of small grains is small due to the effects of dust coagulation, a simplified treatment is possible. Fig.2 shows the ratio $`\stackrel{~}{J}(Z_{\mathrm{gr}}=1)/\stackrel{~}{J}(Z_{\mathrm{gr}}=0)`$. While for very small grain sizes and extremely low temperatures this ratio can be very large, in the region of parameter space studied in this paper the factor is reasonably small. For 0.1$`\mu \mathrm{m}`$, the ratio is already down to 5.7. For larger grains and somewhat higher temperature, the ratio gets closer to one. For simplicity we therefore choose the following approach. We assume that a fraction $`f_{}`$ of the grains is negatively charged, and that $`1f_{}`$ of the grains are neutral. The recombination rate coefficient $`k_{gr}`$ is then given by (Draine & Sutin 1987):
$$k_{gr}=\left(\frac{8kT}{\pi m_H}\right)^{1/2}\pi a_{gr}^2\left(f_{}\stackrel{~}{J}(Z=1)+(1f_{})\stackrel{~}{J}(Z=0)\right)$$
(14)
Choosing $`f_{}=0.3`$, we expect an error in the recombination rates from this approach no larger than a factor of 3.
Finally, because of the uncertainties of the grain size distribution in the midplane of a disk, we use only a single grain size $`a_{\mathrm{gr}}`$ for each model. Assuming a density of the grain material of 2.5 gr cm<sup>-3</sup>, the dust-to-grain ratio in number density is given by:
$$x_{\mathrm{gr}}=3.2\times 10^{12}\left(\frac{f_{\mathrm{d}/\mathrm{g}}}{0.01}\right)\left(\frac{a_{\mathrm{gr}}}{0.1\mu \mathrm{m}}\right)^3$$
(15)
### 2.6 H<sub>2</sub>D<sup>+</sup> and D<sub>2</sub>H<sup>+</sup> line intensities
H<sub>2</sub>D<sup>+</sup> and D<sub>2</sub>H<sup>+</sup> both appear in the para and ortho forms. Gerlich, Herbst and Roueff (2002) have discussed in detail the issue, and found that at low temperatures the ortho to para ratio of H<sub>2</sub>D<sup>+</sup> is close to unity. Subsequently, Walmsley et al. (2004) have shown that the H<sub>2</sub>D<sup>+</sup> ortho to para ratio indeed varies weakly with the density in completely depleted regions, and it is around 0.2 at densities $`10^6`$ cm<sup>-3</sup>. Since the present study focuses on the outer disk midplane where the densities are larger than $`10^6`$ cm<sup>-3</sup>, we do not calculate the H<sub>2</sub>D<sup>+</sup> ortho-to-para ratio from a chemical network, but simply assume it to be 0.3. Finally, Flower et al. (2004) find that the HD$`{}_{}{}^{+}{}_{2}{}^{}`$ ortho-to-para ratio is equal to 10 for large densities and low temperatures.
The ortho and para forms of both H<sub>2</sub>D<sup>+</sup> and HD$`{}_{}{}^{+}{}_{2}{}^{}`$ have ground transitions in the submillimeter to Far-IR (Tera-Hertz) range (Hirao & Amano 2003): 372.4 and 1370.1 GHz for the o-H<sub>2</sub>D<sup>+</sup> and p-H<sub>2</sub>D<sup>+</sup> respectively, and 1476.6 and 691.7 GHz for the o-HD$`{}_{}{}^{+}{}_{2}{}^{}`$ and p-HD$`{}_{}{}^{+}{}_{2}{}^{}`$, respectively. The o-H<sub>2</sub>D<sup>+</sup> and p-HD$`{}_{}{}^{+}{}_{2}{}^{}`$ transitions are observable with ground based telescopes (CSO and JCMT today, APEX and ALMA in the near future), whereas the other two transitions can only be observed with airborne and satellite telescopes, in particular SOFIA and the upcoming HERSCHEL mission. In this study, we report the line intensity of the four transitions, computed with a non-LTE code which treats self-consistently the line optical depth (Ceccarelli et al. 2003)<sup>1</sup><sup>1</sup>1The code has been adapted to the disk geometry., assuming that the disk is seen face-on. Note that, for the disks studied in this article, the line optical depth in the face-on configuration never exceeds unity though, so that the lines are in practice always optically thin. The collisional coefficients are not well known, and we assumed that the critical density of all the H<sub>2</sub>D<sup>+</sup> and HD$`{}_{}{}^{+}{}_{2}{}^{}`$ ground transitions are $`10^6`$ cm<sup>-3</sup> derived using the H<sub>2</sub>D<sup>+</sup> scaled collisional coefficients by Black et al. (1990). Since the regions where the lines originate have relatively large densities ($`10^7`$ cm$`{}_{}{}^{}3`$), the levels are mostly LTE populated, and the uncertainty on the collisional coefficients is, therefore, not of great importance here. The line fluxes are given in erg s<sup>-1</sup>cm<sup>-2</sup> assuming a source distance of 140 pc. For an easy comparison with the presently available observations, we also give the o-H<sub>2</sub>D<sup>+</sup> (at 372 GHz) and p-HD$`{}_{}{}^{+}{}_{2}{}^{}`$ (at 691 GHz) line fluxes in main beam temperature, assuming that the observations are carried out at the CSO and JCMT telescopes respectively.
## 3 Results
In this section we discuss the results of the model calculations, namely the abundances of the H<sub>2</sub>D<sup>+</sup>, HD$`{}_{}{}^{+}{}_{2}{}^{}`$ and D$`{}_{}{}^{+}{}_{3}{}^{}`$ across the disk. One major goal of this article is to give predictions for observations of the two deuterated forms of H$`{}_{}{}^{+}{}_{3}{}^{}`$ that have ground rotational transitions in the submillimeter wavelength range: H<sub>2</sub>D<sup>+</sup> and D<sub>2</sub>H<sup>+</sup>.
We first explore in detail our standard case, which is a disk with canonical values for parameters like dust-to-gas ratio, cosmic ray flux and grain sizes, properties not easily derived from observations. In§ 3.2 we study the dependence on these parameters. In § 3.3 we address system properties that are generally known from other observations: the total dust mass, the disk size, and the stellar luminosity. Finally, in § 3.4 we discuss the influence of one important model assumption, the binding energy of N<sub>2</sub> and CO ice.
### 3.1 The standard model
Our standard model is a disk with a dust disk mass equal to $`2\times 10^4`$ M surrounding a 0.5 L star whose T<sub>eff</sub>=3630 K (see §2.1). The dust-to-gas ratio is 1/100 in mass, and we use the canonical cosmic ray ionization rate $`\zeta =3\times 10^{17}`$ s<sup>-1</sup>. We assume that the disk is $`10^6`$ yr old, and extends up to 400 AU. Figure 1 shows the density and temperature profiles across the disk of our standard case.
Figure 3 shows the abundances of CO and N<sub>2</sub> respectively, across the disk. CO molecules disappear from the gas phase where the dust temperature is around 25 K, at a radius larger than about 100 AU and at heights lower than about 1/4 of the radius. Since the density is large in those cold regions, the drop in the abundance is very sharp, more than a factor 100. The largest CO depletion occurs between about 100 and 200 AU, where it is larger than $`10^4`$ times. At larger radii it is a factor 10 lower. N<sub>2</sub>, however, remains in the gas phase in a much larger region, and, therefore, dominates the destruction rates of H$`{}_{}{}^{+}{}_{3}{}^{}`$ (and its isotopomers). It is only in the very outer regions, at a radius larger than about 300 AU that N<sub>2</sub> freezes-out onto the grains and almost totally disappears from the gas phase. The intermediate region, where CO is depleted but N<sub>2</sub> is not, is where the deuteration of H$`{}_{}{}^{+}{}_{3}{}^{}`$ takes place in a “moderate” form, namely where the H<sub>2</sub>D<sup>+</sup>/H$`{}_{}{}^{+}{}_{3}{}^{}`$ ratio is enhanced with respect to the elemental D/H abundance, but D$`{}_{}{}^{+}{}_{3}{}^{}`$ is not the dominant positive charge carrier.
Figure 4 shows the H<sup>+</sup>, H$`{}_{}{}^{+}{}_{3}{}^{}`$ isotopomers, and electron abundance profiles of the standard disk respectively. We limit the plot of the chemical composition to regions where the CO depletion is larger than a factor of 3. The choice of 3 is a somewhat arbitrary. It represents when the CO abundance is similar to that of N<sub>2</sub>, or, in other words, when the role of these two molecules -believed to be the most abundant gaseous molecules at low temperatures- becomes of similar importance in the H$`{}_{}{}^{+}{}_{3}{}^{}`$ reactions, the H$`{}_{}{}^{+}{}_{3}{}^{}`$ deuteration becomes to be significant, and our simplified model correct.
The electron abundance, is between 3 and $`30\times 10^{10}`$ in the outermost region, at radius $`50`$ AU, increasing going towards the upper layers of the disk. The results are in reasonable agreement with the ionization structure computed by Semenov et al. (2004). In most of the disk, the positive charge is carried by H$`{}_{}{}^{+}{}_{3}{}^{}`$, which becomes less important where also the N<sub>2</sub> disappears from the gas phase, at radius $``$300 AU. H<sub>2</sub>D<sup>+</sup> is the most abundant H$`{}_{}{}^{+}{}_{3}{}^{}`$ isotope, except in the radius $``$300 AU region, where D$`{}_{}{}^{+}{}_{3}{}^{}`$ takes over, and become the most abundant positive ion. H<sup>+</sup> never plays a major role as charge carrier across the entire disk, even though its abundance is comparable to that of H<sub>2</sub>D<sup>+</sup> in outer upper levels of the disk.
The average column densities of a face-on disk of the H$`{}_{}{}^{+}{}_{3}{}^{}`$ isotopes and the electrons in the region where the CO depletion is larger than 3 are reported in Table 2. The D$`{}_{}{}^{+}{}_{3}{}^{}`$ column density is the largest, and it is about $`30\%`$ larger that of H<sub>2</sub>D<sup>+</sup>. The two isotopes together account for almost half of the overall positive charge in the disk midplane. The HD$`{}_{}{}^{+}{}_{2}{}^{}`$ column density is twice smaller than the H<sub>2</sub>D<sup>+</sup> column density. These values are dominated by the outer region, where both CO and N<sub>2</sub> are frozen-out onto the grains, and hence, D$`{}_{}{}^{+}{}_{3}{}^{}`$ becomes the dominant positive charge carrier.
The line intensities of the four ortho and para H<sub>2</sub>D<sup>+</sup> and the HD$`{}_{}{}^{+}{}_{2}{}^{}`$ ground transitions are reported in Table 3.
### 3.2 Ionization rate, dust-to-gas ratio and grain sizes as important factors for the abundances
Important parameters of the model are the cosmic ray ionization rate, the dust-to-gas ratio, and the grain sizes, all of which are relatively ill known. One goal of this theoretical work is to provide predictions of observable quantities which can help to constrain these parameters, which all play a key role in the evolution of the proto-planetary disks and the eventual planet formation (see Introduction).
In order to investigate the influence of the cosmic ray flux and of the dust-to-gas ratio, we have computed a grid of models where we vary these parameters within a factor 100 of the standard value. Since the structure of the disk is fully determined by the dust opacity, changing the dust-to-gas ratio simply implies scaling the gas density while all other factors like gas temperature and grain number density remain constant. Figure 5 shows the column densities as a function of the cosmic ray ionization rate normalized to $`3\times 10^{17}`$, and of the dust-to-gas ratio. It is immediately obvious that the electron column density varies as a simple powerlaw $`\sqrt{\zeta /f_{\mathrm{d}/\mathrm{g}}}\sqrt{\zeta n_{H_2}}`$. This corresponds to an electron abundance dependence $`x_e^{}\sqrt{\zeta /n_{H_2}}`$, indicating that cosmic ray ionization and dissociative recombination of electrons with the charge carrying molecules (H$`{}_{}{}^{+}{}_{3}{}^{}`$ and its isotopes in this case) are determining the electron abundance. This is consistent with the isotopes of H$`{}_{}{}^{+}{}_{3}{}^{}`$ being the main charge carriers. Because throughout the plot, the dust grain density $`n_{\mathrm{gr}}`$ is constant (changing $`f_{\mathrm{d}/\mathrm{g}}`$ only changes the gas density), recombination on grains would lead to a dependence $`x_e^{}\zeta /n_{H_2}`$.
The shape of the contours for the column densities of H<sub>2</sub>D<sup>+</sup> and D<sub>2</sub>H<sup>+</sup> are similar to those of e<sup>-</sup>, again indicating that these molecules are the dominant charge carriers. D$`{}_{}{}^{+}{}_{3}{}^{}`$ shows a similar behavior for very low dust-to-gas ratios (i.e. high gas densities) and normal or low cosmic ray ionization rates. However, for very low gas densities (high dust-to-gas ratios) and for very high ionization rates, the contours become vertical, indicating that the column density becomes independent of the ionization rate. A similar but smaller effect can also be observed for the column densities of H<sub>2</sub>D<sup>+</sup> and HD$`{}_{}{}^{+}{}_{2}{}^{}`$ at high dust-to-gas ratios. This is due to the fact that at low gas densities (corresponding also to low electron densities), the recombination of these molecules on grains starts to become important and leads to a faster decrease in the abundances.
Figures 6 to 7 are similar to Fig. 5, but using a disk model with different grains sizes. We have used grain sizes of 0.01$`\mu \mathrm{m}`$, and 1.0$`\mu \mathrm{m}`$, to study the cases of very little dust growth compared to the interstellar medium, and significant dust growth. Since the total dust mass in the model is constant, reducing the grain size corresponds to an increase in the number density of grains, and also in the total grain surface per cm<sup>3</sup>. This does favor recombination on grains as a destruction route of H$`{}_{}{}^{+}{}_{3}{}^{}`$ and its isotopes. This is indeed visible, in particular at high dust-to-gas ratios, the column densities of all three molecules are strongly decreased, and almost independent of the ionization rate.
A different effect can be observed in the calculation for large grains. In this case, the column density of electrons greatly exceeds that of the H$`{}_{}{}^{+}{}_{3}{}^{}`$ isotopes, in particular in the upper left part of the diagram, at high ionization rates and high gas densities. The electron column density contours are closer together, indicating a linear dependence on ionization rate and gas density. As discussed in § 2.4, this is an indication that H<sup>+</sup> is now the dominating positive charge carrier and that recombination of H<sup>+</sup> on grains has taken over as main electron destruction mechanism. The reason for the significant decrease in deuterated H$`{}_{}{}^{+}{}_{3}{}^{}`$ abundance compared to cases with smaller grains is caused by smaller CO/N<sub>2</sub> depletions. The larger grains provide less grain surface for these molecules to freeze out and shift the abundance ratios away from the deuterated species towards H$`{}_{}{}^{+}{}_{3}{}^{}`$. The column densities of H<sub>2</sub>D<sup>+</sup>, HD$`{}_{}{}^{+}{}_{2}{}^{}`$ and D$`{}_{}{}^{+}{}_{3}{}^{}`$ all have a maximum at normal cosmic ray ionization rates. If the cosmic ray flux is further increased, the column densities degrease again, because the CO/N<sub>2</sub> depletion becomes less efficient. Also for smaller grains, the same effect is expected, but at much higher fluxes.
Figure 8 shows the dependence of the H<sub>2</sub>D<sup>+</sup> and HD$`{}_{}{}^{+}{}_{2}{}^{}`$ line fluxes on the cosmic ray ionization rate and the dust-to-gas ratio. For an easy comparison with observations obtainable at CSO and JCMT we also show plots with the velocity-integrated mean beam temperatures of the o-HD$`{}_{}{}^{+}{}_{2}{}^{}`$ at 372 GHz and p-HD$`{}_{}{}^{+}{}_{2}{}^{}`$ at 691 GHz lines respectively (Figure 9).
The curves of the o-H<sub>2</sub>D<sup>+</sup> and p-HD$`{}_{}{}^{+}{}_{2}{}^{}`$ line intensities run practically parallel, so that the simultaneous observation of the two lines would not depend on the parameters of the model, but the ortho-to-para ratio of each species. They can therefore be “safely” used to derive the relative ortho-to-para ratio of H<sub>2</sub>D<sup>+</sup> with respect to HD$`{}_{}{}^{+}{}_{2}{}^{}`$.
### 3.3 The basic stellar and disk properties
The basic stellar and disk parameters such as stellar luminosity, dust mass in the disk and disk size can normally derived independently. We therefore do not enter into a full parameter study of these quantities. Instead, we limit ourselves to describe the main trends by looking at single-parameter changes from the standard model. Table 4 summarizes the results of these calculations. The stellar and disk parameters mainly change the local densities in the disk, or the temperature. Temperature increase influences the excitation of the lines, and may reduce depletion. Density changes modify the degree of ionization and again the depletion efficiency. The results, summarized in Table 4, show that changes of a factor of two in any of the parameters have only limited effects on the line fluxes, in all cases less than a factor of two. The largest changes occur when the disk radius is changed, mostly because of the larger/smaller integration emitting region.
### 3.4 High N<sub>2</sub> binding energy
As mentioned in §2.3, there is a discrepancy between the measured binding energy of N<sub>2</sub> onto the grains -which is similar to the CO binding energy- and the observations in pre-stellar cores -which show that N<sub>2</sub> molecules condense onto the grains at a larger density, and therefore, suggest a N<sub>2</sub> binding energy lower than the CO one. In our standard model, discussed in the previous paragraphs, we adopted that the N<sub>2</sub> binding energy is 0.65 times that of CO, in agreement with the modeling of the pre-stellar cores observations (e.g. Bergin et al. 1995, 1997). It is also worth noticing that observations of N<sub>2</sub>H<sup>+</sup> in a proto-planetary disk support indeed the lower binding energy for N<sub>2</sub> (Qi et al. 2003). However, for the sake of completeness, here we explore the influence of this assumption on the results.
As shown in Figure 3 and previously discussed, the low N<sub>2</sub> binding energy implies that N<sub>2</sub> molecules remain in the gas phase in a large region where CO is frozen-out onto the grains. It is therefore clear that the presence or not of N<sub>2</sub> in the gas phase will have a strong impact on the modeling results.
Figure 10 shows the column densities of H$`{}_{}{}^{+}{}_{3}{}^{}`$, H<sub>2</sub>D<sup>+</sup>, HD$`{}_{}{}^{+}{}_{2}{}^{}`$ and e<sup>-</sup> as function of the dust-to-gas ratio and the cosmic ionization ratio, similarly to Figure 5. The comparison of the two figures speaks for itself. If N<sub>2</sub> has a binding energy comparable to that of CO (Figure 10), D$`{}_{}{}^{+}{}_{3}{}^{}`$ is the positive charge carrier across most of the outer disk, in the standard case (dust-to-dust ratio equal to 0.01, and cosmic ray ionization rate equal to $`3\times 10^{17}`$ s<sup>-1</sup>), and H<sub>2</sub>D<sup>+</sup> and HD$`{}_{}{}^{+}{}_{2}{}^{}`$ are 30 times less abundant. On the contrary, if N<sub>2</sub> has a reduced binding energy (Figure 5) the positive charge is almost equally shared between D$`{}_{}{}^{+}{}_{3}{}^{}`$, H<sub>2</sub>D<sup>+</sup> and H$`{}_{}{}^{+}{}_{3}{}^{}`$. This is because the N<sub>2</sub> collisions are an important destroyer of the H$`{}_{}{}^{+}{}_{3}{}^{}`$, H<sub>2</sub>D<sup>+</sup> and HD$`{}_{}{}^{+}{}_{2}{}^{}`$, preventing the “total conversion” of H$`{}_{}{}^{+}{}_{3}{}^{}`$ into D$`{}_{}{}^{+}{}_{3}{}^{}`$.
If the CO and N<sub>2</sub> binding energies are similar, increasing and/or decreasing the dust-to-gas ratio does not change much the H<sub>2</sub>D<sup>+</sup> and/or HD$`{}_{}{}^{+}{}_{2}{}^{}`$ column densities. In practice, increasing/decreasing the gas density with respect to the dust density does not change the H<sub>2</sub>D<sup>+</sup> and HD$`{}_{}{}^{+}{}_{2}{}^{}`$ column densities, but just increases/decreases the D$`{}_{}{}^{+}{}_{3}{}^{}`$ column density -the positive charge carrier. This has an important observational consequence, for the H<sub>2</sub>D<sup>+</sup> and/or HD$`{}_{}{}^{+}{}_{2}{}^{}`$ observed column densities cannot constrain the dust-to-gas ratio in a large interval.
However, it is worth noticing that, in principle, the ratio of the derived average column densities can help to discriminate whether N<sub>2</sub> is present in the gas phase or not. Finally, observations of the N<sub>2</sub>H<sup>+</sup> in proto-planetary disks will also help to clarify whether N<sub>2</sub> remains in gas phase after CO condenses (Qi et al. 2003).
## 4 Discussion
### 4.1 Simplified treatment of the chemical network
In this paper we have introduced a simplified chemical network to compute the abundances of deuterated H$`{}_{}{}^{+}{}_{3}{}^{}`$. The network treats only the four isotopes of H$`{}_{}{}^{+}{}_{3}{}^{}`$, electrons, H<sup>+</sup>, CO, and N<sub>2</sub>. Comparison with more extensive networks shows that the abundances of the {H,D}$`{}_{}{}^{+}{}_{3}{}^{}`$ ions can be calculated accurately under the conditions of significant CO and N<sub>2</sub> depletion. The reason for this success is that the chemistry indeed becomes very simple under these conditions. CO and N<sub>2</sub> are the last heavy-elements bearing molecules with significant abundances to leave the gas phase and to freeze out on dust grains. Together with dust grains, they act as the main destroyers of H$`{}_{}{}^{+}{}_{3}{}^{}`$ and its isotopes. The simplified chemical network is, under these conditions, in fact complete in that it treats all important species and reactions. Under the assumption that the freeze-out of CO and N<sub>2</sub> has reached its steady state, the steady state solution for this network can be derived directly, no time integration of the rate equations is necessary. The chemical network, therefore, lends itself for large parameter studies of cold gas.
### 4.2 The main charge carrier in the disk midplane
In the discovery paper of the H<sub>2</sub>D<sup>+</sup> line in DM Tau and TW Hya, Ceccarelli et al (2004) had suggested that H<sub>2</sub>D<sup>+</sup> may be the main charge carrier under the conditions found in a disk, which would allow for a direct derivation of the degree of the column density of positive ions by observing H<sub>2</sub>D<sup>+</sup>. With an H<sub>2</sub> column density derived from dust continuum observations, this number directly leads to the degree of ionization, a much sought-after property of disk gas because of its importance for the magneto-rotational instability, the likely driver of disk turbulence. This conclusion has to be softened in the light of the present results. Our calculations show that H<sub>2</sub>D<sup>+</sup>, HD$`{}_{}{}^{+}{}_{2}{}^{}`$, D$`{}_{}{}^{+}{}_{3}{}^{}`$ and H$`{}_{}{}^{+}{}_{3}{}^{}`$ all carry similar amounts of charge, indicating that the observed column densities of H<sub>2</sub>D<sup>+</sup> and HD$`{}_{}{}^{+}{}_{2}{}^{}`$ provide both an order-of-magnitude estimate of the degree of ionization, and a lower limit. It was shown earlier (Walmsley et al. 2004) that complete depletion in protostellar cores generally leaves H<sup>+</sup> as the main charge carrier. This trend, however, clearly ends at the densities found in disks: H$`{}_{}{}^{+}{}_{3}{}^{}`$ and its isotopes contribute equally or dominate the charge balance.
In this context, the binding energy of N<sub>2</sub> is of critical importance. If N<sub>2</sub> freezes out at the same moment as CO, then the disk midplane are indeed fully depleted of heavy element bearing molecules. Then, there is no important destruction route for D$`{}_{}{}^{+}{}_{3}{}^{}`$ anymore, and the abundance of D$`{}_{}{}^{+}{}_{3}{}^{}`$ sores to dominate the charge balance. If N<sub>2</sub> indeed stays in the gas longer than CO, as suggested by observations in a proto-planetary disk (Qi et al. 2003), H<sub>2</sub>D<sup>+</sup> and HD$`{}_{}{}^{+}{}_{2}{}^{}`$ are produced in significant abundances and do provide a measure of the degree of ionization.
### 4.3 H<sub>2</sub>D<sup>+</sup> and HD$`{}_{}{}^{+}{}_{2}{}^{}`$ as tracers of cold gas
One major question we aim to answer with the current study is: under what conditions are the observable lines of H<sub>2</sub>D<sup>+</sup> and HD$`{}_{}{}^{+}{}_{2}{}^{}`$ tracers of the gas mass in the disk midplane? For these molecules to be a tracer, the observed line intensities must be a strong function of the dust-to-gas ratio. As we have seen in Figures 5 and 10, the answer to this question depends on the behavior of N<sub>2</sub>. If N<sub>2</sub> freezes-out at lower temperatures than CO -as supported by the observations in a proto-planetary disk (Qi et al. 2003)-, then the observed line intensities do depend strongly on the dust-to-gas ratio and can be used to measure it. However, under the conditions of complete depletion, i.e. if also N<sub>2</sub> is completely removed from the midplane gas, the line intensities become insensitive to the dust-to-gas ratio.
Fortunately, observations of N<sub>2</sub>H<sup>+</sup> in disks can discriminate what is the reality, namely whether in the disk midplane CO and N<sub>2</sub> disappear simultaneously from the gas phase or not, and, therefore, whether H<sub>2</sub>D<sup>+</sup> and HD$`{}_{}{}^{+}{}_{2}{}^{}`$ are the main positive charge carriers or D$`{}_{}{}^{+}{}_{3}{}^{}`$ is. In the first case, observations of H<sub>2</sub>D<sup>+</sup> and HD$`{}_{}{}^{+}{}_{2}{}^{}`$ will give a measure of the dust-to-gas ratio, once the cosmic ionization rate is known. This also implies that H<sub>2</sub>D<sup>+</sup> and HD$`{}_{}{}^{+}{}_{2}{}^{}`$ measure the ionization degree in the disk midplane too, as assumed by Ceccarelli et al (2004). Once again, we emphasize that the available observations (Qi et al. 2003) support this case.
### 4.4 Dust coagulation
An important result of this study is the dependence of the chemical structure of the disk midplane on the average dust grain size. Figure 9 shows that the line intensities of the o-H<sub>2</sub>D<sup>+</sup> and p-HD$`{}_{}{}^{+}{}_{2}{}^{}`$ are strongly affected by the average grain sizes, if they are larger than 0.1 $`\mu `$m radius. In practice, if small dust grains are significantly depleted in the midplane, the line intensities decrease dramatically, because CO and N<sub>2</sub> do not freeze-out efficiently on the reduced grain surfaces (see the discussion in §3.2), and no substantial molecular deuteration takes place. The detection of the o-H<sub>2</sub>D<sup>+</sup> and p-HD$`{}_{}{}^{+}{}_{2}{}^{}`$ at the level of the current instrumental sensitivity is, therefore an indication that enough grain surface is still available to account for the molecular depletion. Two scenarios are possible for this. First, coagulation might not be efficient or aggregate destruction is efficient in keeping the number density of small grains at reasonable values. This is in fact in agreement with recent results of grain coagulation which suggest that coagulation at maximum efficiency is inconsistent with the observed opacities and shapes of flaring circumstellar disks (Dullemond & Dominik 2005). Another possibility is that the depletion of CO and N<sub>2</sub> is very efficient in the beginning, before coagulation happens. Efficient coagulation could then form large bodies that bury the ice inside and protect it from cosmic ray desorption. These considerations show that detailed models of coagulation, vertical mixing and freeze-out in combination with observations of H$`{}_{}{}^{+}{}_{3}{}^{}`$ isotopes can be useful diagnostics of grain coagulation.
## 5 Conclusions
We have computed the chemical structure of the midplane of disks surrounding solar type protostars. We have described the derived abundances and the resulting line intensities of the deuterated forms of H$`{}_{}{}^{+}{}_{3}{}^{}`$ in a standard case, and for a wide range of values of the dust-to-gas ratio and cosmic ray ionization rate. We have also discussed in detail the influence of the average grain sizes on the results. Finally, we have reported values also for different stellar luminosities, disk masses and radii, and discussed in detail the case of the N<sub>2</sub> freezing onto the grain mantles simultaneously with CO. Our main conclusions are:
* H$`{}_{}{}^{+}{}_{3}{}^{}`$ deuteration is significant in the midplane of proto-planetary disks around solar type protostars. In our standard case, the positive charge is carried by H$`{}_{}{}^{+}{}_{3}{}^{}`$ in a large zone of the disk midplane, except at the very outer radii, larger than about 300 AU, where D$`{}_{}{}^{+}{}_{3}{}^{}`$ takes over. H<sub>2</sub>D<sup>+</sup> is the most abundant H$`{}_{}{}^{+}{}_{3}{}^{}`$ isotope across most of the disk, at radii less than about 300 AU.
* Contrary to what was earlier assumed, H<sub>2</sub>D<sup>+</sup> is not the dominating positive ion in the standard case. H$`{}_{}{}^{+}{}_{3}{}^{}`$ and its isotopes are equally important. With increasing depletion of CO and N<sub>2</sub>, D$`{}_{}{}^{+}{}_{3}{}^{}`$ becomes more abundant, eventually dominating as the positive charge carrier. Also, H<sup>+</sup> is only of minor importance. While at the pre-stellar-core densities, lower than 10<sup>7</sup>cm<sup>-3</sup>, H<sup>+</sup> dominates (Roberts et al. 2003, 2004; Walmsely et al. 2004), in the disk midplane this is no longer the case.
* The midplane chemical structure and the predicted line intensities of the H<sub>2</sub>D<sup>+</sup> and HD$`{}_{}{}^{+}{}_{2}{}^{}`$ are a strong function of the local cosmic ray ionization rate, which regulates the overall ionization degree, and the CO and N<sub>2</sub> depletion across the disk.
* The chemical structure and line intensities are also sensitive to the dust-to-gas ratio as long as depletion of heavy-element bearing molecules is not complete. This happens if the N<sub>2</sub> binding energy is lower than the CO binding energy, as observed in pre-stellar-cores and possibly in proto-stellar disks. If, on the contrary, N<sub>2</sub> freezes-out simultaneously with CO, as laboratory experiments would rather suggest, since the major charge carrier is D$`{}_{}{}^{+}{}_{3}{}^{}`$, increasing/decreasing the dust-to-gas ratio does not change appreciably the H<sub>2</sub>D<sup>+</sup> and HD<sub>2</sub> column densities and/or line intensities.
* The grain size has strong influence on the line strength expected for H<sub>2</sub>D<sup>+</sup> and D<sub>2</sub>H<sup>+</sup>. Small grains accelerate recombination and reduce the abundances of all positive ions, and therefore decrease the degree of ionization.
This article focuses entirely on theoretical predictions. A forthcoming paper (Dominik et al., in preparation) will analyze in detail the case of DM Tau, where the o-H<sub>2</sub>D<sup>+</sup> line at 372 GHz has been detected, applying the model here developed to a practical case. Likely, with the advent of ALMA, with its great sensitivity and spatial resolution, observations of both the o-H<sub>2</sub>D<sup>+</sup> at 372 GHz and p-HD$`{}_{}{}^{+}{}_{2}{}^{}`$ at 691 GHz will be possible on a routine/systematic base and with spatial resolution. Those observations promise to be very fruitful, and to bring unique information on the physical status of the midplane of the disks surrounding solar type protostars, likely similar to the progenitor of our own Solar System. Besides, observations of p-H<sub>2</sub>D<sup>+</sup> and o-HD$`{}_{}{}^{+}{}_{2}{}^{}`$ by out-of-the-atmosphere instruments will hopefully measure the actual ortho-to-para ratio of these species. Finally, D$`{}_{}{}^{+}{}_{3}{}^{}`$ would be likely the best diagnostics for the gas mass and ionization in low mass disks, but the only possibility to observe it is by absorption at $``$5 $`\mu `$m (Ramanlal & Tennyson 2004; Flower et al. 2004). This requires observations towards disks almost edge-on, but not completely, for the absorption is against the inner dust continuum, which must therefore be detectable. Such disks are hard to find, but new observations by, for example, SPITZER may well discover such sources, and these absorption observations may be possible in the future.
###### Acknowledgements.
It is a pleasure to thank Paola Caselli, Charlotte Vastel and Malcolm Walmsley for comments on the manuscript, and Kees Dullemond for his collaboration on constructing the disk model. We also wish to thank an anonymous referee for careful reading the manuscript. We acknowledge Travel support through the Dutch/French van Gogh program, project VGP 78-387.
|
warning/0506/astro-ph0506053.html
|
ar5iv
|
text
|
# Rotation in Gravitational Lenses
## 1 INTRODUCTION
Gravitational lensing results from the deflection of light under the gravitational influence of all matter, luminous or otherwise. Its physics is clean, and this effect has allowed the measurement of the distribution of dark matter from galaxy-scales (using strong lensing) to the large-scale structure of the universe (using weak lensing).
Usually, many approximations are made to simplify the calculations. One of these is the Born approximation, where one calculates a small deflection along the unperturbed light path. Some effects, such as image rotation due to multi-plane lensing, are not accessible in this approximation. Authors have obtained different results for the magnitude of multi-plane weak lensing rotation (Jain et al 2000, Cooray & Hu 2002, hereafter CH; Hirata & Seljak 2003, hereafter HS). Schneider (1997) showed that a strong lens plus spatially constant weak lens system is mathematically identical to some other single lens plane system, so only differential rotation at the image positions is observable.
In this paper, we apply the multiple plane lensing calculation to real physical systems: quadruply imaged quasars. We show in this physical example how the rotation effect can be measured, why it is physical and real, estimate its magnitude, and show how it can be used to resolve the substructure controversy.
The rotation of images is of current interests in the context of using gravitational lensing to detect substructures predicted by the Cold Dark Matter (CDM) structure formation model. From both semi-analytical studies and numerical simulations, it became clear that hundreds of subhaloes (substructures) are predicted to exist in a Milky-Way type halo (e.g., Kauffmann et al. 1993; Klypin et al. 1999; Moore et al. 1999; Ghigna et al. 2000). In general, about 5-10% of the mass is predicted to be in substructures, with a typical mass spectrum of $`n(M)dMM^{1.8}dM`$. If all the substructures form stars, then the predicted number of satellite galaxies exceeds the observed number in a Milky-Way type halo by a large factor. It is now, however, clear that the correspondence between substructures and visible satellite galaxies is not simple (Gao et al. 2004a,b; see also Springel et al. 2001; Diemand, Moore & Stadel 2004; Nagi & Kravtsov 2005). In particular, if only some substructures house satellite galaxies, then the discrepancy can be alleviated (e.g., Kravtsov et al. 2004). At present, it is not entirely clear whether the internal kinematics of satellite galaxies are consistent with observations (Stoehr et al. 2002; Kazantzidis et al. 2004). Furthermore, the spatial distribution of satellite galaxies in the Milky Way is also somewhat puzzling (Kroupa et al. 2004, but see Kang et al. 2005; Liebeskind et al. 2005; Zentner et al. 2005).
One possible way to detect the (dark) substructures is through the gravitational lensing effect. Simple analytical models in gravitational lenses often fail to reproduce the observed flux ratios (e.g., Kochanek 1991). This discrepancy is commonly referred to as the “anomalous flux ratio problem.” This has been proposed as evidence for substructures in the primary lensing galaxies (e.g., Mao & Schneider 1998; Metcalf & Zhao 2002; Dalal & Kochanek 2002). However, as most of the predicted substructures are in the outer part of the lensing galaxies while the lensed images are typically at a projected distance of only a few kpc from the center, it is unclear whether the predicted amount of substructures in lensing galaxies by CDM is sufficient (Kochanek & Dalal 2004; Mao et al. 2004), so it is important to consider other sources of “substructures.” Recently, Metcalf (2005) proposed that substructures along the line of sight can equally explain the discrepancy. A key question naturally arises: how do we know that substructures are from the primary lens or from elsewhere along the line of sight?
In this paper, we examine the rotation of images induced by structures along the line of sight in gravitational lenses (see also Chen et al. 2003). We use a novel power-spectrum approach and consider the fluctuations of surface densities among different images (separated by few tenths to few arc seconds) and their effects on the magnifications. Throughout this paper, we adopt the “concordance” $`\mathrm{\Lambda }\mathrm{CDM}`$ cosmology (e.g., Ostriker & Steinhardt 1995; Spergel et al. 2003 and references therein), with a density parameter $`\mathrm{\Omega }_\mathrm{m}=0.3`$, a cosmological constant $`\mathrm{\Omega }_\mathrm{\Lambda }=0.7`$, a baryon density parameter $`\mathrm{\Omega }_\mathrm{b}=0.024h^2`$, and the power-spectrum normalization $`\sigma _8=0.9`$. We adopt a Hubble constant of $`H_0=70\mathrm{km}\mathrm{s}^1\mathrm{Mpc}^1`$.
## 2 Lensing by Large Scale Structure in the presence of a strong lens
Many multiply-imaged gravitational lenses on galaxy-scales have been observed. At the time of writing, roughly 100 such systems are known<sup>1</sup><sup>1</sup>1see the CaSTLES database: http://cfa-www.harvard.edu/castles/. The largest systematic survey, the Cosmic Lens All Sky Survey (CLASS) has found 22 new galaxy-scale lenses, approximately one half of which are quadruple lenses (Browne et al. 2003; Myers et al. 2003). Most of these have high-resolution imaging from $`0.1\mathrm{}`$ to mas from MERLIN, HST to VLBI. Several of these lenses are resolved into multiple components. The system 0128+437 provides a good example (Biggs et al. 2004). Each of the image has been resolved into three sub-components with VLBI. Such high-resolution images provide an excellent test bed for lensing models.
In addition to the strong lens system which causes the multiple image splitting, all the matter along the line of sight will further deflect the light and contribute to distortions in the image. One such effect is the apparent rotation of images. It can be shown that a single plane lens has a symmetric amplification matrix, which shears but does not rotate images. With multiple planes, image rotation is possible. Observing it may appear non-trivial, since it would require prior knowledge of the unlensed image alignment. We will show below (see §3) how this can actually be measured if one has a quadruply imaged source with extended structures.
Several geometric configurations can lead to image rotation. A strong lens has shear and convergence of order unity. With sufficiently accurate alignment, a second strong lens could occur along the line of sight. Indeed such an example has already been seen – the JVAS/CLASS lens system, B2114+022 (Augusto et al. 2001; Chae, Mao & Augusto 2001), has two lensing galaxies at redshifts 0.3157 and 0.5883 respectively, within 2 arc seconds of quadruple radio sources<sup>2</sup><sup>2</sup>2It is present unclear whether the radio sources B and C are lensed images..
But in general, the expected variation in surface density due to dark matter is small. Integrating the Limber equation, this leads to large scale structure density variations of order a few percent of the critical surface density (e.g., Jain et al 2000). So a random cluster lens will typically only have weak lenses in its foreground and background.
The typical splitting angle of the strong lens is around one arc second. Large scale structure density fluctuations on such scales are significantly correlated. Since Schneider (1997) had shown that perfectly correlated weak lensing screens do not cause observable rotation, one must compute the differential weak lensing shear.
The cross correlation between two image positions separated by angle $`\mathrm{\Delta }\theta `$ is defined as
$$r\xi _\kappa (\mathrm{\Delta }\theta )/\xi (0),$$
(1)
where $`\xi _\kappa (\mathrm{\Delta }\theta )`$ is the two dimensional Fourier (Bessel) transform of Equation (22) which will be discussed below. The results are shown for four quadruple lenses with known lens and source redshifts, B1422+231 (Patnaik et al. 1992), MG0414+0534 (Hewitt et al. 1992), B1608+656 (Myers et al. 1995), and B2045+265 (Fassnacht et al. 1999). We find a ratio of variances between the difference of two images $`\sigma _{}^2`$ and the individual variances to be
$$\frac{\sigma _{}^2}{\sigma ^2}=2(1r).$$
(2)
If the two images are uncorrelated, the difference will have twice the variances of each individual image. Figure 1 shows the correlation function vs. image separation. As can be seen, typically we have $`r>0.5`$, so the difference mode has a slightly lower variance than each individual image.
We assume that the various images pass through different parts of the strong lens with correspondingly different values of shear and convergence, and also differing weak lens deflectors. We consider a 2-plane lens, $`L_1`$ is the strong lens, $`L_2`$ is a weak lens due to large scale structure, $`O`$ is the observer position and $`S`$ is the source plane. The geometry is shown in Figure 2. We consider a quadratic potential on each lens plane, $`\varphi =ax_1^2+2bx_1x_2+cx_2^2`$, where $`(x_1,x_2)`$ are the (angular) coordinates in the lens plane. The units for the potential are chosen such that $`2\kappa =^2\varphi `$, and the deflection angle is $`\widehat{𝜶}=_x\varphi `$ due to lens, where we use the same notations as in Schneider, Ehlers, & Falco (1992). To distinguish the two lens planes, we will use a prime to denote variables in the $`L_2`$ plane.
For a general anisotropic lens, the deflection angle is a vector. For our quadratic potential, $`\kappa `$ and $`\gamma _{1,2}`$ are constant on the plane,
$$\kappa =\frac{1}{2}(a+c),\gamma _1=\frac{1}{2}(ac),\gamma _2=b.$$
(3)
The quadratic potential is the most general function which leads to constant values of $`\kappa `$ and $`\gamma `$. Their constancy on small scales is inferred from the correlation function shown in Figure 1. In such a potential, the full gravitational lensing effect is straightforward to compute. One can solve the full photon trajectory, which are deflected on the lens planes by the gradient of the potential.
We now denote $`D_1`$ to be the angular diameter distance from the observer to the first lens, $`D_{12}`$ the distance from the first to the second lens, and $`D_{2\mathrm{s}}`$ the distance from the second lens to the source, and so on. We see from Figure 1 that all the relative deflection angles are very small in units of radians ($`10^5`$), so we can expand to first order in the deflection angle (Schneider et al. 1992, chapter 9)
$$𝜼=\frac{D_\mathrm{s}}{D_1}𝝃_1D_{1\mathrm{s}}\widehat{𝜶}_1(𝝃_1)D_{2\mathrm{s}}\widehat{𝜶}_2(𝝃_2),𝝃_2=\frac{D_2}{D_1}𝝃_1D_{12}\widehat{𝜶}_1(𝝃_1).$$
(4)
where $`𝜼`$, $`𝝃_1`$, $`𝝃_2`$ denote the position vectors in the source plane and the first and second lens planes (see Fig. 2), and the deflection angles in the first and second planes are given as $`\widehat{𝜶}_1`$ and $`\widehat{𝜶}_2`$.
Following Schneider et al. (1992), we define two reduced deflection angles
$$𝜶_1=\frac{D_{1\mathrm{s}}}{D_s}\widehat{𝜶}_1,𝜶_2=\frac{D_{2\mathrm{s}}}{D_s}\widehat{𝜶}_2,$$
(5)
With these, eq. (4) can be recast in a very simple form using only angles:
$$𝐲=𝐱_1𝜶_1(𝐱_1)𝜶_2(𝐱_2),𝐱_2=𝐱_1\beta _{12}𝜶_1(𝐱_1),$$
(6)
where $`𝐲=𝜼/D_s`$, $`𝐱_1=𝝃_1/D_1`$, $`𝐱_2=𝝃_2/D_2`$, and $`\beta _{12}=D_{12}D_s/(D_2D_{1\mathrm{s}})`$.
Lensing shear and magnification can be obtained by studying the change of source position $`𝐲`$ resulting from a change of apparent angular position $`𝐱_1`$. In particular, the magnification is given by (Schneider et al. 1992)
$$𝐀=\frac{𝐲}{𝐱_1}=𝐔_1𝐔_2+\beta _{12}𝐔_2𝐔_1,$$
(7)
where $``$ is a unit matrix, $`𝐔_1=𝜶_1/𝐱_1`$, and $`𝐔_2=𝜶_2/𝐱_2`$.
We are considering the combined effects of a strong and a weak lens. The contribution of the weak lens $`𝐔_2`$ is small and we are only interested in its contribution to rotation, so we neglect its linear effect. In the linear (i.e. Born approximation) regime, it is also observationally not possible to distinguish between contributions from the strong and weak lens. In the product term, we can absorb $`\beta _{12}`$ into the definition of $`U_2`$. We define the critical density for the weak lensing large scale structure to be
$$\mathrm{\Sigma }_2^{\mathrm{crit}}\frac{c^2}{4\pi G}\frac{D_{1\mathrm{s}}}{D_{12}D_{2\mathrm{s}}},$$
(8)
which is the lensing strength of the large scale structure as seen by an observer at the strong lens position. This lends a simple observational and computational interpretation of lensing rotation: all distortions visible to an observer at the strong lens position enter linearly into the coupling product of strong and weak lens.
An analogous results holds when the weak lensing plane is in front of the strong lens. In our notation, we have
$$𝐔_1=\left(\begin{array}{cc}\kappa +\gamma _1& \gamma _2\\ \gamma _2& \kappa \gamma _1\end{array}\right),𝐔_2=\left(\begin{array}{cc}\kappa ^{}+\gamma _1^{}& \gamma _2^{}\\ \gamma _2^{}& \kappa ^{}\gamma _1^{}\end{array}\right)$$
(9)
where $`\kappa ^{}\mathrm{\Sigma }_2/\mathrm{\Sigma }_2^{\mathrm{crit}}`$, and so
$`𝐀`$ $`=`$ $`\left(\begin{array}{cc}\gamma _2\gamma _2^{}+(1\kappa +\gamma _1)(1\kappa ^{}+\gamma _1^{})& \gamma _2^{}(1\kappa \gamma _1)+\gamma _2(1\kappa ^{}+\gamma _1^{})\\ \gamma _2^{}(1\kappa +\gamma _1)+\gamma _2(1\kappa ^{}\gamma _1^{})& \gamma _2\gamma _2^{}+(1\kappa \gamma _1)(1\kappa ^{}\gamma _1^{})\end{array}\right)`$ (12)
$``$ $`\left(\begin{array}{cc}1\kappa +\gamma _1& \gamma _2+\gamma _2\gamma _1^{}\\ \gamma _2\gamma _2\gamma _1^{}& 1\kappa \gamma _1\end{array}\right).`$ (15)
In the last approximate equality we used the limit that the weak lensing plane (the primed variables) are much smaller than the strong lens plane, but keeping the anti-symmetric piece which is relevant for rotations. We define the off-diagonal antisymmetric component as $`\omega \gamma _2\gamma _1^{}`$.
For strong lens systems, the shear on $`L_1`$ is of order unity, while the large-structure shear ($`\gamma _1^{}`$, $`\gamma _2^{}`$) is a few percent, corresponding to a rotation of order a degree. Rotation only results when the principal axes of the two amplification matrices are misaligned. We choose the coordinates such that
$$𝐔_1=\left(\begin{array}{cc}\gamma & 0\\ 0& \gamma \end{array}\right),𝐔_2=\left(\begin{array}{cc}\gamma _1^{}& \gamma _2^{}\\ \gamma _2^{}& \gamma _1^{}\end{array}\right).$$
(16)
$`\gamma _1^{}`$ does not contribute to rotation, so we set it to zero. Their product is a pure antisymmetric matrix
$$𝐔_2𝐔_1=\left(\begin{array}{cc}0& \gamma \gamma _2^{}\\ \gamma \gamma _2^{}& 0\end{array}\right).$$
(17)
When added to a unit matrix, this corresponds to a rotation matrix by an angle $`\gamma \gamma _2^{}`$.
In equation (15), we factor the amplification matrix $`𝐀=𝐀_\mathrm{s}𝐑(\varphi )`$, as a product of a symmetric matrix and a pure rotation, where the rotation matrix
$$𝐑=\left(\begin{array}{cc}\mathrm{cos}\varphi & \mathrm{sin}\varphi \\ \mathrm{sin}\varphi & \mathrm{cos}\varphi \end{array}\right).$$
(18)
We have
$$𝐀=\left(\begin{array}{cc}1\kappa +\gamma _1& \gamma _2+\omega \\ \gamma _2\omega & 1\kappa \gamma _1\end{array}\right),$$
(19)
where $`\omega `$ indicates the importance of rotation. Then we find
$$\mathrm{tan}\varphi =\frac{\omega }{1\kappa }.$$
(20)
The non-rotating amplification matrix $`𝐀_s`$ has the same convergence as $`𝐀`$ up to $`𝒪(\omega ^2)`$, but has its shear components rotated by $`\varphi /2`$.
The different images pass through different parts of the strong lens, each with its own values of shear $`\gamma `$. The rotation angle will thus be different for each image. The actual value of the large scale structure shear depends on the non-linear power spectrum of dark matter at small physical scales. This is both a function of the primordial power spectrum and its slope, and non-linear gravitational physics. The most accurate models seem to be a combination of N-body simulations and heuristic models based on stable clustering. Calibrations at larger scales indicate consistency at the better than 20% level on length scales larger than about 100 kpc. In the standard models, the contribution from smaller scales is not dominant, so one would expect forecasts to be good to a factor of two (Huffenberger & Seljak 2003).
To forecast the expected RMS rotation angle, we compute the expected differential variation in the weak lens screen shear, and apply to eqs. (17,20). Mathematically, shear is a polarization field, which can be described as a trace-free spin-2 tensor field. In two spatial dimensions, any trace-free spin-2 tensor field can be decomposed into two kinematic scalars: the “divergence-like” component, also known as “E”-mode which is longitudinal to its Fourier decomposed wave vector, and a pseudo-scalar “curl-like” or “B”-mode which is transverse to its wave vector. In weak gravitational lensing, the “E”-mode is identical to the convergence field $`\kappa `$, and the “B”-mode is zero. To compute the statistics of the shear, it thus suffices to calculate the dimensionless variations in projected matter surface density. Its variation is given by the Limber equation. It involves the projection of a three dimensional non-linear power spectrum
$$\mathrm{\Delta }^2(k,z)\frac{k^3}{2\pi ^2}P(k,z)$$
(21)
to a two dimensional angular power spectrum $`l(l+1)C_l/2\pi `$ (Limber 1954; Kaiser 1992, 1998),
$$\frac{l(l+1)}{2\pi }C_l=\frac{\pi }{l}_{z_i}^{z_f}\mathrm{\Delta }^2(l/\mathrm{\Delta }\chi (z),z)w(z)^2\chi (z)\frac{d\chi }{dz}𝑑z.$$
(22)
For the foreground large scale structure, $`z_i=0,z_f=z_l,\mathrm{\Delta }\chi (z)=\chi (z)`$ where $`z_l`$ is the redshift of the strong lens. For the background large scale structure, $`z_i=z_l,z_f=z_s,\mathrm{\Delta }\chi (z)=\chi (z_f)\chi (z)`$, where $`z_s`$ is the source redshift. We used the Peacock and Dodds (1996) formulation to obtain the non-linear power from the linear transfer function given by Bardeen et al (1986). The comoving angular diameter distance is
$$D_A\chi (z)=c_0^z\frac{dz}{H(z)}$$
(23)
where H(z) is the Hubble constant at redshift z:
$$H(z)=H_0[(1+z)^2(\mathrm{\Omega }_\mathrm{m}z+1)\mathrm{\Omega }_\mathrm{\Lambda }z(z+2)]^{1/2}.$$
(24)
For the comoving angular diameter distance $`\chi `$ we used the fitting formula from Pen (1999). In a cosmological context, it is convenient to use comoving angular diameter distances and conformal time, where light rays propagate as they do in an empty universe (White & Hu 2000). The lensing weight is
$$w(z)=\frac{3}{2}\mathrm{\Omega }_\mathrm{m}H_{0}^{}{}_{}{}^{2}g(z)(1+z)$$
(25)
where
$$g(z)=\frac{[\chi (z_f)\chi (z)][\chi (z)\chi (z_i)]}{\chi (z_f)\chi (z_i)}.$$
(26)
In terms of our previous variables, for the background weak lenses we have $`D_{12}=\chi (z)\chi (z_l)`$, $`D_{2s}=\chi (z_s)\chi (z)`$, etc. The lensing weighting factor $`g`$ corresponds to the distance weighted terms in Equation (8). A similar relation holds for the background lenses.
Table 1 gives the expected rotation angle for a variety of quadruple lenses in the $`\mathrm{\Lambda }\mathrm{CDM}`$ cosmology. Several simplifying assumptions were made. We attribute all the rotation to the furthest image. The change in strong lens $`\gamma `$ between images is taken to be 0.3, which is multiplied by the large scale structure shear in equation (17). What matters is not the change in the absolute value of $`\gamma `$, but the change in each component. We took the variance of the large scale structure shear to be half of the convergence, which corresponds to the variance in $`\gamma _2`$ in the principal axis frame of the strong lens. In practice, only differences in rotation angles are observable, which depends on the alignment angles of the shear at different image positions. There is a contribution to rotation from the structure in the foreground as well as the background of the lens. The variances were simply added. The ratio of the foreground to background variance is listed in column 5 in Table 1. With all these caveats, we expect the expected rms image rotation to be good to about a factor of two, which is comparable to the expected errors on the theoretical lensing power spectrum.
## 3 Measuring rotation
We only consider the simplest case. The source is made up of three components, $`P^1,P^2,P^3`$. We define $`P^3`$ to sit at the origin, $`P^2`$ to sit at $`(P_x^2,P_y^2)`$, and $`P^1`$ at $`(P_x^1,P_y^1)`$, We can always choose our coordinate system this way.
The lensed image appears at positions A, B, C, D also with three components each. We again define the position of the third component to be the origin, and only consider relative distances. The apparent position of $`A^2`$ relative to $`A^3`$ is $`P^2=𝐃_AA^2`$, and similarly $`P^1=𝐃_AA^1`$ where the deflection matrix $`𝐃_A`$ is the inverse of the amplification matrix $`𝐀`$ defined in equation (7) for image A; each image has its own deflection matrix. It will be convenient to concatenate the two position column vectors $`P^1,P^2`$ into a 2$`\times `$2 matrix $`𝐏`$. We can write similar equations for all images, resulting in an apparent 16 equations for 16 unknowns: $`𝐏`$ each has four unknowns, and each symmetric amplification matrix has three unknowns. The equations are linear, and homogeneous:
$$𝐃_A𝐀=𝐏,𝐃_B𝐁=𝐏,𝐃_C𝐂=𝐏,𝐃_D𝐃=𝐏.$$
(27)
It is clear that one could multiply each solution by a constant and obtain a solution, so one must fix one more parameter. Without loss of generality, one could fix $`P_x^2=1`$, which just fixes a length scale. If none of the amplifications are known, the solutions are clearly degenerate between a small image that is strongly magnified and a larger image that is less magnified. A further degeneracy occurs because we can multiply each equation on the left by an arbitrary shear matrix. Since we do not know the intrinsic location of the source substructure positions, this is indistinguishable from a constant shear applied to both the lens and the source. This corresponds to a shear plane between strong lens and source. For a shear between observer and strong lens, one can simularly symmetrize the deflection matrix by multiplying both the lens and the sources by the inverse shear matrix. This is in accordance with Schneider (1997). Thus, we only measure the differences in shears at the image positions.
Then we have 16 equations for 15 unknowns, which is over-determined by one. This allows us to solve for one rotation angle. If one assumes the rotation to be dominated by the most magnified image, say $`A`$, we simply allow $`𝐃_A`$ to be non-symmetric. This allows one to solve for the rotation angle.
In general, however, the large scale structure shear has two independent components $`\gamma _1,\gamma _2`$. To make progress, one can assume them to be Gaussian distributed with standard deviation $`\sigma `$. Their one point function is independent, $`P(\gamma _1,\gamma _2)=N(\gamma _1,\sigma )N(\gamma _2,\sigma )`$. Equation (17) expresses the off-diagonal components of each deflection matrix in terms of the two large-scale structure shear components. If we fix a value of $`\gamma _1`$, we can solve Equation (27) for $`\gamma _2`$, giving us an implicit definition of $`\gamma _2(\gamma _1)`$. Then integrating over all possible values of $`\gamma _1`$ weighted by the probability gives the total likelihood for an assumed $`\sigma `$:
$$L(\sigma )=P(\gamma _1,\gamma _2(\gamma _1))𝑑\gamma _1.$$
(28)
One can then solve for the maximum likelihood value of $`\sigma `$.
## 4 Discussion
If one wishes to observe this effect, several challenges must be overcome. Sources must have at least three localizable components, and the position of each must be measured very precisely, to better than 1% of the component separation. Positions are also affected by changes in the lens from one component position to the next. The latter effect is expected to be significantly smaller than rotation, because it depends on the change in shear on small scales, while the rotation depends on the large scale structure shear itself. Basically, the rotation is of order of the large scale structure shear. The variation of the large scale structure shear on source substructure scales is smaller than the shear itself.
We quantify this as follows: let us assume that the components of the lens have a separation of 10 mas. The model assumed that the shear was constant over the apparent size of the source. The change in shear across the lens is given by $`\xi _\kappa (\mathrm{\Delta }\theta =10\mathrm{m}\mathrm{a}\mathrm{s})`$. From Figure 1 we see that $`\kappa `$ has a differential variance of around $`10^4`$ at this separation, corresponding to a percent change in lensed length scales. This is of comparable magnitude to the rotation effect, which makes it desirable to have more than three components to check. In practice, the actual lensing substructure is suppressed by several factors. Figure 1 shows the weak lensing correlation in the absence of a strong lens. Lines converge behind the lens, which reduces the corresponding physical scales, and thus the total variance. And only the variation in shear principle axes across three sub-image positions affects the rotation, which is suppressed by another dimensionless factor.
It is worth noting that HS proved that the rotation $`\omega `$ is identical to the B-mode of the shear, up to a constant. In this calculation, we only considered the local value of shears, while the E-B decomposition is non-local. We therefore should consider the 4 entries in the magnification matrix to be independent. In principle, if one had a very high number density of background sources, for example from reionization (Pen 2004), one could also solve for the strong lensing E-B decomposed map (Pen 2000). Such a decomposition also allows one to distinguish between structures in the strong lens plane, and contributions from large scale structure along the line of sight. The signal to noise required for such an exercise is difficult to achieve with current technology. Using quadruply imaged sources allows one to reduce the pairwise noise, and thus solve for the rotational component.
We also note that our calculation did not use the Born approximation. We modeled each lens plane by a quadratic potential as inferred from the $`\kappa `$ correlation function, and computed the full deflection trajectories. Recently, calculations of rotations by large scale structure at different lens planes resulted in different answers, depending on whether calculations were done with light rays (HS) or with shears on light bundles (CH). In the CH calculation, ray bundles were propagated on unperturbed trajectories, but distortions were accumulated. The distortions are relative deflections of neighboring rays, which accumulate rotations through the same effect as discussed in this paper. This allows CH to see rotation without considering the explicit deflection of the centers of each ray. In terms of a full Taylor expansion of light rays along unperturbed paths, this includes some of the effects beyond the Born approximation. HS included all leading order corrections to the Born approximation, and thus differed in their answer. In our case, the angular scales involved are very small, and we can assume the rotation to be dominated by the common large scale structure shear at different image positions. The CH and HS approaches do not differ in our model, since the potential is taken to be quadratic, and all second derivatives are constant.
Our result differ from that of Schneider (1997) in the nature of the effect. Schneider (1997) constructs an equivalent single plane strong lens from a strong lens plus constant weak shear. Our calculation circumvents the assumption that the weak lens screen is spatially constant.
Our model assumes a source plane consisting of three compact components with unknown positions. If all three components are quadruply lensed, there are a total of 12 angular positions, i.e., we have 24 observables. We subtract 8 degrees of freedom for the global lens model, because we do not know the macroscopic lens deflection angle. This leaves 16 constraints. Then using Equation (27) we can solve for the relative positions in the source plane modulo a scaling (due to the mass sheet degeneracy), which is three numbers, as well as the values of the amplification matrix at each image position, which is four sets of 3 numbers, plus one rotation, for a total of 16 parameters. This suggests that small scale structure in a single lens is observationally distinguishable from weak shear at a different redshift.
We have made several approximations, which will affect the results at some level. We used the difference in shears at the two furthest image positions. In the analysis, we then assumed that three of the images had no rotation, and only the furthest accounted for the multi-plane rotation. Reality is more complex, and all images have differential weak lensing shear, and thus some level of differential rotation. Simulations are needed to quantify this simplifying assumption.
In this paper, we have concentrated on the rotation induced by the weak large-scale structure. However, for the cases where one has multiple lenses along the line of sight, the rotation can be more significant. For the best-fit model in Chae, Mao & Augusto (2001), we find that the two Jacobian matrices are given by
$$J=\left[\begin{array}{cc}0.176& 0.093\\ 0.258& 0.595\end{array}\right],\left[\begin{array}{cc}0.888& 0.174\\ 0.148& 0.481\end{array}\right],$$
(29)
for images A and D, corresponding to magnifications $`7.78`$ and $`2.49`$ respectively. Clearly for the more highly magnified image $`A`$, the Jacobian is highly asymmetric and the rotation is quite significant.
## 5 Conclusions
We have computed the expected rotation from uncorrelated foreground and background large scale structure in strong lensing systems. We have shown that this effect is in principle observable with precise VLBI imaging of quadruply imaged lens systems. The rotation is physical and observable in the example of a source plane consisting of three point sources. If observed, it can unambiguously determine if the flux anomaly problem is caused by substructure on the lens plane, or by uncorrelated structures along the line of sight.
In addition, it demonstrates how one can extract information about small scale dark matter structure along the line of sight to a lens. A measurement of rotation would measure the variance of $`\kappa `$ from large scale structure at the smallest scales. This affects the scatter in supernovae lensing effects, and may have potential to measure the primordial power on small scales, and possibly a tilt or running spectral index.
We thank Peter Schneider for pointing out a flaw in an earlier version of the paper, T. York and K.-H. Chae for helpful discussions, and Pengjie Zhang for the lensing Limber code.
|
warning/0506/cond-mat0506592.html
|
ar5iv
|
text
|
# Effect of Two-Dimensionality on Step Bunching Induced by the Drift of Adatoms
## I Introduction
The Si(001) surface is reconstructed by dimerization of surface atoms. When its vicinal face is tilted in the $`110`$ direction, the terraces with dimer rows parallel to the steps (we call $`\mathrm{T}_\mathrm{A}`$) and those with dimer rows perpendicular to the steps (we call $`\mathrm{T}_\mathrm{B}`$) appear alternately. Since the surface diffusion along the dimer rows is faster than that perpendicular to the dimer rows, the anisotropy of the surface diffusion changes alternately on consecutive terraces.
On the vicinal face, two types of step instabilities, step wandering and step bunching, occur when a specimen is heated by direct electric current livin-kl91 ; Latyshev-la98ass . The step wandering occurs with step-up current in the region of relatively large inclination (the tilting angle is $`0.08^{}\theta 0.5^{}`$Nielsen-pp01ss . Due to the step wandering, grooves perpendicular to the steps appear on the vicinal face. The step bunching occurs irrespective of the current direction in the region of small inclination ($`\theta 0.08^{}`$). The type of dominant terraces, which separate step bunches, is $`\mathrm{T}_\mathrm{B}`$ with step-down current and $`\mathrm{T}_\mathrm{A}`$ with step-up current. The size of bunch increases with time as $`t^{1/2}`$ Latyshev-la98ass , which is independent of the drift direction. The growth rate of the bunches with step-down current seems slightly slower than that with step-up current.
The step instabilities are caused by drift of adatoms induced by the current. By taking account of alternation of the anisotropic surface diffusion, the step instabilities are theoretically explained. If the repulsive interaction is strong so that the step bunching is suppressed, the step wandering occurs with step-up drift Sato-ushprb03 , and straight grooves parallel to the drift appears by in-phase wandering. The motion of the steps is given by the solution of the nonlinear equation derived by Pierre-Louis and co-workers Pierre-Louis-mskp98prl . The step bunching occurs irrespective of the direction of the drift Stoyanov-90jjap ; Natori-fy92jjap ; Natori-ff92ass ; Sato-us02jcg ; Sato-umh03jpsj ; Sato-muh04jpsj . Since the current and the drift are in the same direction Ichikawa-doi92apl ; Metois-hp99ss , the results are consistent with the experiments livin-kl91 ; Latyshev-la98ass ; Nielsen-pp01ss . In a one-dimensional step flow model Sato-muh04jpsj , the size of bunches increases with time as $`t^\beta `$ with $`\beta `$ slightly smaller than $`1/2`$, consistent with the experiment Latyshev-la98ass . However, the growth rate with step-up drift is slower than that with step-down drift since terraces with fast diffusion in $`y`$-direction are dominant with step-down drift. This is in contradiction with the experiment Latyshev-la98ass . Since the model with alternating diffusion anisotropy has explained the bunching and the wandering instabilities on the Si(001) vicinal face consistently, this disagreement is a major obstacle to the unified understanding.
In the one-dimensional step flow models Sato-umh03jpsj ; Sato-muh04jpsj , the motion of step bunches with step-down drift is similar to that with step-up drift except the time scale. In the Monte Carlo simulation, however, the step pattern is changed by the drift direction Sato-us02jcg ; Sato-ush03prb : the step bunches with step-up drift wander more than that with step-down drift. Such a difference of the step motion in two-dimension may solve the disagreement in the growth rate between the experiment Latyshev-la98ass and the one-dimensional model Sato-umh03jpsj . In this paper, we carry out Monte Carlo simulations and show that the difference of the growth rate vanishes in the two-dimensional model.
## II Model
For simplicity, we use a square lattice model with the lattice constant $`a=1`$. We take $`x`$-axis parallel to the steps and $`y`$-axis in the down-hill direction. Boundary conditions are periodic in the $`x`$-direction and helical in the $`y`$-direction. We forbid two-dimensional nucleation and use solid-on-solid steps, i.e. the step positions are single valued functions of $`x`$.
We repetitively select a solid atom at the step or an adatom on the terrace. We perform the diffusion and solidification trial for the adatom and melting trial for solid atom. In the diffusion trial, the adatom hops to a neighboring site. The anisotropy of the diffusion coefficient and the drift of adatoms are taken into account in the hopping probability. On $`\mathrm{T}_\mathrm{A}`$, where the surface diffusion in the $`x`$-direction is faster, an adatom on the site $`(i,j)`$ moves to $`(i\pm 1,j)`$ with the probability $`1/4`$ and to $`(i,j\pm 1)`$ with the probability $`p_\mathrm{d}(1\pm Fa/k_\mathrm{B}T)/4`$, where $`p_\mathrm{d}(<1)`$ is the ratio of the two diffusion coefficients. $`F`$ is the force to cause the drift. $`F>0`$ represents the drift in the down-hill direction. On $`\mathrm{T}_\mathrm{B}`$, where the surface diffusion in the $`y`$-direction is faster, an adatom on the site $`(i,j)`$ moves to $`(i\pm 1,j)`$ with the probability $`p_\mathrm{d}/4`$ and to $`(i,j\pm 1)`$ with the probability $`(1\pm Fa/k_\mathrm{B}T)/4`$. For a diffusion trial, the time increment is $`\mathrm{\Delta }t=1/4N_\mathrm{a}`$, where $`N_\mathrm{a}`$ is the number of adatoms so that the fast diffusion coefficient is unity.
If the adatom comes in contact with a step from the lower terrace after a diffusion trial, solidification occurs with the probability
$$p_\mathrm{s}=\left[1+\mathrm{exp}\left(\frac{\mathrm{\Delta }E_\mathrm{s}\varphi }{k_\mathrm{B}T}\right)\right]^1.$$
(1)
$`\mathrm{\Delta }E_\mathrm{s}`$ is given by $`\mathrm{\Delta }E_\mathrm{s}=ϵ\times `$(the increment of the step perimeter) and $`\varphi `$ is decrement of the chemical potential by solidification. The step stiffness $`\stackrel{~}{\beta }`$ is related to $`ϵ`$ as
$$\frac{2\stackrel{~}{\beta }}{k_\mathrm{B}T}=\mathrm{sinh}^2\frac{ϵ}{k_\mathrm{B}T}.$$
(2)
If we select a solid atom, melting trial is performed. When an adatom is absent on the top of the solid atom, melting occurs with the probability
$$p_\mathrm{m}=\left[1+\mathrm{exp}\left(\frac{\mathrm{\Delta }E_\mathrm{s}+\varphi }{k_\mathrm{B}T}\right)\right]^1.$$
(3)
There is no extra diffusion barrier over the steps: steps are permeable.
## III Results of simulation
In our simulations, we neglect the long-range repulsive interaction between steps, but take into account a short-range repulsive interaction by forbidding overlap of steps. Impingement of atoms and evaporation are absent.
Figures 1 and 2 represent snapshots of the step bunching. System size is $`256\times 256`$ and the number of steps is $`64`$. The parameters are $`ϵ/k_\mathrm{B}T=1.0`$, $`\varphi /k_\mathrm{B}T=1.5`$, $`Fa/k_\mathrm{B}T=\pm 0.08`$ and $`p_\mathrm{d}=0.5`$. Initially a few adatoms are present on the vicinal face. The dotted lines represent S<sub>A</sub> steps and the solid lines represent S<sub>B</sub> steps.
The vicinal face is unstable with the drift of adatoms. Pairing of $`\mathrm{S}_\mathrm{A}`$ and $`\mathrm{S}_\mathrm{B}`$ occurs in the initial stage. The upper side step in a pair is $`\mathrm{S}_\mathrm{A}`$ with step-up drift and $`\mathrm{S}_\mathrm{B}`$ with step-down drift. Small bunches are formed by coalescence of step pairs. Since the stiffness is small, the bunches wander and connect with each other at many places (Fig. 1). From the figures one may have impression that he step with step-down drift are more straight.
The effect of the drift direction on the form of bunches becomes evident in a late stage (Fig. 2). With step-down drift, the bunches are straight and there are few recombination of bunches. With step-up drift, the wandering width of the bunches is large. The bunches collide with each other and frequent recombination is seen. The difference of the form may affect the time evolution of bunch size.
To test the effect of the wandering and recombination on the growth rate, we carry out simulations with a narrow system (Fig. 3). The number of steps is $`128`$ and the system size is $`16\times 512`$. We use the parameter $`ϵ/k_\mathrm{B}T=0.5`$ for step bunches to wander easily. Other parameters are the same as that in Fig. 2. The wandering of step bunches with step-up drift is suppressed because of the narrow system width. The step bunches are straight irrespective of the drift direction.
In Figure 3, the number $`N_{\mathrm{max}}`$ of steps in the largest bunch at $`x=1`$ is plotted as a function of time. The bunches grow by the collisions of straight bunches due to the fluctuation of position of bunches Sato-umh03jpsj ; Sato-muh04jpsj . The growth rate with step-up drift is slower than that with step-down drift like the one-dimensional model. $`N_{\mathrm{max}}`$ seems to increase in the power law, $`N_{\mathrm{max}}t^\beta `$ with $`\beta 0.4`$ criterion , which is consistent with the one-dimensional result with a combination of $`\mathrm{ln}r`$ and $`r^2`$ potentials Sato-muh04jpsj .
We carry out simulations with a large system size: $`512\times 512`$ (Fig. 4). In contrast to the one-dimensional model Sato-muh04jpsj , the growth of bunch size with step-up drift is as fast as that with step-down drift. Thus the slow diffusion in $`y`$-direction is compensated by the efficient coalescence by the wandering of bunches.
In the experiment Latyshev-la98ass , however, the step bunching with step-up current seems slightly faster than that with step-down current. Additional mechanism to accelerate the bunching with step-up drift may be required. On the Si(001) vicinal face, $`\mathrm{S}_\mathrm{B}`$ steps are rougher than $`\mathrm{S}_\mathrm{A}`$ steps and the kinetic coefficient is probably larger for $`\mathrm{S}_\mathrm{B}`$. We take account of the difference as the probabilities of solidification and melting. At $`\mathrm{S}_\mathrm{A}`$, the probability of solidification is assumed to be $`rp_\mathrm{s}`$ and that of melting is $`rp_\mathrm{m}`$ with $`r1`$.
Figure 5 represents the time evolution of bunch size with $`r=0.1`$. Other parameters and system size are the same as those in Fig. 4. The growth exponent does not change significantly. The growth rate with step-down drift is suppressed as expected from the reduction factor $`r`$, but that with step-up drift is slightly enhanced.
In figure 6 we show that the step patterns. As explained already, the bunches with step-down drift are straight. The bunches with step-up drift are wavy and shows more recombination patterns. The reduction of kinetic coefficient of $`\mathrm{S}_\mathrm{A}`$ seems to enhance the waviness. As a result, the growth rate with step-up drift becomes faster than that with step-down drift, which is in agreement with the experiment Latyshev-la98ass .
## IV Summary and Discussion
We studied the time evolution of bunch size by Monte Carlo simulation. With step-down drift, the step bunches are straight and there are few recombination of bunches. With step-up drift, the bunches wander and recombinations of bunches occur frequently
In the one-dimensional model Sato-umh03jpsj ; Sato-muh04jpsj , the velocity of a bunch is roughly proportional to the diffusion coefficient of $`y`$-direction in large terraces. Bunches with step-down drift grow faster than those with step-up drift, which is in disagreement with the experiment Latyshev-la98ass . In the two-dimensional model, bunches with step-up drift wander and collide with each other more frequently than those with step-down drift. Since coalescence of bunches starts from the connected parts, the step bunching with step-up drift occurs more frequently than that with step-down drift. Also the fast transverse diffusion with step-up drift helps recombination of bunches. Consequently the growth rate of bunches with step-up drift is as fast as that with step-down drift.
When the difference of the kinetic coefficient is taken into account, the bunches with step-up drift grows faster than those with step-down drift. The reversal of the growth rate is thus attributed to the two-dimensional step pattern. Since we have not succeeded in quantifying the effect, it is not clear if the proposed effect is sufficient to explain the experiment Latyshev-la98ass .
###### Acknowledgements.
This work was supported by Grant-in-Aid for Scientific Research from Japan Society for the Promotion of Science. M. U. and Y. S. benefited from the inter-university cooperative research program of the Institute for Materials Research, Tohoku University.
|
warning/0506/hep-ph0506252.html
|
ar5iv
|
text
|
# Status Report of NNLO QCD Calculations
## 1 Introduction
Despite the fact that the theory of strong interactions, quantum chromodynamics (QCD), is today a well-established part of the Standard Model of particle physics, it continues to be an extremely active field of experimental and theoretical research. This is as true for non-perturbative (and often lattice) determinations of meson and baryon spectra, decay constants, or form factors for electric dipole moments or $`B`$-meson decays, as it is for perturbative calculations. Today, the experimental precision of high-energy collider data always requires calculations beyond the leading order (LO) in the strong coupling constant, $`\alpha _s(\mu )`$, and often even beyond the next-to-leading order (NLO) for reliable comparisons and determinations of unknown Standard Model parameters.
After a relatively slow start due to large technical difficulties in the mid-1990s, QCD calculations at next-to-next-to-leading order (NNLO) have recently attained their goals for several phenomenological applications. The aim of this Report is therefore to review recent progress in this field with special emphasis on phenomenologically relevant results, including structure functions, jet, and Higgs boson production. Very recently, twistor methods have received particular attention, as they may offer a route to efficient multiparticle calculations, even beyond tree-level. They will therefore be briefly discussed, before we present our conclusions.
## 2 Structure Functions
Since its start-up in 1992, the DESY $`ep`$ collider HERA has provided a wealth of data on the structure functions of the proton, and thus its parton densities (PDFs), in a large region of Bjorken-$`x`$ and $`Q^2`$ Behnke . The PDFs represent at the same time an important ingredient in the search for new physics, that is currently underway at the Fermilab Tevatron and is an important research goal at the CERN LHC Roeck . In particular, QCD uncertainties in PDF determinations, e.g. from renormalization and factorization scale and scheme Klasen:1996yk ; Martin:2004ir variations, have to be under control before disagreement between theory and experiment can be interpreted as evidence for new physics.
It is thus fortunate that, after completion of the three-loop singlet splitting functions (obtained from the $`1/ϵ`$-poles in dimensional regularization) and longitudinal coefficient function (obtained from the finite terms) Vogt:2004mw , a full NNLO calculation of the structure functions $`F_{1,2}(x,Q^2)`$ (and thus also of $`F_L=F_22xF_1)`$, is now available. These results have been obtained using a large variety of techniques, such as application of the optical theorem, transformation to Mellin space, mapping of diagrams with composite topologies to basic building blocks, and integration by parts. They also relied on recent advances in mathematics (harmonic sums) and computer algebra (QGRAF, FORM). Fortunately, the results can also be applied to other physical observables such as photon structure functions and total cross sections in $`e^+e^{}`$-annihilation.
While the numerical effects of the NNLO QCD corrections are well visible in splitting functions and coefficient functions separately and show clear differences in normalization, although not in shape, from earlier leading-$`\mathrm{ln}(x)`$ estimates (see Fig. 1), the
total effect in the convoluted structure functions is only important for very small or large $`x`$ and small $`Q^2`$. For this reason, the NNLO corrections have not yet been included in the global CTEQ analysis, as the authors find sufficient stability of their fit at NLO Huston:2005jm , whereas the MRST collaboration have now implemented the exact NNLO splitting functions, but find minor effects w.r.t. a DIS-scheme motivated variation of the input gluon density Martin:2004ir or an earlier analysis using approximate NNLO splitting functions Martin:2002dr .
## 3 Jet production in electron-positron annihilation
Calculations of observables less inclusive than deep inelastic scattering (DIS) structure functions and total $`e^+e^{}`$ cross sections do not allow for a straight-forward application of the optical theorem, as they require multiparticle cuts in the three-loop photon propagator. For this reason, NNLO jet calculations in $`e^+e^{}`$-annihilation have been lagging behind, although recently considerable progress has also been achieved in this field. The interference terms of $`13`$ tree-level and two-loop amplitudes are now known, as are the single soft and/or collinear regions of $`14`$ squared one-loop amplitudes. What remains to be calculated are the maximally double-soft and triple-collinear regions of $`15`$ squared tree-level amplitudes. Here, antenna functions have successfully been used for quark-gluon and gluon-gluon final states Gehrmann-DeRidder:2005hi .
A first preliminary numerical result has been obtained for the NNLO contribution to the $`C_F^2`$ color class of the average thrust Gehrmann-DeRidder:2004xe
$`1T`$ $`=`$ $`{\displaystyle (1T)\frac{1}{\sigma _0}\frac{d\sigma }{dT}}=C_F\left[\left({\displaystyle \frac{\alpha _s}{2\pi }}\right)A+\left({\displaystyle \frac{\alpha _s}{2\pi }}\right)^2B+\left({\displaystyle \frac{\alpha _s}{2\pi }}\right)^3C+\mathrm{}\right],`$
where $`A=1.57`$ and $`B=32.3`$ were known and now $`C(C_F^2)=20.4\pm 4`$. The full result will be important for reliable estimates of higher-twist effects in $`e^+e^{}`$ event shapes. Furthermore, analytic continuation allows to apply the results obtained for three-jet production in $`e^+e^{}`$-annihilation also to dijet production in DIS.
## 4 Higgs boson production at hadron colliders
The inclusive NNLO cross section for light/heavy scalar and pseudo-scalar Higgs production in $`gg`$ scattering at hadron colliders has been known for several years, but the relevance of these calculations has been limited by the need for experimental cuts on associated photons, jets or $`b`$-quarks. Fully differential cross sections, such as a recent NNLO calculation for Higgs plus dijet production Anastasiou:2004xq , allow for transverse energy vetos on the additional jets and thus for a suppression of the QCD background in the diphoton (or other) Higgs decay channels (see Fig. 2).
## 5 Jet hadroproduction and twistor methods
For hadron-hadron scattering, the two-loop $`22`$ parton amplitudes have been known for some time, and the result for polarized $`qqqq`$ scattering has recently been confirmed using supersymmetric methods DeFreitas:2004tk . Since the soft/collinear singular regions of one-loop $`23`$ amplitudes are also known, recent work has focused on multiple soft/collinear emissions in tree-level multiparton amplitudes and on extending the dipole subtraction formalism to NNLO, where two alternatives to previously known methods have been proposed Frixione:2004is ; Somogyi:2005xz . Although the subtraction terms are in principle known, the singular regions often overlap and need to be disentangled to avoid double-counting. Furthermore, phase space factorization and integration of the singular matrix elements still needs to be resolved. For this reason, further progress towards a full NNLO calculation for jet hadroproduction has been slow, and numerical integration methods using sector decomposition Binoth:2004jv may have to be developped further.
Multiparticle amplitudes may be calculated efficiently by relating perturbative gauge theory to the $`D`$-instanton expansion for a topological string in twistor space. While originally motivated by $`N=4`$ super Yang-Mills theory, this approach has been shown to work at tree-level for generic QCD amplitudes. $`N=4`$ supersymmetric and cut-constructible one-loop diagrams have also been calculated, and extensions to non-supersymmetric theories and two-loop amplitudes are underway Dixon .
## 6 Conclusion
Although this brief Report is necessarily far from complete, it should be clear that NNLO QCD calculations represent a theoretically challenging, phenomenologically important, and fast-moving field of research allowing for interesting advances in the fields of mathematics, computer science and last, but not least, elementary particle physics.
The author thanks the working group convenors for the kind invitation, the DoE’s INT at the University of Washington for kind hospitality and partial financial support during preparation of this work, and L. Dixon and T. Teubner for useful discussions.
|
warning/0506/math0506499.html
|
ar5iv
|
text
|
# On the Kashiwara-Vergne conjecture
## 1. Introduction
Let $`G`$ be a connected Lie group, with Lie algebra $`𝔤`$. Let $`U(𝔤)`$ denote the universal enveloping algebra, and $`S(𝔤)`$ the symmetric algebra. The Duflo map is the isomorphism of $`𝔤`$-modules,
(1)
$$\mathrm{Duf}:S(𝔤)U(𝔤),$$
obtained by precomposing the symmetrization map with an infinite-order differential operator $`\widehat{J^{1/2}}`$ on the symmetric algebra (viewed as the space of polynomials on $`𝔤^{}`$). Here $`JC^{\mathrm{}}(𝔤)`$ is the analytic function
(2)
$$J(x)=\underset{𝔤}{det}\left(\frac{1e^{\mathrm{ad}_x}}{\mathrm{ad}_x}\right),$$
and $`\widehat{J^{1/2}}:S(𝔤)S(𝔤)`$ is obtained from the Taylor series expansion of $`J^{1/2}`$ by replacing the variables $`x`$ with derivatives $`\frac{}{\mu }`$. Duflo’s theorem states that the map $`\mathrm{Duf}`$ restricts to an *algebra* isomorphism on invariants, $`S(𝔤)^𝔤U(𝔤)^𝔤`$. In Kashiwara and Vergne proposed a conjecture regarding the Campbell-Hausdorff series, which among other things generalizes the Duflo theorem to germs of invariant distributions on the Lie group $`G`$. Here $`S(𝔤)𝒟_{\mathrm{comp}}^{}(𝔤)`$ is identified with the subalgebra (under convolution) of distributions on $`𝔤`$ supported at $`0`$, while $`U(𝔤)`$ is identified with the subalgebra (under convolution) of distributions on $`G`$ supported at $`e`$. This generalization of Duflo’s theorem was proved later in a series of papers using the diagrammatic technique of Kontsevich . The KV conjecture itself was established for $`𝔤`$ solvable , for $`𝔤=𝔰𝔩(2,)`$ in , and in the case of $`𝔤`$ quadratic (that is, $`𝔤`$ carries an invariant nondegenerate symmetric bilinear form) in (see also ).
There are several equivalent formulations of the KV conjecture , one of which is as follows (see Section 2 for details). Let $`𝔤_t`$ denote the family of Lie algebras, obtained from $`𝔤_1=𝔤`$ by rescaling the Lie bracket as $`[,]_t=t[,]`$, and let $`m_t`$ denote the family of products on $`S(𝔤)`$, induced from the products on $`U(𝔤_t)`$ by the Duflo isomorphisms $`\mathrm{Duf}_t:S(𝔤)U(𝔤_t)`$. Thus, $`m_t`$ interpolates between the usual commutative product $`m_0`$ on $`S(𝔤)`$ and the non-commutative product coming from $`U(𝔤)`$. The product maps $`m_t`$ have a natural extension to the space of distributions on $`𝔤`$, with compact support in a sufficiently small open neighborhood $`O`$ of $`0`$,
(3)
$$m_t:𝒟_{\mathrm{comp}}^{}(O^2)𝒟_{\mathrm{comp}}^{}(𝔤).$$
Let $`e_i`$ be a basis of $`𝔤^2=𝔤𝔤`$, and let $`L(e_i)`$ be the Lie derivatives relative to the adjoint action of $`𝔤^2`$ on $`𝒟_{\mathrm{comp}}^{}(O^2)`$. The Kashiwara-Vergne conjecture asserts the existence of an analytic map $`\beta =\beta ^ie_i:O^2𝔤^2`$, vanishing at the origin, such that
(4)
$$\frac{dm_t}{dt}=m_t\underset{i}{}\beta _t^iL(e_i),$$
where $`\beta _t^i(x,y)=t^1\beta ^i(tx,ty)`$. Equation (4) implies that the family of products $`m_t`$ is independent of $`t`$ when restricted to germs of invariant distributions.
In a recent paper, Torossian came very close to a proof of the KV conjecture, again using Kontsevich diagram techniques. In more detail, Torossian proved an analogue of Equation (4), for a certain family of maps $`\stackrel{~}{m}_t:𝒟_{\mathrm{comp}}^{}(O^2)𝒟_{\mathrm{comp}}^{}(𝔤)`$ with $`\stackrel{~}{m}_0=m_0,\stackrel{~}{m}_1=m_1`$. However, $`\stackrel{~}{m}_tm_t`$, and indeed the products defined by $`\stackrel{~}{m}_t`$ are not associative.
Our first main result shows that one obtains a solution $`\beta _t`$ of the KV equation (4) from a solution $`\stackrel{~}{\beta }_t`$ of Torossian’s modified KV property, by solving an interesting ‘zero curvature’ equation. This establishes the Kashiwara-Vergne conjecture in full generality.
Our second result relates the KV property to Lie algebra cohomology, as follows. For any $`𝔤`$-module $`X`$, let $`H^{}(𝔤,X)`$ denote the Lie algebra cohomology of $`𝔤`$ with coefficients in $`X`$. By functoriality, the Duflo map (1) induces a homomorphism of graded vector spaces,
(5)
$$H^{}(𝔤,S(𝔤))H^{}(𝔤,U(𝔤)).$$
In , Shoikhet announced a joint result with Kontsevich that the linear map (5) is an isomorphism of graded algebras. Shortly later, a proof was published by Pevzner-Torossian . In his paper, Shoikhet raises the question \[13, Remark 3.2\] whether this result would follow from the KV conjecture. We show that this is indeed the case: Let
(6)
$$M_t:𝒟_{\mathrm{comp}}^{}(O^2)(𝔤^2)^{}𝒟_{\mathrm{comp}}^{}(𝔤)𝔤^{}$$
denote the family of products on Chevalley-Eilenberg complexes, induced by the maps $`m_t`$ (3). We prove that $`\beta :O^2𝔤^2`$ solves the KV conjecture if and only if
(7)
$$\frac{dM_t}{dt}=[\text{d},M_t\underset{i}{}\beta _t^i\iota (e_i)].$$
Here d is the Chevalley-Eilenberg differential, and $`\iota (e_i)`$ are contraction operators. Equation (7) implies that the product maps $`M_t`$ are chain homotopic. Hence, the induced map in cohomology is independent of $`t`$.
The organization of this paper is as follows. In Section 2, we give three different formulations I,II,III of the KV conjecture. The algebraic formulation III is the main version of the conjecture, as stated in \[8, page 250\]. The other two versions are more geometric. They are implicit in \[8, Section 3\], although we add some details to show that they are indeed equivalent to III. In Section 3 we show that Torossian’s theorem implies the KV conjecture. Section 4 gives yet another reformulation IV of the KV conjecture, as a homotopy formula for Chevalley-Eilenberg complexes. In this context, we formulate a new conjecture extending the Kashiwara-Vergne approach to higher homotopies.
Notational conventions. For any manifold $`M`$, we denote by $`𝔛(M)`$ the Lie algebra of vector fields (viewed as derivations of $`C^{\mathrm{}}(M)`$), and by $`\mathrm{\Omega }^{}(M)`$ the algebra of differential forms. We follow the convention that the flow $`F_t`$ of a time dependent vector field $`X_t`$ is given by $`(X_tf)(F_t^1(x))=\frac{}{t}f(F_t^1(x))`$. The Lie derivative on $`\mathrm{\Omega }(M)`$ is thus given by $`F_t^{}L_{X_t}=\frac{}{t}F_t^{}`$. A family of forms $`\alpha _t`$ is intertwined by the flow if $`F_t^{}\alpha _t`$ is independent of $`t`$; in terms of the Lie derivative this means $`(\frac{}{t}L(X_t))\alpha _t=0`$.
Acknowledgments. We would like to thank C. Torossian for explaining his work . We are grateful to M. Vergne for comments, and to H. Maire and E. Petracci for helpful discussions. We are indebted to the referees of this paper for valuable remarks. The research of A.A. was supported in part by the Swiss National Science Foundation.
## 2. Review of the Kashiwara-Vergne conjecture
### 2.1. Preliminaries
Throughout this paper, $`r_t^V`$ will denote scalar multiplication by a real parameter $`t`$ on some vector space $`V`$. We omit the superscript if $`V`$ is clear from the context. We will frequently use the following elementary fact. Suppose $`R_t:WW,0<t1`$ are automorphisms of a topological vector space, with $`R_{t_1t_2}=R_{t_1}R_{t_2}`$. Assume that $`wW`$ be a smooth vector, i.e. $`w_t=R_t(w)`$ is differentiable in $`t`$. Then the derivative $`\dot{w}_t=\frac{w_t}{t}`$ satisfies
(8)
$$\dot{w}_t=t^1R_t(\dot{w}_1).$$
For example, suppose $`O`$ is a star-shaped open subset of a vector space $`V`$, and take $`W`$ to be the space of $`k`$-forms on $`O`$. Let $`\alpha \mathrm{\Omega }^k(O)`$ and $`a`$. Then the derivative of $`\alpha _t:=t^ar_t^{}\alpha `$ scales according to $`\dot{\alpha }_t=t^{a1}r_t^{}\dot{\alpha }_1`$.
### 2.2. The Duflo map on distributions
Let $`𝔤`$ be a finite-dimensional Lie algebra over $``$, and $`G`$ the corresponding connected, simply connected Lie group. Let $`\mathrm{exp}:𝔤G`$ denote the exponential map. The Jacobian $`JC^{\mathrm{}}(𝔤)`$ of the exponential map (using left trivialization of the tangent bundle of $`G`$) is given by formula (2). Let $`𝒟_{\mathrm{comp}}^{}(𝔤)`$ denote the commutative algebra of compactly supported distributions on $`𝔤`$, with product given by convolution. The symmetric algebra $`S(𝔤)`$ is identified with the subalgebra of distributions supported at $`0`$. Similarly, let $`𝒟_{\mathrm{comp}}^{}(G)`$ denote the non-commutative convolutions algebra of distributions on $`G`$, and identify $`U(𝔤)`$ with distributions supported at the group unit $`eG`$. Assume temporarily that $`J`$ admits a global square root $`J^{1/2}`$, equal to $`1`$ at the origin. (This is not the case for all $`G`$, but holds true, for instance, if $`𝔤`$ admits an invariant non-degenerate symmetric bilinear form.) Then the Duflo map $`\mathrm{Duf}:S𝔤U𝔤`$ extends to a homomorphism of $`𝔤`$-modules,
(9)
$$\mathrm{Duf}=\mathrm{exp}_{}J^{1/2}:𝒟_{\mathrm{comp}}^{}(𝔤)𝒟_{\mathrm{comp}}^{}(G),$$
and again this is an algebra homomorphism on invariants. For a general non-compact Lie group, this extension may not be very interesting since there may be very few invariant distributions of compact support. However, one obtains an interesting generalization by working instead with *germs* of invariant distributions. Let $`𝔇^{}(𝔤,A)`$ denote the space of germs (at the origin $`0𝔤`$) of distributions on $`𝔤`$, with asymptotic support <sup>1</sup><sup>1</sup>1Let $`x`$ be a manifold, and $`𝔲`$ the germ of a distribution at $`xM`$. The (asymptotic) support of $`𝔲`$ is the closed subset $`\mathrm{supp}_x(𝔲)T_xM`$, defined as follows: Let $`\gamma (t)`$ denote a smooth curve with $`\gamma (0)=x`$. Then the tangent vector $`\dot{\gamma }(0)\mathrm{supp}_x(𝔲)`$ if and only if $`\gamma (t)\mathrm{supp}(𝔲)`$ for $`t>0`$ sufficiently small. in the closed cone $`A𝔤`$. Let $`𝔇^{}(G,A)`$ be the space of germs of distributions at $`eG`$, given as the image of $`𝒟^{}(𝔤,A)`$ under $`\mathrm{exp}_{}`$. For closed cones $`A_1,A_2𝔤`$ with $`A_1A_2=\{0\}`$, there are well-defined convolution products
(10)
$$𝔇^{}(𝔤,A_1)𝔇^{}(𝔤,A_2)𝔇^{}(𝔤,A_1+A_2),$$
and
(11)
$$𝔇^{}(G,A_1)𝔇^{}(G,A_2)𝔇^{}(G,A_1+A_2).$$
For instance, these products are well-defined if $`A_1=\{0\}`$ or $`A_2=\{0\}`$. The natural extension of Duflo’s theorem asserts that the Duflo map intertwines (10) (restricted to $`𝔤`$-invariants), with (11) (restricted to $`𝔤`$-invariants). This extension of Duflo’s theorem was put forward by Kashiwara-Vergne in their paper, and proved for the case of solvable Lie algebras. The general case was proved many years later in .
### 2.3. The Kashiwara-Vergne approach
The Kashiwara-Vergne approach is based on the idea that the Duflo theorem (and its extension to germs of distributions) should follow from a more general property of the Campbell-Hausdorff series
(12)
$$\mathrm{\Phi }(x,y):=\mathrm{log}(\mathrm{exp}(x)\mathrm{exp}(y))=x+y+\frac{1}{2}[x,y]+\mathrm{}.$$
Recall that the function $`\mathrm{\Phi }(x,y)`$ is well-defined and analytic for $`x,y`$ in a sufficiently small neighborhood of the origin. To be specific, let $`U𝔤`$ be a star-shaped neighborhood of the origin such that the exponential map is a diffeomorphism over $`U`$, and let $`OU`$ be a smaller star-shaped open neighborhood such that $`\mathrm{exp}(x)\mathrm{exp}(y)\mathrm{exp}(U)`$ for all $`x,yO`$. Then $`\mathrm{\Phi }:O^2𝔤`$ is well-defined and takes values in $`U`$. Since $`J>0`$ on $`U`$, the square root $`J^{1/2}`$ is a well-defined analytic function on $`U`$. Under the Duflo map, the convolution product on $`𝒟_{\mathrm{comp}}^{}(G)`$ gives rise to a non-commutative product on distributions on $`𝔤`$ with support in $`O`$. In terms of the map $`\mathrm{\Phi }`$ this product can be written,
(13)
$$m=\mathrm{\Phi }_{}\kappa :𝒟_{\mathrm{comp}}^{}(O^2)𝒟_{\mathrm{comp}}^{}(𝔤)$$
where $`\kappa C^{\mathrm{}}(O^2)`$ is the function,
(14)
$$\kappa (x,y)=\frac{J^{1/2}(x)J^{1/2}(y)}{J^{1/2}(\mathrm{\Phi }(x,y))}.$$
Denote by $`𝔤_t`$ the Lie algebra, obtained from $`𝔤`$ by rescaling the Lie bracket,
(15)
$$[,]_t=t[,].$$
Let $`G_t`$ denote the connected, simply connected Lie group with Lie algebra $`𝔤_t`$. For $`t0`$, the map $`r_t:𝔤𝔤`$ defines a Lie algebra isomorphism $`𝔤_t𝔤_1`$, while $`𝔤_0`$ is the vector space $`𝔤`$ with the zero bracket. The exponential map $`\mathrm{exp}_t:𝔤_tG_t`$ is a diffeomorphism over $`U_t=r_{t^1}U`$, and the counterpart $`\mathrm{\Phi }_t:O_t𝔤`$ of the map $`\mathrm{\Phi }`$ is well-defined over the rescaled neighborhood $`O_t=r_{t^1}O`$. One has $`\mathrm{\Phi }_t=t^1r_t^{}\mathrm{\Phi }`$ for $`t0`$, i.e.
(16)
$$\mathrm{\Phi }_t(x,y)=x+y+\frac{t}{2}[x,y]+\mathrm{},$$
while $`\mathrm{\Phi }_0(x,y)=x+y`$. The Jacobian of $`\mathrm{exp}_t`$ is $`J_t=r_t^{}J`$. From the family of Lie brackets $`[,]_t`$, we obtain a family of multiplication maps $`m_t=(\mathrm{\Phi }_t)_{}\kappa _t`$ with $`\kappa _t=r_t^{}\kappa `$. For $`t0`$,
(17)
$$m_t=(r_{t^1})_{}m_1(r_t)_{}.$$
The map $`m_0`$ is the commutative product given by convolution on the vector space $`𝔤`$. The $`t`$-derivative $`\dot{m}_t=\frac{m_t}{t}`$ scales according to
(18)
$$\dot{m}_t=t^1(r_{t^1})_{}\dot{m}_1(r_t)_{}.$$
The generalized Duflo theorem is equivalent to the statement that the derivative $`\dot{m}_t`$ vanishes on $`𝒟^{}(𝔤\times 𝔤,A_1\times A_2)^{𝔤\times 𝔤}`$, where $`A_1(A_2)=\{0\}`$. Kashiwara-Vergne formulated the following stronger property of $`m_t`$.
### 2.4. The Kashiwara-Vergne conjecture
Let $`e_i`$ be a basis of $`𝔤^2`$, and let $`L_i=L(e_i)`$ be the Lie derivatives for the $`𝔤^2`$-action on $`𝒟_{\mathrm{comp}}^{}(O^2)`$.
Kashiwara-Vergne conjecture, version I: Taking $`O`$ smaller if necessary, there exists a $`𝔤`$-equivariant smooth function $`\beta =_i\beta ^ie_i:O^2𝔤^2`$, with $`\beta (0,0)=0`$, such that $`\beta _t=t^1r_t^{}\beta `$ satisfies,
(19)
$$\frac{m_t}{t}=m_t\underset{i}{}\beta _t^iL_i.$$
Duflo’s theorem, as well as its extension to germs of invariant distributions, are a direct consequence of (19) since the right hand side vanishes on invariants.
###### Remark 2.1.
It is not difficult to show that $`\dot{m}_t`$ can be written in the form
(20)
$$\dot{m}_t=m_tP_t$$
where $`P_t`$ are first order differential operators on $`𝒟_{\mathrm{comp}}^{}(O^2)`$. In light of Duflo’s theorem, the idea that one can take $`P_t`$ of the form $`_i\beta _t^iL_i`$ may therefore appear rather natural. Note also that the operator $`_i\beta _t^iL_i`$ on $`𝒟^{}(O^2)`$ is different, in general, from the Lie derivative $`L(X^{\beta _t})`$ in the direction of the vector field $`X^{\beta _t}`$ generated by $`\beta _t`$, since on distributions $`L(X^{\beta _t})=_iL_i\beta _t^i`$.
Another version of the conjecture states the KV property in terms of the derivatives of the function $`\mathrm{\Phi }_t`$ and $`\kappa _t`$.
Kashiwara-Vergne conjecture, version II: Taking $`O`$ smaller if necessary, there exists a $`𝔤`$-equivariant smooth function $`\beta :O^2𝔤^2`$, with $`\beta (0,0)=0`$, such that $`\beta _t=t^1r_t^{}\beta `$ satisfies the following two equations:
(21)
$$\frac{\mathrm{\Phi }_t}{t}L(X^{\beta _t})\mathrm{\Phi }_t=0,$$
(22)
$$\frac{\kappa _t}{t}\left(\underset{i}{}L_i\beta _t^i\right)\kappa _t=0.$$
###### Remark 2.2.
The geometric interpretation of the first equation is that the vector field $`X^{\beta _t}`$ intertwines the product maps, $`\mathrm{\Phi }_t`$. To interpret the second equation, choose a translation invariant volume form $`\mathrm{\Gamma }_0`$ on $`𝔤^2`$, and consider the family of volume forms $`\mathrm{\Gamma }_t=\kappa _t\mathrm{\Gamma }_0`$. Then the second equation is equivalent to
$$\begin{array}{cc}\hfill \dot{\mathrm{\Gamma }}_t& =L(X^{\beta _t})\mathrm{\Gamma }_t\underset{i}{}\beta _t^i\kappa _tL_i\mathrm{\Gamma }_0\hfill \\ & =L(X^{\beta _t})\mathrm{\Gamma }_t\underset{i}{}\beta _t^i\mathrm{tr}_{𝔤\times 𝔤}(e_i)\mathrm{\Gamma }_t\hfill \end{array}$$
where we have used that $`L(X^{\beta _t})=_iL_i\beta _t^i`$ on volume forms. If $`𝔤`$ is *unimodular* (i.e., $`\mathrm{tr}_𝔤(\mathrm{ad}(x))=0`$ for all $`x`$), the equation says that the vector field $`X^{\beta _t}`$ intertwines the volume forms $`\mathrm{\Gamma }_t`$, while in the general case $`\mathrm{\Gamma }_t`$ is a relative invariant for the unimodular character.
The third version of the Kashiwara-Vergne conjecture is more algebraic, and no longer involves the variable $`t`$. This is the ‘main’ version of the conjecture, as formulated in .
Kashiwara-Vergne conjecture, version III: Taking $`O`$ smaller if necessary, there exist smooth $`𝔤`$-equivariant functions $`\beta ^1,\beta ^2:O^2𝔤`$, vanishing at $`(0,0)`$, such that the following two equations are satisfied:
(23)
$$(1e^{\mathrm{ad}_x})\beta ^1(x,y)+(e^{\mathrm{ad}_y}1)\beta ^2(x,y)=x+y\mathrm{log}(\mathrm{exp}(y)\mathrm{exp}(x)),$$
(24)
$$\mathrm{tr}\left(\mathrm{ad}_x_x\beta ^1(x,y)+\mathrm{ad}_y_y\beta ^2(x,y)\right)=\frac{1}{2}\mathrm{tr}\left(\frac{\mathrm{ad}_x}{e^{\mathrm{ad}_x}1}+\frac{\mathrm{ad}_y}{e^{\mathrm{ad}_y}1}\frac{\mathrm{ad}_z}{e^{\mathrm{ad}_z}1}1\right),$$
with $`z=\mathrm{log}(\mathrm{exp}(x)\mathrm{exp}(y))`$. Here $`\mathrm{tr}`$ denotes the trace in the adjoint representation, $`_x\beta ^1(x,y)\mathrm{End}(𝔤)`$ is the linear map
(25)
$$\xi \frac{}{u}|_{u=0}\beta ^1(x+u\xi ,y),$$
and similarly for $`_y\beta ^2(x,y)`$.
###### Remark 2.3.
A *Lie polynomial* in variables $`x_1,\mathrm{},x_l`$ is an element of the free Lie algebra $`𝔩=𝔩(x_1,\mathrm{},x_l)`$ with generators $`x_i`$. Recall that the free Lie algebra has a grading $`𝔩=_k𝔩^k`$, where the degree $`k`$ component consists of linear combinations of *Lie monomials* $`[z_{i_1},[z_{i_2},\mathrm{},[z_{i_{k1}},z_{i_k}],\mathrm{}]]`$ where $`z_j\{x_1,\mathrm{},x_l\}`$. A *Lie series* is an element of the degree completion $`\widehat{𝔩}=_k𝔩^k`$. If $`x_i`$ are elements of a given Lie algebra $`𝔤`$, one obtains a Lie algebra homomorphism $`𝔩𝔤`$ taking $`x_i𝔩`$ to $`x_i𝔤`$.
The Campbell-Hausdorff theorem (see e.g. \[7, Chapter I.3\]) states that if $`dim𝔤<\mathrm{}`$, the function $`\mathrm{\Phi }(x,y)=\mathrm{log}(\mathrm{exp}(x)\mathrm{exp}(y))`$ is a Lie series in $`x,y`$. Dynkin’s formula gives this series explicitly as a *universal* element in $`\widehat{𝔩}(x,y)`$ (i.e. not depending on the Lie algebra). It is hence natural to require that similarly, $`\beta ^1,\beta ^2`$ are analytic functions in $`x,y`$ given as universal Lie series.
### 2.5. Equivalence of versions I-III of the KV conjecture
The implication $`IIIIII`$ was shown in \[8, Section 3\]. Closer examination of their argument shows that the three statements are in fact equivalent:
#### 2.5.1. Equivalence $`III`$
We have to show that $`\dot{m}_t=m_t_i\beta _t^iL_i`$ if and only if $`\beta _t`$ solves the pair of equations (21),(22). In fact, this equivalence does not involve the particular form of $`\mathrm{\Phi }_t,\kappa _t`$. Thus let $`O𝔤`$ be an open neighborhood of the origin, and let $`\mathrm{\Phi }_t:O^2𝔤,\kappa _t:O^2`$ be *arbitrary* families of smooth functions. Define a family of maps $`𝒟_{\mathrm{comp}}^{}(O^2)𝒟_{\mathrm{comp}}^{}(𝔤)`$ by $`m_t=(\mathrm{\Phi }_t)_{}\kappa _t`$. Let $`\beta _t:O^2𝔤^2`$ be a smooth family of functions.
###### Lemma 2.4.
The equation $`\dot{m}_t=m_t_i\beta _t^iL_i`$ holds if and only if
(26)
$$\dot{\mathrm{\Phi }}_tL(X^{\beta _t})\mathrm{\Phi }_t=0,\dot{\kappa }_t\left(\underset{i}{}L_i\beta _t^i\right)\kappa _t=0.$$
###### Proof.
Recall that linear combinations of delta-distributions are dense in the space of distributions. Hence $`\dot{m}_t=m_t_i\beta ^iL_i`$ holds if and only if it holds on $`\delta _{(q,p)}`$, for all $`(q,p)O^2`$. Let us work out the resulting identities. Taking the $`t`$-derivative of
(27)
$$m_t(\delta _{(q,p)})=\kappa _t(q,p)\delta _{\mathrm{\Phi }_t(q,p)},$$
we obtain
(28)
$$\dot{m}_t(\delta _{(q,p)})=\dot{\kappa }_t(q,p)\delta _{\mathrm{\Phi }_t(q,p)}\kappa _t(q,p)\underset{a}{}\dot{\mathrm{\Phi }}_t^a(q,p)\frac{}{x^a}\delta _{\mathrm{\Phi }_t(q,p)}.$$
Here $`e_a`$ is a basis of $`𝔤`$ and $`x^a`$ are the corresponding coordinates. On the other hand,
$`\left(m_t{\displaystyle \underset{i}{}}\beta _t^iL_i\right)(\delta _{(q,p)})`$ $`=`$ $`(\mathrm{\Phi }_t)_{}\left({\displaystyle \underset{i}{}}\kappa _t\beta _t^iL_i\delta _{(q,p)}\right)`$
$`=`$ $`(\mathrm{\Phi }_t)_{}\left({\displaystyle \underset{i}{}}\left(L_i(\kappa _t\beta _t^i)\right)\delta _{(q,p)}+{\displaystyle \underset{i}{}}(\kappa _t\beta _t^i)(q,p)L_i\delta _{(q,p)}\right)`$
$`=`$ $`{\displaystyle \underset{i}{}}L_i(\kappa _t\beta _t^i)(q,p)\delta _{\mathrm{\Phi }_t(q,p)}+{\displaystyle \underset{i}{}}(\kappa _t\beta _t^i)(q,p)(\mathrm{\Phi }_t)_{}(L_i\delta _{(q,p)})`$
$`=`$ $`{\displaystyle \underset{i}{}}L_i(\kappa _t\beta _t^i)(q,p)\delta _{\mathrm{\Phi }_t(q,p)}+{\displaystyle \underset{ia}{}}(\kappa _t\beta _t^iL_i(\mathrm{\Phi }_t^a))(q,p){\displaystyle \frac{}{x^a}}\delta _{\mathrm{\Phi }_t(q,p)}.`$
Comparing the coefficients of $`\delta _{\mathrm{\Phi }(p,q)},\frac{}{x^a}\delta _{\mathrm{\Phi }_t(q,p)}`$ with those in Equation (28), we find that $`\dot{m}_t=m_t_i\beta ^iL_i`$ is equivalent to
(29)
$$\dot{\mathrm{\Phi }}_t^a=\underset{i}{}\beta _t^iL_i(\mathrm{\Phi }_t^a),\dot{\kappa }_t=\underset{i}{}L_i(\kappa _t\beta _t^i),$$
which is (26). ∎
#### 2.5.2. Equivalence $`IIIII`$.
Let $`A(x,y),A^{}(x,y)`$ denote the left-, right-hand side of (23), and write $`A_t=t^1r_t^{}A`$ and $`A_t^{}=t^1r_t^{}A^{}`$. In the proof of \[8, Lemma 2.3\], Kashiwara-Vergne compute
$$\begin{array}{cc}\hfill \dot{\mathrm{\Phi }}_t(x,y)& =\frac{1}{t}\frac{}{ϵ}|_{ϵ=0}\mathrm{log}\left(e^{tx}e^{ϵA_t^{}(x,y)}e^{ty}\right)\hfill \\ \\ \hfill (L(X^{\beta _t})\mathrm{\Phi }_t)(x,y)& =\frac{1}{t}\frac{}{ϵ}|_{ϵ=0}\mathrm{log}\left(e^{tx}e^{ϵA_t(x,y)}e^{ty}\right).\hfill \end{array}$$
It follows that the first condition (21) of version II is equivalent to $`A_t=A_t^{}`$, hence to $`A=A^{}`$ which is the first condition (23) of version III.
Assuming (21), we now show that (22) is equivalent to (24). Let $`B(x,y)`$, $`B^{}(x,y)`$ denote the left-, right-hand side of Equation (24), and let $`B_t=t^1r_t^{}B`$, $`B_t^{}=t^1r_t^{}B^{}`$. Then
(30)
$$B_t=\underset{i}{}(L_i\beta _t^i),$$
and Equation (22) may be written in the form
(31)
$$0=\dot{\kappa }_tL(X^{\beta _t})\kappa _t+\underset{i}{}(L_i\beta _t^i)\kappa _t=\left(\frac{}{t}L(X^{\beta _t})\right)\kappa _tB_t\kappa _t.$$
From (21), we have
$$\left(\frac{}{t}L(X^{\beta _t})\right)J_t(\mathrm{\Phi }_t(x,y))=\dot{J}_t(\mathrm{\Phi }_t(x,y)).$$
On the other hand, $`L(X^{\beta _t})J_t(x)=L(X^{\beta _t})J_t(y)=0`$ since $`J`$ is $`\mathrm{ad}`$-invariant. Recalling the formula \[8, Lemma 3.3\]
(32)
$$J_t(z)^1\dot{J}_t(z)=\mathrm{tr}_𝔤\left(\frac{t\mathrm{ad}_z}{e^{t\mathrm{ad}_z}1}1\right),z𝔤$$
this gives,
(33)
$$\dot{\kappa }_tL(X^{\beta _t})\kappa _t=B_t^{}\kappa _t.$$
We thus obtain $`B_t=B_t^{}`$, i.e. $`B=B^{}`$.
## 3. Torossian’s theorem implies the KV conjecture
### 3.1. Torossian’s theorem
In his paper, Torossian \[14, page 601\] proves the following statement, similar to version II of the KV conjecture.
###### Theorem 3.1 (Torossian).
There are families of smooth functions
(34)
$$\stackrel{~}{\mathrm{\Phi }}_s:O^2𝔤,\stackrel{~}{\kappa }_s:O^2,\stackrel{~}{\beta }_s:O^2𝔤^2,s[0,1]$$
with the following properties:
1. The Taylor expansion of $`\stackrel{~}{\mathrm{\Phi }}_s(x,y)`$ at the origin $`(0,0)`$ is of the form $`\stackrel{~}{\mathrm{\Phi }}_s(x,y)=x+y+\mathrm{}`$. Furthermore, $`\stackrel{~}{\kappa }_s(0,0)=1,\stackrel{~}{\beta }_s(0,0)=0`$,
2. For $`s=0`$,
$$\stackrel{~}{\mathrm{\Phi }}_0(x,y)=x+y,\stackrel{~}{\kappa }_0(x,y)=1,$$
while for $`s=1`$,
$$\stackrel{~}{\mathrm{\Phi }}_1(x,y)=\mathrm{\Phi }(x,y),\stackrel{~}{\kappa }_1(x,y)=\kappa (x,y).$$
3. The following two equations are satisfied:
$$\dot{\stackrel{~}{\mathrm{\Phi }}}_sL(X^{\stackrel{~}{\beta }_s})\stackrel{~}{\mathrm{\Phi }}_s=0,\dot{\stackrel{~}{\kappa }}_s\left(\underset{i}{}L_i\stackrel{~}{\beta }_s^i\right)\stackrel{~}{\kappa }_s=0.$$
The functions $`\stackrel{~}{\mathrm{\Phi }}_s,\stackrel{~}{\kappa }_s,\stackrel{~}{\beta }_s`$ are constructed using the diagrammatic technique of Kontsevich . See for details.
Using $`\stackrel{~}{\mathrm{\Phi }}_s`$ and $`\stackrel{~}{\kappa }_s`$ one can define maps $`\stackrel{~}{m}_s=(\stackrel{~}{\mathrm{\Phi }}_s)_{}\stackrel{~}{\kappa }_s`$ as before. However, $`\stackrel{~}{m}_sm_s`$, and in fact the ‘products’ $`\stackrel{~}{m}_s`$ are not associative, in general. Nonetheless, according Lemma 2.4 one still has
(35)
$$\frac{\stackrel{~}{m}_s}{s}=\stackrel{~}{m}_s\underset{i}{}\stackrel{~}{\beta }_s^iL_i,$$
which as before implies the Duflo theorem and its generalization to germs of invariant distributions.
### 3.2. Torossian $``$ KV
We will now show how to obtain a solution $`\beta _t`$ of the original KV problem from Torossian’s function $`\stackrel{~}{\beta }_s`$. Let us first state the relevant argument in a more general setting. Let $`𝔨`$ be a Lie algebra, acting on a manifold $`M`$. The vector fields defining this action are a Lie algebra homomorphism $`𝔨𝔛(M),\xi X^\xi `$. For $`\beta 𝔨_M:=C^{\mathrm{}}(M,𝔨)`$ define $`X^\beta 𝔛(M)`$ by $`X^\beta (x)=X^{\beta (x)}(x)`$. Then
(36)
$$𝔨_M𝔛(M),\beta X^\beta $$
is a Lie algebra homomorphism for the following Lie bracket on $`𝔨_M`$,
(37)
$$[\beta ,\gamma ](x)=(L(X^\beta )\gamma )(x)(L(X^\gamma )\beta )(x)+[\beta (x),\gamma (x)].$$
The following Lemma will be proved in the appendix.
###### Lemma 3.2.
Suppose $`\gamma _{s,t}𝔨_M`$ is a given smooth 2-parameter family, with $`\gamma _{s,t}(x_0)=0`$ for a given point $`x_0M`$. Replacing $`M`$ by a smaller neighborhood of $`x_0`$ if necessary, there exists a unique 2-parameter family $`\beta _{s,t}𝔨_M`$ with $`\beta _{0,t}=0`$, such that the following zero curvature equation is satisfied:
(38)
$$\frac{\beta _{s,t}}{s}\frac{\gamma _{s,t}}{t}+[\beta _{s,t},\gamma _{s,t}]=0.$$
Let us describe another general feature of the Lie algebra $`𝔨_M`$. For $`\beta 𝔨_M`$, consider the first order differential operator on $`𝒟^{}(M)`$ (cf. (22))
(39)
$$V(\beta )=\underset{i}{}\beta ^iL(e_i)$$
The operator $`V(\beta )`$ is different from the Lie derivative $`L(X^\beta )`$, which equals $`_iL_i\beta ^i`$ on distributions. However, one verifies that the difference $`L(X^\beta )V(\beta )=\tau (\beta )`$, where
(40)
$$\tau :𝔨_MC^{\mathrm{}}(M),\beta \underset{i}{}(L_i\beta ^i),$$
is a Lie algebra cocycle: $`\tau ([\beta ,\gamma ])=L(X^\beta )\tau (\gamma )L(X^\gamma )\tau (\beta )`$. This gives
(41)
$$V([\beta ,\gamma ])=[V(\beta ),V(\gamma )].$$
(Note that for the function $`\beta `$ from the Kashiwara-Vergne conjecture (version III), the cocycle $`\tau (\beta )`$ appears as the left hand side of (24) (see also (30)).)
###### Theorem 3.3.
Any solution $`\stackrel{~}{\beta }_s`$ of Torossian’s modified KV problem (cf. Theorem 3.1) determines a solution $`\beta `$ of the KV problem.
###### Proof.
Introduce a 2-parameter family of maps $`\stackrel{~}{\beta }_{s,t}=r_t^{}\stackrel{~}{\beta }_s,s,t[0,1]`$. Since $`\stackrel{~}{\beta }_s(0,0)=0`$, we have $`\stackrel{~}{\beta }_{s,t}(0,0)=0`$. Similarly, rescale $`\stackrel{~}{\mathrm{\Phi }}_{s,t}=t^1r_t^{}\stackrel{~}{\mathrm{\Phi }}_s`$ and $`\stackrel{~}{\kappa }_{s,t}=r_t^{}\stackrel{~}{\kappa }_s`$. For each $`s,t`$ the ‘product’ $`\stackrel{~}{m}_{s,t}=(\stackrel{~}{\mathrm{\Phi }}_{s,t})_{}\stackrel{~}{\kappa }_{s,t}`$ satisfies the equation,
$$\frac{\stackrel{~}{m}_{s,t}}{s}=\stackrel{~}{m}_{s,t}\underset{i}{}\stackrel{~}{\beta }_{s,t}^iL_i.$$
Now let $`\beta _{s,t}`$ be the 2-parameter family of maps $`\beta _{s,t}:O^2𝔤^2,\beta _{0,t}=0`$ obtained by solving the zero curvature equation, Lemma 3.2, with $`\gamma _{s,t}=\stackrel{~}{\beta }_{s,t}`$. The scaling property $`\stackrel{~}{\beta }_{s,t}=r_t^{}\stackrel{~}{\beta }_s`$ implies the scaling property for the derivative, $`\frac{\stackrel{~}{\beta }_{s,t}}{t}=t^1r_t^{}\left(\frac{\stackrel{~}{\beta }_{s,u}}{u}|_{u=1}\right)`$, and hence (by Equation (38)) $`\beta _{s,t}=t^1r_t^{}\beta _{s,1}`$. We compute,
$$\begin{array}{ccc}\multicolumn{3}{c}{\frac{}{s}\left(\frac{\stackrel{~}{m}_{s,t}}{t}+\stackrel{~}{m}_{s,t}V(\beta _{s,t})\right)}\\ & =\frac{}{t}\frac{\stackrel{~}{m}_{s,t}}{s}+\frac{\stackrel{~}{m}_{s,t}}{s}V(\beta _{s,t})+\stackrel{~}{m}_{s,t}V\left(\frac{\beta _{s,t}}{s}\right)\hfill \\ & =\frac{}{t}\left(\stackrel{~}{m}_{s,t}V(\stackrel{~}{\beta }_{s,t})\right)\stackrel{~}{m}_{s,t}V(\stackrel{~}{\beta }_{s,t})V(\beta _{s,t})+\stackrel{~}{m}_{s,t}V\left(\frac{\beta _{s,t}}{s}\right)\hfill \\ & =\frac{\stackrel{~}{m}_{s,t}}{t}V(\stackrel{~}{\beta }_{s,t})\stackrel{~}{m}_{s,t}\left(V\left(\frac{\stackrel{~}{\beta }_{s,t}}{t}\right)V\left(\frac{\beta _{s,t}}{s}\right)+V(\stackrel{~}{\beta }_{s,t})V(\beta _{s,t})\right)\hfill \\ & =\left(\frac{\stackrel{~}{m}_{s,t}}{t}+\stackrel{~}{m}_{s,t}V(\beta _{s,t})\right)V(\stackrel{~}{\beta }_{s,t})\hfill \end{array}$$
where in the last step we used (38) with $`\gamma _{s,t}=\stackrel{~}{\beta }_{s,t}`$. This equality is an ordinary first order differential equation (in $`s`$) for the map
(42)
$$_{s,t}:=\frac{\stackrel{~}{m}_{s,t}}{t}+\stackrel{~}{m}_{s,t}V(\beta _{s,t}).$$
Since $`\stackrel{~}{m}_{0,t}=m_0`$ and $`\beta _{0,t}=0`$, one has the initial condition $`_{0,t}=0`$. Hence, by uniqueness of solutions of ordinary differential equations it follows that $`_{s,t}=0`$ for all $`s`$. In particular, $`_{1,t}=0`$ gives the desired equation
$$\frac{m_t}{t}+m_tV(\beta _t)=0$$
where $`\beta _t=\beta _{1,t}`$ has the scaling property $`\beta _t=t^1r_t^{}\beta `$. ∎
### 3.3. Analyticity
As pointed out in Remark 2.3, it is natural to impose the additional condition on the KV solution $`\beta `$, that its components $`\beta ^1,\beta ^2`$ are given by (universal) analytic Lie series. We will now verify that the solution $`\beta `$ constructed in the last Section does indeed have this property.
We introduce the following terminology. Let $`𝔩=_{n=1}^{\mathrm{}}𝔩^n`$ be the free Lie algebra on generators $`x,y`$, and $`\widehat{𝔩}`$ its degree completion. An element $`l𝔩^n`$ will be called a *Lie word of length $`n`$* if it is obtained by bracketing $`n`$ elements $`x_{i_1},\mathrm{},x_{i_n}\{x,y\}`$, in any order. For instance, $`[y,[[x,y],x]]`$ is a Lie word of length $`4`$.
###### Definition 3.4.
Given a Lie series $`l=_{n=1}^{\mathrm{}}l_n\widehat{𝔩}`$, let $`C_n(l)`$ denote the smallest possible $`_i|c_{n,i}|`$ for presentations
(43)
$$l_n=\underset{i}{}c_{n,i}l_{n,i}$$
where $`l_{n,i}`$ are Lie words of length $`n`$. The Lie series $`l`$ is called strongly convergent if $`C_n(l)D^n`$ for some $`D>0`$.
The Campbell-Hausdorff series is an example of a strongly convergent Lie series. For any finite-dimensional Lie algebra $`𝔤`$, a strongly convergent Lie series $`l`$ defines an analytic function $`l(x,y)`$ with values in $`𝔤`$ on some neighborhood of the origin in $`𝔤\times 𝔤`$. Strongly convergent Lie series form a Lie subalgebra of $`\widehat{𝔩}`$.
Motivated by (37), we introduce a new bracket on $`\widehat{𝔩}\times \widehat{𝔩}`$, as follows. For any $`\beta =(\beta ^1,\beta ^2)`$, let $`L_\beta `$ denote the derivation of $`\widehat{𝔩}`$ given on generators by $`L_\beta x=[\beta ^1,x],L_\beta y=[\beta ^2,y]`$. If $`\gamma =(\gamma ^1,\gamma ^2)`$ is a pair of Lie series, we set $`L_\beta \gamma =(L_\beta \gamma ^1,L_\beta \gamma ^2)`$. Now put
(44)
$$[\beta ,\gamma ]=L_\beta \gamma L_\gamma \beta +[\beta ,\gamma ]_0.$$
where $`[,]_0`$ denotes the original (componentwise) bracket. The Jacobi identity for the new bracket is a simple corollary of the derivation property for $`L_\beta `$.
Let us write $`\mathrm{ad}_\beta =[\beta ,]`$, and denote by $`C_n(\beta )`$ the maximum of $`C_n(\beta ^1)`$ and $`C_n(\beta ^2)`$.
###### Lemma 3.5.
Suppose $`\beta ,\gamma \widehat{𝔩}\times \widehat{𝔩}`$ with $`C_n(\beta ),C_n(\gamma )D^n`$ for some $`D>0`$. Then $`C_n([\beta ,\gamma ])n^2D^n`$. More generally, If $`C_n(\beta _i),C_n(\gamma )D^n`$ then
(45)
$$C_n(\mathrm{ad}_{\beta _k}\mathrm{}\mathrm{ad}_{\beta _1}\gamma )\frac{(2n^2)^k}{(2k1)!!}D^n.$$
Here $`(2k1)!!=(2k1)(2k3)\mathrm{}1`$ for all $`k>0`$.
###### Proof.
Note that if $`\beta ,\gamma `$ are homogeneous of degree $`n_1,n_2`$, then $`[\beta ,\gamma ]`$ is homogeneous of degree $`n=n_1+n_2`$, and
$$C_n([\beta ,\gamma ])(n+1)C_{n_1}(\beta )C_{n_2}(\gamma ).$$
This follows since $`C_n(L_\beta \gamma )n_2C_{n_1}(\beta )C_{n_2}(\gamma )`$, and similarly $`C_n(L_\gamma \beta )n_1C_{n_1}(\beta )C_{n_2}(\gamma )`$, while $`C_n([\beta ,\gamma ]_0)C_{n_1}(\beta )C_{n_2}(\gamma )`$. We now obtain (45) by induction on $`k`$. The case $`k=0`$ is trivial, and for $`k>0`$ we find, using the induction hypothesis for $`k1`$,
$$\begin{array}{cc}\hfill C_n(\mathrm{ad}_{\beta _k}\mathrm{}\mathrm{ad}_{\beta _1}\gamma )& (n+1)\underset{j=1}{\overset{n1}{}}C_{nj}(\beta _k)C_j(\mathrm{ad}_{\beta _{k1}}\mathrm{}\mathrm{ad}_{\beta _1}\gamma )\hfill \\ & (n+1)\frac{1}{(2k3)!!}\underset{j=1}{\overset{n1}{}}(2j^2)^{(k1)}D^n\hfill \\ & 2^{k1}(n+1)\frac{n^{2k1}}{(2k1)!!}D^n\frac{2^kn^{2k}}{(2k1)!!}D^n.\hfill \end{array}$$
(In the last line, we used $`_{j=1}^{n1}j^{2k2}_1^nj^{2k2}𝑑j`$.) ∎
We will need one more remark. Suppose $`l=_{n=1}^{\mathrm{}}l_n\widehat{𝔩}`$ is a strongly convergent Lie series. Then $`l_t=r_t^{}l=_{n=1}^{\mathrm{}}t^nl_n`$ is strongly convergent for all $`0t1`$, and since $`t^n1`$ one has the uniform estimate $`C_n(l_t)C_n(l)D^n`$. Similarly, $`\dot{l}_t=\frac{l_t}{t}`$ is strongly convergent, with a uniform estimate on $`C_n(\dot{l}_t)`$.
###### Proposition 3.6.
Let $`\gamma _s\widehat{𝔩}\times \widehat{𝔩}`$, with coefficients depending continuously on $`s[0,1]`$, and such that $`C_n(\gamma _s)D^n`$. Let $`\gamma _{s,t}=r_t^{}\gamma _s`$. Then the formal sum
(46)
$$\beta _{s,t}=\underset{k=0}{\overset{\mathrm{}}{}}_{0s_0s_1\mathrm{}s_ks}𝑑s_0\mathrm{}𝑑s_k\left(\mathrm{ad}_{\gamma _{s_k,t}}\mathrm{}\mathrm{ad}_{\gamma _{s_1,t}}\frac{\gamma _{s_0,t}}{t}\right)$$
defines a strongly convergent Lie series for $`(s,t)[0,1]^2`$, with $`\beta _{0,t}=0`$. Moreover, $`\beta _{s,t}=t^1r_t^{}\beta _s`$ and the following zero curvature equation holds true,
(47)
$$_s\beta _{s,t}_t\gamma _{s,t}+[\beta _{s,t},\gamma _{s,t}]=0.$$
###### Proof.
Note first of all that the right hand side of (46) defines a Lie series, since only indices $`kn`$ contribute to the term of degree $`n`$ in $`\beta _{s,t}`$. Also, $`\beta _{0,t}=0`$ since each term on the right hand side of (46) vanishes at $`s=0`$. The flatness condition (47) is satisfied as an equality of formal Lie series. To show that the series $`\beta _{s,t}`$ is strongly convergent, choose $`D>0`$ with $`C_n(\gamma _{s,t}),C_n(\dot{\gamma }_{s,t})D^n`$. Using the estimate (45), we obtain
$$\begin{array}{cc}\hfill C_n(\beta _{s,t})& \underset{k=0}{\overset{\mathrm{}}{}}_{0s_0s_1\mathrm{}s_ks}𝑑s_0\mathrm{}𝑑s_kC_n\left(\mathrm{ad}_{\gamma _{s_k,t}}\mathrm{}\mathrm{ad}_{\gamma _{s_1,t}}\frac{\gamma _{s_0,t}}{t}\right)\hfill \\ & \underset{k=0}{\overset{\mathrm{}}{}}\frac{s^k}{k!}\frac{2^kn^{2k}}{(2k1)!!}D^n\underset{k=0}{\overset{\mathrm{}}{}}\frac{2^{2k}n^{2k}}{(2k)!}D^ne^{2n}D^n.\hfill \end{array}$$
The scaling property $`\beta _{s,t}=t^1r_t^{}\beta _s`$ holds since it is satisfied by each term in the sum (46). ∎
We now apply these results to the Torossian function, $`\gamma _s=\stackrel{~}{\beta }_s`$. The diagrammatic technique from Torossian’s paper gives $`\stackrel{~}{\beta }_s`$ directly as a universal Lie series. (The family of maps $`\stackrel{~}{\beta }_s`$ is however not canonical, since it depends on the choice of a suitable path in $`\overline{C}_{2,0}^+`$, the compactified configuration space of two points in the upper half plane.) The estimates of (see also Remark after Theorem 4.2 in ) show that this Lie series is strongly convergent (with coefficients depending on the parameter $`s`$), and with $`C(\stackrel{~}{\beta }_s)`$ uniformly bounded for $`s[0,1]`$. Define $`\beta _{s,t}`$ by the Lie series (46), and identify $`\beta _{s,t}`$ and $`\stackrel{~}{\beta }_{s,t}`$ with the corresponding analytic functions on $`𝔤\times 𝔤`$ (defined near the origin). By the uniqueness property for solutions of the zero curvature equation (cf. Lemma 3.2), the function $`\beta _{1,t}=t^1r_t^{}\beta `$ coincides with the KV solution from Theorem 3.3. We have hence shown that our KV solution is given by a universal, strongly convergent Lie series.
###### Remark 3.7.
One can replace any solution $`\beta =(\beta ^1,\beta ^2)`$ of the KV conjecture with a symmetric solution $`\beta _{\mathrm{sym}}=(\beta _{\mathrm{sym}}^1,\beta _{\mathrm{sym}}^2)`$ where
$$\begin{array}{cc}\hfill \beta _{\mathrm{sym}}^1(x,y)& =\frac{1}{2}\left(\beta ^1(x,y)+\beta ^2(y,x)\right),\hfill \\ \hfill \beta _{\mathrm{sym}}^2(x,y)& =\frac{1}{2}\left(\beta ^2(x,y)+\beta ^1(y,x)\right).\hfill \end{array}$$
This solution has the additional property $`\beta _{\mathrm{sym}}^1(x,y)=\beta _{\mathrm{sym}}^2(y,x)`$.
While Theorem 3.3 settles the KV conjecture in the form stated in \[8, page 250\], there remain some interesting open problems in this context.
1. The algebraic version III of the KV conjecture can also be considered for Lie algebras over $``$. We do not know, however, whether the coefficients of our solution $`\beta (x,y)`$ are rational, for a suitable choice of Torossian’s function $`\stackrel{~}{\beta }_s(x,y)`$.
2. In a recent paper, Alekseev and Petracci proved that for universal symmetric solutions $`\beta `$ of the KV problem (for arbitrary finite-dimensional Lie algebras over $``$), the linear terms in $`x`$ or in $`y`$ (i.e. the terms $`(_x\beta )(0,y)`$ and $`(_y\beta )(x,0)`$) are uniquely determined. The linear terms in Kashiwara and Vergne’s solution for the case of solvable Lie algebras \[8, Equation (5.2)\] are as prescribed by this result. It is unknown whether their formula might in fact give a solution for all Lie algebras.
## 4. Lie algebra cohomology and the KV conjecture
In this Section, we will give yet another version of the Kashiwara-Vergne conjecture. This reformulation depends on a general fact in Lie algebra cohomology, stated in Proposition 4.1 below.
### 4.1. Lie algebra cohomology
For any $`x𝔤`$, we denote by $`\iota ^{}(x)`$ the derivation of $`𝔤^{}`$ given by contraction. For $`\mu 𝔤^{}`$ we denote by $`ϵ(\mu )`$ the operator of exterior multiplication on $`𝔤^{}`$.
For any $`𝔤`$-module $``$, let $`L^{}(x)\mathrm{End}()`$ denote the action of $`x𝔤`$. The Lie algebra cohomology $`H^{}(𝔤,)`$ of $`𝔤`$ with coefficients in $``$ is the cohomology of the Chevalley-Eilenberg complex
(48)
$$C^k(𝔤,):=^k𝔤^{},\text{d}=1\text{d}^{}+\underset{a}{}L^{}(e_a)ϵ(e^a).$$
Here $`e_a𝔤`$ and $`e^a𝔤^{}`$ are dual bases, and $`\text{d}^{}=\frac{1}{2}_aϵ(e^a)L^{}(e_a)`$ is the Lie algebra differential on $`𝔤^{}`$. Suppose $`𝔨`$ is a another Lie algebra, with basis $`f_i`$, and that $`𝒩`$ is a $`𝔨`$-module. Consider the linear map
(49)
$$V:\mathrm{End}(𝒩)𝔨\mathrm{End}(𝒩),R=\underset{i}{}R^if_i\underset{i}{}R^iL^𝒩(f_i).$$
If $`\psi :𝔤𝔨`$ is a Lie algebra homomorphism and $`R\mathrm{End}(𝒩)𝔨`$ is $`𝔤`$-invariant, then $`V(R)\mathrm{End}(𝒩)`$ is $`𝔤`$-equivariant. Hence it defines a chain map, $`V(R)\psi ^{}:C(𝔨,𝒩)C(𝔤,𝒩)`$. We will need the following fact:
###### Proposition 4.1.
Let $`𝒩`$ be a $`𝔨`$-module, and $`\psi :𝔤𝔨`$ a Lie algebra homomorphism. Suppose $`R\mathrm{End}(𝒩)𝔨`$ is $`𝔤`$-invariant. Then the chain map $`V(R)\psi ^{}`$ is homotopic to the trivial map. In fact,
(50)
$$V(R)\psi ^{}=(1\psi ^{})[\text{d},\iota (R)]$$
with $`\iota (R)=_iR^i\iota (f_i)`$.
###### Proof.
We have to show that $`1\psi ^{}:𝒩𝔨^{}𝒩𝔤^{}`$ annihilates the expression, $`[\text{d},\iota (R)]V(R)1`$. Let $`f^i𝔨^{}`$ be the dual basis to $`f_i𝔨`$. We compute,
$`[\text{d},\iota (R)]V(R)1`$ $`=`$ $`[1\text{d}^{}+{\displaystyle \underset{j}{}}L^𝒩(f_j)ϵ(f^j),{\displaystyle \underset{i}{}}R^i\iota (f_i)]V(R)1`$
$`=`$ $`{\displaystyle \underset{i}{}}R^iL^{}(f_i)+{\displaystyle \underset{ij}{}}L^𝒩(f_j)(R^i)\left(ϵ(f^j)\iota (f_i)\right).`$
Since $`_j(\psi ^{}f^j)f_j=_ae^a\psi (e_a)`$, we have the following equalities of maps $`𝔨^{}𝔤^{}`$,
$`\psi ^{}L^{}(f_i)`$ $`=`$ $`{\displaystyle \underset{j}{}}\psi ^{}ϵ(f^j)\iota ([f_j,f_i]_𝔨)`$
$`=`$ $`{\displaystyle \underset{j}{}}ϵ(\psi ^{}f^j)\psi ^{}\iota ([f_j,f_i]_𝔨)`$
$`=`$ $`{\displaystyle \underset{a}{}}ϵ(e^a)\psi ^{}\iota ([\psi (e_a),f_i]_𝔨).`$
One the other hand, using the $`𝔤`$-equivariance of $`R`$,
$`(1\psi ^{})\left({\displaystyle \underset{ij}{}}L^𝒩(f_j)(R^i)ϵ(f^j)\iota (f_i)\right)`$ $`=`$ $`{\displaystyle \underset{ia}{}}L^𝒩(\psi (e_a))(R^i)ϵ(e^a)\psi ^{}\iota (f_i)`$
$`=`$ $`{\displaystyle \underset{ia}{}}R^i\left(ϵ(e^a)\psi ^{}\iota ([\psi (e_a),f_i]_𝔨)\right).`$
### 4.2. Reformulation of the KV conjecture
We will apply this Proposition to the diagonal embedding $`\psi :𝔤𝔤^2`$. The map $`\psi ^{}:𝔤^{}𝔤^{}=(𝔤^{}𝔤^{})𝔤^{}`$ is just the product map in the exterior algebra. View $`𝒩=𝒟_{\mathrm{comp}}^{}(O^2)`$ as a $`𝔤^2`$-module, and $`=𝒟_{\mathrm{comp}}^{}(𝔤)`$ as a $`𝔤`$-module. We obtain a family of chain maps,
(51)
$$M_t=m_t\psi ^{}:C(𝔤^2,𝒟_{\mathrm{comp}}^{}(O^2))C(𝔤,𝒟_{\mathrm{comp}}^{}(𝔤)).$$
Version I of the Kashiwara-Vergne conjecture says
(52)
$$\frac{m_t}{t}=m_tV(\beta _t),$$
where $`\beta _tC^{\mathrm{}}(M,𝔨)`$. (The components $`\beta _t^i`$ are identified with the corresponding operators of multiplication on $`𝒟_{\mathrm{comp}}^{}(O^2)`$.) Tensoring with $`\psi ^{}:^2𝔤^{}𝔤^{}`$, and using Proposition 4.1, we obtain the following equivalent reformulation of the KV conjecture:
Kashiwara-Vergne conjecture, version IV: Taking $`O`$ smaller if necessary, there exists a $`𝔤`$-equivariant function $`\beta :O^2𝔤^2`$, with $`\beta (0,0)=0`$, such that $`\beta _t=t^1r_t^{}\beta `$ satisfies
(53)
$$\frac{M_t}{t}=M_t[\text{d},\iota (\beta _t)].$$
The algebra isomorphism $`H(𝔤,S𝔤)H(𝔤,U𝔤)`$ is an immediate consequence of this conjecture. More generally, one obtains a similar statement for convolution of germs of distributions. Let $`A_1,A_2𝔤`$ be closed cones with $`A_1A_2=\{0\}`$, and consider the diagram
$$\begin{array}{ccc}C(𝔤,𝔇^{}(𝔤,A_1))\times C(𝔤,𝔇^{}(𝔤,A_1))& & C(𝔤,𝔇^{}(𝔤,A_1+A_2))\\ & & & & \\ C(𝔤,𝔇^{}(G,A_1))\times C(𝔤,𝔇^{}(G,A_1))& & C(𝔤,𝔇^{}(G,A_1+A_2))\end{array}$$
where the horizontal maps are product maps, and the vertical maps are induced by the Duflo maps. Version IV of the Kashiwara-Vergne conjecture implies that this diagram commutes up to a chain homotopy. In particular, passing to cohomology one obtains a commutative diagram
$$\begin{array}{ccc}H(𝔤,𝔇^{}(𝔤,A_1))H(𝔤,𝔇^{}(𝔤,A_2))& & H(𝔤,𝔇^{}(𝔤,A_1+A_2))\\ & & & & \\ H(𝔤,𝔇^{}(G,A_1))H(𝔤,𝔇^{}(G,A_2))& & H(𝔤,𝔇^{}(G,A_1+A_2))\end{array}$$
### 4.3. Higher homotopies
Version IV of the KV conjecture shows that each map $`\mathrm{Duf}_t\mathrm{id}:S(𝔤)𝔤^{}U(𝔤)𝔤^{}`$ is an algebra isomorphism up to homotopy. Solutions of the KV problem provide infinitesimal homotopies for family of products $`M_t`$. One can ask whether in fact $`\mathrm{Duf}_t\mathrm{id}`$ extend to an $`A_{\mathrm{}}`$-morphisms (for a definition see ) and to construct infinitesimal homotopies for its higher components. Indeed, let
(54)
$$M_t^{(3)}=M_t(M_t1):C(𝔤^3,𝒟_{\mathrm{comp}}^{}(O^3))C(𝔤,𝒟_{\mathrm{comp}}(𝔤))$$
denote the triple product map (defined for $`O`$ sufficiently small). Let $`\beta _t`$ denote a solution of the KV-problem. Taking the $`t`$-derivative of $`M_t^{(3)}=M_t(M_t1)`$, we obtain $`\dot{M}_t^{(3)}=[\text{d},h_t^{(3)}]`$ where $`h_t^{(3)}`$ is the homotopy operator
(55)
$$h_t=M_t^{(3)}(\iota (\beta _t)1)+M_t\iota (\beta _t)(M_t1).$$
On the other hand, starting with $`M_t^{(3)}=M_t(1M_t)`$, we get another homotopy operator
(56)
$$\stackrel{~}{h}_t=M_t^{(3)}(1\iota (\beta _t))+M_t\iota (\beta _t)(1M_t).$$
The differences $`\stackrel{~}{h}_th_t`$ are cochain maps of degree $`1`$. In the spirit of version $`IV`$ of the KV conjecture, one can formulate the following new conjecture: For $`O𝔤`$ a sufficiently small neighborhood of the origin, there exists a $`𝔤`$-equivariant map $`\beta ^{(3)}:O^3^2𝔤^3`$, with $`\beta ^{(3)}(0)=0`$, such that $`\beta _t^{(3)}=t^1r_t^{}\beta ^{(3)}`$ satisfies
(57)
$$\stackrel{~}{h}_th_t=M_t^{(3)}[d,\iota (\beta _t^{(3)})].$$
In a similar fashion, one can conjecture the existence of higher homotopies given by equivariant functions $`\beta ^{(n)}:O^n^{n1}𝔤^n`$.
## Appendix A The zero curvature equation
In this appendix we prove Lemma 3.2, the zero curvature equation for the Lie algebra $`𝔨_M`$. Let us first recall the more general zero curvature equation for vector fields on a manifold $`M`$.
###### Lemma A.1.
Let $`Y_{s,t}𝔛(M)`$ a given 2-parameter family of vector fields, depending smoothly on $`s,t[0,1]`$, and assume that $`Y_{s,t}`$ vanishes at some given point $`x_0M`$ for all $`s,t`$. Then there exists an open neighborhood $`U`$ of $`x_0`$, and a unique 2-parameter family of vector fields $`X_{s,t}`$, $`s,t[0,1]`$, satisfying the zero curvature equation
(58)
$$\frac{X_{s,t}}{s}\frac{Y_{s,t}}{t}+[X_{s,t},Y_{s,t}]=0$$
with initial condition $`X_{0,t}=0`$.
###### Proof.
Let $`F_{s,t}`$ denote the flow of $`Y_{s,t}`$, viewed as an $`s`$-dependent vector field depending on $`t`$ as a parameter. In terms of the action on smooth functions $`f`$,
(59)
$$(Y_{s,t}f)F_{s,t}^1=\frac{}{s}(fF_{s,t}^1);F_{0,t}=\mathrm{id}.$$
Since $`Y_{s,t}(x_0)=0`$, there exists an open neighborhood $`U`$ of $`x_0`$ such that $`F_{s,t}(x)`$ is defined for all $`xU`$ and $`s,t[0,1]`$. Define a vector field $`X_{s,t}`$ by
(60)
$$(X_{s,t}f)F_{s,t}^1=\frac{}{t}(fF_{s,t}^1).$$
Note that $`X_{0,t}=0`$ since $`F_{0,t}=\mathrm{id}`$. Subtracting the $`s`$-derivative of (60) from the $`t`$-derivative of (59) one finds that $`X_{s,t}`$ solves (58). On the other hand, any two solutions of (58) differ by a 2-parameter family of vector fields $`Z_{s,t}`$ with $`\frac{}{s}Z_{s,t}+[Z_{s,t},Y_{s,t}]=0`$, or equivalently $`\frac{}{s}\left((F_{s,t}^1)^{}Z_{s,t}\right)=0`$. Integrating with initial condition $`Z_{0,t}=0`$ one obtains $`Z_{s,t}=0`$, proving uniqueness. ∎
Lemma 3.2 (the zero curvature equation in $`𝔨_M`$) may be viewed as a special case of Lemma A.1. Indeed, $`𝔨_M`$ may be realized as a Lie algebra of vector fields on $`K\times M`$, where $`K`$ is a Lie group with Lie algebra $`𝔨`$. To see this, lift the Lie algebra $`𝔨`$-action on $`M`$ to an action on $`K\times M`$ by $`\xi \widehat{X}^\xi =(\xi ^L,X^\xi )`$ where $`\xi ^L𝔛(K)`$ is the left-invariant vector field generated by $`\xi `$. Accordingly, the homomorphism $`𝔨_M𝔛(M)`$ lifts to a Lie algebra homomorphisms
(61)
$$𝔨_M𝔛(K\times M),\beta \widehat{X}^\beta .$$
It is clear that the map (61) is injective. Its image consists of vector fields on $`K\times M`$ that are invariant under the $`K`$-action by left-multiplication on the first factor, and are tangent to the foliation defined by the $`𝔨`$-action $`\xi \widehat{X}^\xi `$ on $`K\times M`$. (Equivalently, the flow of such vector fields is $`K`$-equivariant, and preserves the leaves of the $`𝔨`$-action.) Using (61) to identify elements of $`𝔨_M`$ with vector fields, Lemma 3.2 is now a direct consequence of (58).
|
warning/0506/gr-qc0506086.html
|
ar5iv
|
text
|
# Explicit multipole moments of stationary axisymmetric spacetimes
## 1 Introduction
The relativistic multipole moments of stationary spacetimes have been defined by Hansen . This definition is an extension of the static case considered by Geroch , and apart from a slightly different setup due to the possible angular momentum, the recursive definitions of the moments in and are the same. The Hansen formulation reduces to the Geroch formulation in the static case but with a different potential. In section 7 we conclude that these two potentials indeed give the same multipole moments in the general axisymmetric static case. Beig defined a generalisation of centre of mass so that the expansion of the Hansen moments around this ’point’ determines the multipole moments uniquely. Thorne gave another definition of multipole moments which is known to be equivalent to the Hansen formulation if the spacetime has non-zero mass. There are also other definitions of multipole moments , , , which will not be considered here. See for instance for further details about these moments.
The recursive definition of multipole moments of Geroch and Hansen (1) takes place in a conformal compactification of the 3-manifold of Killing-trajectories. The recursion produces a family of totally symmetric and trace-free tensors, which are to be evaluated at a certain point, and the values will then provide the moments of the spacetime in question. Even in the case of axisymmetric spacetimes, the actual calculations of the tensors in (1) are non-trivial.
In , it was shown how the moments in the axisymmetric static case can be obtained through a set of recursively defined real valued functions $`\{f_n\}_{n=0}^{\mathrm{}}`$ on $``$. The moments are then given by the values $`\{f_n(0)\}_{n=0}^{\mathrm{}}`$. In this way, one can easily calculate ‘any’ desired number of moments. By exploring the conformal freedom of the construction, it was also shown how all moments could be captured in one real valued function $`y`$, where the moments appeared as the derivatives of $`y`$ at $`0`$. In this paper we show that a similar scalar recursion can be found in the stationary axisymmetric case. The family of real valued functions will be replaced by a family of complex valued functions, allowing for the angular moment parts. It will also be possible to collect all moments into a complex valued function $`y`$, where the moments appear as the derivatives of $`y`$ at $`0`$. Again, the complex part of $`y`$ is related to the angular moment parts of the multipole moments. The derivation in used a two-surface $`S`$, which reflected the axisymmetry of the spacetime. In this paper we will produce the scalar recursion directly, with the methods presented in section 3.
As an application, we will calculate the multipole moments of the Kerr solution. Another issue is the choice of potential. We will show that the potential used by Hansen and the Ernst potential , (as well as a large class of other potentials) give the same multipole-moments.
## 2 Multipole moments of stationary spacetimes
In this section we quote the definition given by Hansen in . We thus consider a stationary spacetime $`(M,g_{ab})`$ with time-like Killing vector field $`\xi ^a`$. We let $`\lambda =\xi ^a\xi _a`$ be the norm, and define the twist $`\omega `$ through $`_a\omega =ϵ_{abcd}\xi ^b^c\xi ^d`$. If $`V`$ is the 3-manifold of trajectories, the metric $`g_{ab}`$ (with signature $`(,+,+,+)`$) induces the positive definite metric
$$h_{ab}=\lambda g_{ab}+\xi _a\xi _b$$
on $`V`$. It is required that $`V`$ is asymptotically flat, i.e., there exists a 3-manifold $`\stackrel{~}{V}`$ and a conformal factor $`\mathrm{\Omega }`$ satisfying
* $`\stackrel{~}{V}=V\mathrm{\Lambda }`$, where $`\mathrm{\Lambda }`$ is a single point
* $`\stackrel{~}{h}_{ab}=\mathrm{\Omega }^2h_{ab}`$ is a smooth metric on $`\stackrel{~}{V}`$
* At $`\mathrm{\Lambda }`$, $`\mathrm{\Omega }=0,\stackrel{~}{D}_a\mathrm{\Omega }=0,\stackrel{~}{D}_a\stackrel{~}{D}_b\mathrm{\Omega }=2\stackrel{~}{h}_{ab}`$,
where $`\stackrel{~}{D}_a`$ is the derivative operator associated with $`\stackrel{~}{h}_{ab}`$. On $`M`$, and/or $`V`$ one defines the scalar potential
$$\varphi =\varphi _M+i\varphi _J,\varphi _M=\frac{\lambda ^2+\omega ^21}{4\lambda },\varphi _J=\frac{\omega }{2\lambda }$$
The multipole moments of $`M`$ are then defined on $`\stackrel{~}{V}`$ as certain derivatives of the scalar potential $`\stackrel{~}{\varphi }=\varphi /\sqrt{\mathrm{\Omega }}`$ at $`\mathrm{\Lambda }`$. More explicitly, following , let $`\stackrel{~}{R}_{ab}`$ denote the Ricci tensor of $`\stackrel{~}{V}`$, and let $`P=\stackrel{~}{\varphi }`$. Define the sequence $`P,P_{a_1},P_{a_1a_2},\mathrm{}`$ of tensors recursively:
$$P_{a_1\mathrm{}a_n}=C[\stackrel{~}{D}_{a_1}P_{a_2\mathrm{}a_n}\frac{(n1)(2n3)}{2}\stackrel{~}{R}_{a_1a_2}P_{a_3\mathrm{}a_n}],$$
(1)
where $`C[]`$ stands for taking the totally symmetric and trace-free part. The multipole moments of $`M`$ are then defined as the tensors $`P_{a_1\mathrm{}a_n}`$ at $`\mathrm{\Lambda }`$. The requirement that all $`P_{a_1\mathrm{}a_n}`$ be totally symmetric and trace-free makes the actual calculations non-trivial. In the axisymmetric case, however, we will see that the tensorial recursion can be replaced by a scalar recursion.
### 2.1 Multipole moments of axisymmetric spacetimes
If, in addition to the requirement that $`M`$ is stationary and asymptotically flat, we also impose the condition that $`M`$ is axisymmetric, the metric can be written in the following canonical form
$$ds^2=\lambda (dtWd\phi )^2+\lambda ^1(R^2d\phi ^2+e^{2\beta }(dR^2+dZ^2)),$$
(2)
where $`\xi ^a=(\frac{}{t})^a`$ and $`(\frac{}{\phi })^a`$ are the timelike and axial Killing vectors. This implies that the metric on $`V`$ is
$$h_{ab}=\lambda g_{ab}+\xi _a\xi _bR^2d\phi ^2+e^{2\beta }(dR^2+dZ^2).$$
To conformally compactify $`V`$, we define new variables $`\stackrel{~}{\rho }`$, $`\stackrel{~}{z}`$ and $`r,\theta `$ via $`\stackrel{~}{\rho }=\frac{R}{R^2+Z^2}=r\mathrm{sin}\theta `$, $`\stackrel{~}{z}=\frac{Z}{R^2+Z^2}=r\mathrm{cos}\theta `$ and put $`\widehat{\mathrm{\Omega }}=r^2e^\beta `$. We then get the rescaled metric, i.e., the metric on $`\stackrel{~}{V}`$ as
$$\widehat{h}_{ab}=\widehat{\mathrm{\Omega }}^2h_{ab}\stackrel{~}{\rho }^2e^{2\beta }d\phi ^2+d\stackrel{~}{\rho }^2+d\stackrel{~}{z}^2=r^2\mathrm{sin}^2\theta e^{2\beta }d\phi ^2+dr^2+r^2d\theta ^2.$$
(3)
Therefore we can assume that the rescaled metric has the form (3), where the infinity point $`\mathrm{\Lambda }`$ corresponds to the point $`r=0`$. However other choices of variables and conformal factors may also give the rescaled metric the form (3). Here we only require that the rescaled metric has the form (3), but we do not require that the original metric has the form (2).
The conformal factor $`\widehat{\mathrm{\Omega }}`$ is not uniquely determined. One can make a further conformal transformation of $`\stackrel{~}{V}`$, using as conformal factor $`e^\kappa `$, where $`\kappa `$ is any smooth function on $`\stackrel{~}{V}`$ with $`\kappa (\mathrm{\Lambda })=0`$. Thus $`\kappa `$ reflects the freedom in choosing $`\widehat{\mathrm{\Omega }}`$. Of particular importance is the value of $`(_a\kappa )(\mathrm{\Lambda })=\kappa ^{}(0)`$. Namely, under a change $`\widehat{\mathrm{\Omega }}\mathrm{\Omega }=\widehat{\mathrm{\Omega }}e^\kappa `$, a non-zero $`\kappa ^{}(0)`$ changes the moments defined by (1) in a way which corresponds to a ‘translation’ of the physical space . With this extra conformal factor the metric becomes
$$\stackrel{~}{h}_{ab}e^{2\kappa }(\stackrel{~}{\rho }^2e^{2\beta }d\phi ^2+d\stackrel{~}{\rho }^2+d\stackrel{~}{z}^2)=e^{2\kappa }(r^2\mathrm{sin}^2\theta e^{2\beta }d\phi ^2+dr^2+r^2d\theta ^2).$$
(4)
By choosing $`\kappa ^{}(0)`$ such that the expansion is taken around the generalised centre of mass , the multipole moments are fixed and invariant under the restricted remaining conformal freedom. Also, from their very construction, the multipole moments are coordinate-independent.
## 3 Multipole moments through a scalar recursion on $`^2`$
In this section, we will show how the assumption of axisymmetry allows us to replace the tensors in (1) by family $`f_n`$ of recursively defined functions on $`^2`$. The multipole moments of $`M`$ will appear as the values of $`f_n`$ at the origin point. The reason that this works is that the following lemma holds:
###### Lemma 1
Suppose $`\stackrel{~}{V}`$ is a 3-manifold with metric given by (4). Then there exists a regularly direction dependent (at $`\mathrm{\Lambda }`$) vector field $`\eta ^a`$ with the following properties:
a) For all tensors $`T_{a_1\mathrm{}a_n}`$, $`\eta ^{a_1}\mathrm{}\eta ^{a_n}T_{a_1\mathrm{}a_n}=\eta ^{a_1}\mathrm{}\eta ^{a_n}C[T_{a_1\mathrm{}a_n}]`$,
b) At $`\mathrm{\Lambda }`$, $`P_{a_1\mathrm{}a_n}`$ is determined by $`\eta ^{a_1}\mathrm{}\eta ^{a_n}P_{a_1\mathrm{}a_n}`$
c) $`\eta ^a\stackrel{~}{D}_a\eta ^b`$ is parallel to $`\eta ^b`$.
Proof:
a) It is sufficient to require that $`\eta _a\eta ^a=0`$, i.e., that $`\eta ^a`$ is a complex null vector on $`\stackrel{~}{V}`$. To get the totally symmetric and trace-free part of a tensor $`T_{a_1\mathrm{}a_n}`$ we first symmetrize and define $`S_{a_1\mathrm{}a_n}=T_{(a_1\mathrm{}a_n)}`$. We then subtract all traces to get $`C[T_{a_1\mathrm{}a_n}]=S_{a_1\mathrm{}a_n}\gamma (n)\stackrel{~}{h}_{(a_1a_2}S_{a_3\mathrm{}a_n)}{}_{}{}^{b}{}_{b}{}^{}\delta (n)\stackrel{~}{h}_{(a_1a_2}h_{a_3a_4}S_{a_5\mathrm{}a_n)}{}_{b}{}^{b}{}_{}{}^{c}{}_{c}{}^{}\mathrm{}`$, where $`\gamma (n),\delta (n),\mathrm{}`$ have their appropriate values and where all terms except the first contain $`\stackrel{~}{h}_{a_ia_j}`$ for some $`i,j`$. From $`\eta _a\eta ^a=0`$ it follows that $`\eta ^{a_1}\mathrm{}\eta ^{a_n}C[T_{a_1\mathrm{}a_n}]=\eta ^{a_1}\mathrm{}\eta ^{a_n}S_{a_1\mathrm{}a_n}=\eta ^{a_1}\mathrm{}\eta ^{a_n}T_{a_1\mathrm{}a_n}`$, where the last equality follows from the fact that $`\eta ^{a_1}\mathrm{}\eta ^{a_n}`$ is totally symmetric.
b) Define
$$\eta ^a=(\frac{}{\stackrel{~}{z}})^ai(\frac{}{\stackrel{~}{\rho }})^a=e^{i\theta }\left((\frac{}{r})^a\frac{i}{r}(\frac{}{\theta })^a\right).$$
(5)
We see that $`\eta _a\eta ^a=0`$ so a) is valid. The axisymmetry implies that at $`\mathrm{\Lambda }`$, $`P_{a_1a_2\mathrm{}a_n}`$ is proportional to $`C[z_{a_1}z_{a_2}\mathrm{}z_{a_n}]`$, i.e.,
$$P_{a_1a_2\mathrm{}a_n}(\mathrm{\Lambda })=m_nC[z_{a_1}z_{a_2}\mathrm{}z_{a_n}],$$
(6)
where $`z_a=(d\stackrel{~}{z})_a`$ is the direction along the symmetry axis. Hence, at $`\mathrm{\Lambda }`$,
$$\eta ^{a_1}\mathrm{}\eta ^{a_n}P_{a_1\mathrm{}a_n}=m_n\eta ^{a_1}\mathrm{}\eta ^{a_n}z_{a_1}\mathrm{}z_{a_n}=m_n.$$
(7)
Therefore, at $`\mathrm{\Lambda }`$, $`P_{a_1\mathrm{}a_n}`$ is determined by $`\eta ^{a_1}\mathrm{}\eta ^{a_n}P_{a_1\mathrm{}a_n}`$.
c) With $`\eta ^a`$ defined by (5), and using the metric (4), a direct calculation gives
$$\eta ^a\stackrel{~}{D}_a\eta ^b=2\eta ^b\eta ^c\stackrel{~}{D}_c\kappa .$$
(8)
Note that the special case $`\kappa =0`$, i.e., when the metric is of the form (3), gives $`\eta ^a\stackrel{~}{D}_a\eta ^b=\eta ^a\widehat{D}_a\eta ^b=0`$. It is also important to note that when we interpret $`\kappa ,\beta ,\stackrel{~}{\varphi }`$ as functions of two variables, say, $`\stackrel{~}{z},\stackrel{~}{\rho }`$, that they are defined in the half-space $`\stackrel{~}{\rho }0`$, but that they can be naturally extended to all $`\stackrel{~}{\rho }`$ via $`\beta (\stackrel{~}{z},\stackrel{~}{\rho })=\beta (\stackrel{~}{z},\stackrel{~}{\rho })`$, etc., so that they are even in $`\stackrel{~}{\rho }`$. Moreover, $`\eta ^a`$ is direction dependent at $`\mathrm{\Lambda }`$ when regarded as a vector field on $`\stackrel{~}{V}`$, but not when regarded as a vector field on the half-space $`\stackrel{~}{\rho }0`$. Moreover, $`\eta ^a`$ can be naturally and smoothly extended to $`\stackrel{~}{\rho }<0`$, and in particular, derivatives like $`\eta ^a\stackrel{~}{}_a\beta `$ will then be smooth at $`\mathrm{\Lambda }`$ but not necessarily even when extended to all values of $`\stackrel{~}{\rho }`$.
We are now ready to simplify the recursion (1). We start by defining
$$f_n=\eta ^{a_1}\eta ^{a_2}\mathrm{}\eta ^{a_n}P_{a_1a_2\mathrm{}a_n},n=0,1,2,\mathrm{}$$
(9)
In particular, $`f_0=P=\stackrel{~}{\varphi }=\mathrm{\Omega }^{\frac{1}{2}}\varphi `$. By contracting (1) with $`\eta ^a`$ we get the following theorem.
###### Theorem 2
Let $`\eta ^a`$ have the properties given by lemma 1, and let $`f_n`$ be defined by (9). We then have the recursion
$$f_n=\eta ^a\stackrel{~}{D}_af_{n1}2(n1)f_{n1}\eta ^a\stackrel{~}{D}_a\kappa \frac{(n1)(2n3)}{2}\eta ^a\eta ^b\stackrel{~}{R}_{ab}f_{n2},$$
(10)
where the moments $`m_n`$ are given by $`m_n=f_n(\mathrm{\Lambda })`$.
Proof:
Let
$$T_{a_1\mathrm{}a_n}=\stackrel{~}{D}_{a_1}P_{a_2\mathrm{}a_n}\frac{(n1)(2n3)}{2}\stackrel{~}{R}_{a_1a_2}P_{a_3\mathrm{}a_n}.$$
(11)
Using the recursion (1) and the properties of $`\eta ^a`$ we get
$`f_n`$ $`=\eta ^{a_1}\mathrm{}\eta ^{a_n}C[T_{a_1\mathrm{}a_n}]=\eta ^{a_1}\mathrm{}\eta ^{a_n}T_{a_1\mathrm{}a_n}`$
$`=\eta ^{a_2}\mathrm{}\eta ^{a_n}\eta ^{a_1}\stackrel{~}{D}_{a_1}P_{a_2\mathrm{}a_n}\frac{(n1)(2n3)}{2}\eta ^{a_1}\eta ^{a_2}\stackrel{~}{R}_{a_1a_2}\eta ^{a_3}\mathrm{}\eta ^{a_n}P_{a_3\mathrm{}a_n}`$
$`=\eta ^a\stackrel{~}{D}_af_{n1}P_{a_2\mathrm{}a_n}\eta ^a\stackrel{~}{D}_a(\eta ^{a_2}\mathrm{}\eta ^{a_n})\frac{(n1)(2n3)}{2}\eta ^a\eta ^b\stackrel{~}{R}_{ab}f_{n2}.`$
Using (8) we get (10). By comparing (7) and (9), we see that $`m_n=f_n(\mathrm{\Lambda })`$.
Note that the proof was carried out in $`\stackrel{~}{V}`$, but the resulting scalar recursion is most conveniently regarded as defined on $`^2`$ with Cartesian coordinates $`\stackrel{~}{z},\stackrel{~}{\rho }`$ or polar coordinates $`r,\theta `$. In $`\eta ^a\eta ^b\stackrel{~}{R}_{ab}`$, $`\stackrel{~}{R}_{ab}`$ refers to the Ricci tensor of $`\stackrel{~}{V}`$; using the metric (4) one readily finds
$$\eta ^a\eta ^b\stackrel{~}{R}_{ab}=\eta ^a\stackrel{~}{D}_a(\eta ^b\stackrel{~}{D}_b\beta )(\eta ^a\stackrel{~}{D}_a\beta )^2\frac{2i}{\stackrel{~}{\rho }}\eta ^a\stackrel{~}{D}_a\beta \eta ^a\stackrel{~}{D}_a(\eta ^b\stackrel{~}{D}_b\kappa )+(\eta ^a\stackrel{~}{D}_a\kappa )^2.$$
(12)
Remark:
As well as different sign conventions some authors use the convention $`M_n=\frac{1}{n!}\stackrel{~}{z}^{a_1}\mathrm{}\stackrel{~}{z}^{a_n}P_{a_1\mathrm{}a_n}(\mathrm{\Lambda })`$. This implies that $`M_n=\pm \frac{2^nn!}{(2n)!}m_n`$.
## 4 Multipole moments through a scalar recursion on $``$
In this section we show that we can reduce the problem from scalar fields of two variables to scalar fields of one variable. So far we have not required analyticity of the fields. There are several proofs of analyticity for the potential and the metric after a special rescaling in the case with non-zero mass . However, since we don’t want to exclude spacetimes with zero mass, analyticity will be assumed explicitly. In this article, analyticity will always be taken in the ’real’ sense, i.e. that the fields are given by power series with positive radii of convergence. In the cases where the results of this paper are used to compute the multipole moments for a certain spacetime, it is easy to check analyticity.
###### Lemma 3
Let $`\beta `$ come from (3), $`\stackrel{~}{\varphi }`$ be the rescaled (axisymmetric) potential, and suppose that $`\beta `$ and $`\stackrel{~}{\varphi }`$ are analytic in a neighbourhood of $`\mathrm{\Lambda }\stackrel{~}{V}`$. Then $`\beta (\stackrel{~}{z},\stackrel{~}{\rho })`$ contains a factor $`\stackrel{~}{\rho }^2`$. Furthermore, viewed as a functions on $`^2`$, $`\eta ^a\eta ^b\stackrel{~}{R}_{ab}`$ and $`f_n`$ are analytic in $`(\stackrel{~}{z},\stackrel{~}{\rho })`$ in a neighbourhood of $`(0,0)`$.
Proof:
First we establish that $`\beta `$ contains a factor $`\stackrel{~}{\rho }^2`$. Consider an orbit $`\mathrm{\Gamma }:\{\stackrel{~}{\rho }=\rho _0\}`$ in the surface $`\stackrel{~}{z}=z_0`$. The circumference $`C(z_0,\rho _0)=_\mathrm{\Gamma }𝑑s=_0^{2\pi }\rho _0e^{\beta (z_0,\rho _0)}𝑑\varphi =2\pi \rho _0e^{\beta (z_0,\rho _0)}`$. We also know that $`C(z_0,\rho _0)=2\pi \rho _0\frac{\pi K_0\rho _0^3}{3}+𝒪(\rho _0^4)`$, where $`K_0`$ is the scalar curvature at $`\stackrel{~}{\rho }=0`$. Hence
$`{\displaystyle \frac{\pi K_0}{3}}`$ $`=\underset{\rho _00^+}{lim}{\displaystyle \frac{2\pi \rho _0C(z_0,\rho _0)}{\rho _0^3}}`$
$`=\underset{\rho _00^+}{lim}{\displaystyle \frac{2\pi }{\rho _0^2}}(1e^{\beta (z_0,\rho _0)})=\underset{\rho _00^+}{lim}{\displaystyle \frac{2\pi \beta (z_0,\rho _0)}{\rho _0^2}}(1+𝒪(\beta (z_0,\rho _0))).`$
Since $`\stackrel{~}{z}=z_0`$ is a smooth manifold, $`K_0`$ is finite, and we have that $`\beta (\stackrel{~}{z},\stackrel{~}{\rho })`$ contains the factor $`\stackrel{~}{\rho }^2=r^2\mathrm{sin}^2\theta `$. From this we see that $`\frac{2i}{\stackrel{~}{\rho }}\eta ^a\stackrel{~}{D}_a\beta `$ is analytic in $`(\stackrel{~}{z},\stackrel{~}{\rho })`$ and consequently all terms in (12) are analytic in $`(\stackrel{~}{z},\stackrel{~}{\rho })`$. The analyticity of $`f_0=\stackrel{~}{\varphi }`$ is assumed. It follows inductively from (10) that all $`f_n`$ are analytic in $`(\stackrel{~}{z},\stackrel{~}{\rho })`$.
To obtain a scalar recursion on $``$, we must extract/define functions of one real variable from our given functions on $`^2`$, where the latter are supposed to be analytic in terms of $`(\stackrel{~}{z},\stackrel{~}{\rho })`$.
###### Definition 4
Suppose that $`g:^2`$ is an analytic function of two variables in a neighbourhood of the origin. We then define the leading order function $`g_L`$ on $``$ via $`g_L(r)=g(r,ir)`$.
Note that the analyticity condition allows us to write $`g(\stackrel{~}{z},\stackrel{~}{\rho })=_0^{\mathrm{}}_0^{\mathrm{}}a_{nk}\stackrel{~}{z}^n\stackrel{~}{\rho }^k`$ for $`(\stackrel{~}{z},\stackrel{~}{\rho })`$ in some disc $`B_\delta (0)`$ centred at the origin. Thus $`\stackrel{~}{z},\stackrel{~}{\rho }`$ can be allowed to be complex in $`B_\delta (0)`$ and the definition $`g_L(r)=g(r,ir)`$ makes sense as long as $`r\sqrt{2}<\delta `$.
We will now fix $`\eta ^a`$ to be precisely $`\eta ^a=(\frac{}{\stackrel{~}{z}})^ai(\frac{}{\stackrel{~}{\rho }})^a`$. That the recursion (9) can be replaced by a recursion of the leading order functions, will then follow from the following lemma:
###### Lemma 5
If $`g`$ satisfies the conditions of definition 4, $`(\eta ^a\stackrel{~}{D}_ag)_L(r)=g_L^{}(r)`$.
Proof: Both sides equal $`\frac{g}{\stackrel{~}{z}}(r,ir)i\frac{g}{\stackrel{~}{\rho }}(r,ir)`$.
We are now ready to derive the promised scalar recursion on $``$.
###### Theorem 6
Suppose that the conditions in lemma 3 are met, and define $`y_n(r)=(f_n(\stackrel{~}{z},\stackrel{~}{\rho }))_L(r)`$. Then the multipole moments $`m_n`$ are given through the recursion
$$y_n=y_{n1}^{}2(n1)\kappa _L^{}y_{n1}\frac{(n1)(2n3)}{2}My_{n2},$$
(13)
where
$$M(r)=\beta _L^{\prime \prime }(\beta _L^{})^2+\frac{2}{r}\beta _L^{}\kappa _L^{\prime \prime }+(\kappa _L^{})^2$$
(14)
and $`m_n=y_n(0)`$.
Proof: Taking the leading order part of (10), putting $`M(r)=(\eta ^a\eta ^b\stackrel{~}{R}_{ab})_L(r)`$, and applying lemma 5, we get (13). Also, $`m_n=f_n(0,0)=y_n(0)`$.
## 5 All moments from one scalar function
The recursion (13) would simplify considerably if $`M=0`$, since we then get a recursion of depth one. As is seen below, it is possible to choose $`\kappa _L`$ such that $`M=0`$. The simplified recursion is then solved by choosing a different radial parameter $`\rho `$. The choice of $`\kappa _L`$ in order for $`M=0`$ to hold in (13) can be viewed in two ways. Either, one starts with the spacetime cast in the form (4) and chose a particular $`\kappa `$, or else one uses the metric $`\widehat{h}_{ab}`$ in the form (3) as the starting point. In the latter case, starting with $`\widehat{h}_{ab}`$, we have the potential $`\widehat{\varphi }=\varphi /\sqrt{\mathrm{\Omega }}`$, and perform a second conformal rescaling $`\mathrm{\Omega }\mathrm{\Omega }e^\kappa `$ where $`\kappa `$ is chosen such that $`\kappa _L`$ solves equation (14) with $`M=0`$. The new potential $`\stackrel{~}{\varphi }`$ is then related to $`\widehat{\varphi }`$ via $`\stackrel{~}{\varphi }=e^{\kappa /2}\widehat{\varphi }`$. Although equivalent, the latter viewpoint can be advantageous when explicit spacetimes are considered (cf. section 6).
###### Theorem 7
Suppose that $`M`$ is a stationary axisymmetric asymptotically flat spacetime, with $`\stackrel{~}{V}`$ a conformal rescaling of the manifold of time-like Killing-trajectories, where the metric $`\widehat{h}_{ab}`$ on $`\stackrel{~}{V}`$ has the form (3). Let $`\widehat{\varphi }`$ be the conformally rescaled potential. Furthermore, assume that $`\widehat{\varphi }`$ and $`\beta `$ are analytic in a neighbourhood of $`\mathrm{\Lambda }\stackrel{~}{V}`$. Choose $`\kappa `$ such that, for some constant $`C`$,
$$\kappa _L(r)=\mathrm{ln}(1r_0^r\frac{e^{2\beta _L(r)}1}{r^2}𝑑rrC)+\beta _L(r),$$
(15)
thereby inducing the metric $`\stackrel{~}{h}_{ab}=e^{2\kappa }\widehat{h}_{ab}`$. Define $`r(\rho )`$ implicitly by $`\rho (r)=re^{\kappa _L\beta _L}`$. Put $`y(\rho )=\stackrel{~}{\varphi }_L(r(\rho ))=e^{\kappa _L(r(\rho ))/2}\widehat{\varphi }_L(r(\rho ))`$. Then the multipole moments $`m_0,m_1,`$ of $`M`$ are given by $`m_n=\frac{d^ny}{d\rho ^n}(0)`$.
Remark:
In the expression for $`\kappa _L`$, the constant $`C`$ is seen to equal $`\kappa _L^{}(0)=\kappa ^{}(\mathrm{\Lambda })`$. In particular one can choose $`\kappa _L^{}(0)`$ such that the expansion is taken around the generalised centre of mass. With this choice the multipole moments are unique. This is accomplished in the following way: If $`k`$ is the first integer such that $`m_k0`$, we choose $`\kappa _L^{}(0)`$ so that $`m_{k+1}=0`$. This is equivalent to the definition of generalised centre of mass: $`0=P_{a_1\mathrm{}a_{k+1}}(\mathrm{\Lambda })P^{a_1\mathrm{}a_k}(\mathrm{\Lambda })=m_{k+1}m_kC[z_{a_1}\mathrm{}z_{a_{k+1}}]C[z^{a_1}\mathrm{}z^{a_k}]m_{k+1}m_kz_{a_{k+1}}`$. Note also that the non-leading terms of $`\kappa `$ can be chosen arbitrarily.
Proof:
From lemma 3, it follows that $`\beta _L(0)=\beta _L^{}(0)=0`$. Taking this into account, a direct insertion of (15) into (14) shows that $`M(r)=0`$, i.e., this choice of $`\kappa _L`$ solves the differential equation $`M(r)=0`$. Therefore, with this choice of $`\kappa _L`$ the recursion (13) reduces to
$$y_n=y_{n1}^{}2(n1)\kappa _L^{}y_{n1}.$$
(16)
Put $`z_n(r)=e^{2n\kappa _L(r)}y_n(r)`$ so that $`z_n(0)=y_n(0)=m_n`$. As in , (16) becomes
$$z_n(r)=e^{2\kappa _L(r)}z_{n1}^{}(r)$$
(17)
If $`\rho `$ is such that $`\frac{d\rho (r)}{dr}=e^{2\kappa _L(r)},`$ we find that $`z_n=\frac{dz_{n1}}{d\rho }=\frac{d^nz_0}{d\rho ^n}`$. The specific choice (15) of $`\kappa _L`$ implies that $`r(\kappa _L^{}\beta _L^{})+1=e^{\kappa _L+\beta _L}`$. Hence,
$$\rho (r)=_0^re^{2\kappa _L}𝑑r=_0^r(r(\kappa _L^{}\beta _L^{})+1)e^{\kappa _L\beta _L}𝑑r=re^{\kappa _L\beta _L}.$$
(18)
Thus equation (18) implicitly defines $`r(\rho )`$ in a neighbourhood of $`\rho =0`$. We can now define $`y(\rho )=z_0(r(\rho ))=\stackrel{~}{\varphi }_L(r(\rho ))`$ and get all the moments from the function $`y(\rho )`$:
$$m_n=\frac{d^ny}{d\rho ^n}(0).$$
(19)
## 6 The Kerr solution
As an example of how our method can be used, we compute the multipole moments for the Kerr solution. Here we will not use the Weyl canonical coordinates because the expressions would be much longer than necessary. Instead we will follow Hansen . We begin with Boyer-Lindquist coordinates $`(\stackrel{~}{r},\theta ,\phi )`$ and the metric
$$ds^2=\frac{\stackrel{~}{r}^22m\stackrel{~}{r}+a^2\mathrm{cos}^2\theta }{\stackrel{~}{r}^22m\stackrel{~}{r}+a^2}d\stackrel{~}{r}^2+(\stackrel{~}{r}^22m\stackrel{~}{r}+a^2\mathrm{cos}^2\theta )d\theta ^2+(\stackrel{~}{r}^22m\stackrel{~}{r}+a^2)d\phi ^2.$$
(20)
Following Hansen we define a new radial coordinate $`r`$ through $`\stackrel{~}{r}=r^1(1+mr+\frac{1}{4}(m^2a^2)r^2)`$ and use the conformal factor $`\widehat{\mathrm{\Omega }}=\frac{r^2}{\sqrt{(1\frac{1}{4}(m^2a^2)r^2)^2a^2r^2\mathrm{sin}^2\theta }}`$. Define $`\stackrel{~}{z}=r\mathrm{cos}\theta `$ and $`\stackrel{~}{\rho }=r\mathrm{sin}\theta `$. We then get
$`\beta (\stackrel{~}{z},\stackrel{~}{\rho })`$ $`={\displaystyle \frac{1}{2}}\mathrm{ln}\left(1{\displaystyle \frac{(4a\stackrel{~}{\rho })^2}{(4(m^2a^2)(\stackrel{~}{z}^2+\stackrel{~}{\rho }^2))^2}}\right),`$ (21)
$`\beta _L(r)`$ $`={\displaystyle \frac{1}{2}}\mathrm{ln}(1+a^2r^2),`$ (22)
$`\widehat{\varphi }(\stackrel{~}{z},\stackrel{~}{\rho })`$ $`={\displaystyle \frac{m(1+\frac{1}{4}(m^2a^2)(\stackrel{~}{z}^2+\stackrel{~}{\rho }^2)ia\stackrel{~}{z})}{((1\frac{1}{4}(m^2a^2)(\stackrel{~}{z}^2+\stackrel{~}{\rho }^2)^2)^2a^2\stackrel{~}{\rho }^2)^{\frac{3}{4}}}},`$ (23)
$`\widehat{\varphi }_L(r)`$ $`={\displaystyle \frac{m(1iar)}{(1+a^2r^2)^{3/4}}}.`$ (24)
Furthermore, from theorem 7 we get
$`\kappa _L(r)`$ $`={\displaystyle \frac{1}{2}}\mathrm{ln}\left({\displaystyle \frac{(r^2a^2+r\kappa ^{}(0)1)^2}{r^2a^2+1}}\right),`$ (25)
$`\rho (r)`$ $`={\displaystyle \frac{r}{1r\kappa ^{}(0)r^2a^2}},`$ (26)
$`r(\rho )`$ $`={\displaystyle \frac{\sqrt{(\rho \kappa ^{}(0)+1)^2+4a^2\rho ^2}\rho \kappa ^{}(0)1}{2\rho a^2}},`$ (27)
$`y(\rho )`$ $`=e^{\kappa _L(r)/2}\widehat{\varphi }_L(r)={\displaystyle \frac{m\sqrt{1r\kappa ^{}(0)a^2r^2}}{1+iar}}={\displaystyle \frac{m}{\sqrt{1+(2ia+\kappa ^{}(0))\rho }}}.`$ (28)
Comparing with the Schwarzschild solution, $`a=0`$, we see that the Kerr solution is in a sense an imaginary translation of the Schwarzschild spacetime. Similar descriptions can be found in and .
With $`\kappa ^{}(0)=0`$, we get the expansion around the centre of mass:
$$\underset{n=0}{\overset{\mathrm{}}{}}\frac{m_n\rho ^n}{n!}=y(\rho )=\underset{n=0}{\overset{\mathrm{}}{}}\frac{m(2ia\rho )^n\sqrt{\pi }}{n!\mathrm{\Gamma }(\frac{1}{2}n)}.$$
(29)
With the conventions used by Hansen we have
$$M_n=\frac{2^nn!}{(2n)!}m_n=\frac{2^nn!m(2ia)^n\sqrt{\pi }}{(2n)!\mathrm{\Gamma }(\frac{1}{2}n)}=(i)^nma^n.$$
(30)
Thus our calculation agrees with Hansen . (In , the moments were given without a proof, see however .)
## 7 Potentials
So far we have only used the complex version of the potential defined by Hansen <sup>1</sup><sup>1</sup>1We have changed sign compared to
$$\varphi _H=\frac{\lambda ^21+\omega ^2+2i\omega }{4\lambda },$$
(31)
and as Hansen points out, the choice of potential is not unique. This freedom of choice was addressed in using a different but equivalent definition of multipole moments compared to the Geroch-Hansen definition. Here we will however only use theorem 6 to investigate the freedom in the choice of potential.
For instance, in the special case when the stationary axisymmetric spacetime is actually static, the Hansen potential reduces to
$$\varphi _H=\frac{1\lambda ^2}{4\lambda }.$$
(32)
This potential is not the same as the one given by Geroch, who uses<sup>2</sup><sup>2</sup>2Again, we have changed sign. The factor $`\lambda ^{1/4}`$ comes from different definitions of the metric for $`V`$.
$$\varphi _G=\lambda ^{1/4}(1\sqrt{\lambda }).$$
(33)
It is not obvious that (32) and (33) give the same sequence of multipole moments. This was however shown in . The following theorem proves the same thing in the axisymmetric case using a different technique.
###### Theorem 8
Suppose that the spacetime $`M`$ is static and axisymmetric. Then the potentials (32) and (33) produce the same moments.
Proof:
We may assume that the coordinates are chosen in the way described in section 2.1 with $`\kappa =0`$, i.e., $`\stackrel{~}{\mathrm{\Omega }}=r^2e^\beta =(\stackrel{~}{\rho }^2+\stackrel{~}{z}^2)e^\beta `$; in other words we use Weyl coordinates. With this choice of coordinates, $`\stackrel{~}{\alpha }=\frac{\mathrm{ln}\lambda }{2r}`$ is (flat-) harmonic with respect to $`(\stackrel{~}{z},\stackrel{~}{\rho },\phi )`$ taken as cylindrical coordinates in $`^3`$. From this it follows that $`\frac{\varphi _G}{r}=\frac{2}{r}\mathrm{sinh}\frac{\stackrel{~}{\alpha }r}{2}`$ is analytic near $`\mathrm{\Lambda }\stackrel{~}{V}`$, since $`\frac{2}{r}\mathrm{sinh}\frac{\stackrel{~}{\alpha }r}{2}`$ is even in $`r`$. It also follows from the explicit form of $`\beta `$ in that $`\beta `$ is analytic and thus $`\mathrm{\Omega }`$ is analytic with $`\mathrm{\Omega }_L=0`$. From the definitions (32) and (33) we see that
$$\stackrel{~}{\varphi }_H=\frac{\stackrel{~}{\varphi }_G}{4}(\mathrm{\Omega }\stackrel{~}{\varphi }_G^2+2)\sqrt{\mathrm{\Omega }\stackrel{~}{\varphi }_G^2+4},$$
(34)
so that $`\stackrel{~}{\varphi }_H`$ is also analytic. Taking leading order functions on both sides implies $`(\stackrel{~}{\varphi }_H)_L=(\stackrel{~}{\varphi }_G)_L`$. Referring to theorem 6, this is sufficient for both potentials to generate the same multipole moments.
Now consider the stationary case again. Assume that $`\stackrel{~}{\varphi }_H`$ and $`\mathrm{\Omega }`$ are analytic functions in a neighbourhood of $`\mathrm{\Lambda }`$. The same will then hold for $`\stackrel{~}{\varphi }_M`$ and $`\stackrel{~}{\varphi }_J`$, the real and imaginary part of $`\stackrel{~}{\varphi }_H`$ respectively. Asymptotic flatness implies that $`\mathrm{\Omega }_L=0`$. Using theorem 6 we see that we can in fact choose any well behaved potential that satisfies $`(\stackrel{~}{\varphi })_L=(\stackrel{~}{\varphi }_H)_L`$ to get the same moments. For instance if we wanted to use the Ernst potential , instead,
$$\varphi _E=\frac{1\lambda i\omega }{1+\lambda +i\omega },$$
(35)
we see that
$$\stackrel{~}{\varphi }_E=\frac{2\stackrel{~}{\varphi }_H}{1+\sqrt{4\mathrm{\Omega }\stackrel{~}{\varphi }_M^2+4\mathrm{\Omega }\stackrel{~}{\varphi }_J^2+1}}.$$
(36)
Taking leading order functions of both sides, we find that $`(\stackrel{~}{\varphi }_E)_L=(\stackrel{~}{\varphi }_H)_L`$, and thus the potentials produce the same moments. This proves the following theorem.
###### Theorem 9
Suppose that the spacetime $`M`$ is stationary and axisymmetric. Then the potentials (31) and (35) produce the same moments.
Remark
This result could also be obtained easily by the method in .
## 8 Discussion
We have shown that the relatively complicated tensorial recursion (1) of Geroch and Hansen can, in the stationary axisymmetric case, be reduced to a recursion of scalar functions. Furthermore, we have demonstrated how a careful choice of conformal factor collects all moments into one complex valued function on $``$, where the moments appear as the derivatives at $`0`$. These results reduce to the results in in the static case. The main ideas in and in this paper are similar, but here we have simplified the proofs as well as generalised the results. As an application of this method, we have shown how easily the moments of the Kerr solution follow. We have also seen that there is a great freedom in the choice of the potential. This can possibly be used to simplify the field equations expressed in terms of the potential.
In it was shown how to obtain the metric for a static axisymmetric spacetime with prescribed multipole moments. With the methods presented here it is natural to try to extend these results to the stationary case. Such results would be very desirable due to the physical relevance of the multipole moments. For instance, axisymmetric stationary solutions are good approximations for astrophysical objects.
It would also be interesting to address the growth condition of the multipole moments for existence of stationary spacetimes. Another natural extension of the work presented here is of course the general non-axisymmetric case.
|
warning/0506/hep-th0506191.html
|
ar5iv
|
text
|
# Contents
## 1 Introduction
Heterotic strings have long provided the most promising candidate for unified description of phenomenology despite persisting problems. It was realized in that some of these problems can be resolved if one considers their strongly coupled limit, given by M-theory on an interval . The corresponding four-dimensional compactification , called heterotic M-theory, has received a great deal of attention. Especially interesting, in view of the astronomical observations indicating a positive cosmological constant and an exponential expansion of the early universe, are the recently found de Sitter and assisted inflation solutions. Heterotic M-theory has the very distinctive feature that unlike the weakly coupled case it does not allow vanishing background flux.
In string theory, nonzero fluxes play a significant role in the resolution of the moduli stabilization problem. The latter occurs in purely geometric compactifications due to the lack of a potential for the many scalar fields that originate from deformations of the internal Calabi-Yau manifold. These moduli are of two types depending on whether they parametrize the complex structure or the Kähler structure deformations. Stabilizing them is essential for predictability of the four-dimensional coupling constants and also for avoiding decompactification of the internal space. It was realized in the context of type IIB that background fluxes generically lift the flat directions of the complex structure moduli by generating a superpotential for them. But this superpotential does not depend on the Kähler moduli. So in order to stabilize the latter one has to resort to quantum corrections.<sup>1</sup><sup>1</sup>1In type IIA, however, all moduli can be stabilized classically . There are two kinds of nonperturbative effects that can create a potential for the Kahler moduli: D-brane instantons and gaugino condensation. It was argued in that using these and nonzero NS-NS and RR fluxes one can fix all moduli.
Another modification of the Kähler potential is due to $`\alpha ^{}`$ corrections , which appear as higher derivative terms in the string effective action. Typically though, their contribution was expected to be suppressed in the large volume limit. However, this was shown to be too naive in . That work argued that, as classically the Kähler moduli are flat directions of the potential, the perturbative, $`\alpha ^{}`$, corrections are generically the leading ones even at large volume. Since they dominate the non-perturbative contributions, their presence alters qualitatively the structure of the scalar potential.
In M-theory there are also higher derivative corrections to the eleven-dimensional supergravity action.<sup>2</sup><sup>2</sup>2They can be deduced from duality with ten-dimensional string theory or from superparticle scattering amplitudes in $`d=11`$ . When compactified on $`\text{}^{1,9}\times \text{𝕊}^1/\text{}_2`$, the theory has an expansion in powers of $`\kappa ^{2/3}`$ with $`\kappa `$ being the gravitational coupling constant. Completing the effective action at orders $`\kappa ^{4/3}`$ and higher encounters technical problems that may be overcome in the approach of . Our main focus in this work is the eleven-dimensional $`R^4`$ term which appears at $`𝒪(\kappa ^{4/3})`$. Naturally, it is expected to correct the Kähler potential for the moduli fields of the appropriate Calabi-Yau three-fold compactification to four dimensions. We compute this correction as well as its implications for the soft supersymmetry breaking terms in the four-dimensional effective theory. As in , the scalar soft masses still do not receive any tree level contribution for supersymmetry breaking in the $`T`$-modulus direction.
The $`R^4`$ term should also affect the geometry of the background solution. To first order in $`\kappa ^{2/3}`$, the solution of heterotic M-theory was studied in . Due to the $`E_8`$ gauge fields propagating on the two boundaries, the Bianchi identity of the supergravity three-form is modified, leading to nonzero background flux and a generically non-Kähler deformation of the initial $`CY(3)`$. Clearly, this is the strong-coupling description of the weakly coupled heterotic string compactifications with torsion .<sup>3</sup><sup>3</sup>3Recent improvement in their understanding is due to , where it was shown that up to order $`𝒪(\alpha ^2)`$ the supersymmetry conditions and Bianchi identities imply the field equations (recall that generically this is true only for maximally supersymmetric backgrounds), and also , where the role of gaugino condensation for the effective four-dimensional superpotential was clarified. Progress in the explicit construction of the latter non-Kähler backgrounds was achieved only very recently .<sup>4</sup><sup>4</sup>4Strictly speaking, the work of gives backgrounds only for the $`SO(32)`$ heterotic string. But presumably one can use similar methods for the $`E_8\times E_8`$ case. We thank Radu Tatar for a discussion on that issue. On the other hand, not much is known about the explicit form of their strongly coupled lift at orders higher than $`\kappa ^{2/3}`$. One can simplify things by considering only warp factor deformations of the eleven-dimensional metric, which in particular means only conformal deformations of the Calabi-Yau.<sup>5</sup><sup>5</sup>5Of course, that implies certain restrictions on the three-form flux. In this case, a non-linear background containing corrections of all orders in $`\kappa ^{2/3}`$ was obtained in , without taking into account any higher derivative terms in the eleven-dimensional effective action. This solution was used in an essential way in . As the eleven-dimensional $`R^4`$ term appears already at second order in the $`\kappa ^{2/3}`$ expansion, a valid question is how it modifies the solution of . We will see that certain relations between warp factors have to be different in the present case. Also, the presence of higher derivative terms opens up the possibility of turning on simultaneously different flux components while still preserving the only-warp-factor character of the geometric deformation, which was not possible before. We find the anticipated shift of the Calabi-Yau volume by a constant proportional to the Euler number, although it is unlikely that this would resolve the singularity of the non-linear background of , appearing at a point along the interval $`\text{𝕊}^1/\text{}_2`$ at which the Calabi-Yau volume shrinks to zero .
The non-linear solution of was used in to find de Sitter vacua in heterotic M-theory by balancing gaugino condensation against membrane instantons. One might expect that, similarly to the string theory case , here too the $`R^4`$ corrections will be dominant over these non-perturbative effects. However, this does not happen essentially because the no scale structure of the Kähler potential for the $`T`$-modulus is preserved by the $`R^4`$ term. On the other hand, taking into account higher derivative corrections means that one can not integrate the solution to all orders in $`\kappa ^{2/3}`$ but instead should always work only to the appropriate level of accuracy. Therefore, it is worth revisiting the considerations of , that led to the existence of de Sitter vacua, in the context of the $`R^4`$ corrected effective action. We will see that de Sitter vacua still exist although they appear to have a much smaller cosmological constant as a result of employing the truncated to $`𝒪(\kappa ^{4/3})`$ solution. The $`R^4`$ induced corrections are of the order of a small percentage and strengthen/weaken the positive value of the energy density for a positive/negative value of the CY Euler number.
The present paper is organized as follows. In Section 2 we review necessary material about the linear solution of and write down the correction to the Kähler potential due to the $`R^4`$ term. A detailed derivation of this correction is given in Appendix A. In Section 3 we calculate the $`R^4`$ induced contributions to the soft supersymmetry breaking masses of the gravitino, gaugino and scalar fields and also to the trilinear couplings. In section 4 we consider the influence of the $`R^4`$ correction on the geometric background. In Subsection 4.1 we find solutions for the case of a metric deformation given by warp factors only. In Subsection 4.2 we address generic non-Kähler deformations of the initial Calabi-Yau. We elaborate more on that in Appendix B, where we derive the appropriate generalized Hitchin flow equations, that constitute the first steps towards finding explicit solutions for the strongly coupled limit of heterotic strings on generic non-Kähler manifolds. Finally, Section 5 is dedicated to the investigation of the minima of the scalar potential in the presence of the $`R^4`$ terms.
## 2 $`R^4`$ terms and Horava-Witten theory
The strongly coupled limit of the heterotic $`E_8\times E_8`$ string theory was argued in to be given at low energies by eleven-dimensional supergravity on $`\text{}^{1,9}\times \text{𝕊}^1/\text{}_2`$. To obtain an effective field theory on four-dimensional Minkowski space, one further compactifies on a Calabi-Yau three-fold. The eleven-dimensional action has an expansion in powers of $`\kappa ^{2/3}`$, where $`\kappa `$ is the gravitational coupling constant. More precisely, it is of the form<sup>6</sup><sup>6</sup>6It has been argued in that the expansion in powers of $`\kappa ^{2/3}`$ is in fact an expansion in the dimensionless quantity $`ϵ=\frac{2\pi L}{3V_v^{2/3}}(\frac{\kappa }{4\pi })^{\frac{2}{3}}`$, where $`L`$ is the length of the interval and $`V_v`$ is the Calabi-Yau volume at the visible boundary.
$$S_{11d}=\frac{1}{\kappa ^2}\left(S^{(0)}+\kappa ^{2/3}S^{(1)}+\kappa ^{4/3}S^{(2)}+\mathrm{}\right)$$
(2.1)
with $`S^{(0)}/\kappa ^2`$ being the well-known Cremmer-Julia-Scherk action :
$$S_0=\frac{1}{2\kappa _{11}^2}d^{11}x(R1\frac{1}{2}G_{(4)}G_{(4)}\frac{1}{6}C_{(3)}G_{(4)}G_{(4)}),$$
(2.2)
where $`G_{(4)}=dC_{(3)}`$. The reduction to four dimensions was performed in completely to order $`\kappa ^{2/3}`$, meaning the term $`S^{(1)}`$, and partially at order $`\kappa ^{4/3}`$, meaning that only some of the contributions to $`S^{(2)}`$ were found. Since this will be important in the following, let us recall the main features of the results of .
To start with, the presence of the two boundaries in eleven dimensions modifies the Bianchi identity of the three-form $`C_{(3)}`$ and hence a vanishing C-field background is not a solution anymore. As a consequence of this, the eleven-dimensional metric acquires nontrivial warp factors. In other words, the initial direct product $`M_4\times CY(3)\times \text{𝕊}^1/\text{}_2`$ gets deformed to a nontrivial fibration of a (in general non-Kähler) 6d manifold over the interval $`\text{𝕊}^1/\text{}_2`$. To first order the deformed metric has the form :
$$ds_{11d}^2=g_{IJ}^{(0)}dx^Idx^J+b\eta _{\mu \nu }dx^\mu dx^\nu +h_{mn}dx^mdx^n+k(dx^{11})^2,$$
(2.3)
where $`I,J=1,\mathrm{},11`$; $`\mu ,\nu =1,\mathrm{},4`$ and $`m,n=5,\mathrm{},10`$. The universal moduli<sup>7</sup><sup>7</sup>7As in , these are the only moduli we will concentrate on. (meaning those that are independent of the particular $`CY(3)`$ that one is considering) of the original direct product metric $`g_{IJ}^{(0)}`$ are the volume of the Calabi-Yau space and the size of the interval (orbifold) $`\text{𝕊}^1/\text{}_2`$. Hence one can write
$$g_{IJ}^{(0)}dx^Idx^J=g_{\mu \nu }dx^\mu dx^\nu +e^{2a}\stackrel{~}{g}_{mn}dx^mdx^n+e^{2c}(dx^{11})^2,$$
(2.4)
where the Calabi-Yau volume and the interval size are parametrized in terms of the scalar fields $`a(x^\mu ),c(x^\mu )`$. The deformed metric (2.3) has the same moduli, but the dependence on them is more complicated since generically:
$$b=b(a,c),h_{mn}=h_{mn}(a,c),k=k(a,c).$$
(2.5)
We will not need the explicit form of the first order solution (2.3), found in . We only note that it is such that there are no order $`𝒪(\kappa ^{2/3})`$ corrections to the effective action that are due to the deformed metric or $`C`$-field . As shown in , the only contribution at that order comes from the gauge multiplets propagating on the ten-dimensional boundaries. The contribution of the latter at order $`\kappa ^{4/3}`$ is the $`𝒪(\kappa ^{4/3})`$ part of the action that was found in . To complete the effective action at this order one has to take into account two more contributions: from higher derivative terms and from higher order deformations of the background metric. The first will be our immediate concern; the second we will address in Section 4.
The lowest order higher derivative corrections to the eleven-dimensional supergravity action are the $`R^4`$ term and its superpartner $`CR^4`$ :<sup>8</sup><sup>8</sup>8Actually, in Horava-Witten theory there is another contribution: Gauss-Bonnet $`R^2`$ terms localized on the two ten-dimensional boundaries . However, they give rise to higher (than two) derivative terms in the four-dimensional effective action and similarly to we omit such contributions in the following.
$$S_1=b_1T_2d^{11}x\sqrt{g}\left[t_8t_8RRRR\frac{1}{4}E_84ϵ_{11}C_{(3)}(trR^4\frac{1}{4}(trR^2)^2)\right],$$
(2.6)
where
$$E_8=\frac{1}{3!}ϵ_{IJKM_1\mathrm{}M_8}ϵ^{IJKN_1\mathrm{}N_8}R_{}^{M_1M_2}{}_{N_1N_2}{}^{}R_{}^{M_3M_4}{}_{N3N_4}{}^{}R_{}^{M_5M_6}{}_{N_5N_6}{}^{}R_{}^{M_7M_8}{}_{N_7N_8}{}^{}.$$
(2.7)
The constants and parameters in the action $`S_1`$ are defined by<sup>9</sup><sup>9</sup>9We use the notation and conventions of .
$$2\kappa _{11}^2=(2\pi )^5l_{11}^9,l_{11}=(2\pi g_s)^{1/3}\alpha _{}^{}{}_{}{}^{1/2},T_2=\frac{1}{2\pi l_{11}^3},b_1=\frac{1}{(2\pi )^43^22^{13}}.$$
(2.8)
For future convenience we also introduce the notation $`\epsilon b_1T_2\kappa ^2`$. Clearly $`T_2=(2\pi )^{2/3}(2\kappa _{11}^2)^{1/3}`$ and hence $`S_1`$ appears at order $`\kappa ^{4/3}`$ in the expansion (2.1). In the following, our goal will be to extract its contribution to the moduli space metric of the four-dimensional effective theory.
So far we have introduced only half of the universal moduli, namely $`a`$ and $`c`$. The other two scalar fields, $`\sigma _S`$ and $`\sigma _T`$, are axions that arise from the eleven-dimensional 3-form $`C`$ via the identifications
$$C_{mn11}^{(0)}=\frac{1}{6\sqrt{2}}\sigma _TJ_{mn},\sqrt{2}_4dB=d\sigma _S\mathrm{with}B_{\mu \nu }=6C_{\mu \nu 11}^{(0)},$$
(2.9)
where $`J_{mn}`$ is the Kähler form of the $`CY(3)`$. In the four-dimensional $`N=1`$ theory the 4 universal moduli make up the bosonic components of two chiral superfields:
$$S=e^{6a}+i\sigma _S,T=e^{\widehat{c}}+i\sigma _T,$$
(2.10)
where we have redefined $`\widehat{c}=c+2a`$ for later purposes. Due to the superfield structure and the independence of the Kähler potential $`K`$ on $`ImS`$ and $`ImT`$ (see for example ), in order to deduce the correction to $`K(S,T)`$ it is sufficient to only keep track of the kinetic terms for $`a,c`$. Therefore from now on we ignore the $`CR^4`$ term and concentrate on the dimensional reduction of the remaining terms in the action (2.6).
Using that $`t_8t_8R^4+1/4E_8`$ can be recast up to Ricci terms in the form $`64(12ZRS+12R_{IJ}S^{IJ})`$ , we rewrite the part of the action that we want to reduce as
$$S_1=b_1T_2d^{11}x\sqrt{g}\left[2^6(12ZRS+12R_{IJ}S^{IJ})\frac{1}{2}E_8\right]+\mathrm{},$$
(2.11)
where
$`Z_{IJ}=R_{IKLR}R_{JMN}{}_{}{}^{R}(R^K{}_{P}{}^{}{}_{}{}^{M}QR^{NPLQ}{\displaystyle \frac{1}{2}}R^{KN}{}_{PQ}{}^{}R_{}^{MLPQ}),`$
$`S_{IJ}=2R_I{}_{}{}^{MKL}R_{J}^{}{}_{K}{}^{P}{}_{}{}^{Q}R_{LPMQ}^{}+{\displaystyle \frac{1}{2}}R_I{}_{}{}^{MKL}R_{JMPQ}^{}R_{KL}^{PQ}`$
$`R_I{}_{J}{}^{K}{}_{}{}^{L}R_{KMNQ}^{}R_L{}_{}{}^{MNQ},`$
$`Z=Z_{IJ}g^{IJ},S=S_{IJ}g^{IJ}.`$ (2.12)
We will also use that, up to Ricci terms, $`S=12(2\pi )^3Q`$ with the Euler density defined as
$$_Xd^6x\sqrt{det(g_{IJ})}Q=\chi ,$$
(2.13)
where $`\chi `$ is the Euler number of the Calabi-Yau three-fold $`X`$.
Since the action (2.11) is already of order $`\kappa ^{4/3}`$ and going to higher orders in the $`\kappa `$ expansion of the eleven-dimensional action is beyond the scope of this paper, clearly we have to reduce (2.11) on the zeroth order solution for the eleven-dimensional metric, namely the one given in (2.4). To obtain canonical Einstein term for the four-dimensional action, we have to rescale the metric:
$$g_{\mu \nu }=e^{6ac}\overline{g}_{\mu \nu }.$$
(2.14)
The details of the reduction are given in Appendix A. Here we only record the final result for the $`R^4`$ induced correction to the Kähler potential:
$$K=3\mathrm{ln}(T+\overline{T})\mathrm{ln}(S+\overline{S})\frac{b_2\kappa ^{4/3}\chi }{6(S+\overline{S})},$$
(2.15)
where $`b_2`$ is a numerical coefficient given in (A.14).
## 3 Soft supersymmetry breaking terms
The low energy effective action of a four-dimensional theory with $`N=1`$ supersymmetry is determined by the Kähler potential $`\widehat{K}`$, the superpotential $`W`$ and the gauge kinetic function $`f`$. For the case of the above compactification of the strongly coupled heterotic string, these functions can be read off from the action of , which is complete to order $`\kappa ^{2/3}`$ but contains only some of the $`𝒪(\kappa ^{4/3})`$ contributions. Expanding to second order in the charged matter fields $`C^p`$, one has :
$$\widehat{K}=\kappa _4^2K(S,T,\overline{S},\overline{T})+Z_{p\overline{q}}(S,T,\overline{S},\overline{T})C^p\overline{C}^{\overline{q}},$$
(3.1)
where
$$K=\mathrm{ln}(S+\overline{S})3\mathrm{ln}(T+\overline{T}),Z_{p\overline{q}}=\left(\frac{3}{T+\overline{T}}+\frac{\beta }{S+\overline{S}}\right)\delta _{p\overline{q}},$$
(3.2)
and
$$W=\frac{1}{3}\stackrel{~}{Y}d_{pqr}C^pC^qC^r+W_{\mathrm{non}\mathrm{pert}.},f=S+\beta T.$$
(3.3)
The four-dimensional gravitational constant is $`\kappa _4^2=\kappa _{11}^2/Vl`$ and $`\beta 𝒪(\kappa ^{2/3})`$. The superfields $`S`$ and $`T`$ were already introduced in the previous section.<sup>10</sup><sup>10</sup>10Abusing notation, we denote both a superfield and its bosonic component with the same letter. The remaining ones, $`C^p`$, are chiral superfields that arise from the ten-dimensional $`E_8`$ gauge fields and are charged under the unbroken gauge group in the visible sector. For phenomenological purposes, $`C^p`$ are often taken to transform in the 27 of $`E_6`$.
Since the action of is incomplete at order $`\kappa ^{4/3}`$ so are the above formulae (3.1)-(3.3). As we saw in Section 2, the additional contribution, coming from the higher derivative $`R^4`$ correction to the eleven-dimensional effective action, is:
$$\delta K=\frac{b_2\kappa ^{4/3}\chi }{6(S+\overline{S})}.$$
(3.4)
This correction has further implications for the soft supersymmetry breaking terms of the four-dimensional theory. Leaving aside for the moment the issue of finding (meta)stable minima of the potential, one can parametrize the supersymmetry breaking by the auxiliary components of the superfields $`S`$ and $`T`$, denoted respectively by $`F^S`$,$`F^T`$. The general form of the soft terms arising from spontaneous supersymmetry breaking due to non-perturbative effects in the hidden sector was derived in . In a Kähler covariant language, the tree level formulae for the masses of the gravitino, gaugino, scalar fields and Yukawa couplings are :
$`m_{3/2}^2={\displaystyle \frac{1}{3}}K_{i\overline{j}}F^iF^{\overline{j}},m_{1/2}={\displaystyle \frac{1}{2}}F^i_i\mathrm{ln}(\mathrm{Re}f),`$
$`m_{p\overline{q}}^2=m_{3/2}^2Z_{p\overline{q}}F^iF^{\overline{j}}R_{i\overline{j}p\overline{q}},A_{pqr}=F^iD_iY_{pqr},`$ (3.5)
where $`i,j`$ run over the fields $`S,T`$ and
$`R_{i\overline{j}p\overline{q}}`$ $`=`$ $`_i_{\overline{j}}Z_{p\overline{q}}\mathrm{\Gamma }_{ip}^sZ_{s\overline{t}}\mathrm{\Gamma }_{\overline{j}\overline{q}}^{\overline{t}},`$
$`D_iY_{pqr}`$ $`=`$ $`_iY_{pqr}+{\displaystyle \frac{1}{2}}K_iY_{pqr}\mathrm{\Gamma }_{i(p}^sY_{qr)s}`$ (3.6)
with
$$Y_{pqr}=e^{K/2}\stackrel{~}{Y}d_{pqr},\mathrm{\Gamma }_{ip}^s=Z^{s\overline{t}}_iZ_{\overline{t}p}.$$
(3.7)
The corrections due to the terms proportional to $`\beta `$ in $`Z_{p\overline{q}}`$ and $`f`$ were computed in . The new contributions coming from the $`R^4`$ induced change in the Kähler potential (3.4) are:
$`\delta m_{3/2}^2`$ $`=`$ $`{\displaystyle \frac{\stackrel{~}{b}|F^S|^2}{9(S+\overline{S})^3}},\delta m_{1/2}=0,`$
$`\delta m_{p\overline{q}}^2`$ $`=`$ $`{\displaystyle \frac{\stackrel{~}{b}|F^S|^2}{3(S+\overline{S})^3(T+\overline{T})}}\delta _{p\overline{q}},`$
$`\delta A_{pqr}`$ $`=`$ $`\stackrel{~}{b}\left({\displaystyle \frac{F^S}{4(S+\overline{S})^2}}{\displaystyle \frac{F^T}{2(S+\overline{S})(T+\overline{T})}}\right)Y_{pqr},`$ (3.8)
where $`\stackrel{~}{b}=b_2\kappa ^{4/3}\chi `$. It may seem surprising that we have obtained corrections of order $`\kappa ^{4/3}`$, whereas those of were of order $`\kappa ^{2/3}`$, since in both cases one starts with an $`𝒪(\kappa ^{4/3})`$ correction to the eleven-dimensional action. The resolution of this seeming puzzle is in a rescaling of the matter fields $`C^p`$, involving a power of $`\kappa `$ , that was needed for the proper normalization of the chiral superfields.
The most significant difference in the patterns of supersymmetry breaking of the weak and strong coupling limits occurs for breaking in the direction of the $`T`$-modulus, i.e. when $`F^T0`$ and $`F^S=0`$. In that case, in the weakly coupled heterotic string description only the gravitino mass acquires a non-vanishing value at tree level. On the other hand, in the strongly coupled heterotic M-theory limit this also happens for $`m_{1/2}`$ and $`A_{pqr}`$ . As we see from (3), for $`F^S=0`$ the $`R^4`$ term induces an additional contribution to the trilinear couplings $`A_{pqr}`$, whereas the scalar field masses $`m_{p\overline{q}}`$ remain small as in the case without higher derivative corrections .
Another observation following from (3) is that the changes we have computed to the soft supersymmetry breaking terms do not spoil universality. Recall that the phenomenological requirement for suppression of flavor changing neutral interactions is satisfied if the masses of the scalar superpartners of the observable fermions (which usually differ from the fermionic components of the charged matter superfields $`C^p`$ due to various mixing angles) are all almost equal to each other . This condition is achieved for $`m_{p\overline{q}}^2\delta _{p\overline{q}}`$ and also for $`A_{pqr}Y_{pqr}`$ with all components $`Y_{pqr}`$ having the same moduli dependence. Actually, the latter criteria for universality were expected to hold, even without an explicit calculation, due to the general form of the soft terms (3) and the fact that in the present case $`Z_{p\overline{q}}\delta _{p\overline{q}}`$. Let us also note that models with more than one Kähler modulus are generically non-universal and it seems clear that higher derivative corrections would only worsen that effect. An obvious remedy, proposed in and used in many other references, is to consider only Calabi-Yau manifolds with $`h^{1,1}=1`$.
## 4 Deformation of the background geometry
In this section we analyze the effect that the higher derivative $`R^4`$ terms have on the geometric background and $`G`$-flux of Horava-Witten theory.
Recall that the existence of a solution to linear order in $`\kappa ^{2/3}`$ was shown in and its explicit form was found in . In a non-linear solution was obtained which contains corrections of all orders in $`\kappa ^{2/3}`$. However, none of these works took into account higher derivative corrections to eleven-dimensional supergravity. Adding the $`R^4`$ term to the action though, clearly modifies both the equations of motion and the supersymmetry variations of the theory. The complete supersymmetry transformations are not yet known despite significant progress in that direction . However, the modified gravitino transformation rule was derived on a case-by-case basis for compactifications on the following special holonomy manifolds: $`CY(3)`$ , $`G_2`$ , $`Spin(7)`$ .<sup>11</sup><sup>11</sup>11 Ref. also provides a detailed account of the $`R^4`$-corrected bosonic equations of motions in the $`Spin(7)`$ case. In each case, the Einstein equations in the presence of the $`R^4`$ term were recovered from the integrability of the proposed supersymmetry variation by using in an essential way properties particular to the special holonomy manifold under consideration. Nonetheless, the final result for the gravitino variation turned out to be the same for all cases, when written in purely Riemannian form, i.e. without the use of any special structures. Namely, it was found that the Killing spinor equation is
$$\delta \mathrm{\Psi }_I=\left(D_I+\epsilon (^JR_{IKM_1M_2})R_{JLM_3M_4}R^{KL}{}_{M_5M_6}{}^{}\mathrm{\Gamma }_{}^{M_1\mathrm{}M_6}\right)\eta =0,$$
(4.1)
where $`\epsilon `$ is a numerical constant times $`\alpha ^{\mathrm{\hspace{0.17em}3}}`$ in string theory and a numerical constant times $`\kappa ^{4/3}`$ in M-theory. The derivative $`D_I`$ is the appropriate extension of the covariant derivative $`_I`$ with flux terms, i.e. for the case of interest to us:
$$D_I=_I+\frac{\sqrt{2}}{288}\left(\mathrm{\Gamma }_{IJKLM}8g_{IJ}\mathrm{\Gamma }_{KLM}\right)G^{JKLM}.$$
(4.2)
In the following we will analyze the solutions of (4.1). As the order of accuracy to which we work is $`\kappa ^{4/3}`$, we do not need to perform any checks on whether this is also the correct modification for Horava-Witten theory, the reason being that due to $`\epsilon 𝒪(\kappa ^{4/3})`$ the three curvature tensors have to be of zeroth order, i.e. they are the curvatures for the direct product $`M_4\times CY(3)\times \text{𝕊}^1/\text{}_2`$. In fact, it is most convenient to use the form of (4.1) adapted to the Calabi-Yau case:
$$\delta \mathrm{\Psi }_m=(D_m+P_m)\eta ,P_m=\frac{i}{2}\epsilon J_{mn}^nQ\mathrm{\Gamma }_{11},$$
(4.3)
where $`m,n=1,\mathrm{},6`$ , $`Q`$ is as before the Euler density in six dimensions, $`J_{mn}`$ is the Kähler form of the Calabi-Yau and $`\mathrm{\Gamma }_{11}`$ is the $`\mathrm{\Gamma }`$-matrix in the interval direction. Recall that in terms of the constants in the action (2.6) $`\epsilon =b_1T_2\kappa ^2`$, where we remind the reader that $`T_2𝒪(\kappa ^{2/3})`$. Since the $`\text{}_2`$ projection implies $`\mathrm{\Gamma }_{11}\eta =\eta `$ and also the non-vanishing components of the Kähler form are $`J_{a\overline{b}}=J_{\overline{b}a}=ig_{a\overline{b}}`$ with $`a=1,2,3`$ being the holomorphic indices, we have<sup>12</sup><sup>12</sup>12Actually, the positive chirality projection condition $`\mathrm{\Gamma }_{11}\eta =\eta `$ is modified by the warp factor in front of $`(dx^{11})^2`$ in the full geometry. However, that warp factor is of $`𝒪(\kappa ^{2/3})`$ and higher whereas our computation is reliable to $`𝒪(\kappa ^{4/3})`$. So in $`P_m`$ we can simply take $`\mathrm{\Gamma }_{11}\eta =\eta `$.
$$P_a=\frac{\epsilon }{2}_aQ,P_{\overline{a}}=\frac{\epsilon }{2}_{\overline{a}}Q.$$
(4.4)
Now let us explore the consequences of this additional term to the gravitino variation.
### 4.1 Warp factor deformations
Before addressing general deformations of the initial Calabi-Yau manifold, we will start with the simpler metric ansatz:
$$ds^2=e^{b(x^m,x^{11})}\eta _{\mu \nu }dx^\mu dx^\nu +e^{f(x^m,x^{11})}g_{ln}(x^m)dx^ldx^n+e^{k(x^m,x^{11})}(dx^{11})^2,$$
(4.5)
where the deformation due to fluxes and $`R^4`$ terms is entirely encoded in three warp factors.
This is precisely the form of the non-linear background of valid in the absence of $`R^4`$ corrections. Although in principle we should expand the exponentials and keep terms only up to second order in $`\kappa ^{2/3}`$, we leave (4.5) for now as it is for easier comparison with the solution of . Generically, the covariantly constant spinor of the full metric, $`\eta `$, is a deformation of the Calabi-Yau one, $`\eta _0`$, and we write
$$\eta =e^{\psi (x^m,x^{11})}\eta _0.$$
(4.6)
The Killing spinor equation $`\delta \mathrm{\Psi }_I=0`$ implies relations between the functions $`b,f,k,\psi `$ and the $`G`$-flux components, which are conveniently written in terms of the following quantities<sup>13</sup><sup>13</sup>13We note in passing that this parametrization of the four-form flux, first introduced in , is very reminiscent of the recent G-structure approach to classification of supergravity solutions in various dimensions . More precisely, looking at the decomposition of G as $`G=\frac{Q}{3}JJ+JA+v(JW+U)`$ in (3.10) of , one can identify up to numerical coefficients $`\alpha `$ with $`Q`$, $`\beta _l`$ with the 1-form $`W`$ and $`\mathrm{\Theta }_{lm}`$ with the 2-form $`A\frac{2}{3}QJ+2vW`$ (see eq. (3.12) of ). On the other hand, the 3-form $`U`$ is given by the components $`G_{lmn11}`$ such that $`G_{lmn11}J^{mn}=0`$. The remaining terms in (3.10) of vanish due to (3.13) there, which is a consequence of supersymmetry.:
$$\alpha =G_{lmnp}J^{lm}J^{np},\beta _l=G_{lmn11}J^{mn},\mathrm{\Theta }_{lm}=G_{lmnp}J^{np}.$$
(4.7)
It is straightforward to find out how the additional $`P_m`$ term in (4.3) modifies the supersymmetry conditions derived in . The result for the warp factors is:
$`8_a\psi =2_ab=_ak=i{\displaystyle \frac{\sqrt{2}}{3}}e^{k/2f}\beta _a=_af+2\epsilon _aQ,`$
$`4_{11}\psi =_{11}b=_{11}f={\displaystyle \frac{\sqrt{2}}{24}}e^{k/22f}\alpha .`$ (4.8)
The $`G`$-flux conditions remain the same as before:
$$\mathrm{\Theta }^{\overline{b}}{}_{\overline{a}}{}^{}=0,\overline{b}\overline{a};\mathrm{\Theta }^{\overline{a}}{}_{\overline{a}}{}^{}=\mathrm{\Theta }^{\overline{b}}{}_{\overline{b}}{}^{}a,b;G^{\overline{c}}{}_{\overline{a}\overline{b}11}{}^{}=0,\overline{c}\overline{a},\overline{b}.$$
(4.9)
Equations (4.9) imply that $`\mathrm{\Theta }_{lm}`$ is completely determined by $`\alpha `$ and hence there are only two flux parameters: $`\alpha `$ and $`\beta _l`$. This is due to the special metric ansatz (4.5) in which the deformation of the Calabi-Yau three-fold is conformally Calabi-Yau. For a generic deformation, resulting in a non-Kähler manifold, all three kinds of flux components will be independent.
As is well-known, solving the supersymmetry conditions alone is not enough to ensure that the equations of motion will be satisfied unless the backgrounds of interest are maximally supersymmetric. Therefore, in our case we have to consider also the Bianchi identity and field equation of the four-form $`G`$ (the Einstein equation is guaranteed to follow from these and supersymmetry ). From (2.2) and (2.6), the $`G`$ field equation is:
$$dG=\frac{1}{2}GG+1152(2\pi )^4\epsilon X_8,$$
(4.10)
where $`\epsilon `$ is the same $`𝒪(\kappa ^{4/3})`$ constant as in (4.4) and $`X_8`$ is the eight-form made up of four Riemann tensors that multiplies $`C_{(3)}`$ in $`S_1`$ (see (2.6)). We are looking for solutions that preserve the Poincare invariance of the four-dimensional external space and so only components of $`G`$ along the seven internal directions can be non-vanishing. Therefore the eight-form $`GG=0`$ in the backgrounds of interest. Also, to $`𝒪(\kappa ^{4/3})`$ $`X_8`$ consists of the curvatures for the direct product $`M_4\times CY(3)\times \text{𝕊}^1/\text{}_2`$. Writing this space as $`\text{}^{1,2}\times Y`$, where $`Y=\text{}\times CY(3)\times \text{𝕊}^1/\text{}_2`$, we see that $`X_8(\text{}^{1,2}\times Y)=X_8(Y)`$ due to $`\text{}^{1,2}`$ being flat. Now, recall that up to a numerical coefficient $`X_8`$ equals $`p_2\frac{1}{4}p_1^2`$, where $`p_i`$ is the $`i`$-th Pontryagin class.<sup>14</sup><sup>14</sup>14Strictly speaking, topological invariants are mathematically well-defined only for compact manifolds. But since for our purposes we are only interested in their differential form representation, we can disregard that technicality. In addition, on an eight-manifold admitting a nowhere vanishing spinor the combination $`p_2\frac{1}{4}p_1^2`$ is proportional to the Euler class . And since $`Y`$ is a direct product of a six-manifold and two flat dimensions, the Euler class vanishes identically. Recapitulating, in our case $`X_8=0`$ too.<sup>15</sup><sup>15</sup>15This conclusion will most certainly change at higher (than $`\kappa ^{4/3}`$) orders when the $`X_8`$ term starts feeling the warped background.
So we are left with the same field equation $`D^IG_{IJKL}=0`$ and Bianchi identity, considered in . Although that work studied the linearized approximation (i.e. $`𝒪(\kappa ^{2/3})`$), the part of its analysis that we need is still valid. More precisely, the $`G`$ field equation and Bianchi identity imply that the fluxes $`\alpha `$ and $`\beta _a`$ are constrained by
$$D_{11}\beta _a=\frac{i}{4}_a\alpha ,D^m\beta _m=0.$$
(4.11)
It is worth emphasizing that the above flux relations are exact to all orders in $`\kappa `$ as long as the eleven-dimensional action contains no higher derivative terms. However, when taking into account the $`R^4`$ term and its superpartner $`CX_8`$, they will be modified beyond $`𝒪(\kappa ^{4/3})`$. Note also that, for the warp factor deformation of the metric considered in this section, we have $`D_{11}\beta =_{11}\beta `$ in (4.11).
Let us now turn to the analysis of the supersymmetry conditions. As in , we can extract from (4.1) that $`\psi =b/4`$. However, the relation $`f(x^m,x^{11})=k(x^m,x^{11})+F(x^{11})`$ of (see eq. (2.18) there) can not be true in the presence of the $`R^4`$ correction. Instead, we have
$$k(x^m,x^{11})=2b(x^m,x^{11}),$$
(4.12)
where we have set to zero the undetermined function of $`x^{11}`$ by a reparametrization of the eleventh coordinate. At this stage one is left with two warp factors $`f`$ and $`b`$. As observed in , the equations for them in (4.1) are generically incompatible. The compatibility conditions arise from $`_a_{11}f=_{11}_af=0`$ and a similar equation for $`b`$:
$$_{11}\left(e^{bf}\beta _a\right)=_a\left(e^{b2f}\alpha \right)=0.$$
(4.13)
Without the higher derivative correction (i.e. dropping the $`Q`$ term in (4.1)), nontrivial solutions exist only for $`\alpha =0`$, $`\beta _a0`$ or for $`\alpha 0`$, $`\beta _a=0`$. This can be seen as follows. From
$`_a\alpha `$ $`=`$ $`\alpha _a(b+2f)={\displaystyle \frac{i\sqrt{2}}{2}}e^{bf}\beta _a\alpha `$
$`_{11}\beta _a`$ $`=`$ $`\beta _a_{11}(bf)={\displaystyle \frac{\sqrt{2}}{12}}e^{b2f}\beta _a\alpha ,`$ (4.14)
and $`_{11}\beta _a=\frac{i}{4}_a\alpha `$, we find
$$i\frac{\sqrt{2}}{8}\beta _a\alpha \left(e^{bf}+\frac{2}{3}e^{b2f}\right)=0.$$
(4.15)
The conclusion is that either $`\alpha `$ or $`\beta _a`$ has to vanish in order to have a warp-deformed supersymmetric solution. Notice the essential role in the above argument of the flux constraint (4.11), which is a consequence of the field equations (the latter were not considered in ).
Let us now reinstate the correction to the integrability conditions (4.13) due to the higher derivative terms (which also means that we have to truncate to order $`\kappa ^{4/3}`$):
$`_{11}\beta _a+{\displaystyle \frac{\sqrt{2}}{12}}\alpha \beta _a=0`$
$`_a\alpha +\alpha ({\displaystyle \frac{i\sqrt{2}}{2}}\beta _a4\epsilon _aQ)=0.`$ (4.16)
Notice that in (4.16) the term induced by the $`R^4`$-modified supersymmetry gravitino variation is, in fact, of higher order than the accuracy that we are working at, since $`\alpha \epsilon \kappa ^{(2+4)/3}`$.
We proceed to investigate in turn what happens with the backgrounds found by , corresponding to either $`\alpha =0`$ or $`\beta _a=0`$, in the presence of the $`R^4`$ terms.
Taking $`\alpha =0`$, $`\beta _a0`$ we see that all $`x^{11}`$-dependence disappears and one can write
$$8\psi (x^m)=2b(x^m)=k(x^m)$$
(4.17)
as in . However, the answer for $`f`$ is different now. Namely, $`_{11}f=0=_{11}b`$ and $`_af=2_ab2\epsilon _aQ`$ imply that
$$f=2b2\epsilon Q.$$
(4.18)
Using (4.17), (4.18) and equation $`2_ab=i\frac{\sqrt{2}}{3}e^{k/2f}\beta _a`$ in (4.1), one finds the same relation between warp factor and flux as in thus recovering the weakly coupled heterotic string result.<sup>16</sup><sup>16</sup>16In doing this one has to remember that we are working with accuracy to order $`\kappa ^{4/3}`$ and so should discard higher orders in $`2_ab=i\frac{\sqrt{2}}{3}e^{k/2f}\beta _a`$, in particular $`e^{2\epsilon Q}\beta _a\beta _a`$. The novelty, due to the $`R^4`$ correction, comes when one considers the volume of the Calabi-Yau manifold:
$$V_{CY}=e^{3f}\sqrt{g}d^6x.$$
(4.19)
Using (4.18), we obtain
$$V_{CY}=d^6x\sqrt{g}e^{6b}(16\epsilon Q+\mathrm{})=V_{CY}^06\epsilon \chi +\mathrm{},$$
(4.20)
where we have kept only terms up to first order in $`\epsilon 𝒪(\kappa ^{4/3})`$. For the same reason the term with the Calabi-Yau Euler number $`\chi `$ in (4.20) is not multiplied by $`e^{6b}`$ (recall that $`b𝒪(\kappa ^{2/3})`$ and higher).
Equation (4.20) shows explicitly that the $`R^4`$ correction to the effective action induces a shift of the Calabi-Yau volume as argued in based on a field redefinition of the dilaton. It was hoped in that the same phenomenon might resolved the curvature singularity which appears in their non-linear background at a point along the interval direction at which the warp factor vanishes. We will see below that things are not that straightforward with regard to the proposed singularity resolution mechanism and also that there are other implications. In particular, it will turn out that solutions exist with both $`\alpha 0`$ and $`\beta _a0`$.
Let us now consider the case $`\alpha 0`$, $`\beta _a=0`$, which is exactly the kind of flux that the strongly coupled non-linear background of has. Clearly, we can make the identification
$$4\psi (x^{11})=b(x^{11}),$$
(4.21)
as in . However, since $`_ak=0`$ whereas $`_af0`$ the relation $`f=k`$ that the background considered in satisfies can not be true once the higher derivative correction is taken into account. We should note though, that for the present case, i.e. $`\beta _a=0`$, equations (4.1) do not imply anything about the relationship between $`b(x^{11})`$ and $`k(x^{11})`$. Actually, they leave $`k(x^{11})`$ completely arbitrary and so by a suitable redefinition of $`x^{11}`$ one can set $`k(x^{11})=b(x^{11})`$ as in . On the other hand, we find it most natural to redefine $`x^{11}`$ such that $`e^{k(x^{11})}=1`$. Clearly, this freedom has no bearing on the physics of the solution. Specializing the relevant equations in (4.1) to the present case, we find
$$f(x^m,x^{11})=2\epsilon Q(x^m)b(x^{11}).$$
(4.22)
Hence expanding to order $`\kappa ^{4/3}`$ the equation that determines the $`x^{11}`$ dependence of $`f`$,
$$_{11}f=\frac{\sqrt{2}}{24}e^{k/22f}\alpha ,$$
(4.23)
we end up with
$$e^{2f(x^m,x^{11})}=1\frac{\sqrt{2}}{12}_0^{x^{11}}𝑑z\alpha (z)4\epsilon Q(x^m)+𝒪(\kappa ^{6/3}).$$
(4.24)
The above solution is consistent since when $`\beta _a=0`$ the constraints (4.11) imply $`_a\alpha =0`$. Taking the flux sources to be localized at the two ends of the interval, the Bianchi identity is solved by $`G_{(4)}=\mathrm{\Theta }(x^{11})𝒮_{(4)}(x^m)`$, where $`𝒮_{(4)}(x^m)`$ is a closed four-form. The corresponding value of the $`\alpha `$ flux is:
$$\alpha =\mathrm{\Theta }(x^{11})𝒮_{mnpq}J^{mn}J^{pq}=4\mathrm{\Theta }(x^{11})𝒮_{a\overline{b}c\overline{d}}J^{a\overline{b}}J^{c\overline{d}}4\sqrt{2}\mathrm{\Theta }(x^{11})𝒮.$$
(4.25)
So we find for the volume of the Calabi-Yau:
$$V_{CY}(x^{11})=d^6x\sqrt{g}e^{3f}=V_{CY}^0\left(1x^{11}\mathrm{\Theta }(x^{11})𝒮+\frac{1}{6}(x^{11}\mathrm{\Theta }(x^{11})𝒮)^2\right)6\epsilon \chi .$$
(4.26)
We see again the expected constant shift proportional to the Euler number.
Finally, let us note that we can construct to the same level of accuracy solutions which have both $`\alpha 0`$ and $`\beta _a0`$, for example by taking $`\beta =\beta (x^m)\kappa ^{4/3}`$ meaning also that $`\alpha =\alpha (x^{11})`$. In particular, for
$$\beta _a=iC\epsilon _aQ,$$
(4.27)
where $`C`$ is an arbitrary real number, one can solve both the supersymmetry conditions (4.1) and the flux constraints (4.11). Indeed, writing $`f(x^a,x^{11})=f_1(x^a)+f_2(x^{11})`$ and $`b(x^a,x^{11})=b_1(x^a)+b_2(x^{11})`$, one finds $`b_1=2f_1`$, $`b_2=f_2`$ and:
$$f_1(x^a)=\left(\frac{\sqrt{2}C}{3}2\right)\epsilon Q,e^{f_2(x^{11})}=e^{f_2(0)}\frac{\sqrt{2}}{24}_0^{x^{11}}𝑑z\alpha (z),$$
(4.28)
where we have used the relation $`k(x^a,x^{11})=2b(x^a,x^{11})`$, (4.12), that has to be satisfied when both $`\alpha 0`$ and $`\beta _a0`$.
### 4.2 General case
Let us turn now to the metric ansatz
$`ds^2=e^{b(x^m,x^{11})}\eta _{\mu \nu }dx^\mu dx^\nu +[g_{ln}(x^m)+h_{ln}(x^m,x^{11})]dx^ldx^n+e^{k(x^m,x^{11})}(dx^{11})^2`$
$`=\widehat{g}_{MN}dx^Mdx^N,`$ (4.29)
which allows more general flux. Since generically the deformation $`h_{mn}(x^m,x^{11})`$ does not preserve the Kähler property of the Calabi-Yau metric $`g_{mn}(x^m)`$, we can not describe the $`G`$-flux in terms of the same quantities as in (4.7). A convenient new parametrization is given by
$$G=\widehat{g}^{a\overline{b}}\widehat{g}^{c\overline{d}}G_{a\overline{b}c\overline{d}},G_m=\widehat{g}^{b\overline{c}}G_{mb\overline{c}11},G_{mn}=\widehat{g}^{c\overline{d}}G_{mnc\overline{d}}.$$
(4.30)
Splitting the vielbein of the internal six-dimensional space as $`\widehat{e}^{\underset{¯}{l}}{}_{n}{}^{}(x^m,x^{11})`$ $`=e^{\underset{¯}{l}}{}_{n}{}^{}(x^m)+f^{\underset{¯}{l}}{}_{n}{}^{}(x^m,x^{11})`$, where $`e^{\underset{¯}{l}}{}_{n}{}^{}(x^m)`$ is the vielbein for $`g_{ln}`$, one finds for the spin-connection :
$`\mathrm{\Omega }_{l\underset{¯}{m}\underset{¯}{n}}(\widehat{e})=\mathrm{\Omega }_{l\underset{¯}{m}\underset{¯}{n}}(e)+\mathrm{\Omega }_{l\underset{¯}{m}\underset{¯}{n}}^{(d)}(e,f),\mathrm{\Omega }_{11\underset{¯}{l}\underset{¯}{11}}(\widehat{e})={\displaystyle \frac{1}{2}}\widehat{e}_{\underset{¯}{l}}{}_{}{}^{l}\widehat{e}_{\underset{¯}{11},11}^{}_lk,`$
$`\mathrm{\Omega }_{l\underset{¯}{m}\underset{¯}{11}}(\widehat{e})={\displaystyle \frac{1}{2}}\widehat{e}_{\underset{¯}{11}}{}_{}{}^{11}(_{11}f_{\underset{¯}{m}l}+\widehat{e}_{\underset{¯}{m}}{}_{}{}^{m}\widehat{e}_{}^{\underset{¯}{l}}{}_{l}{}^{}_{11}^{}f_{\underset{¯}{l}m}),\mathrm{\Omega }_{11\underset{¯}{l}\underset{¯}{m}}(\widehat{e})=\widehat{e}_{[\underset{¯}{l}}{}_{}{}^{l}_{|11|}^{}f_{\underset{¯}{m}]l},`$
$`\mathrm{\Omega }_{\mu \underset{¯}{\nu }\underset{¯}{l}}(\widehat{e})={\displaystyle \frac{1}{2}}\widehat{e}_{\underset{¯}{l}}{}_{}{}^{m}\widehat{e}_{\underset{¯}{\nu }\mu }^{}_mb,\mathrm{\Omega }_{\mu \underset{¯}{\nu }11}(\widehat{e})={\displaystyle \frac{1}{2}}\widehat{e}_{\underset{¯}{\nu }\mu }\widehat{e}_{\underset{¯}{11}}{}_{}{}^{11}_{11}^{}b.`$ (4.31)
In the above formulae $`\mathrm{\Omega }_{l\underset{¯}{m}\underset{¯}{n}}(e)`$ is the spin-connection of the initial $`CY(3)`$ whereas $`f^{\underset{¯}{l}}_m`$ and $`\mathrm{\Omega }_{l\underset{¯}{m}\underset{¯}{n}}^{(d)}(e)`$ measure the deviation from it.
It is easy to compute the contribution of the $`R^4`$ induced term in the gravitino variation to the supersymmetry conditions (7.93)-(7.103) of . We write down only the single equation that changes:
$$\mathrm{\Omega }_{ab}^{(d)}{}_{}{}^{b}\mathrm{\Omega }_{a\overline{b}}^{(d)}{}_{}{}^{\overline{b}}=\frac{2\sqrt{2}}{3}e^{k/2}G_a+2\epsilon _aQ.$$
(4.32)
As explained in , one can find the $`x^{11}`$ dependence of the Calabi-Yau volume by using $`_M\sqrt{\widehat{g}}=\sqrt{\widehat{g}}\mathrm{\Gamma }^N{}_{NM}{}^{}(\widehat{g})`$. Relating $`\mathrm{\Gamma }^N_{N11}`$ to $`\mathrm{\Omega }^N_{N11}`$ via the vielbein postulate one finds:
$$_{11}\sqrt{\widehat{g}_{CY}}=\sqrt{\widehat{g}_{CY}}\mathrm{\Omega }^m{}_{m11}{}^{}(\widehat{e}).$$
(4.33)
Since the higher derivative correction appears only in the equation for $`\mathrm{\Omega }_{lmn}^{(d)}`$ but not those for $`\mathrm{\Omega }^m_{n11}`$ (or any of the warp factors $`b,k,\psi `$), we conclude from (4.33) that this correction does not affect the $`x^{11}`$ dependence of $`V_{CY}`$. This is consistent with the constant shift of the Calabi-Yau volume that it was inducing for the special flux of the previous subsection. Unfortunately though, in the present general case one can not be more explicit than this.<sup>17</sup><sup>17</sup>17Similarly to $`_{11}\sqrt{\widehat{g}_{CY}}`$, one can consider $`_a\sqrt{\widehat{g}_{CY}}`$ but the expression for it contains $`\mathrm{\Omega }_{ab}^{(d)}{}_{}{}^{b}+\mathrm{\Omega }_{a\overline{b}}^{(d)}^{\overline{b}}`$ instead of $`\mathrm{\Omega }_{ab}^{(d)}{}_{}{}^{b}\mathrm{\Omega }_{a\overline{b}}^{(d)}^{\overline{b}}`$ as in (4.32) and so no definite statement can be made about the role of the $`\epsilon _aQ`$ term.
The search for deformed backgrounds allowing generic flux may be significantly facilitated by the $`G`$-structures approach. So far we have not been able to find new solutions, but in Appendix B we present some preliminary considerations based on the results of . More precisely, we derive generalized Hitchin flow equations for non-Kähler six-manifolds fibered over an interval.<sup>18</sup><sup>18</sup>18These considerations assume that the internal manifold of the heterotic string compactification has $`SU(3)`$ structure. However, in the presence of higher derivative corrections to the effective action the internal manifold might even be noncomplex (see the discussion in Section 4 of ). This case would be very difficult to analyze and we have nothing to say about it at present.
As a last remark in the current section, we note that the change in $`\mathrm{\Omega }^{(d)}`$, implied by equation (4.32), means that the $`R^4`$ correction has an impact on the holonomy group $``$ of the deformed (generically non-Kähler) six-dimensional manifold since $``$ is generated by parallel transport, determined by the full spin-connection $`\mathrm{\Omega }(\widehat{e})=\mathrm{\Omega }(e)+\mathrm{\Omega }^{(d)}(e,f)`$, around all closed curves in the manifold.
## 5 Scalar potential
Let us now combine ingredients from previous sections to address the role of the $`R^4`$ correction in shaping the scalar potential of the four-dimensional effective theory. More precisely, we will be interested in the fate of the de Sitter vacua of .
It may seem that such questions are out of reach if one uses only a finite number of terms in a perturbative expansion, as we have done. However, recall that the minima of occur for $`ϵ=\frac{2\pi L}{3V_v^{2/3}}(\frac{\kappa }{4\pi })^{\frac{2}{3}}𝒪(1)`$ and even $`ϵ<1`$, which is just at the border of validity of the perturbative expansion. In addition, since higher derivative terms in M-theory like $`R^4`$, the higher order $`D^4R^4`$ <sup>19</sup><sup>19</sup>19The notation $`D^4R^4`$ is symbolic; the index contractions are different from those in $`R^4`$ or $`X_8`$. etc. change the supersymmetry conditions and field equations (in particular, the constraints (4.11)), it is not a priori clear whether the better approximation is to extend to all orders a solution that neglects them, as in , or to keep only terms up to the appropriate order of accuracy (for us $`𝒪(\kappa ^{4/3})`$). The latter option merits an investigation on its own, which we now turn to.
The scalar potential in $`N=1`$ $`d=4`$ supergravity is given by
$$U=e^K\left(K^{A\overline{B}}D_AWD_{\overline{B}}\overline{W}3|W|^2\right)+U_D,$$
(5.1)
where $`K`$ is the Kähler potential, $`W`$ \- the superpotential, $`K^{A\overline{B}}`$ \- the inverse moduli space metric i.e. the inverse of $`K_{A\overline{B}}_A_{\overline{B}}K`$, the derivative $`D_AW_AW+K_AW`$ and $`U_D`$ are D-terms for the charged matter fields $`C^p`$ that originate from the gauge multiplets localized on the boundaries of the eleven-dimensional space-time. In principle, the indices $`A,B`$ in (5.1) range over all scalar fields in the effective action. However, as in we do not address here the stabilization of the Calabi-Yau complex structure moduli and the vector bundle moduli.<sup>20</sup><sup>20</sup>20The latter have been shown to be significantly suppressed . Presumably this can be achieved similarly to the weakly coupled case and the novelty is in the stabilization of the orbifold length. Again as in , we take $`h^{1,1}=1`$. One can show that in the present case too the matter fields $`C^p`$ can be stabilized at a nonzero but strongly suppressed value and so they can be safely neglected for the purposes of minimization of the scalar potential $`U`$ w.r.t. to the remaining fields, which we do from now on.
The $`R^4`$ corrections in the effective action change the Kähler potential, but not the superpotential.<sup>21</sup><sup>21</sup>21Indeed, in our derivation of the Kähler potential in Appendix A we have only neglected terms with more than two derivatives but not terms that would contribute to the superpotential. That such do not arise from the $`R^4`$ term is clear from the fact that each curvature component in (A) contains two derivatives. Recall that one expects higher derivative terms to not modify the superpotential for reasons similar to the original argument of about $`\alpha ^{}`$ corrections in string theory. Namely, the Peccei-Quinn symmetries of the low energy theory (the shift symmetries of the axions $`\sigma _S`$ and $`\sigma _T`$ in (2.9)), that are inherited from the three-form gauge transformation $`CC+d\omega `$ with $`\omega `$ being a 2-form, are broken only by non-perturbative effects, like M2-brane instantons wrapping topologically non-trivial 3-cycles . Whereas, if higher derivative terms contributed $`S,T`$ dependence to the superpotential, that would break perturbatively the axionic shift symmetries.
From (3.4) we see that the change in the Kähler potential, due to the $`R^4`$ corrections, induces the following change in the scalar potential to $`𝒪(\kappa ^{4/3})`$:
$$\delta U=b_2\kappa ^{4/3}\chi \left(\frac{1}{6(S+\overline{S})}U_0+\frac{1}{6}e^{K_0}(\overline{W}D_S^0W+W\overline{D_S^0W})+\frac{(S+\overline{S})}{3}e^{K_0}D_S^0W\overline{D_S^0W}\right),$$
(5.2)
where $`K_0=\mathrm{ln}(S+\overline{S})3\mathrm{ln}(T+\overline{T})`$, $`D_S^0=_S+K_{0,S}`$ and $`U_0(K,W)=U(K_0,W)`$. Note that, although the Kähler potential for the superfield $`T`$ does not change, the potential for it does, as the superpotential is a function of both $`S`$ and $`T`$: $`W=W(S,T)`$. The non-perturbative part of the superpotential $`W_{\mathrm{non}\mathrm{pert}}`$ is given by the sum of two contributions: from open membrane instantons, $`W_{OM}`$, and from gaugino condensation, $`W_{GC}`$.
Without the charged matter sector, the perturbative contribution to the superpotential vanishes and one is left with :
$$W=W_{OM}+W_{GC}=he^T+ge^{(S\gamma T)/C_H},$$
(5.3)
where $`g=C_H(2M_{GUT}l_{11})^3`$ with $`C_H`$ being the dual Coxeter number of the hidden gauge group. The open membrane instanton Pfaffian $`h`$ is bounded by $`|h|(2M_{GUT}l_{11})^3`$. Finally, the slope $`\gamma `$ is determined in terms of the the length of the interval $``$ and the flux on the visible boundary $`𝒮`$ via $`\gamma =𝒮`$. For convenience and easier comparison with , from now on we introduce the notation:
$$S=𝒱+i\sigma _S,T=𝒱_{OM}+i\sigma _T,$$
(5.4)
where $`𝒱=𝒱()`$ is the average volume of the Calabi-Yau and $`𝒱_{OM}=𝒱_{OM}()`$ is the volume of the open membrane instanton:
$$𝒱()=\frac{1}{}_0^Ldx^{11}e^{k/2}V_{CY}(x^{11}),𝒱_{OM}()=𝒱^{1/3}().$$
(5.5)
We differ from in our averaging over the orbifold direction; by introducing the measure $`dx^{11}e^{k/2}`$, and defining the orbifold length with respect to the same measure $`=_0^L𝑑x^{11}e^{k/2}`$, we ensure that the average volume is independent of redefinitions of the $`x^{11}`$ coordinate (or equivalently, $`𝒱()`$ is the same irrespective of the freedom we noted in Subsection 4.1 in choosing the value of the warp factor $`e^{k(x^{11})}`$). We should also recall that, whereas employed a “fully” non-linear background derived by ignoring all higher derivative corrections to the eleven-dimensional supergravity action, we must perform an expansion in $`\kappa `$. It follows from the previously derived Calabi-Yau volume dependence on $`x^{11}`$ (4.26) that, to the order of accuracy we are working to, the average volume of the Calabi-Yau is:
$$𝒱=𝒱_v\left[\left(1\frac{1}{2}𝒮+\frac{1}{18}^2𝒮^2\right)6\epsilon \chi \right].$$
(5.6)
where we denoted, as before, $`\epsilon =b_1T_2\kappa ^2`$.
For the zeroth order (direct product $`CY(3)\times \text{𝕊}^1/\text{}_2`$) metric, (5.4) reduces to $`S=e^{6a}+i\sigma _S`$ and $`T=e^{c+2a}+i\sigma _T`$. But the warping of the strongly coupled background introduces $``$-dependence of the Calabi-Yau volume $`𝒱`$. As in , we take for the modulus $`𝒱_v`$ (the volume of the Calabi-Yau at the visible boundary) a value dictated by phenomenology and study the resulting minima for $``$.<sup>22</sup><sup>22</sup>22The stabilization of $`𝒱_v`$ could presumably be achieved by compactification on a 6d non-Kähler manifold . The existence of a minimum for the orbifold length $``$ is guaranteed, as observed in , by the two opposing nonperturbative effects: the runaway behavior of the scalar potential, due to open membrane instantons, is being offset by the contribution coming from gaugino condensation on the hidden boundary.
The scalar potential, including the $`R^4`$ induced correction (5.2), is given by
$`U`$ $`=`$ $`{\displaystyle \frac{1}{2^4𝒱_{OM}^3𝒱}}({\displaystyle \frac{1}{C_H^2}}|W_{GC}|^2(4𝒱^2+{\displaystyle \frac{b_2\chi 𝒱}{3}})+|{\displaystyle \frac{\gamma }{C_H}}W_{GC}W_{OM}|^2({\displaystyle \frac{4𝒱_{OM}^2}{3}}{\displaystyle \frac{b_2\chi 𝒱_{OM}^2}{9𝒱}})`$ (5.7)
$`+`$ $`|W_{OM}+W_{GC}|^2(1{\displaystyle \frac{b_2\chi }{12𝒱}}){\displaystyle \frac{1}{C_H}}(W_{GC}(\overline{W_{GC}}+\overline{W_{OM}})+c.c.)(2𝒱)`$
$`+`$ $`(({\displaystyle \frac{\gamma }{C_H}}W_{GC}W_{OM})\overline{(W_{OM}+W_{GC})}+c.c.)(2𝒱_{OM}+{\displaystyle \frac{b_2\chi 𝒱_{OM}}{6𝒱}})).`$
All volumes and lengths are defined to be dimensionless by appropriately factorizing the eleven-dimensional Planck scale, $`l_{11}`$. For the same reason, we dropped the $`\kappa ^{4/3}`$ factor from $`\delta U`$. The scalar potential (5.7) can be written explicitly in terms of the moduli as
$`U`$ $`=`$ $`{\displaystyle \frac{1}{2^4}}[g^2e^{2(𝒱\gamma 𝒱_{OM})/C_H}({\displaystyle \frac{4𝒱^2}{C_H^2}}+{\displaystyle \frac{4\gamma ^2𝒱_{OM}^2}{3C_H^2}}+{\displaystyle \frac{4𝒱}{C_H}}{\displaystyle \frac{4\gamma 𝒱_{OM}}{C_H}}+{\displaystyle \frac{b_2\chi 𝒱}{3}}{\displaystyle \frac{b_2\chi \gamma ^2𝒱_{OM}^2}{9C_H^2𝒱}})`$
$`+`$ $`|h|^2e^{2𝒱_{OM}}\left({\displaystyle \frac{4𝒱_{OM}^2}{3}}+4𝒱_{OM}{\displaystyle \frac{b_2\chi 𝒱_{OM}^2}{9𝒱}}\right)`$
$``$ $`2ge^{𝒱_{OM}(𝒱\gamma 𝒱_{OM})/C_H}(Re(h)\mathrm{cos}({\displaystyle \frac{\sigma _S}{C_H}}({\displaystyle \frac{\gamma }{C_H}}+1)\sigma _T)`$
$`+`$ $`Im(h)\mathrm{sin}({\displaystyle \frac{\sigma _S}{C_H}}({\displaystyle \frac{\gamma }{C_H}}+1)\sigma _T))({\displaystyle \frac{4\gamma 𝒱_{OM}^2}{3C_H}}{\displaystyle \frac{2𝒱}{C_H}}+{\displaystyle \frac{2\gamma 𝒱_{OM}}{C_H}}{\displaystyle \frac{b_2\chi \gamma 𝒱_{OM}^2}{9C_H𝒱}})]{\displaystyle \frac{1}{𝒱_{OM}^3𝒱}},`$
where we have kept only the leading and next-to-leading order terms in an expansion in powers of $`1/𝒱`$, $`1/𝒱_{OM}`$.<sup>23</sup><sup>23</sup>23It is perhaps worth noting that (5.7) contains all contributions in $`1/𝒱`$, $`1/𝒱_{OM}`$ of the original potential, i.e. with $`\chi =0`$. This truncation is justified since near the minimum one must have $`𝒱>>1`$ and $`𝒱_{OM}>>1`$ for the validity of the supergravity approximation to M-theory. Due to the same condition (namely, $`𝒱_{OM}>>1`$), the contributions of multiply wrapped membranes, which are not well-understood, can be neglected.
From (LABEL:uuu) one can see that the $`R^4`$ corrections appear at the next-to-leading order in $`1/𝒱`$, $`1/𝒱_{OM}`$ compared to the leading terms (the first line in (5.7) with $`b_2\chi =0`$) that were kept in . Hence, non-surprisingly their effect is only a few percentage as we will see below.<sup>24</sup><sup>24</sup>24One might have thought that, similarly to the string case , the $`R^4`$ corrections would be dominant if the complex structure moduli were assumed to be stabilized at a supersymmetric minimum by a classical superpotential, for example from compactifying on a 6d non-Kähler manifold. This does not happen however, due to the fact that the no-scale structure of the Kähler potential for the $`T`$-modulus is preserved in our case. The main reason for the differences we find is in the expansion of the background to order $`\kappa ^{4/3}`$.
Extremizing with respect to the axionic scalars $`\sigma _{S,T}`$ yields the same condition as in
$$\mathrm{tan}(\frac{\sigma _S}{C_H}(\frac{\gamma }{C_H}+1)\sigma _T)=\frac{Imh}{Reh}.$$
(5.9)
Substituting the latter in (LABEL:uuu), to leading order in the expansion in powers of $`1/𝒱`$ and $`1/𝒱_{OM}`$, the scalar potential is a sum of squares:
$$U=\frac{1}{2^4𝒱𝒱_{OM}^3}[\frac{4𝒱_{OM}^2}{3}(\frac{\gamma }{C_H}ge^{(𝒱\gamma 𝒱_{OM})/C_H}(\pm )|h|e^{𝒱_{OM}})^2+\frac{4𝒱^2}{C_H^2}g^2e^{2(𝒱\gamma 𝒱_{OM})/C_H}]+\mathrm{}.$$
(5.10)
This positivity of the vacuum energy led the authors of to the conclusion that meta-stable<sup>25</sup><sup>25</sup>25Clearly, the global minimum is given by $`U=0`$ and is obtained in the decompactification limit $`𝒱\mathrm{}`$, $`𝒱_{OM}\mathrm{}`$. de Sitter vacua exist in heterotic M-theory. The two sign choices in (5.10) follow from solving for the axionic scalars from the extremum condition (5.9).
Keeping the sub-leading terms of the expansion in $`1/𝒱`$, $`1/𝒱_{OM}`$ as well as the $`R^4`$ induced corrections, we proceed next to extremize the scalar potential with respect to the orbifold length. The equation $`U/=0`$ can not be solved analytically. We have studied its solutions numerically for the same values of the various parameters as in : $`|h|=10^8,𝒱_v=800`$, $`C_H=8`$ corresponding to a hidden gauge group $`SO(10)`$, $`𝒮=(\frac{6}{10𝒱_v})^{1/3}`$. The $`R^4`$ induced correction is proportional to the Euler number of the Calabi-Yau manifold.
Figure 1 depicts the average Calabi-Yau volume $`𝒱`$ as a function of the orbifold length $``$ for the $`\kappa ^{4/3}`$ expansion of the “exact” background of . The solid line corresponds to a negative Euler number, whereas the dotted one - to $`\chi =0`$ meaning also no $`R^4`$ correction. As it is transparent from the second plot of Figure 1, both volumes $`𝒱,𝒱_{OM}`$ are sufficiently large near the minimum to justify the use of supergravity as an effective action. At the same time, the condition $`𝒱>>𝒱_{OM}`$ is satisfied to a reasonable degree, which justifies the use of the perturbative expansion in $`\kappa ^{2/3}`$.
We would also like to comment on the positivity of the volume of the Calabi-Yau: from the analysis of , it might be construed that the volume is positive for all values of $`x^{11}`$ as a consequence of using the “fully” non-linear background. With the choice of the warp factor $`k(x^{11})=f(x^{11})`$, possible in the absence of the higher derivative corrections as discussed extensively in Section 4.1, the authors of have found that the volume of the Calabi-Yau is equal to $`V_{CY}^0=\sqrt{e^{6f}}=\left[(1x^{11}\mathrm{\Theta }(x^{11})𝒮)^{2/3}\right]^3`$. Hence, $`V_{CY}^0=(1x^{11}\mathrm{\Theta }(x^{11})𝒮)^2`$, and using this expression of the volume, the authors of solved for the warp factors in terms of roots of a quantity that is manifestly positive, $`V_{CY}^0`$. We believe this to be misleading since the warp factors were derived first, using the Killing spinor equations: $`e^{3/2f}=1x^{11}\mathrm{\Theta }(x^{11})𝒮`$ according to eq (5.47) in . This implies that $`x^{11}`$ cannot be defined in this coordinate system beyond $`x_{\mathrm{max}}^{11}=1/𝒮`$, otherwise the warp factors will become negative. Moreover, the positivity of the volume of the CY for all values of $`x^{11}`$ is a coordinate dependent statement: with $`k=0`$, for instance, one finds $`V_{CY}^0=(1x^{11}\mathrm{\Theta }(x^{11})𝒮)^{3/2}`$, which carries the same implications, namely $`x^{11}1/𝒮`$. On the other hand, the average volume of the Calabi-Yau (5.5), which depends only on the orbifold length $``$, is coordinate independent, i.e. it is the same function $`𝒱()`$ irrespective of the choice of the warp factor $`k(x^{11})`$.
In Figure 2, we plotted the scalar potential near its minimum as a function of the orbifold length, for both signs arising from the extremization with respect to the axionic scalars. For each choice of sign we plotted both the truncated $`𝒪(\kappa ^{4/3})`$ background scalar potential, and the $`R^4`$ corrected potential, assuming a negative Euler number. From Figure 2 we see that still there is a de Sitter minimum, although the value of the scalar potential at the minimum is much smaller than the one reported in , which was of order $`10^{58}`$ for the same set of parameters. Leaving aside the difference in defining the average volume, this is due solely to the effect of truncating the exact background to order $`\kappa ^{4/3}`$, i.e. performing an expansion in the background flux up to second order and ignoring terms of order $`𝒪(𝒮^3)`$ and higher. Another consequence of this truncation is that the minimum is shifted towards bigger values of the orbifold length $``$.
Finally, Figure 3 shows the dependence of the scalar potential on the Euler number. Clearly, the correction to the volume and that to the scalar potential have opposite signs: for negative Euler numbers, the average volume is increasing, whereas the value of the scalar potential at the minimum shifts towards zero.
## 6 Summary and outlook
In the present paper we have studied various implications of the eleven-dimensional $`R^4`$ term for heterotic M-theory. Working to order $`\kappa ^{4/3}`$, we derived the change in the Kähler potential of the universal moduli and the ensuing corrections to the four-dimensional soft supersymmetry breaking terms. In particular, we observed that the induced $`R^4`$ corrections do not spoil the universality of the soft terms.
Next, we performed a detailed analysis of warp-factor geometric deformations of the background metric in the presence of the $`R^4`$ term, and commented on the generic-type deformations. We used the Killing spinor equations to extract the modified differential equations obeyed by the warp factors. We have presented a few explicit solutions, depending on the nature of the background fluxes. This has enabled us to find the correction to the volume of the Calabi-Yau three-fold, which turned out to be proportional to the Calabi-Yau Euler number. Our careful averaging over the $`x^{11}`$ direction also showed that the positive definitness of the Calabi-Yau volume in the solution of is a coordinate dependent statement. Thus, although we have found the expected shift of the volume of the Calabi-Yau manifold, it is unlikely that this can serve as a singularity resolution mechanism as suggested in . As shown in Subsection 4.1, when neglecting higher derivative corrections and considering the warp-factor metric deformations to all orders in $`\kappa `$, one cannot turn on simultaneously the two flux components: the boundary flux $`\beta _a`$, which does not introduce any $`x^{11}`$ dependence in the solution, and the flux $`\alpha `$ which induces an $`x^{11}`$ fibration of the Calabi-Yau. However, taking into account higher derivative terms like $`R^4`$, $`D^4R^4`$ etc. induces changes beyond the order $`\kappa ^{4/3}`$ to the supersymmetry variations and field equations. And so in principle this opens up the possibility to find solutions to higher orders which have both $`\alpha 0`$ and $`\beta _a0`$. This would be of great interest, since both components are important: the $`\alpha `$-flux for stabilization of the orbifold length, while the $`\beta _a`$ flux for generating a superpotential that would allow the stabilization of the complex structure moduli of the Calabi-Yau. At the same time, it is likely that having turned on both $`\alpha ,\beta _a`$ fluxes one has to abandon the warp-factor deformations, and consider generic non-Kähler deformations of the Calabi-Yau background.
The understanding of compactifications on six-dimensional non-Kähler manifolds is necessary for the stabilization of the complex structure moduli. Therefore, one natural direction for future research would be finding explicit non-Kähler backgrounds for the weakly coupled heterotic $`E_8\times E_8`$ string with methods similar to those of for the $`SO(32)`$ case. Another direction would be to study their strong coupling limit by trying to solve the generalized Hitchin flow equations we derived in Appendix B together with the appropriate $`G`$-field equation and Bianchi identity.
We have also studied the effective scalar potential in the context of an expansion of the background to order $`\kappa ^{4/3}`$. Since the no scale structure of the Kähler potential for the $`T`$-modulus is not violated by the $`R^4`$ term, the main effect is not due to the higher derivative correction, as was the case in string theory , but rather to the expansion of the background. We have found that de Sitter vacua still exist, although with a much smaller cosmological constant. It is worth investigating whether even higher order corrections, like $`D^4R^4`$ compactified on the zeroth order background or $`R^4`$ compactified on the deformed background etc., would violate the no scale structure of the Kähler potential, thus leading to qualitative changes along the lines of .
It would also be interesting to see how far one can get in building the eleven-dimensional action of Horava-Witten theory at higher orders, with the approach of . Lastly, one could also address the issue of how a truncation to order $`\kappa ^{4/3}`$ affects the assisted inflation solution of heterotic M-theory .
## Acknowledgements
We would like to thank K. Dasgupta, P. de Medeiros, M. Haack, A. Krause and A. Tseytlin for useful conversations. This work was supported in part by DOE grant DE-FG02-95ER40899.
## Appendix A Derivation of the Kähler potential
The Christoffel symbols for the metric
$$ds^2=e^{6ac}\overline{g}_{\mu \nu }dx^\mu dx^\nu +e^{2a}\stackrel{~}{g}_{mn}dx^mdx^n+e^{2c}(dx^{11})^2$$
(A.1)
are
$`\mathrm{\Gamma }_{np}^\mu =e^{8a+c}\stackrel{~}{g}_{np}^\mu a,\mathrm{\Gamma }_{n\rho }^m=\delta _n^m_\rho a,\mathrm{\Gamma }_{\mathrm{11\hspace{0.17em}11}}^\mu =e^{6a+3c}^\mu c,`$
$`\mathrm{\Gamma }_{11\mu }^{\mathrm{\hspace{0.17em}\hspace{0.17em}11}}=_\mu c,\mathrm{\Gamma }_{\mu \nu }^\rho =\overline{\mathrm{\Gamma }}_{\mu \nu }^\rho \delta _{(\mu }^\rho _{\nu )}(6a+c)+{\displaystyle \frac{1}{2}}\eta _{\mu \nu }^\rho (6a+c).`$ (A.2)
Note that also $`\mathrm{\Gamma }_{\mathrm{11\hspace{0.17em}11}}^{\mathrm{\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}11}}0`$, but it will not be needed.<sup>26</sup><sup>26</sup>26All partial derivatives in (A.2) and below are w.r.t. the metric $`\overline{g}_{\mu \nu }`$, i.e. they do not include the warp factor: all dependence on the moduli $`a`$ and $`c`$ is written down explicitly.
The nonzero curvature components are
$`R^\mu _{m\nu n}`$ $`=`$ $`e^{8a+c}\stackrel{~}{g}_{mn}[7_\nu a^\mu a+_\nu ^\mu a+{\displaystyle \frac{1}{2}}(_\nu c^\mu a+_\nu a^\mu c)`$
$``$ $`{\displaystyle \frac{1}{2}}\delta _\nu ^\mu ^\rho a_\rho (6a+c)],`$
$`R^m_{\mu n\nu }`$ $`=`$ $`\delta _n^m\left[7_\mu a_\nu a+_\mu _\nu a+_{(\mu }a_{\nu )}c{\displaystyle \frac{1}{2}}\eta _{\mu \nu }_\rho a^\rho (6a+c)\right],`$
$`R^k_{mnp}`$ $`=`$ $`\stackrel{~}{R}^k{}_{mnp}{}^{}e^{8a+c}_\rho a^\rho a[\stackrel{~}{g}_{pm}\delta _n^k\stackrel{~}{g}_{mn}\delta _p^k],`$
$`R^\mu _{\nu \rho \sigma }`$ $`=`$ $`\overline{R}^\mu {}_{\nu \rho \sigma }{}^{}+\mathrm{},`$
$`R^{11}_{\mu 11\nu }`$ $`=`$ $`2_\mu c_\nu c_\mu _\nu c6_{(\mu }c_{\nu )}a+{\displaystyle \frac{1}{2}}\eta _{\mu \nu }_\rho c^\rho (6a+c),`$
$`R^\mu _{11\nu 11}`$ $`=`$ $`e^{6a+3c}[2_\nu c^\mu c_\nu ^\mu c3(_\nu c^\mu a+^\mu c_\nu a)`$
$`+`$ $`{\displaystyle \frac{1}{2}}\delta _\nu ^\mu ^\rho c_\rho (6a+c)],`$
$`R^{11}_{m11n}`$ $`=`$ $`e^{8a+c}\stackrel{~}{g}_{mn}^\rho a_\rho c,`$
$`R^m_{11n11}`$ $`=`$ $`e^{6a+3c}\delta _n^m^\rho a_\rho c.`$ (A.3)
The expression $`\mathrm{}`$ in $`R^\mu _{\nu \rho \sigma }`$ is somewhat messy, so let us write down only what we need, namely its contribution to the scalar curvature:
$$R^\mu {}_{\nu \rho \sigma }{}^{}\delta _{\mu }^{\rho }e^{6a+c}\overline{g}^{\nu \sigma }=e^{6a+c}\left[R^{(4)}+3(6_\mu ^\mu a+_\mu ^\mu c)\frac{3}{2}(6a+c)^2\right].$$
(A.4)
Now let us see how the action $`S_0+S_1`$ in (2.2), (2.6) reduces to an effective four-dimensional theory of gravity and the scalars $`a,c`$. Keeping only terms that are at most quadratic in derivatives and ignoring the flux contribution, we find from $`S_0`$:<sup>27</sup><sup>27</sup>27For the reduction of the CJS action to fourdimensions, terms like $`\mathrm{}a`$, $`\mathrm{}c`$ can be dropped as they are simply total derivatives. But later on, when considering the $`R^4`$ correction, such terms will appear multiplied by nontrivial functions and so will have to be kept.
$$d^{11}x\sqrt{g}R^{(11)}=Vol_7d^4x\sqrt{\overline{g}}\left[R^{(4)}24(a)^2\frac{3}{2}(c)^26ac\right],$$
(A.5)
where $`Vol_7`$ is the volume of the 6d space times the length of the interval; we have dropped the total derivative term $`\mathrm{}(6a+c)`$. Upon the field redefinition $`\widehat{c}=c+2a`$ one recovers the action of .
Now let us consider the contribution of $`S_1`$. First, let us look at the following part of the integrand in (2.11):
$$\sqrt{g}\mathrm{\hspace{0.17em}2}^6(12ZRS+12R_{IJ}S^{IJ}).$$
(A.6)
We can evaluate its contribution to the kinetic terms of the four-dimensional scalars $`a`$ and $`c`$, using that as in
$$Z=24(2\pi )^3e^{6a+c}(a)^2Q$$
(A.7)
and computing the Ricci tensor and scalar curvature from (A.3), (A.4).<sup>28</sup><sup>28</sup>28The term $`R_{IJ}S^{IJ}`$ contributes only when the indices $`I,J`$ range over the 6d manifold, i.e. $`I,J=m,n`$. We obtain
$$(2\pi )^3\sqrt{\overline{g}\stackrel{~}{g}}Q\mathrm{\hspace{0.17em}2}^6(512^2(a)^2+18(c)^212R^{(4)}).$$
(A.8)
To get to the last expression, we have partially integrated terms of the form $`Q\mathrm{}a`$, $`Q\mathrm{}c`$ using that $`_\mu Q=6Q_\mu a`$.
Finally, let us turn to the remaining term in $`S_1`$, $`\frac{1}{2}E_8`$. The only index contractions in $`E_8`$ which give terms at most quadratic in derivatives are:
$`E_8`$ $`=`$ $`{\displaystyle \frac{1}{3!}}4(3ϵ^{\mu _1\mu _2\mu _3\mu _4}ϵ_{\mu _1\mu _2\nu _3\nu _4}R_{\mu _3\mu _4}{}_{}{}^{\nu _3\nu _4}+4ϵ^{\mu _1\mu _2\mu _3\mu }ϵ_{\mu _1\mu _2\mu _3\nu }R_{\mu 11}{}_{}{}^{\nu 11})`$ (A.9)
$`ϵ^{m_1m_2\mathrm{}m_6}ϵ_{n_1n_2\mathrm{}n_6}R_{m_1m_2}{}_{}{}^{n_1n_2}R_{m_3m_4}^{}{}_{}{}^{n_3n_4}R_{m_5m_6}^{}^{n_5n_6}`$
$`=`$ $`(2^4)(122^3\pi ^3)Q{\displaystyle \frac{1}{3!}}4(34R_{\mu \nu }{}_{}{}^{\mu \nu }+46R_{\mu 11}{}_{}{}^{\mu 11}),`$
where we have used that
$`ϵ^{m_1m_2\mathrm{}m_6}ϵ_{n_1n_2\mathrm{}n_6}R_{m_1m_2}{}_{}{}^{n_1n_2}R_{m_3m_4}^{}{}_{}{}^{n_3n_4}R_{m_5m_6}^{}{}_{}{}^{n_5n_6}=`$ (A.10)
$`2^4(R_{m_1m_2}{}_{}{}^{m_3m_4}R_{m_3m_4}^{}{}_{}{}^{m_5m_6}R_{m_5m_6}^{}{}_{}{}^{m_1m_2}2R_{m_1}{}_{m_2}{}^{m_3}{}_{}{}^{m_4}R_{m_3}^{}{}_{m_4}{}^{m_5}{}_{}{}^{m_6}R_{m_5}^{}{}_{m_6}{}^{m_1}{}_{}{}^{m_2})`$
$`=`$ $`2^4(122^3\pi ^3)Q.`$
From (A.3), (A.4) we see that
$$R_{\mu \nu }{}_{}{}^{\mu \nu }+2R_{\mu 11}{}_{}{}^{\mu 11}=e^{6a+c}[R^{(4)}54(a)^2\frac{3}{2}(c)^26ac+18\mathrm{}a+\mathrm{}c]$$
(A.11)
and hence
$`{\displaystyle \frac{1}{2}}\sqrt{g}E_8`$ $`=`$ $`2^6(12(2\pi )^3)\sqrt{g}Q(R_{\mu \nu }{}_{}{}^{\mu \nu }+2R_{\mu 11}{}_{}{}^{\mu 11})=`$ (A.12)
$`=`$ $`2^6(12(2\pi )^3)\sqrt{\overline{g}\stackrel{~}{g}}Q[R^{(4)}+54(a)^2{\displaystyle \frac{3}{2}}(c)^2],`$
where we have partially integrated the terms containing $`Q\mathrm{}a`$ and $`Q\mathrm{}c`$.
Now assembling (A.5), (A.8) and (A.12) we find the action
$`S_0+S_1`$ $`=`$ $`{\displaystyle \frac{Vl}{2\kappa _{11}^2}}{\displaystyle }d^4x\sqrt{\overline{g}}\{R^{(4)}[24{\displaystyle \frac{6b_2\kappa ^{4/3}\chi e^{6a}}{V}}](a)^2`$ (A.13)
$``$ $`.{\displaystyle \frac{3}{2}}(c)^26ac\},`$
where
$$b_2=122^6(2\pi )^3b_1T_2(2\kappa _{11}^{2/3})$$
(A.14)
and the factor of $`e^{6a}`$ is due to the fact that $`Q`$ in (2.13) was defined w.r.t. the metric $`e^{2a}\stackrel{~}{g}_{mn}`$ whereas the volume element we were now left with was just $`\sqrt{\stackrel{~}{g}}`$. Note that $`b_2`$ is a constant independent of $`\kappa _{11}`$ since $`T_2=(2\pi )^{2/3}(2\kappa _{11}^2)^{1/3}`$. For convenience, from now on we normalize $`Vd^6x\sqrt{\stackrel{~}{g}}=1`$. Clearly we can diagonalize the kinetic terms with the same field redefinition as before, i.e. $`\widehat{c}=c+2a`$. The result is:
$$S_0+S_1=\frac{l}{2\kappa _{11}^2}d^4x\sqrt{\overline{g}}\left\{R^{(4)}\mathrm{\hspace{0.17em}6}[3b_2\kappa ^{4/3}\chi e^{6a}](a)^2\frac{3}{2}(\widehat{c})^2\right\}.$$
(A.15)
Now we are ready to read off the new Kähler potential. Recall that the fields $`a`$ and $`c`$ make up the real parts of two chiral superfields with bosonic components
$$S=e^{6a}+i\sigma _S,T=e^{\widehat{c}}+i\sigma _T,$$
(A.16)
where the axionic scalars $`\sigma _S`$, $`\sigma _T`$ originate from the eleven-dimensional three-form field $`C`$. For convenience we will use from now on $`S`$, $`T`$ to denote the full superfields. The kinetic terms of the four-dimensional $`N=1`$ effective action for these fields are of the standard form
$$2K_{S\overline{S}}dSd\overline{S}+2K_{T\overline{T}}dTd\overline{T}.$$
(A.17)
Since the $`(\widehat{c})^2`$ term in (A.15) does not receive any correction, the corresponding part of the Kähler potential is as before:
$$K_{(T)}=3\mathrm{ln}(T+\overline{T}).$$
(A.18)
On the other hand, the new kinetic term for $`a`$ can be reproduced from the following Kähler potential:
$$K_{(S)}=\mathrm{ln}(S+\overline{S})\frac{b_2\kappa ^{4/3}\chi }{6(S+\overline{S})}.$$
(A.19)
We also record the derivatives of the zero-th order Kähler potential
$$K_0=\mathrm{ln}[(T+\overline{T})^3(S+\overline{S})]\mathrm{ln}(2^4𝒱_{OM}^3𝒱)$$
(A.20)
which are needed in the evaluation of the scalar potential
$$K_{0,S}=\frac{1}{2𝒱},K_{0,T}=\frac{3}{2𝒱_{OM}},K_{0,S\overline{S}}=\frac{1}{4𝒱^2},K_{0,T\overline{T}}=\frac{3}{4𝒱_{OM}^2},$$
(A.21)
where we introduced the notation $`S+\overline{S}=2𝒱`$, $`T+\overline{T}=2𝒱_{OM}`$.
## Appendix B Generalized Hitchin flow equations
Dall’Agata and Prezas studied the conditions for $`N=1`$ compactifications of M-theory on seven-dimensional manifolds with $`SU(3)`$ structure . Recall that requiring $`G_2`$ holonomy is too restrictive, meaning that it leads to trivial warp factors and no fluxes . On the other hand $`G_2`$ structure, considered in , carries less information than the $`SU(3)`$ structure case.
Let us summarize the relevant results of . The eleven-dimensional metric is of the form:
$$ds_{11}^2=e^{2\mathrm{\Delta }}\eta _{\mu \nu }dx^\mu dx^\nu +ds_7^2,$$
(B.1)
where $`\mathrm{\Delta }`$ depends only on the coordinates of the internal 7d space. Since we are interested in an internal metric which is a warped product of an interval and a six-dimensional non-Kähler manifold whose three torsion classes $`W_3`$, $`W_4`$, $`W_5`$ are non-vanishing<sup>29</sup><sup>29</sup>29Recall that 6d non-Kähler manifolds are classified by five torsion classes $`W_i`$, $`i=1,\mathrm{},5`$ and that in heterotic string compactifications supersymmetry requires $`W_{1,2}=0`$, whereas generically all three $`W_3`$, $`W_4`$, $`W_5`$ can be nonzero. For more details see ., we take for $`ds_7^2`$ the form given in Section 4.2 of :
$$ds_7^2=e^{p\varphi (y,t)}ds_6^2(y,t)+e^{2\varphi (y,t)}dt^2,$$
(B.2)
where $`p`$ is a real number and $`t`$ parametrizes the interval $`I`$. The 4-form flux $`G`$ has decomposition as in footnote 10 in terms of the 1-form $`v`$, 2-form $`J`$ and three-form $`\mathrm{\Psi }`$ that define an $`SU(3)`$ structure in $`d=7`$. In fact, it will be more convenient to use the 7d Hodge dual with decomposition (see (3.11) of ):<sup>30</sup><sup>30</sup>30As recalled in footnote 10, conditions (3.13) of imply that $`c_1^{}=0`$, $`c_2^{}=0`$, $`V=0`$ in (3.11) there.
$$_7G=\frac{Q}{3}Jv+vA\frac{3}{2}J\sigma +S.$$
(B.3)
The supersymmetry conditions can be rewritten as equations relating $`v`$, $`J`$, $`\mathrm{\Psi }`$ and the $`G`$-flux components $`Q,A,\sigma ,S`$.<sup>31</sup><sup>31</sup>31The notation in this appendix is the same as in and should not be confused with the notation in the rest of the present paper. Recall that the six-dimensional derivatives of the three $`SU(3)`$-structure defining forms determine the non-Kähler manifold whereas their $`_t`$ derivatives determine its fibration along the seventh dimension. Let us now write down these generalized Hitchin flow equations for arbitrary $`p`$ (unlike the case $`p=\frac{1}{2}`$ in eqs. (4.37), (4.38) of ) so that all of the torsion classes $`W_i`$, $`i=3,4,5`$, are non-vanishing. Using
$$v=e^\varphi dt,J=e^{p\varphi }\widehat{J},\mathrm{\Psi }=e^{\frac{3}{2}p\varphi }\widehat{\mathrm{\Psi }},$$
(B.4)
where $`\widehat{}`$ denotes a six-dimensional quantity, and also (3.20), (3.21), (4.34) of we find for the six-dimensional space:
$`\widehat{d}\widehat{J}`$ $`=`$ $`2e^{p\varphi }S\left(p{\displaystyle \frac{1}{2}}\right)\widehat{d}\varphi \widehat{J}`$
$`\widehat{d}\widehat{\mathrm{\Psi }}`$ $`=`$ $`{\displaystyle \frac{3}{2}}(p1)\widehat{d}\varphi \widehat{\mathrm{\Psi }}`$ (B.5)
and for the dependence on the interval:
$`_t\widehat{J}`$ $`=`$ $`p\dot{\varphi }\widehat{J}+{\displaystyle \frac{2}{3}}Qe^\varphi \widehat{J}2e^{(p1)\varphi }A`$
$`_t\widehat{\mathrm{\Psi }}`$ $`=`$ $`{\displaystyle \frac{3}{2}}p\dot{\varphi }\widehat{\mathrm{\Psi }}+Qe^\varphi \widehat{\mathrm{\Psi }}.`$ (B.6)
Clearly $`W_4`$ vanishes if $`p=\frac{1}{2}`$. And also one can easily see that substituting $`p=\frac{1}{2}`$ in (B) and (B), gives exactly (4.37), (4.38) of . From (B) we can read off the torsion classes of the 6d manifold:
$$W_3=2e^{p\varphi }S,W_4=\left(p\frac{1}{2}\right)\widehat{d}\varphi ,W_5=\frac{3}{2}(p1)\widehat{d}\varphi .$$
(B.7)
If this space is to be a solution of the weakly coupled heterotic string<sup>32</sup><sup>32</sup>32Strictly speaking, this is not necessary in order to have a solution of the eleven-dimensional theory. But backgrounds that have this property are most easily interpreted as eleven-dimensional lifts of ten-dimensional heterotic string solutions., then supersymmetry requires $`W_5=2W_4`$ (see ) and so fixes $`p=\frac{5}{7}`$.
However, the ansatz (B.2) is not the most general one. Compatibility of the structure of the solutions in and requires that $`\stackrel{~}{\varphi }=2\mathrm{\Delta }`$, where $`\stackrel{~}{\varphi }`$ is the warp factor in front of $`dt^2`$ and $`\mathrm{\Delta }`$ is the one in front of the four-dimensional space $`\eta _{\mu \nu }dx^\mu dx^\nu `$. Note that this is the same condition as (4.12). On the other hand, there is no reason why the warp factor in front of $`ds_6^2(y,t)`$ would be related to any of the other two. So let us consider metric of the following form:
$$ds_{11}^2=e^{2\mathrm{\Delta }}\eta _{\mu \nu }dx^\mu dx^\nu +e^{p\varphi (y,t)}ds_6^2(y,t)+e^{4\mathrm{\Delta }}dt^2,$$
(B.8)
where $`\mathrm{\Delta }`$ and $`\varphi `$ are unrelated to each other. It is easy to check that instead of (4.34) and (4.35) of we have now $`\sigma =\widehat{d}\mathrm{\Delta }`$, $`\dot{\mathrm{\Delta }}=\frac{1}{3}Qe^{2\mathrm{\Delta }}`$, $`\widehat{d}Q=2Q\widehat{d}\mathrm{\Delta }`$. Therefore using $`v=e^{2\mathrm{\Delta }}dt`$, $`J=e^{p\varphi }\widehat{J}`$, $`\mathrm{\Psi }=e^{\frac{3}{2}p\varphi }\widehat{\mathrm{\Psi }}`$ and (3.20), (3.21) of , we derive for the generalizations of (B) and (B):
$`\widehat{d}\widehat{J}`$ $`=`$ $`2e^{p\varphi }Sp\widehat{d}\varphi \widehat{J}\widehat{d}\mathrm{\Delta }\widehat{J}`$
$`\widehat{d}\widehat{\mathrm{\Psi }}`$ $`=`$ $`{\displaystyle \frac{3}{2}}p\widehat{d}\varphi \widehat{\mathrm{\Psi }}3\widehat{d}\mathrm{\Delta }\widehat{\mathrm{\Psi }}`$ (B.9)
and
$`_t\widehat{J}`$ $`=`$ $`p\dot{\varphi }\widehat{J}+e^{2\mathrm{\Delta }}\left({\displaystyle \frac{2}{3}}Q\widehat{J}2e^{p\varphi }A\right)`$
$`_t\widehat{\mathrm{\Psi }}`$ $`=`$ $`{\displaystyle \frac{3}{2}}p\dot{\varphi }\widehat{\mathrm{\Psi }}+e^{2\mathrm{\Delta }}Q\widehat{\mathrm{\Psi }}.`$ (B.10)
From (B) we read off the following torsion classes:
$$W_3=2e^{p\varphi }S,W_4=p\widehat{d}\varphi \widehat{d}\mathrm{\Delta },W_5=\frac{3}{2}p\widehat{d}\varphi 3\widehat{d}\mathrm{\Delta }.$$
(B.11)
If one wants the 6d non-Kähler manifold to be a solution of heterotic strings, i.e. to satisfy $`W_5=2W_4`$, then one must impose a linear relation between $`\widehat{d}\mathrm{\Delta }`$ and $`\widehat{d}\varphi `$. Taking $`\widehat{d}\mathrm{\Delta }=k\widehat{d}\varphi `$ for some constant $`k`$, we find that $`k=\frac{7}{10}p`$. It is easy to see that this is consistent with our previous considerations: for $`p=\frac{5}{7}`$ we recover the relation $`\widehat{d}\mathrm{\Delta }=\frac{1}{2}\widehat{d}\varphi `$ valid for the ansatz (B.2) .
|
warning/0506/cs0506071.html
|
ar5iv
|
text
|
# Signal transmission on lossy lines as a dissipative quantum state propagation.
## I Introduction
The electric signals transmission on distributed RLC-lines will be solved analytically. Our goals are deformation of a signal shape and time delay. These questions are usually considered by the calculation technique, made for lumped lines \[1-5\]. On the other hand, in the works \[6-9\] people build models, based on the Maxwell- or Schrödinger-like equations.
The physical approach, developed here, is based on the introduction of electric impulses on the lossy line as a dissipative quasi particle. The space-time propagation of it is described by the Klein-Gordon equation - the equation of motion of free relativistic scalars (see, for example, ). The parameters of an electric signal have a direct particle-like meaning. For example, the time decay rate plays the role of a negative mass square.
## II Equations, solutions. Information transfer
The lossy interconnect line can be introduced as consisting of a row of infinitesimally small RLC-cells. Each cell of length $`dl`$ has a local resistance $`dR,`$ inductance $`dL`$ and capacitance $`dC:`$
$$dR=r(l)dl,dL=\mathrm{\pounds }(l)dl,dC=c(l)dl,$$
(1)
determined by the linear densities as:
$$r(l)=dR/dl,\mathrm{\pounds }(l)=dL/dl,c(l)=dC/dl.$$
(2)
The target voltage of cell $`U_C=`$ $`U_{out}(t)`$ is expressed through entrance one - $`U_{in}(t)`$ by the condition:
$$U_{in}=U_L+U_R+U_{out},$$
(3)
where:
$$U_L=dL\frac{J}{t},U_R=dRJ,\frac{J}{l}=c\frac{U}{t}.$$
(4)
So, the voltage $`U(l,t),`$ as a function of time and distance, obeys the equation:
$$c\mathrm{\pounds }\frac{^2U}{t^2}\frac{^2U}{l^2}+rc\frac{U}{t}=0.$$
(5)
This equation is known as the one-dimensional telegraph equation and describes the wave distribution with damping. For the case $`\mathrm{\pounds }=0`$ it has been treated in . The space variable can be normalized by replacing $`l`$ with $`x=l/v`$. Then, using the notations: $`m=r/2\mathrm{\pounds }`$ $`,v^2=1/c\mathrm{\pounds }`$, the function $`\mathrm{\Phi }(x,t)`$ is introduced: $`U(x,t)=\mathrm{exp}(mt)\mathrm{\Phi }(x,t).`$ It obeys the equation:
$$\frac{^2\mathrm{\Phi }}{t^2}\frac{^2\mathrm{\Phi }}{x^2}m^2\mathrm{\Phi }=0.$$
(6)
This is the Klein-Gordon equation\* with the negative mass square - so-called ”tachyons”. Earlier we dealt with such an exotic object in a real situation . This representation looks like the motion of the tachyon quasi particle in the constant external dissipative field.
The basis solution of the main equation (5) with the wave vector $`k`$
$$\phi (\omega _0x,t)=e^{mti\left(\omega _0tkx\right)},\text{ }\omega _0(k)=\sqrt{k^2m^2}$$
(7)
is a plane wave, damped in time at the rate $`m`$, identical for all frequencies. A macro-packet - $`U(x,t)`$ can be built as a linear superposition of basis states:
$$U(x,t)=\underset{\underset{¯}{\omega }}{\overset{\overline{\omega }}{}}d\omega _0U(\omega _0)\phi (\omega _0x,t).$$
(8)
The wave $`\phi (\omega _0x,t)`$ is a not a stationary state, but its energy loss in time as the square of amplitude:
$$\widehat{\omega }\phi (\omega _0x,t)=\omega (t)\phi (\omega _0x,t)=e^{2mt}\omega _0\phi (\omega _0x,t).$$
(9)
The frequency-width of a packet: $`\mathrm{\Delta }\omega =\overline{\omega }\underset{¯}{\omega }`$ also falls down in time as $`e^{2mt}.`$ The packet does not spread in time, but agglomerates in frequencies. The wave $`\phi (\omega _0x,t)`$ has a phase velocity:
$$v_{ph}=\omega _0/k=\omega _0/\sqrt{\omega _0^2+m^2}.$$
(10)
But the information transfer velocity- the group velocity of a packet, defined as: $`u\omega _0/k,`$ for the dispersion law (7) results:
$$v_{gr}=\sqrt{\omega _0^2+m^2}/\omega _0=k/\omega _0.$$
(11)
On the first view this velocity (in the natural units: $`v_{gr}=vk/\omega _0`$) is higher than the light speed in the wire: $`v_{gr}>v`$. But the packet decays in time as $`e^{mt}`$. In a short time $`dt`$ it loses the part: $`dx_2\frac{\pi m}{\omega _0}dt`$ from the forward edge, which amplitude becomes $`e^1`$ of initial. This part must be subtracted from the whole packet relocation: $`dx_1=v_{gr}dt`$. So, the effective speed of the packet can be approximated as:
$$uv_{gr}\frac{\pi m}{\omega _0}=(\sqrt{\omega _0^2+m^2}\pi m)/\omega _0<1.$$
(12)
The treatment of electric impulses on the lossy line as a quasi particle gives a physical restriction on the measurement accuracy of the time delay. Following the uncertainty relations, the measurement of the time-slice $`\mathrm{\Delta }t`$ is accompanied with the energy variation: $`\mathrm{\Delta }\omega _01/\mathrm{\Delta }t.`$ At the signal frequency $`\omega _0`$ the delay time $`\delta `$ can be measured up to accuracy:$`\mathrm{\Delta }\delta `$ $`1/\omega _0.`$ This inequality gives a minimal error, physically permissible by the measurements of time-slices. This restriction has a fundamental character, and does not depend on the chosen approach.
## III Green-function. Integral representation
The Green function for the equation (5) can be built from the solutions (7) as:
$$G(x,t)=\frac{e^{mt}}{(2\pi )^2}\underset{\mathrm{}}{\overset{}{}}d\omega \underset{\mathrm{}}{\overset{}{}}dk\frac{\mathrm{exp}[i(\omega tkx)]}{m^2\omega ^2+k^2}.$$
(13)
The integrand has two poles at $`\omega =\omega _{1/2}`$, which form two complex resonance frequencies:
$$\omega _{1/2}(k)=mi\pm \omega _0(k).$$
(14)
At $`k<m`$ frequency $`\omega _0`$ becomes imaginary- the wave does not propagate. But for all $`\mathrm{}<k<\mathrm{}`$ the both poles lay below the real $`\omega `$-axis. To build a retarded Green function, the integration contour can be kept on the real $`\omega `$-axis. At $`t>0`$ the contour closes in the low $`\omega `$-half-plane and contains all possible positions of poles. The retarded Green function is the sum of the residues in the both poles $`\omega _{1/2}`$:
$$G^{ret}(x,t)=\frac{e^{mt}\mathrm{\Theta }\left(t\right)}{2\pi }\underset{\mathrm{}}{\overset{}{}}dk\frac{\mathrm{sin}\omega _0t}{\omega _0}e^{ikx},$$
(15)
where: $`\mathrm{\Theta }(t)`$\- is a Heaviside function. The real range of the integration in (15) is restricted by the condition: $`|k|m`$. Such ”damped tachyon” Green function in the coordinate representation is the analytical continuation of the scalar one into the area $`m^2m^2`$. Using the integrals, calculated in , this function can be expressed as:
$$G^{ret}(x,t)=\mathrm{\Theta }\left(t\right)\mathrm{\Theta }\left(\lambda \right)e^{mt}I_0(m\sqrt{\lambda }),$$
(16)
where: $`\lambda =t^2x^2,`$ and $`I_0(x)`$ modified Bessel - function.
Supposing that the source voltage $`U_{in}(t)`$ was applied to the RLC-line during the limited time $`0tt_{ac}`$ at the point $`x=0:`$
$$U_{in}(x=0,t)=u_0(t).$$
(17)
The cross-section of the wire in the point $`x=0`$ can be treated as a 2-dimensional boundary $`(S)`$. The unit vector of an external normal to it is denoted as $`\stackrel{}{n}.`$ The solution of homogeneous equations (5) at the boundary condition (17) is given by the integral formula:
$$U(x,t)=\underset{0}{\overset{𝜏}{}}dt\stackrel{´}{}G^S(x,t\stackrel{´}{})u_0(tt\stackrel{´}{}),\text{ }\tau =\mathrm{max}(t,t_{ac}),$$
(18)
where $`G^S(x,t)`$ -is the Green function of the 2nd kind, expressed through $`G^{ret}`$ as: $`G^S(x,t)\frac{dG^{ret}}{dn}.`$ Taking into account the property of Bessel - functions and noting, that $`\stackrel{}{n}`$ indicates the positive $`x`$-direction, it becomes:
$$G^S(x,t)=2xe^{mt}[\delta \left(\lambda \right)+\mathrm{\Theta }\left(\lambda \right)I_1(m\sqrt{\lambda })].$$
(19)
## IV Delays. Leading transfer mode
The representation (19) can be applied to the macro-packet (8). The amplitude of a signal, propagating in the positive (negative-) $`x`$-direction- $`U_r(U_l)`$, is equal:
$$U_{r(l)}(x,t)=e^{mx}u_0(tx)+x\underset{0}{\overset{\tau ^2x^2}{}}d\lambda \frac{e^{m\sqrt{\lambda +x^2}}}{\sqrt{\lambda +x^2}}$$
(20)
$$I_1(m\sqrt{\lambda })u_0(t\sqrt{\lambda +x^2}),\text{ }x<\tau .$$
Such a representation repeats the result, given in . The input signal $`U_{in}`$ in the point $`x`$ achieves the level $`U_{in}=bU_{\mathrm{max}}`$, $`b<1`$ in the delay time $`\delta `$, determined by the equation:
$$U_r(x,\delta )=bU_{\mathrm{max}}.$$
(21)
The frequent properties of the transient response, defining by the Green function (15), are following. The signal of frequency $`\omega _{in}`$ is transmitted by the leading transfer mode (LTM) of the same frequency: $`\omega _0=\omega _{in}.`$ For example, at small frequencies: $`\omega _{in}0,u_0=const,`$ the asymptotic of the Green-function, also corresponding to the DC propagation, has a view:
$$G_0^{ret}(x,t)\frac{te^{mt}}{\pi }\frac{\mathrm{sin}mx}{x},\text{ }\omega _00.$$
(22)
For the constant input signal $`u_0`$ equation for the delay time in the point $`x`$ is:
$$\frac{te^{mt}}{\pi }[\frac{\mathrm{sin}mx}{x^2}\frac{m\mathrm{cos}mx}{x}]\left[(t+1/m)e^{mt}1/m\right]=b$$
(23)
## V Reflections
For the finite RLC-lines, length $`\overline{l}`$ $`(`$ $`\overline{x}=l/v),`$ reflections occur. The wave packet is now a superposition of incoming and reflected waves. The resulting voltage with $`N_r`$\- reflections is:
$$U_{ref}(x,t)=\underset{s=0}{\overset{N_r}{}}\left[\mathrm{\Gamma }^{2s}U_r(x+2s\overline{x},t)+\mathrm{\Gamma }^{2s+1}U_l(2s\overline{x}x,t)\right],$$
(24)
where $`\mathrm{\Gamma }`$ is the reflection coefficient. The flying signal decays on the length: $`\overline{l}_dv/m`$. For the line of length $`\overline{l},`$ the reflection number is the ratio: $`N_r\overline{l}_d/\overline{l}.`$ Equation: $`U_{ref}(x,\delta )=bU_{\mathrm{max}}`$ gives now the signal delay time $`\delta .`$
In the modern interconnection RLC-lines, for example, at the parameters values as in : $`r=37.8`$ $`\mathrm{\Omega }/cm,`$ $`c=3.28e13F/cm,`$ $`Z_0=\sqrt{\mathrm{\pounds }/c}=266.5\mathrm{\Omega }`$ and the length of the line: $`\stackrel{\_}{l}=3.6cm,`$ we have: $`\mathrm{\pounds }=2.3e8Hn/cm,`$ the decay rate:$`m=8.1e+08Hz`$. For the signal frequency $`\nu 3GHz`$ the decay length is: $`\overline{l}_d14.1cm`$ \- and, at least: $`N_r4`$ reflections are observable.
## VI Multilevel networks
For generalizing the obtained results on the multilevel network, the network can be represented as the $`N`$-component space formed from $`N`$ $`RLC`$-lines. An orthogonal basis will be put on the directions of RLC-lines: $`\{e\}=\{\stackrel{}{e}_1,\stackrel{}{e}_2\mathrm{}\stackrel{}{e}_N\},`$ $`\stackrel{}{e}_i\stackrel{}{e}_k=\delta ^{ik}`$. The voltage impulse, flowing through the network, becomes now a vector in this space. For example, for the network, consisting of three lines, the voltage vector is: $`\stackrel{}{U}(l,t)=[U_1(l,t),U_2(l,t),U_3(l,t)].`$ The vector $`\stackrel{}{U}`$ obeys the equation (5) in the matrix form :
$$\widehat{𝐜}\widehat{\mathrm{\pounds }}\frac{^2\stackrel{}{U}}{t^2}\frac{^2\stackrel{}{U}}{l^2}+r\widehat{𝐜}\frac{\stackrel{}{U}}{t}=0,$$
(25)
where the matrixes $`\widehat{c}`$ and $`\widehat{\mathrm{\pounds }},`$ are:
$$\widehat{𝐜}=\left[\begin{array}{ccc}2c_{grd}+c_m& c_m& 0\\ c_m& 2c_{grd}+2c_m& c_m\\ 0& c_m& 2c_{grd}+c_m\end{array}\right],$$
(26)
$$\widehat{\mathrm{\pounds }}=\left[\begin{array}{ccc}\mathrm{\pounds }_{11}& \mathrm{\pounds }_{12}& \mathrm{\pounds }_{13}\\ \mathrm{\pounds }_{12}& \mathrm{\pounds }_{22}& \mathrm{\pounds }_{23}\\ \mathrm{\pounds }_{13}& \mathrm{\pounds }_{23}& \mathrm{\pounds }_{33}\end{array}\right],$$
(27)
$`c_{grd}`$ \- line-to-ground capacitance, $`c_m`$\- line-to-line capacitance; and $`\mathrm{\pounds }_{ik}`$-self- and mutual between-conductor-inductances. The non-diagonal components of tensors describe interline transmission. It was pointed out in , the product of the matrixes $`\widehat{𝐜}`$ and $`\widehat{\mathrm{\pounds }}`$ is proportional to the unit matrix: $`\widehat{𝐜}\widehat{\mathrm{\pounds }}=\frac{1}{v^2}\widehat{𝐈}.`$. It is natural as voltage obeys the inhomogeneous Maxwell equation, containing, as well as (6), D‘Alamber operator. Then the space variable $`l`$ can again be normalized by: $`x=l/v`$ and the particle mass (which is now a tensor) can be introduced: $`\widehat{𝐦}=\frac{rv^2}{2}\widehat{𝐜}=\frac{r}{2}\widehat{\mathrm{\pounds }}^1.`$ The multilevel network with distributed RLC-parameters appears as an anisotropic solid medium. The situation with the quasi particle parameters, depending on direction, is here well known. By substitution: $`\stackrel{}{U}(x,t)=`$ $`\mathrm{exp}(\widehat{𝐦}t)\stackrel{}{\mathrm{\Phi }}(x,t)`$ the matrix analog of the equation (6) for $`\stackrel{}{\mathrm{\Phi }}`$ can be obtained:
$$\frac{^2\stackrel{}{\mathrm{\Phi }}}{t^2}\frac{^2\stackrel{}{\mathrm{\Phi }}}{x^2}\widehat{𝐦}^2\stackrel{}{\mathrm{\Phi }}=0.$$
(28)
The vectors $`\stackrel{}{U}(x,t)`$ and $`\stackrel{}{\mathrm{\Phi }}(x,t)`$ in general case are not parallel, so as the decay operator $`\mathrm{exp}(\widehat{𝐦}t)`$ mix’s the components. It seems that the last term in (28) describes the self-interaction of the field $`\stackrel{}{\mathrm{\Phi }}.`$ But the numerical tensor $`\widehat{𝐦}`$, generally, can be diagonalized and thus the field components will be split. With it the Green function will also be diagonalized. Then, the proper directions of the tensor $`\widehat{𝐦}`$: $`\{ϵ\}=\{\stackrel{}{ϵ}_1,\stackrel{}{ϵ}_2,\stackrel{}{ϵ}_3\}`$ can be chosen as the new basis in the network-space. The same quasi-free propagation occurs along the each of new orts as in the case of the single RLC-line:
$$U_i(x,t)=\mathrm{exp}(m_it)\mathrm{\Phi }_i(x,t).$$
(29)
The decay rates $`(m_1,m_2,m_3)`$ are different for each ort of this basis. Now the further reasons and results of the work can be reproduced for multilevel networks. The Green function has the same form as (15):
$$\stackrel{}{U}_{(m)}(x,t)=\underset{0}{\overset{𝜏}{}}dt\stackrel{´}{}G^{ret}(m_ix,tt\stackrel{´}{})\stackrel{}{U}_{(in)}(t\stackrel{´}{}),$$
(30)
where $`\stackrel{}{U}_{(in)}(t)`$\- is incoming voltage vector, in the new basis $`\{ϵ\}`$. The traveling voltage, propagating in the real space on the RLC-lines, becomes the projection of $`\stackrel{}{U}_{(m)},`$ calculated by (30), on the corresponding unit vector of the old basis $`\{e\}`$.
## VII Conclusion
The suggested representation for the transient response of the distributed RLC-line has allowed to obtain the exact formulas for the Green function and time delay. The simple description of signals propagation by the frequency characteristics in the LTM-approximation is obtained. The natural generalization on the multilevel networks of arbitrary dimension is made. The physical approach can be helpful at the multilevel networks design and in the quantum information.
###### Acknowledgements.
The author expresses gratitude to professor D. Robaschik for useful discussions.
———————————————
\*Indeed, it is necessary to multiple the equation by factor $`\mathrm{}^2`$/$`c^2,`$ but in the field theory the system of unit is adopted, where: $`\mathrm{}`$=1, $`c=1.`$
|
warning/0506/quant-ph0506035.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
Entanglement is presumably the most non-classical feature of quantum mechanics. It is in the heart of Einstein-Podolsky-Rosen paradox ; of the so-called *which way* interferometers , in which the interference pattern is washed out by the entanglement of the interferometric particle with a “which way discriminator”; of teleportation of quantum states ; of some cryptographic protocols ; and of the most important quantum algorithms . It is not difficult to *define* when a quantum state is entangled: for pure states they can only be (in bipartite case) factorizable, when can be written as $`|\psi |\varphi `$, or entangled. The multipartite case is a little bit subtler, since the parts can be put together. For example, for three qubits ($`A`$, $`B`$, and $`C`$), the state given by a Bell state $`|\mathrm{\Psi }^\pm =\left\{|01\pm |10\right\}/\sqrt{2}`$ of the pair $`AB`$ together with a pure state of $`C`$ is said *biseparable*, in the sense that, with respect to the bipartition $`ABC`$, it has no entanglement at all. One usually says that in such a state there is no *genuine* tripartite entanglement . For mixed states, whenever $`\rho `$ can be written as an ensemble of pure states which do not have some kind of entanglement, the mixed state is also said not to have such kind of entanglement. However, it is not a simple task to identify whether a given state has some kind of entanglement. Peres obtained a good criterion by noting that for separable states, the *partial transposition* lead to valid density matrices, whereas for some entangled state a non-positive matrix is generated . The Horodeckis put this contribution in the context of *positive maps which fail to be completely positive* acting on density operators , and showed that the criterion is sufficient only for two qubits or for one qubit and one qutrit. Examples of bipartite entangled states which have positive partial transpose (PPT) are known .
Another way of identifying entanglement comes from the fact that, for a given type of entanglement, the set of (mixed) states which are free from such entanglement is a closed convex set. Any point in the complement of a closed convex subset of a vector space can be separated from it by a hyperplane . Equivalently, there is a linear functional over this space which is positive in all separable states $`\sigma `$ (i.e.: those which do not pursue the inspected kind of entanglement), and negative in the specific entangled state $`\rho `$ (i.e.: the point out of the closed convex subset). This functional is called an *entanglement witness*, and is represented by the operator $`𝒲`$ . In linear algebraic context, it reads: $`\mathrm{Tr}(𝒲\rho )<0`$ and $`\mathrm{Tr}(𝒲\sigma )0`$, for all separable $`\sigma `$. Once again, it is simple *in theory*, but given a state which one wants to decide whether it shows some kind of entanglement or not, it is not easy to find the appropriated witness. A very important class of entanglement witnesses (EW) is the optimal entanglement witnesses (OEW). An OEW is the EW that best witnesses the entanglement of $`\rho `$, in the sense that it reaches the maximum value of $`\mathrm{Tr}(𝒲\rho )`$ . Unfortunately, finding an OEW is a difficult task because it involves a hard process of optimization . However some improvements has been done in the sense of finding OEWs for certain kinds of states .
For three qubits, Dür, Vidal, and Cirac showed that there are two different genuine tripartite entangled states, in the sense that, even in a statistical sense, they can not be locally converted one into another . These states are the so called GHZ state, which appeared on the literature in Ref. , $`|GHZ=\left\{|000+|111\right\}/\sqrt{2}`$, and the W state , $`|W=\left\{|001+|010+|100\right\}/\sqrt{3}`$. Ishizaka and Plenio showed that even with PPT-operations (those which send PPT operators into PPT operators), GHZ and W remain inequivalent , however, stochastically there is a protocol to convert GHZ in W via PPT-operations with approximately $`75\%`$ of success .
## 2 Fooling the GHZ/W-criterion
A systematic way to construct entanglement witnesses to pure states is looking for the optimal element of the set of EWs that have the specific form
$$𝒲_\psi =\mathrm{\Lambda }|\psi \psi |,$$
(1)
with $`\mathrm{\Lambda }`$ and multiplication by the identity operator omitted in notation. By the condition that $`\mathrm{Tr}(W\sigma )0`$ if $`\sigma `$ is separable (denote $`𝒮`$ this set), we can see that
$$\mathrm{\Lambda }=\underset{\sigma 𝒮}{\mathrm{max}}\sigma \psi ^2.$$
(2)
As an example, for the specific case of three qubits, the OEW for the GHZ-like states,
$$|GHZ(\varphi )=\frac{(|000+e^{i\varphi }|111)}{\sqrt{2}}$$
(3)
and for the W-like states,
$$|W(\gamma ,\beta )=\frac{(|001+e^{i\gamma }|010+e^{i\beta }|100)}{\sqrt{3}},$$
(4)
are
$$𝒲_{GHZ(\varphi )}=\frac{1}{2}|GHZ(\varphi )GHZ(\varphi )|$$
(5)
and
$$𝒲_{W(\gamma ,\beta )}=\frac{2}{3}|W(\gamma ,\beta )W(\gamma ,\beta )|,$$
(6)
respectively. Motivated by the existence of the two classes of genuine three qubit entanglement, some authors have used the GHZ-witness and the W-witness as a test of existence of genuine tripartite entanglement. We call this test the GHZ/W-criterion and, as we shall see, this criterion is incomplete in the sense that there are genuine entangled states which are not detected by it.
In the context of generalizing the Schmidt decomposition, Acín et al. showed that all three-qubit pure states can be written in the following way :
$$|\psi =\lambda _0|000+\lambda _1e^{i\alpha }|001+\lambda _2|010+\lambda _3|100+\lambda _4|111,$$
(7)
where $`0\lambda _i`$, $`0\alpha \pi `$, and $`_i\lambda _i^2=1`$, by appropriate choices of local basis. Let us see what are the conditions for the GHZ$`(\varphi )`$-witness and the W$`(\gamma ,\beta )`$-witness indicating entanglement in a general pure state.
GHZ$`(\varphi )`$-witness:
$$\mathrm{Tr}(𝒲_{GHZ(\varphi )}|\psi \psi |)=\frac{1}{2}\frac{1}{2}(\lambda _0^2+\lambda _4^2+2\lambda _0\lambda _4\mathrm{cos}(\varphi \alpha ))<0.$$
(8)
As $`\lambda _0,\lambda _40`$, it is sufficient to our aim to consider $`\varphi =\alpha `$. Thus,
$$(\lambda _0+\lambda _4)^2>1,$$
(9)
must hold for the entanglement of $`|\psi `$ to be detected by GHZ$`(\varphi )`$-witness.
W$`(\gamma ,\beta )`$-witness:
$`\mathrm{Tr}(𝒲_{W(\gamma ,\beta )}|\psi \psi |)`$ $`=`$ $`{\displaystyle \frac{2}{3}}{\displaystyle \frac{1}{3}}\{\lambda _1^2+\lambda _2^2+\lambda _3^2+2\lambda _1\lambda _2\mathrm{cos}(\gamma \varphi )+`$ (10)
$`2\lambda _1\lambda _3\mathrm{cos}(\beta \varphi )+2\lambda _2\lambda _3\mathrm{cos}(\gamma \beta )\}<0.`$
Again, we consider the extremal case $`\gamma =\varphi =\beta `$. Thus,
$$(\lambda _1+\lambda _2+\lambda _3)^2>2.$$
(11)
Within these conditions, it is easy to see that the state
$$|\xi =\frac{1}{\sqrt{5}}(|000+|001+|010+|100+|111)$$
(12)
is neither witnessed by W$`(\gamma ,\beta )`$-witness nor by GHZ$`(\varphi )`$-witness. Note that $`|\xi `$ has genuine-tripartite entanglement, i.e.: it cannot be written as a biseparable state. In fact, $`|\xi `$ is a particular case of a whole family of pure states which are not witnessed by $`𝒲_{GHZ(\varphi )}`$ and $`𝒲_{W(\gamma ,\beta )}`$. The states
$$|\xi ^{}=a|GHZ(\varphi )+b|W(\gamma ,\beta ),$$
(13)
with $`|a|^2+|b|^2=1`$, reach
$$𝒲_{GHZ(\varphi )}=\frac{1}{2}|a|^2,$$
(14)
and
$$𝒲_{W(\gamma ,\beta )}=\frac{2}{3}|b|^2.$$
(15)
It means that $`|\xi ^{}`$ is witnessed if $`|a|^2>\frac{1}{2}`$ or $`|b|^2>\frac{2}{3}`$. However these conditions exclude all states which satisfy $`\frac{1}{3}|a|^2\frac{1}{2}`$ (see figure 2).
Furthermore statistic mixtures of unwitnessed pure states are also unwitnessed. To prove it we can take a look at the state:
$$\rho =\underset{i}{}p_i|\xi _i^{}\xi _i^{}|,$$
(16)
where the states $`|\xi _i^{}`$ are not witnessed by the GHZ/W-criterion. Applying the criterion to $`\rho `$ we have:
$$\mathrm{Tr}(𝒲_{GHZ(\varphi )}\rho )=\underset{i}{}p_i\mathrm{Tr}(𝒲_{GHZ(\varphi )}|\xi _i^{}\xi _i^{}|)0$$
(17)
and
$$\mathrm{Tr}(𝒲_{W(\gamma ,\beta )}\rho )=\underset{i}{}p_i\mathrm{Tr}(𝒲_{W(\gamma ,\beta )}|\xi _i^{}\xi _i^{}|)0,$$
(18)
where we have used the linearity of trace and that $`p_i0`$, $`\mathrm{Tr}(𝒲_{GHZ(\varphi )}|\xi _i^{}\xi _i^{}|)0`$, and $`\mathrm{Tr}(𝒲_{W(\gamma ,\beta )}|\xi _i^{}\xi _i^{}|)0`$. For sure, it is harder to guarantee which such combinations remain tripartite genuine entangled, however, continuity is enough to guarantee that some of them do remain.
## 3 Conclusion and Speculations
In conclusion, it was shown that testing the GHZ$`(\varphi )`$-witnesses and the W$`(\gamma ,\beta )`$-witnesses is not a sufficient condition to attest if a general state is a genuine-entangled state. We have presented a family of pure states which fools the GHZ/W-criterion. For sure, this must not be a great surprise, since entanglement witness are taylored to indicate entanglement, not to deny it. We believe that our work can avoid mistakes on the detection of three-qubit entanglement in future works.
An interesting question raised by this incomplete criterion is to determine whether there is a finite set of witness operators that can determine if any state is a genuine-entangled state. Maybe it could be done through recent results on searching procedures for OEW . It can be viewed as a detection counterpart of the problem of determining the minimal set of entangled states that can generate all entangled states under certain operations (e.g.: LOCC) . It could help, for instance, in understanding the geometry behind the set of separable states with respect to each kind of entanglement, in a complementary way to Ref. . Another elucidating research would consist in finding what are the states witnessed by the class of witnesses reached through of different local basis of the GHZ and W states.
Acknowledgements
D. Cavalcanti acknowledges financial support of the Brazilian agency CNPq. The authors thank Fernando Brandão for stimulating discussions on the theme.
References
|
warning/0506/hep-th0506118.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
The AdS/CFT correspondence provides a non-perturbative definition of string theory in asymptotically AdS space . In principle all bulk observables are encoded in correlation functions of local operators in the CFT. In practice, however, many of the quantum gravity questions we would like to address are not simply related to local boundary correlators. These questions include: how does a quasi-local bulk spacetime emerge from the CFT? How does the region behind a horizon get encoded in the CFT? What is the CFT description (or perhaps resolution) of a black hole singularity?
In this paper we develop a set of tools for recovering local bulk physics from the CFT. We use the Lorentzian AdS/CFT correspondence developed in . The basic idea is to express local operators in the bulk in terms of non-local operators on the boundary. We work in the leading semiclassical approximation – meaning both large $`N`$ and large ’t Hooft coupling – and consider free scalar fields in AdS. The fields are taken to have normalizable fall-off near the boundary of AdS.
$$\varphi (z,x)z^\mathrm{\Delta }\varphi _0(x).$$
Here $`z`$ is a radial coordinate which vanishes at the boundary. General AdS/CFT considerations imply that the boundary behavior of the field corresponds to an operator of conformal dimension $`\mathrm{\Delta }`$ in the CFT.
$$\varphi _0(x)𝒪(x).$$
This implies a correspondence between local fields in the bulk and non-local operators in the CFT.
$$\varphi (z,x)𝑑x^{}K(x^{}|z,x)𝒪(x^{}).$$
We will refer to the kernel $`K(x^{}|z,x)`$ as a smearing function. Bulk-to-bulk correlation functions, for example, are then equal to correlation functions of the corresponding non-local operators in the dual CFT
$$\varphi (z_1,x_1)\varphi (z_2,x_2)=𝑑x_1^{}𝑑x_2^{}K(x_1^{}|z_1,x_1)K(x_2^{}|z_2,x_2)𝒪(x_1^{})𝒪(x_2^{}).$$
In this paper we construct smearing functions and show how certain aspects of bulk physics are encoded by these non-local operators. We will use AdS<sub>2</sub> as our main example, although many of the results presented here generalize to higher dimensions . Smearing functions were discussed in , and smearing functions in AdS<sub>5</sub>, as well as in some non-conformal variants, have been computed by Bena . An algebraic formulation of the correspondence between local bulk fields and non-local boundary observables was developed in . Other studies of bulk locality and causality include .
To avoid any possible confusion, we note that Witten (see also ) introduced a bulk-to-boundary propagator whose Lorentzian continuation can be used to represent bulk fields having a prescribed non-normalizable behavior near the boundary . Such bulk fields are dual to sources that deform the CFT action. Our approach is quite different: we work directly with the undeformed CFT, and introduce non-local operators that are dual to normalizable fluctuations in the bulk of AdS.<sup>1</sup><sup>1</sup>1In Witten’s approach one can relate expectation values of local operators in the CFT to the normalizable boundary behavior of fields in AdS .
An outline of this paper is as follows. In section 2 we construct smearing functions in AdS<sub>2</sub> in global coordinates. In section 3 we construct smearing functions in Poincaré coordinates and show how the Poincaré horizon appears in the CFT. In section 4 we discuss the way in which bulk-to-bulk correlators are recovered from the CFT and show how coincident singularities arise. In section 5 we work in Rindler coordinates and discuss the appearance of black hole horizons. In section 6 we extend the picture to general AdS black holes. We conclude in section 7.
## 2 Global smearing functions
### 2.1 AdS<sub>2</sub> generalities
We begin by reviewing a few standard results; for more details see or appendix A. In global coordinates the AdS<sub>2</sub> metric is
$$ds^2=\frac{R^2}{\mathrm{cos}^2\rho }\left(d\tau ^2+d\rho ^2\right)$$
(1)
where $`R`$ is the radius of curvature and $`\mathrm{}<\tau <\mathrm{}`$, $`\pi /2\rho <\pi /2`$. It is convenient to introduce a distance function
$$\sigma (\tau ,\rho |\tau ^{},\rho ^{})=\frac{\mathrm{cos}(\tau \tau ^{})\mathrm{sin}\rho \mathrm{sin}\rho ^{}}{\mathrm{cos}\rho \mathrm{cos}\rho ^{}}$$
(2)
which is invariant under AdS isometries. Points that can be connected by a geodesic necessarily lie in the unit cell $`\pi <\tau \tau ^{}<\pi `$. For such points
$$\sigma =\{\begin{array}{cc}\mathrm{cos}(s/R)\hfill & \text{timelike (}s\text{ = geodesic proper time)}\hfill \\ 1\hfill & \text{null}\hfill \\ \mathrm{cosh}(d/R)\hfill & \text{spacelike (}d\text{ = geodesic proper distance) .}\hfill \end{array}$$
Points in the unit cell with $`\sigma <1`$ are timelike separated but are not connected by a geodesic. A free scalar field of mass $`m`$ can be expanded in a complete set of normalizable modes<sup>2</sup><sup>2</sup>2For certain masses there are inequivalent sets of normalizable modes, a complication analyzed in which we will not consider further here.
$$\varphi (\tau ,\rho )=\underset{n=0}{\overset{\mathrm{}}{}}a_ne^{i\omega _n\tau }\mathrm{cos}^\mathrm{\Delta }\rho C_n^\mathrm{\Delta }(\mathrm{sin}\rho )+\mathrm{h}.\mathrm{c}.$$
(3)
where $`\omega _n=n+\mathrm{\Delta }`$, $`\mathrm{\Delta }=\frac{1}{2}+\sqrt{\frac{1}{4}+m^2R^2}`$ is the conformal dimension of the corresponding operator, and $`C_n^\mathrm{\Delta }(x)`$ is a Gegenbauer polynomial. We have not bothered normalizing the modes.
The field vanishes at the boundary of AdS. In global coordinates we define the right boundary value of the field by
$$\varphi _0^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},R}(\tau )=\underset{\rho \pi /2}{lim}\frac{\varphi (\tau ,\rho )}{\mathrm{cos}^\mathrm{\Delta }\rho }.$$
(4)
Similarly the left boundary value is
$$\varphi _0^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},L}(\tau )=\underset{\rho \pi /2}{lim}\frac{\varphi (\tau ,\rho )}{\mathrm{cos}^\mathrm{\Delta }\rho }.$$
(5)
Some special simplifications occur when $`\mathrm{\Delta }`$ is a positive integer. First of all, in this case the field is single-valued on the AdS<sub>2</sub> hyperboloid (meaning that we can identify $`\tau \tau +2\pi `$). Also we define the antipodal map on AdS<sub>2</sub>
$$A:(\tau ,\rho )(\tau +\pi ,\rho ).$$
(6)
Note that $`\sigma (x|Ax^{})=\sigma (x|x^{})`$. When $`\mathrm{\Delta }`$ is a positive integer we have
$$\varphi (Ax)=(1)^\mathrm{\Delta }\varphi (x)$$
in which case the boundary values are related by
$$\varphi _0^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},L}(\tau )=(1)^\mathrm{\Delta }\varphi _0^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},R}(\tau +\pi ).$$
(7)
### 2.2 Green’s function approach
In this subsection we construct smearing functions for AdS<sub>2</sub> in global coordinates starting from a suitable Green’s function.
The Green’s function should satisfy
$$\left(\mathrm{}m^2\right)G(x|x^{})=\frac{1}{\sqrt{g}}\delta ^2(xx^{}),$$
(8)
where $`\delta ^2(xx^{})`$ is defined on the universal cover of AdS, $`\pi /2\rho <\pi /2`$, $`\mathrm{}<\tau <\mathrm{}`$. We want a smearing function that is non-zero only at spacelike separation, so we make the ansatz
$$G(x|x^{})=f\left(\sigma (x|x^{})\right)\theta \left((\rho \rho ^{})^2(\tau \tau ^{})^2\right).$$
Here $`\sigma `$ is the AdS invariant distance defined in (2). Due to the step function $`G`$ is non-zero only at spacelike separation. By direct substitution one can check that (8) is satisfied provided that $`f(\sigma )`$ satisfies the homogeneous AdS-invariant wave equation
$$(\sigma ^21)f^{\prime \prime }(\sigma )+2\sigma f^{}(\sigma )\mathrm{\Delta }(\mathrm{\Delta }1)f(\sigma )=0$$
(9)
with the boundary condition $`f(1)=1/4`$.<sup>3</sup><sup>3</sup>3To see this use the AdS invariance to set $`\tau ^{}=\rho ^{}=0`$, work in light-front coordinates $`x^\pm =\tau \pm \rho `$, and write the step function as $`\theta (x^+)\theta (x^{})+\theta (x^{})\theta (x^+)`$. The solution
$$G(x|x^{})=\frac{1}{4}P_{\mathrm{\Delta }1}(\sigma )\theta \left((\rho \rho ^{})^2(\tau \tau ^{})^2\right)$$
is given by a Legendre function. It is worth emphasizing some curious properties of this Green’s function. First of all, by construction it is non-zero only at spacelike separation. It is finite (but discontinuous) on the light cone, with $`G1/4`$ as the light cone is approached from a spacelike direction. However it is non-normalizable near the boundary of AdS, with
$$G(x|x^{})\frac{\mathrm{\Gamma }(2\mathrm{\Delta }1)}{2^{\mathrm{\Delta }+1}\mathrm{\Gamma }(\mathrm{\Delta })^2}\sigma ^{\mathrm{\Delta }1}$$
at large spacelike separation. In AdS<sub>2</sub> we can simplify the discussion by working with a Green’s function that is non-zero only in the right-hand part of the light cone.<sup>4</sup><sup>4</sup>4We could have chosen the left part of the light cone. In either case, this step is not possible in higher dimensions. With similar arguments it is easy to see that
$$G(x|x^{})=\frac{1}{2}P_{\mathrm{\Delta }1}(\sigma )\theta (\rho \rho ^{})\theta (\rho \rho ^{}|\tau \tau ^{}|)$$
(10)
is a suitable Green’s function.
Having constructed a Green’s function that is non-zero only in the right light cone we can make use of Green’s identity
$$\varphi (x^{})=_{\mathrm{}}^{\mathrm{}}𝑑\tau \sqrt{g}\left[\varphi (\tau ,\rho )^\rho G(\tau ,\rho |x^{})G(\tau ,\rho |x^{})^\rho \varphi (\tau ,\rho )\right]|_{\rho =\rho _0}.$$
We are interested in sending the regulator $`\rho _0\pi /2`$. In this limit only the leading behavior of both the field and the Green’s function contributes, and we have
$$\varphi (x^{})=_{\mathrm{}}^{\mathrm{}}𝑑\tau K(\tau |x^{})\varphi _0^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},R}(\tau )$$
(11)
where the smearing function can be variously expressed as
$`K(\tau |\tau ^{},\rho ^{})`$ $`=`$ $`(2\mathrm{\Delta }1)\underset{\rho \pi /2}{lim}\mathrm{cos}^{\mathrm{\Delta }1}\rho G(\tau ,\rho |\tau ^{},\rho ^{})`$ (12)
$`=`$ $`{\displaystyle \frac{2^{\mathrm{\Delta }1}\mathrm{\Gamma }(\mathrm{\Delta }+1/2)}{\sqrt{\pi }\mathrm{\Gamma }(\mathrm{\Delta })}}\underset{\rho \pi /2}{lim}(\sigma \mathrm{cos}\rho )^{\mathrm{\Delta }1}\theta (\rho \rho ^{}|\tau \tau ^{}|)`$
$`=`$ $`{\displaystyle \frac{2^{\mathrm{\Delta }1}\mathrm{\Gamma }(\mathrm{\Delta }+1/2)}{\sqrt{\pi }\mathrm{\Gamma }(\mathrm{\Delta })}}\left({\displaystyle \frac{\mathrm{cos}(\tau \tau ^{})\mathrm{sin}\rho ^{}}{\mathrm{cos}\rho ^{}}}\right)^{\mathrm{\Delta }1}\theta \left({\displaystyle \frac{\pi }{2}}\rho ^{}|\tau \tau ^{}|\right).`$
(14)
These smearing functions have several important properties.
* The smearing function has compact support on the boundary of AdS: $`K`$ is non-zero on the boundary only within the right lightcone of the point $`(\tau ^{},\rho ^{})`$. Of course we could have chosen to construct a smearing function that was non-zero only in the left lightcone. Note that a local bulk operator near the left boundary could be described with a highly localized smearing function on the left boundary, or with a delocalized smearing function on the right boundary.
* The whole set-up is AdS covariant, since $`\varphi _0^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},R}`$ transforms as a primary field with dimension $`\mathrm{\Delta }`$ under conformal transformations. This is clear when $`K`$ is written in the form (2.2): the factor $`\mathrm{cos}^{\mathrm{\Delta }1}\rho `$ appearing in that expression cancels the conformal weight of the field together with the conformal weight of the measure $`𝑑\tau `$.
* As can be seen explicitly in (14), the smearing function has a finite limit as the regulator is removed, $`\rho _0\pi /2`$.
Note that these properties all follow from the fact that we began with a Green’s function that is non-normalizable near the boundary and non-zero only at spacelike separation. We have plotted the $`\mathrm{\Delta }=3`$ smearing function in Fig. 1. As a simple case, note that for a massless field in AdS<sub>2</sub> one has $`\mathrm{\Delta }=1`$ and the general expression reduces to
$$K=\frac{1}{2}\theta \left(\frac{\pi }{2}\rho ^{}|\tau \tau ^{}|\right).$$
That is, a massless bulk field is expressed in terms of its boundary value by
$$\varphi (\tau ^{},\rho ^{})=\frac{1}{2}_{\tau ^{}(\pi /2\rho ^{})}^{\tau ^{}+(\pi /2\rho ^{})}𝑑\tau \varphi _0^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},R}(\tau ).$$
### 2.3 Mode sum approach
In this subsection we take a different point of view and construct global smearing functions from a mode sum. We begin with integer conformal dimension then generalize.
Let us first suppose that $`\mathrm{\Delta }`$ is a positive integer. Then given an on-shell bulk field with mode expansion (3) we can reconstruct the bulk field from its right boundary value using
$$a_n=\frac{1}{C_n^\mathrm{\Delta }(1)}\frac{d\tau }{2\pi }e^{i\omega _n\tau }\varphi _0^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},R}(\tau ).$$
(15)
The integral is over any $`2\pi `$ interval on the boundary. Plugging this back into the bulk mode expansion, we can write (as an operator identity!)
$$\varphi (\tau ^{},\rho ^{})=𝑑\tau K(\tau |\tau ^{},\rho ^{})\varphi _0^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},R}(\tau )$$
(16)
where the smearing function
$$K(\tau |\tau ^{},\rho ^{})=\frac{1}{2\pi }e^{i\mathrm{\Delta }(\tau \tau ^{})}\mathrm{cos}^\mathrm{\Delta }\rho ^{}\underset{n=0}{\overset{\mathrm{}}{}}e^{in(\tau \tau ^{}+iϵ)}\frac{C_n^\mathrm{\Delta }(\mathrm{sin}\rho ^{})}{C_n^\mathrm{\Delta }(1)}+\mathrm{c}.\mathrm{c}.$$
(17)
We have inserted an $`iϵ`$ to keep the mode sum convergent. Note that $`K`$ is periodic in $`\tau `$, with the same periodicity as the underlying modes, unlike the Green’s function (10). To make contact with the results of the previous section we will eventually choose the range of integration in (16) to be $`\pi <\tau \tau ^{}<\pi `$.
It is important to note that the smearing functions are not unique. For example we could equally well have constructed a smearing function on the left boundary. More importantly, from (3) note that $`\varphi _0^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},R}`$ does not have Fourier components with frequencies in the range $`\mathrm{\Delta }+1,\mathrm{},\mathrm{\Delta }1`$, so we are free to drop any Fourier components of $`K`$ with frequencies in this range.
To evaluate $`K`$ we use the integral representation
$$\frac{C_n^\mathrm{\Delta }(x)}{C_n^\mathrm{\Delta }(1)}=\frac{\mathrm{\Gamma }(\mathrm{\Delta }+1/2)}{\sqrt{\pi }\mathrm{\Gamma }(\mathrm{\Delta })}_0^\pi 𝑑\theta \mathrm{sin}^{2\mathrm{\Delta }1}\theta \left(x+\sqrt{x^21}\mathrm{cos}\theta \right)^n.$$
Performing the sum on $`n`$ gives
$$K(\tau |0,\rho ^{})=\frac{1}{2\pi }e^{i\mathrm{\Delta }\tau }\mathrm{cos}^\mathrm{\Delta }\rho ^{}\frac{\mathrm{\Gamma }(\mathrm{\Delta }+1/2)}{\sqrt{\pi }\mathrm{\Gamma }(\mathrm{\Delta })}_0^\pi 𝑑\theta \frac{\mathrm{sin}^{2\mathrm{\Delta }1}\theta }{1e^{i(\tau +iϵ)}(\mathrm{sin}\rho ^{}+i\mathrm{cos}\rho ^{}\mathrm{cos}\theta )}+\mathrm{c}.\mathrm{c}..$$
The integral is a polynomial in $`e^{i\tau }`$ plus a logarithm . The polynomial only involves Fourier modes which do not appear in $`\varphi _0^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙}}`$, so we may drop it leaving
$`K(\tau |0,\rho ^{})`$ $``$ $`{\displaystyle \frac{2^{\mathrm{\Delta }1}\mathrm{\Gamma }(\mathrm{\Delta }+1/2)}{\sqrt{\pi }\mathrm{\Gamma }(\mathrm{\Delta })}}\left({\displaystyle \frac{\mathrm{cos}\tau \mathrm{sin}\rho ^{}}{\mathrm{cos}\rho ^{}}}\right)^{\mathrm{\Delta }1}`$
$`\times {\displaystyle \frac{i}{2\pi }}\mathrm{log}\left({\displaystyle \frac{1ie^{i(\tau \rho ^{}+iϵ)}}{1+ie^{i(\tau \rho ^{}iϵ)}}}{\displaystyle \frac{1ie^{i(\tau +\rho ^{}iϵ)}}{1+ie^{i(\tau +\rho ^{}+iϵ)}}}\right).`$
Here $``$ means up to terms whose time dependence is such that they vanish when integrated against $`\varphi _0^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙}}`$. At this point it helps to note that $`f(x)=i\mathrm{log}\frac{1+e^{i(x+iϵ)}}{1+e^{i(xiϵ)}}`$ is a sawtooth function, $`f(x)=x`$ for $`\pi <x<\pi `$ and $`f(x+2\pi )=f(x)`$. Again some Fourier modes do not contribute – only the discontinuities of the sawtooth function matter – and one is left with
$$K(\tau |\tau ^{},\rho ^{})\frac{2^{\mathrm{\Delta }1}\mathrm{\Gamma }(\mathrm{\Delta }+1/2)}{\sqrt{\pi }\mathrm{\Gamma }(\mathrm{\Delta })}\left(\frac{\mathrm{cos}(\tau \tau ^{})\mathrm{sin}\rho ^{}}{\mathrm{cos}\rho ^{}}\right)^{\mathrm{\Delta }1}\theta \left(\mathrm{cos}(\tau \tau ^{})\mathrm{cos}\left(\frac{\pi }{2}\rho ^{}\right)\right).$$
(18)
Choosing the range of integration in (16) to be $`\pi <\tau \tau ^{}<\pi `$, the step function in (18) reduces to the step function in (14), so this expression reproduces the result we obtained in the previous subsection starting from a Green’s function.<sup>5</sup><sup>5</sup>5The smearing function defined in (17) does not transform nicely under AdS isometries. One can see this, for example, by doing the sum explicitly for a massless field. Fortunately the non-uniqueness of the smearing function allowed us to construct an AdS covariant expression.
What if $`\mathrm{\Delta }`$ is not an integer? Recall we are working on the universal cover of AdS where $`\mathrm{}<\tau <\mathrm{}`$. In this case the mode functions are no longer periodic. This means (15) is no longer valid, since for general $`\mathrm{\Delta }`$ the positive and negative frequency modes are not orthogonal on the interval $`\pi <\tau <\pi `$. The trick is to first decompose the field into positive and negative frequency pieces, $`\varphi (\tau ,\rho )=\varphi _+(\tau ,\rho )+\varphi _{}(\tau ,\rho )`$ where
$$\varphi _+(\tau ,\rho )=\underset{n=0}{\overset{\mathrm{}}{}}a_ne^{i\omega _n\tau }\mathrm{cos}^\mathrm{\Delta }\rho C_n^\mathrm{\Delta }(\mathrm{sin}\rho )$$
(19)
and $`\varphi _{}=\varphi _+^{}`$. We can recover $`\varphi _\pm `$ from their boundary values, integrating over only the range $`\pi \tau \tau ^{}<\pi `$, via $`\varphi _\pm (\tau ^{},\rho ^{})=𝑑\tau K_\pm (\tau |\tau ^{},\rho ^{})\varphi _{0\pm }^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},R}(\tau )`$ where
$$K_\pm (\tau |\tau ^{},\rho ^{})=\frac{1}{2\pi }\mathrm{cos}^\mathrm{\Delta }\rho ^{}\underset{n=0}{\overset{\mathrm{}}{}}e^{\pm i\omega _n(\tau \tau ^{}\pm iϵ)}\frac{C_n^\mathrm{\Delta }(\mathrm{sin}\rho ^{})}{C_n^\mathrm{\Delta }(1)}.$$
(20)
These positive and negative frequency smearing functions are highly non-unique, since for example we can add Fourier modes to $`K_+`$ that $`e^{i(\mathrm{\Delta }1)(\tau \tau ^{})}`$, $`e^{i(\mathrm{\Delta }2)(\tau \tau ^{})}`$, $`\mathrm{}`$. By making use of this freedom we can put $`K_\pm `$ into the form of an image sum,
$$K_\pm \underset{k=\mathrm{}}{\overset{\mathrm{}}{}}e^{\pm i2\pi k\mathrm{\Delta }}K(\tau |\tau ^{}+2\pi k,\rho ^{})$$
(21)
where for any $`\mathrm{\Delta }`$
$$K(\tau |\tau ^{},\rho ^{})\frac{2^{\mathrm{\Delta }1}\mathrm{\Gamma }(\mathrm{\Delta }+1/2)}{\sqrt{\pi }\mathrm{\Gamma }(\mathrm{\Delta })}\left(\frac{\mathrm{cos}(\tau \tau ^{})\mathrm{sin}\rho ^{}}{\mathrm{cos}\rho ^{}}\right)^{\mathrm{\Delta }1}\theta \left(\frac{\pi }{2}\rho ^{}|\tau \tau ^{}|\right)$$
(22)
is defined over the range $`\mathrm{}<\tau \tau ^{}<\mathrm{}`$. One can verify (21) by doing an inverse Fourier transform using a complete set of skew-periodic modes $`e^{\pm i(n+\mathrm{\Delta })(\tau \tau ^{})}`$, $`n`$, and showing that the coefficients of the $`n0`$ modes appear in (20). On the restricted interval $`\pi <\tau \tau ^{}<\pi `$ only the $`k=0`$ term contributes, so the smearing functions all agree and
$$\varphi =_{\tau ^{}\pi }^{\tau ^{}+\pi }𝑑\tau \left(K_+\varphi _{0+}^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},R}+K_{}\varphi _0^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},R}\right)=_{\tau ^{}\pi }^{\tau ^{}+\pi }𝑑\tau K\varphi _0^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},R}$$
reproduces our previous Green’s function expression.
## 3 Poincaré coordinates
### 3.1 Smearing functions
It is worth asking how these non-local operators look in different coordinate systems. For example one can introduce Poincaré coordinates
$`Z={\displaystyle \frac{R\mathrm{cos}\rho }{\mathrm{cos}\tau +\mathrm{sin}\rho }}T={\displaystyle \frac{R\mathrm{sin}\tau }{\mathrm{cos}\tau +\mathrm{sin}\rho }}`$ (23)
$`0<Z<\mathrm{}\mathrm{}<T<\mathrm{}`$
in which
$$ds^2=\frac{R^2}{Z^2}\left(dT^2+dZ^2\right).$$
These coordinates only cover an interval $`\pi <\tau <\pi `$ of global time on the right boundary. The mode expansion reads
$$\varphi (T,Z)=_0^{\mathrm{}}𝑑\omega a_\omega e^{i\omega T}\sqrt{Z}J_\nu (\omega Z)+\mathrm{c}.\mathrm{c}.$$
where $`\nu =\mathrm{\Delta }\frac{1}{2}`$. The field vanishes as $`Z0`$. In Poincaré coordinates it is convenient to define the boundary value
$$\varphi _0^{\mathrm{𝑃𝑜𝑖𝑛𝑐𝑎𝑟𝑒}}(T)=\underset{Z0}{lim}\frac{\varphi (T,Z)}{Z^\mathrm{\Delta }}.$$
(24)
Note that this differs from the definition (4) we used in global coordinates.
It is straightforward to compute the smearing function in Poincaré coordinates directly from a mode sum. Given an on-shell bulk field we have
$$a_\omega =_{\mathrm{}}^{\mathrm{}}𝑑Te^{i\omega T}\frac{1}{2\pi \sqrt{Z}J_\nu (\omega Z)}\varphi (T,Z),$$
or taking the limit $`Z0`$
$$a_\omega =\frac{\mathrm{\Gamma }(\nu +1)}{2\pi (\omega /2)^\nu }_{\mathrm{}}^{\mathrm{}}𝑑Te^{i\omega T}\varphi _0^{\mathrm{𝑃𝑜𝑖𝑛𝑐𝑎𝑟𝑒}}(T).$$
Plugging this back into the bulk mode expansion we can write
$$\varphi (T^{},Z^{})=_{\mathrm{}}^{\mathrm{}}𝑑TK(T|T^{},Z^{})\varphi _0^{\mathrm{𝑃𝑜𝑖𝑛𝑐𝑎𝑟𝑒}}(T)$$
where the Poincaré smearing function is
$$K(T|T^{},Z^{})=\frac{1}{\pi }2^\nu \mathrm{\Gamma }(\nu +1)\sqrt{Z^{}}_0^{\mathrm{}}𝑑\omega \frac{1}{\omega ^\nu }J_\nu (\omega Z^{})\mathrm{cos}\omega (TT^{}).$$
(25)
The integral is well defined without an $`iϵ`$ prescription. It vanishes identically for $`|TT^{}|>Z^{}`$, while for $`|TT^{}|<Z^{}`$ we have
$$K=\frac{\mathrm{\Gamma }(\mathrm{\Delta }+1/2)}{\sqrt{\pi }\mathrm{\Gamma }(\mathrm{\Delta })}Z^{\mathrm{\Delta }1}F(\frac{1}{2},1\mathrm{\Delta },\frac{1}{2},\frac{(TT^{})^2}{Z^{}^2}).$$
Thus in general
$$K=\frac{\mathrm{\Gamma }(\mathrm{\Delta }+1/2)}{\sqrt{\pi }\mathrm{\Gamma }(\mathrm{\Delta })}\left(\frac{Z^{}{}_{}{}^{2}(TT^{})^2}{Z^{}}\right)^{\mathrm{\Delta }1}\theta (Z^{}|TT^{}|).$$
(26)
where we used $`F(\alpha ,\beta ,\alpha ,x)=(1x)^\beta `$. This expression is valid for any positive $`\mathrm{\Delta }`$. Note that, unlike the global mode sum (17), the Poincaré mode sum is non-zero only at spacelike separation. It is also AdS covariant since
$$K(T|T^{},Z^{})=\frac{2^{\mathrm{\Delta }1}\mathrm{\Gamma }(\mathrm{\Delta }+1/2)}{\sqrt{\pi }\mathrm{\Gamma }(\mathrm{\Delta })}\underset{Z0}{lim}\left(Z\sigma (T,Z|T^{},Z^{})\right)^{\mathrm{\Delta }1}\theta (\sigma 1)$$
(27)
where the AdS invariant distance (2) expressed in Poincaré coordinates is
$$\sigma (Z,T|Z^{},T^{})=\frac{Z^2+Z^{}{}_{}{}^{2}(TT^{})^2}{2ZZ^{}}.$$
(28)
Upon changing variables from Poincaré time to global time, this is equivalent to our previous result (2.2). To see this, one merely has to note that near the boundary
$$Z^\mathrm{\Delta }\varphi _0^{\mathrm{𝑃𝑜𝑖𝑛𝑐𝑎𝑟𝑒}}\mathrm{cos}^\mathrm{\Delta }\rho \varphi _0^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},R}$$
while the change of integration measure is
$$\frac{dT}{Z}=\frac{d\tau }{\mathrm{cos}\rho }.$$
### 3.2 Going behind the Poincaré horizon
In the previous subsection we showed how local bulk fields in the Poincaré patch are represented by smeared operators on the boundary. But what if the bulk point is outside the Poincaré patch? Can it still be represented as an operator on the boundary of the Poincaré patch? The answer turns out to be yes: we can still work in Poincaré coordinates on the boundary, but we have to use a different smearing function.
To obtain the correct smearing function our strategy is to start with the global smearing function, manipulate it so that it is non-zero only on the boundary of the Poincaré patch, and then convert to Poincaré coordinates. We first assume that $`\mathrm{\Delta }`$ is a positive integer. In this case $`\varphi _0^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},R}(\tau )`$ is periodic in the global time coordinate $`\tau `$ with period $`2\pi `$, so one can take the global smearing function (18) and translate whatever part of it has left the Poincaré patch by a multiple of $`2\pi `$ in order to get it back inside the Poincaré patch. This is illustrated in Fig. 2.
This can be expressed quite simply in terms of the invariant distance (2). Noting that $`\sigma `$ is $`2\pi `$ periodic in global time, for a general point $`P`$ (not necessarily inside the Poincaré patch) we can express the smearing function in a form that, upon changing to Poincaré coordinates, looks identical to (27):
$`\varphi (P)={\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑TK(T|P)\varphi _0^{\mathrm{𝑃𝑜𝑖𝑛𝑐𝑎𝑟𝑒}}(T)`$
$`K(T|P)={\displaystyle \frac{2^{\mathrm{\Delta }1}\mathrm{\Gamma }(\mathrm{\Delta }+1/2)}{\sqrt{\pi }\mathrm{\Gamma }(\mathrm{\Delta })}}\underset{Z0}{lim}\left(Z\sigma (T,Z|P)\right)^{\mathrm{\Delta }1}\theta (\sigma 1).`$
Note that the signal of a bulk operator approaching the future (past) Poincaré horizon is that the smearing function extends to Poincaré time $`T+\mathrm{}`$ ($`T\mathrm{}`$).
What happens if $`\mathrm{\Delta }`$ is not an integer? The trick is to note that, although the field itself is not periodic in $`\tau `$, its positive and negative frequency components (19) are periodic up to a phase:
$$\varphi _\pm (\tau +2\pi ,\rho )=e^{i2\pi \mathrm{\Delta }}\varphi _\pm (\tau ,\rho ).$$
For a bulk point $`P`$ with global coordinates $`(\tau ^{},\rho ^{})`$ we set $`\tau ^{}=\tau _0^{}+2\pi n`$ with $`\pi <\tau _0^{}<\pi `$ and $`n`$ and write
$$\varphi (\tau ^{},\rho ^{})=𝑑\tau K_{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙}}(\tau |\tau _0^{},\rho ^{})\left(e^{i2\pi n\mathrm{\Delta }}\varphi _{0+}^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},R}(\tau )+e^{i2\pi n\mathrm{\Delta }}\varphi _0^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},R}(\tau )\right).$$
(30)
Converting to Poincaré coordinates this becomes
$$\varphi (P)=_{\mathrm{}}^{\mathrm{}}𝑑TK_{\mathrm{𝑃𝑜𝑖𝑛𝑐𝑎𝑟𝑒}}(T|P_0)\left(e^{i2\pi n\mathrm{\Delta }}\varphi _{0+}^{\mathrm{𝑃𝑜𝑖𝑛𝑐𝑎𝑟𝑒}}(T)+e^{i2\pi n\mathrm{\Delta }}\varphi _0^{\mathrm{𝑃𝑜𝑖𝑛𝑐𝑎𝑟𝑒}}(T)\right).$$
(31)
Here $`K_{\mathrm{𝑃𝑜𝑖𝑛𝑐𝑎𝑟𝑒}}`$ is given in (3.2) and $`P_0`$ is the bulk point with global coordinates $`(\tau _0^{},\rho ^{})`$. In deriving this we used the fact that the global and Poincaré vacua are identical . This equivalence is discussed in more detail in appendix B.
When expressed in terms of $`\varphi _{0+}^{\mathrm{𝑃𝑜𝑖𝑛𝑐𝑎𝑟𝑒}}`$ and $`\varphi _0^{\mathrm{𝑃𝑜𝑖𝑛𝑐𝑎𝑟𝑒}}`$ the smearing function consists of one or two disconnected blobs on the boundary.<sup>6</sup><sup>6</sup>6One blob if the bulk point is inside the Poincaré patch or any of its images under $`\tau \tau +2\pi `$, two blobs otherwise. However expressing $`\varphi _{0\pm }^{\mathrm{𝑃𝑜𝑖𝑛𝑐𝑎𝑟𝑒}}`$ in terms of $`\varphi _0^{\mathrm{𝑃𝑜𝑖𝑛𝑐𝑎𝑟𝑒}}`$ is completely non-local. For example the positive frequency part of the Poincaré boundary field is
$`\varphi _{0+}^{\mathrm{𝑃𝑜𝑖𝑛𝑐𝑎𝑟𝑒}}(T^{})={\displaystyle 𝑑TP_+(T|T^{})\varphi _0^{\mathrm{𝑃𝑜𝑖𝑛𝑐𝑎𝑟𝑒}}(T)}`$ (32)
$`P_+(T|T^{})={\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{d\omega }{2\pi }}e^{i\omega (TT^{}+iϵ)}={\displaystyle \frac{i}{2\pi (TT^{}+iϵ)}}`$
For a point outside the Poincaré patch the smearing function in terms of $`\varphi _0^{\mathrm{𝑃𝑜𝑖𝑛𝑐𝑎𝑟𝑒}}`$ is non-zero everywhere. Just as for integer $`\mathrm{\Delta }`$, the signal of a bulk operator approaching the future (past) Poincaré horizon is that the smearing function extends to Poincaré time $`T+\mathrm{}`$ ($`T\mathrm{}`$).
## 4 Recovering bulk correlators
In the previous section we defined a set of smearing functions which enable us to reconstruct a normalizable bulk field from its behavior near the boundary of AdS. We obtained these smearing functions by solving a wave equation in AdS space, so our expressions are valid for any state of the field provided we are in the limit of semiclassical supergravity where back-reaction of the field on the geometry can be neglected.
Assuming the existence of a dual CFT, we can identify local operators in the bulk supergravity with non-local operators in the CFT.
$$\varphi (\tau ^{},\rho ^{})𝑑\tau K(\tau |\tau ^{},\rho ^{})𝒪(\tau ).$$
This correspondence should hold for any state of the field (equivalently, any state of the CFT) provided we are in the limit of semiclassical supergravity where back-reaction is negligible; note that in this limit the smearing functions are independent of the state. We can use the correspondence to recover bulk supergravity correlation functions from the CFT, for example
$${}_{S}{}^{}\psi |\varphi (\tau ,\rho )\varphi (\tau ^{},\rho ^{})|\psi _{S}^{}=dsds^{}K(s|\tau ,\rho )K(s^{}|\tau ^{},\rho ^{}){}_{C}{}^{}\psi |𝒪(s)𝒪(s^{})|\psi _{C}^{}.$$
(33)
Here $`|\psi _S`$ is any supergravity state and $`|\psi _C`$ is the corresponding CFT state. Although true by construction,<sup>7</sup><sup>7</sup>7This is perhaps easiest to see if one represents the smearing functions and correlators using mode sums. at first sight this is a rather surprising identity. For example in the global AdS<sub>2</sub> vacuum the bulk and boundary Wightman functions are
$`\varphi (\tau ,\rho )\varphi (\tau ^{},\rho ^{})={\displaystyle \frac{\mathrm{\Gamma }(\mathrm{\Delta })}{2^{\mathrm{\Delta }+1}\sqrt{\pi }\mathrm{\Gamma }(\mathrm{\Delta }+1/2)}}\sigma ^\mathrm{\Delta }F({\displaystyle \frac{\mathrm{\Delta }}{2}},{\displaystyle \frac{\mathrm{\Delta }+1}{2}};{\displaystyle \frac{2\mathrm{\Delta }+1}{2}};{\displaystyle \frac{1}{\sigma ^2}})`$
$`𝒪(s)𝒪(s^{})={\displaystyle \frac{(1)^\mathrm{\Delta }\mathrm{\Gamma }(\mathrm{\Delta })}{2^{2\mathrm{\Delta }+1}\sqrt{\pi }\mathrm{\Gamma }(\mathrm{\Delta }+\frac{1}{2})\mathrm{sin}^{2\mathrm{\Delta }}\left(\frac{ss^{}}{2}\right)}}.`$ (34)
When $`\sigma <1`$ (bulk points not connected by a geodesic) both the left and right hand sides of (33) are unambiguous. To continue into the regime $`\sigma >1`$ one has to check that the $`iϵ`$ prescriptions go through correctly. The bulk Wightman function is defined by a $`\tau \tau iϵ`$ prescription, and fortunately by translation invariance of $`K(s|\tau ,\rho )`$ this is equivalent to the correct $`ssiϵ`$ prescription for the boundary Wightman function.
Given (33) we are guaranteed that – in the strict semiclassical limit – two non-local boundary operators will commute whenever the bulk points are spacelike separated. Bena checked explicitly that the commutator vanishes from the boundary point of view.
### 4.1 Light-cone singularities in the bulk
Correlation functions of these non-local boundary operators diverge whenever the corresponding bulk points are coincident or light-like separated. In this section we explain how these singularities arise from the boundary point of view.
It may seem a little surprising that the correlators diverge at all, since in field theory one usually introduces smeared operators to avoid singular correlators. To see what is going on let us look at how these operators are constructed. It is simplest to work in terms of a mode sum in frequency space. From (17) or (25) one sees that the smearing is done by multiplying each mode of the boundary operator by a frequency dependent phase. These phases act as a regulator which makes the correlator with most other operators non-singular. However for a given smeared operator one can find certain other smeared operators for which the phases cancel. For two such operators the correlator has a UV divergence.
To make this explicit consider the correlator of two bulk operators in the Poincaré patch. Working in frequency space on the boundary the correlator is
$$\varphi (T,Z)\varphi (T^{},Z^{})=_{\mathrm{}}^{\mathrm{}}𝑑\omega K(\omega |T,Z)K(\omega |T^{},Z^{})𝒪(\omega )𝒪(\omega ).$$
(35)
The Poincaré smearing function in frequency space can be read off from (25),
$`K(\omega |T,Z)`$ $`=`$ $`2^\nu \mathrm{\Gamma }(\nu +1){\displaystyle \frac{\sqrt{Z}}{|\omega |^\nu }}J_\nu (|\omega |Z)e^{i\omega T}`$
$``$ $`{\displaystyle \frac{1}{|\omega |^{\nu +1/2}}}\mathrm{cos}(|\omega |Z+\mathrm{const}.)e^{i\omega T}`$
at large $`|\omega |`$. For an operator of dimension $`\mathrm{\Delta }`$ the boundary correlator in frequency space behaves as $`𝒪(\omega )𝒪(\omega )|\omega |^{2\nu }`$. For generic bulk points the integrand in (35) has oscillating phases which regulate the behavior at large $`\omega `$. But whenever $`(T,Z)`$ and $`(T^{},Z^{})`$ either coincide or are light-like separated the phases cancel and the integral diverges logarithmically in the UV.
It is important to note that, even in the center of AdS, the lightcone singularities of the bulk theory arise from the UV behavior of the boundary theory. This means, for example, that any attempt to put a UV cutoff on the boundary theory will modify locality everywhere in the bulk.
## 5 AdS<sub>2</sub> black holes
### 5.1 Smearing functions
One can also introduce Rindler coordinates on AdS<sub>2</sub>
$`{\displaystyle \frac{r}{R\sqrt{M}}}={\displaystyle \frac{\mathrm{cos}\tau }{\mathrm{cos}\rho }}\mathrm{tanh}{\displaystyle \frac{t\sqrt{M}}{R}}={\displaystyle \frac{\mathrm{sin}\tau }{\mathrm{sin}\rho }}`$
$`R\sqrt{M}<r<\mathrm{}\mathrm{}<t<\mathrm{}`$ (36)
$`ds^2=\left({\displaystyle \frac{r^2}{R^2}}M\right)dt^2+\left({\displaystyle \frac{r^2}{R^2}}M\right)^1dr^2.`$
The metric looks like a black hole,<sup>8</sup><sup>8</sup>8In fact it is the dimensional reduction of a BTZ black hole . but the “mass” $`M`$ is just an arbitrary dimensionless parameter which enters through the change of coordinates. These coordinates only cover an interval $`\pi /2<\tau <\pi /2`$ of global time on the right boundary, but they can be extended in the usual way to cover an identical interval on the left boundary.
What do smearing functions look like in Rindler coordinates? This depends on whether the bulk point is inside or outside the Rindler horizon. For a bulk point in the right Rindler wedge there is no difficulty: we merely have to transform the global smearing function function (12) to Rindler coordinates. This gives a Rindler smearing function which is non-zero only at spacelike separation on the right boundary. Likewise for a bulk point in the left Rindler wedge we can construct a smearing function that is non-zero only at spacelike separation on the left boundary. But what if the bulk point is inside the horizon? Then we need to modify the smearing function. Just as in Poincaré coordinates, our strategy will be to start with the global smearing function, manipulate it so that it is non-zero only on the boundary of the Rindler patch, and then convert to Rindler coordinates.
The analysis is simplest when $`\mathrm{\Delta }`$ is an integer. Then we can use (7) to take those parts of the global smearing function that have left the boundary of the Rindler patch and bring them back inside. However one obtains a smearing function with support on both the left and right boundaries of AdS. This is illustrated in Fig. 3.
This can be expressed quite simply in terms of the invariant distance $`\sigma `$. Recall that $`\sigma `$ is $`2\pi `$ periodic in global time, while under the antipodal map
$$\sigma (x|Ax^{})=\sigma (x|x^{})\varphi (Ax)=(1)^\mathrm{\Delta }\varphi (x).$$
By starting with the global smearing function (12), decomposing it into pieces that are inside / outside the Rindler patch, and moving the outside part to the other boundary we have
$`\varphi (P)={\displaystyle _{\pi /2}^{\pi /2}}𝑑\tau \left(K_{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙}}^R(\tau |P)\varphi _0^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},R}(\tau )+(1)^\mathrm{\Delta }K_{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙}}^L(\tau |P)\varphi _0^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},L}(\tau )\right)`$
$`K_{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙}}^R(\tau |P)={\displaystyle \frac{2^{\mathrm{\Delta }1}\mathrm{\Gamma }(\mathrm{\Delta }+1/2)}{\sqrt{\pi }\mathrm{\Gamma }(\mathrm{\Delta })}}\underset{\rho \pi /2}{lim}\left(\sigma (\tau ,\rho |P)\mathrm{cos}\rho \right)^{\mathrm{\Delta }1}\theta (\sigma 1)`$ (37)
$`K_{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙}}^L(\tau |P)={\displaystyle \frac{2^{\mathrm{\Delta }1}\mathrm{\Gamma }(\mathrm{\Delta }+1/2)}{\sqrt{\pi }\mathrm{\Gamma }(\mathrm{\Delta })}}\underset{\rho \pi /2}{lim}\left(\sigma (\tau ,\rho |P)\mathrm{cos}\rho \right)^{\mathrm{\Delta }1}\theta (\sigma 1).`$
Here $`P`$ is any point inside the Rindler horizon.<sup>9</sup><sup>9</sup>9In fact this expression is valid for any $`P`$. However if $`P`$ is in the left Rindler wedge it is simpler to work with a smearing function that is non-zero only at spacelike separation on the left boundary. To express this in Rindler coordinates it is convenient to define the left and right Rindler boundary fields $`\varphi _0^{\mathrm{𝑅𝑖𝑛𝑑𝑙𝑒𝑟},L/R}(t)=lim_r\mathrm{}r^\mathrm{\Delta }\varphi (t,r)`$ where $`L`$ ($`R`$) refers to the left (right) Rindler wedge. Near the boundary
$$\frac{\varphi _0^{\mathrm{𝑅𝑖𝑛𝑑𝑙𝑒𝑟},L/R}}{r^\mathrm{\Delta }}\mathrm{cos}^\mathrm{\Delta }\rho \varphi _0^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},L/R}$$
while the change of integration measure is
$$\frac{d\tau }{\mathrm{cos}\rho }=\frac{rdt}{R^2}.$$
Putting this all together we have the final expression in Rindler coordinates
$`\varphi (P)={\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑t\left(K_{\mathrm{𝑅𝑖𝑛𝑑𝑙𝑒𝑟}}^R(t|P)\varphi _0^{\mathrm{𝑅𝑖𝑛𝑑𝑙𝑒𝑟},R}(t)+(1)^\mathrm{\Delta }K_{\mathrm{𝑅𝑖𝑛𝑑𝑙𝑒𝑟}}^L(t|P)\varphi _0^{\mathrm{𝑅𝑖𝑛𝑑𝑙𝑒𝑟},L}(t)\right)`$
$`K_{\mathrm{𝑅𝑖𝑛𝑑𝑙𝑒𝑟}}^R(t|P)={\displaystyle \frac{2^{\mathrm{\Delta }1}\mathrm{\Gamma }(\mathrm{\Delta }+1/2)}{\sqrt{\pi }\mathrm{\Gamma }(\mathrm{\Delta })R^2}}\underset{r\mathrm{}}{lim}(\sigma /r)^{\mathrm{\Delta }1}\theta (\sigma 1)`$ (38)
$`K_{\mathrm{𝑅𝑖𝑛𝑑𝑙𝑒𝑟}}^L(t|P)={\displaystyle \frac{2^{\mathrm{\Delta }1}\mathrm{\Gamma }(\mathrm{\Delta }+1/2)}{\sqrt{\pi }\mathrm{\Gamma }(\mathrm{\Delta })R^2}}\underset{r\mathrm{}}{lim}(\sigma /r)^{\mathrm{\Delta }1}\theta (\sigma 1).`$
Here $`\sigma `$ is the invariant distance from $`P`$ to the point $`(t,r)`$ on the appropriate boundary,
$$\sigma (t,r|t^{},r^{})=\frac{1}{M}\left[\frac{rr^{}}{R^2}\pm \left(\frac{r^2}{R^2}M\right)^{1/2}\left(M\frac{r^{}^2}{R^2}\right)^{1/2}\mathrm{sinh}\frac{\sqrt{M}(tt^{})}{R}\right]$$
where $`P`$ is inside the horizon with coordinates $`(t^{},r^{})`$ and the upper (lower) sign applies for $`(t,r)`$ near the right (left) boundary. Note that the smearing function on the left boundary has support only on points that are not connected to $`P`$ by a geodesic (points with $`\sigma <1`$).
What happens if $`\mathrm{\Delta }`$ is not an integer? Although we no longer have (7), a similar property holds separately for the positive and negative frequency components of the boundary field (19):
$$\varphi _{0\pm }^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},R}(\tau )=e^{\pm i\pi \mathrm{\Delta }}\varphi _{0\pm }^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},L}(\tau +\pi ).$$
(39)
We can use this to rewrite the global smearing function in a form similar to (37), where it is non-zero only on the boundary of the Rindler patch. For a bulk point $`P`$ inside the horizon with global coordinates $`(\tau ^{},\rho ^{})`$ we set $`\tau ^{}=\tau _0^{}+2\pi n`$, $`n`$ and
$`\varphi (P)={\displaystyle _{\pi /2}^{\pi /2}}d\tau [K_{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙}}^R(\tau |P)(e^{i2\pi n\mathrm{\Delta }}\varphi _{0+}^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},R}(\tau )+e^{i2\pi n\mathrm{\Delta }}\varphi _0^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},R}(\tau ))`$
$`+K_{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙}}^L(\tau |P)(e^{i\pi (2n+1)\mathrm{\Delta }}\varphi _{0+}^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},L}(\tau )+e^{i\pi (2n+1)\mathrm{\Delta }}\varphi _0^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},L}(\tau ))]`$ (40)
where $`K_{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙}}^{R,L}(\tau |P)`$ appear in (37). Expressed in terms of positive and negative frequency global fields this smearing function consists of two disconnected blobs, one on the left boundary and one on the right.<sup>10</sup><sup>10</sup>10For a bulk point in the right Rindler wedge or any of its images under $`\tau \tau +2\pi `$ one gets a single blob on the right boundary. For a bulk point in the left wedge or any of its images it is more natural to use a smearing function with a single blob on the left boundary. One can switch to Rindler coordinates, however for points inside the horizon in general one obtains a smearing function that is completely non-local in terms of $`\varphi _0^{\mathrm{𝑅𝑖𝑛𝑑𝑙𝑒𝑟},L/R}`$. This is worked out in appendix B.
### 5.2 Thermofield interpretation and black holes
We can regard the AdS<sub>2</sub> metric in Rindler coordinates as a prototype for a black hole. A standard Euclidean calculation shows that the Hawking temperature is $`1/\beta =\frac{\sqrt{M}}{2\pi R}`$. By keeping modes in both the left and right Rindler wedges, the Hartle-Hawking state can be understood as a thermofield double .<sup>11</sup><sup>11</sup>11Despite appearances, the Hartle-Hawking state is independent of $`M`$. In fact the global vacuum, the Poincaré vacuum and the Hartle-Hawking state are identical .
How does this look on the boundary? As discussed in an eternal AdS black hole is dual to two copies of the CFT in an entangled thermofield state. As we show in appendix A the thermofield Hamiltonian is
$$H_{TF}=\frac{\sqrt{M}}{R}(\widehat{S}_1\text{1}\text{ }\text{1}_2\text{1}\text{ }\text{1}_1\widehat{S}_2).$$
Here $`\widehat{S}`$ is a non-compact conformal generator on the boundary. The thermofield state is annihilated by $`H_{TF}`$, and can be formally expressed in terms of $`\widehat{S}`$ eigenstates as
$$|\psi =\frac{1}{\sqrt{Z(\beta )}}\underset{n}{}e^{\beta E_n/2}|n_1|n_2.$$
Since the global vacuum is $`SL(2,)`$ invariant, this state should also be annihilated by the global Hamiltonian $`\widehat{R}\text{1}\text{ }\text{1}+\text{1}\text{ }\text{1}\widehat{R}`$ and the Poincaré Hamiltonian $`\text{1}\text{ }\text{1}\widehat{H}`$.
What does this mean as far as constructing local bulk operators? The expressions we derived in section 5.1 are perfectly applicable. They imply that to put a bulk operator behind the horizon of the black hole we need boundary operators that act non-trivially on both copies of the Hilbert space. Related observations were made in .
Let us summarize the picture we have developed. A bulk operator outside the horizon corresponds to a non-local operator that acts on a single copy of the Hilbert space. As the bulk point approaches the future (past) horizon the smearing function extends to cover an infinite range of coordinate time on the boundary: it has support as $`t+\mathrm{}`$ ($`t\mathrm{}`$).<sup>12</sup><sup>12</sup>12A similar property held in the Poincaré patch: the signal of the Poincaré horizon was that the smearing function extended to Poincaré time $`T\pm \mathrm{}`$. To insert a bulk operator behind the horizon we need a non-local operator that acts on both copies of the Hilbert space.
What does this mean from the point of view of an observer outside the black hole, who can only interact with a single copy of the CFT? Such an observer must trace over the second copy of the CFT. If no operators are inserted behind the horizon then the trace leads to a precisely thermal density matrix that describes the black hole. But operator insertions behind the horizon act on the other copy of the CFT, and modify the resulting density matrix. In general these modifications will not have a thermal character. Thus, from the point of view of an outside observer, operator insertions behind the horizon are seen as non-thermal modifications to the black hole density matrix.
## 6 Higher dimensional black holes
In the previous section we saw that in order to describe an object inside the Rindler horizon of a two-dimensional black hole one needs a non-local operator that acts on both copies of the Hilbert space.<sup>13</sup><sup>13</sup>13Equivalently one needs an operator that acts both at real time $`t`$ and at complex time $`t+i\beta /2`$. In this section we show that a similar property holds for a general AdS black hole. Analogous arguments can be made for bulk points outside of the Poincaré patch.
A local field $`\varphi (U,V,\mathrm{\Omega })`$ anywhere in the extended Kruskal diagram can be expanded in terms of Kruskal creation and annihilation operators.
$$\varphi =\underset{i}{}f_i(U,V,\mathrm{\Omega })a_K^i+f_i^{}(U,V,\mathrm{\Omega })a_K^i$$
(41)
The Kruskal creation and annihilation operators can be expressed in terms of left and right creation and annihilation operators using Bogolubov coefficients.
$`a_K^i=\alpha _{ij}^La_L^j+\beta _{ij}^La_L^j+\alpha _{ij}^Ra_R^j+\beta _{ij}^Ra_R^j`$ (42)
$`a_K^i=\alpha _{ij}^L{}_{}{}^{}a_{L}^{j}+\beta _{ij}^L{}_{}{}^{}a_{L}^{j}+\alpha _{ij}^R{}_{}{}^{}a_{R}^{j}+\beta _{ij}^R{}_{}{}^{}a_{R}^{j}`$
Since the left and right creation and annihilation operators can be written as a Fourier transform of operators in one of the copies of the CFT,
$`a_L^j{\displaystyle 𝑑td^dxe^{i\omega _jt+ik_jx}𝒪_L(t,x)}`$ (43)
$`a_R^j{\displaystyle 𝑑td^dxe^{i\omega _jt+ik_jx}𝒪_R(t,x)}`$
we see that a local bulk field anywhere in the Kruskal diagram can be represented as a linear combination of two operators, one acting on the left and one acting on the right. If the bulk point is in either the left or the right region then its representation reduces to a single operator on the appropriate copy.
## 7 Conclusions
In this paper we related local bulk fields in AdS to non-local operators on the boundary. In global coordinates we found AdS-covariant smearing functions with support purely at spacelike separation. We then showed how to represent bulk operators that are inserted behind the Poincaré horizon, or inside the horizon of a black hole. By construction these boundary operators reproduce all bulk correlation functions; they therefore respect bulk locality. Light-cone singularities in the bulk arise from the UV behavior of the boundary theory. Although we concentrated on AdS<sub>2</sub>, similar results hold in higher dimensions .
It is curious that local operators in the interior of a black hole correspond to boundary operators that act on both copies of the thermofield double. Note that in the thermal AdS phase, such operators do not exist: any operator that acts on both copies of the CFT will at best be bilocal in the (disconnected) bulk spacetime. It is only in the black hole phase that operators which act on both copies of the CFT can be local in the bulk.
One can even give a boundary description of the fact that the black hole time coordinate switches from timelike to spacelike at the horizon. In the right Rindler wedge, or more generally whenever the Killing vector $`\frac{}{t}`$ is timelike, chronology in the bulk corresponds to chronology on the boundary: bulk operators which are inserted at the same spatial position but different values of $`t`$ correspond to smeared operators on a single boundary that are related by time translation. This holds in the left and right Rindler wedges, inside the Poincaré horizon, or even globally if one uses global time. But for bulk points inside the horizon, a Rindler time translation will move the left and right boundary operators in opposite directions. This corresponds to the fact that $`\frac{}{t}`$ is spacelike inside the horizon.
There are many open questions and directions for future work. Our construction is applicable in the limit of semiclassical supergravity; it would be extremely interesting to understand $`1/N`$ and $`\alpha ^{}`$ corrections. Non-local operators in the CFT should provide a new tool to study bulk phenomena such as black hole singularities , the interplay of holography and locality , or even non-local deformations of AdS . But at a more fundamental level, one might ask: purely from the boundary CFT point of view, what if anything would allow one to identify a particular set of non-local operators as appropriate for describing bulk spacetime?
Acknowledgements
We are grateful to Nori Iizuka and Lenny Susskind for valuable discussions. AH and DK are supported by DOE grant DE-FG02-92ER40699. The research of DL is supported in part by DOE grant DE-FE0291ER40688-Task A. DK, DL and GL are supported in part by US–Israel Bi-national Science Foundation grant #2000359.
## Appendix A AdS isometries
We review the relationship between isometries of AdS<sub>2</sub> and conformal transformations on the boundary. For a more extensive discussion see .
AdS<sub>2</sub> can be embedded as a hypersurface
$$(X^0)^2(X^1)^2+(X^2)^2=R^2$$
in $`^{2,1}`$ with a $`(+)`$ metric. Isometries of AdS<sub>2</sub> arise from Lorentz transformations of the embedding space, with generators
$$J_{\mu \nu }=\eta _{\mu \sigma }X^\sigma _\nu \eta _{\nu \sigma }X^\sigma _\mu .$$
For example the $`SL(2,)`$-invariant distance function is
$$\sigma (x|x^{})=\frac{1}{2R^2}(XX^{})_\mu (XX^{})^\nu +1$$
and the antipodal map is simply $`A:X^\mu X^\mu `$.
We would like to understand the action of these isometry generators on the boundary in different coordinate systems (global, Poincaré and Rindler). These coordinate systems are defined as follows :
* Global: $`\{\begin{array}{ccc}X^0\hfill & =& R\mathrm{sec}\rho \mathrm{cos}\tau \hfill \\ X^1\hfill & =& R\mathrm{sec}\rho \mathrm{sin}\tau \hfill \\ X^2\hfill & =& R\mathrm{tan}\rho \hfill \end{array}`$
with metric: $`ds^2=\frac{R^2}{\mathrm{cos}^2\rho }(d\tau ^2+d\rho ^2)`$,
* Poincaré: $`\{\begin{array}{ccc}X^0\hfill & =& (R^2+Z^2T^2)/2Z\hfill \\ X^1\hfill & =& RT/Z\hfill \\ X^2\hfill & =& (R^2Z^2+T^2)/2Z\hfill \end{array}`$
with metric: $`ds^2=\frac{R^2}{Z^2}(dT^2+dZ^2)`$,
* Rindler (with arbitrary dimensionless M): $`\{\begin{array}{ccc}X^0\hfill & =& r/\sqrt{M}\hfill \\ X^1\hfill & =& \sqrt{\frac{r^2}{M}R^2}\mathrm{sinh}(\sqrt{M}t/R)\hfill \\ X^2\hfill & =& \sqrt{\frac{r^2}{M}R^2}\mathrm{cosh}(\sqrt{M}t/R)\hfill \end{array}`$
with metric: $`ds^2=\left(\frac{r^2}{R^2}M\right)dt^2+\left(\frac{r^2}{R^2}M\right)^1dr^2`$.
Near the boundary ($`\rho \pi /2,z0,r\mathrm{}`$) the isometry generators approach
* Global: $`\{\begin{array}{ccc}J_{01}\hfill & =& _\tau \hfill \\ J_{02}\hfill & =& \mathrm{sin}\tau _\tau \hfill \\ J_{12}\hfill & =& \mathrm{cos}\tau _\tau \hfill \end{array}`$
* Poincaré: $`\{\begin{array}{ccc}J_{01}\hfill & =& \frac{1}{2R}(T^2+R^2)_T\hfill \\ J_{02}\hfill & =& T_T\hfill \\ J_{12}\hfill & =& \frac{1}{2R}(T^2R^2)_T\hfill \end{array}`$
* Rindler: $`\{\begin{array}{ccc}J_{01}\hfill & =& \frac{R}{\sqrt{M}}\mathrm{cosh}(\sqrt{M}t/R)_t\hfill \\ J_{02}\hfill & =& \frac{R}{\sqrt{M}}\mathrm{sinh}(\sqrt{M}t/R)_t\hfill \\ J_{12}\hfill & =& \frac{R}{\sqrt{M}}_t.\hfill \end{array}`$
Here we are keeping only the leading (divergent) behavior of the vector field near the boundary.
In Poincaré coordinates the isometries give rise to the usual representation for the conformal generators on a line,
$`\widehat{H}`$ $`=`$ $`i_T=(i/R)(J_{01}+J_{12})`$ (44)
$`\widehat{D}`$ $`=`$ $`iT_T=iJ_{02}`$
$`\widehat{K}`$ $`=`$ $`iT^2_T=iR(J_{01}J_{12}).`$
Rather than adopting $`\widehat{H}`$ as the Hamiltonian one can use a different linear combination of the conformal generators to evolve in time . This is exactly what is done when using the other two coordinate systems:
Global: $`i_\tau =\frac{1}{2}(R\widehat{H}+\frac{1}{R}\widehat{K})=\widehat{R}`$
Rindler: $`i_t=\frac{\sqrt{M}}{2R}(R\widehat{H}\frac{1}{R}\widehat{K})=\frac{\sqrt{M}}{R}\widehat{S}.`$
Here $`\widehat{R}`$ (not to be confused with the AdS radius) and $`\widehat{S}`$ are compact rotation and non-compact boost generators, respectively. One can show that when evolving in time using non-compact generators (such as $`\widehat{S}`$), one cannot cover the entire range $`\mathrm{}<T<\mathrm{}`$. This is exactly what happens in Rindler coordinates – Rindler time $`t`$ only covers half of the boundary of the Poincaré patch. When evolving with compact generators such as $`\widehat{R}`$, one does cover the entire range of $`T`$ (as global coordinates do) .
## Appendix B AdS vacua
At various points in the paper we made use of the equivalence between the global, Poincaré and Hartle-Hawking vacua . In this appendix we review these results and show how they relate different boundary fields.
We begin with the equivalence between the global and Poincaré vacua. Bulk modes that are positive frequency with respect to global time are also positive frequency with respect to Poincaré time . This implies that the two bulk vacua are equivalent. To understand the corresponding relationship between the global and Poincaré boundary fields, let us start with a global boundary field that is purely positive frequency
$$\varphi _{0+}^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},R}(\tau )=\underset{n=0}{\overset{\mathrm{}}{}}c_ne^{i\omega _n\tau }.$$
(45)
Near the boundary
$$Z^\mathrm{\Delta }\varphi _0^{\mathrm{𝑃𝑜𝑖𝑛𝑐𝑎𝑟𝑒}}\mathrm{cos}^\mathrm{\Delta }\rho \varphi _0^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},R}$$
so the corresponding Poincaré boundary field is equal to the global field times a Jacobian
$$\varphi _0^{\mathrm{𝑃𝑜𝑖𝑛𝑐𝑎𝑟𝑒}}(T)=\underset{\rho \pi /2}{lim}\left(\frac{\mathrm{cos}\rho }{Z}\right)^\mathrm{\Delta }\varphi _{0+}^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},R}(\tau )=\left(\frac{2R}{T^2+R^2}\right)^\mathrm{\Delta }\varphi _{0+}^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},R}(\tau ).$$
(46)
Here $`\tau =2\mathrm{tan}^1(T/R)=i\mathrm{log}\frac{1+iT/R}{1iT/R}`$ so
$$\varphi _0^{\mathrm{𝑃𝑜𝑖𝑛𝑐𝑎𝑟𝑒}}(T)=\left(\frac{2}{R}\right)^\mathrm{\Delta }\underset{n=0}{\overset{\mathrm{}}{}}c_n\frac{(1iT/R)^n}{(1+iT/R)^{n+2\mathrm{\Delta }}}.$$
(47)
The Poincaré boundary field is analytic in the lower half of the complex $`T`$ plane, so its Fourier transform
$$\stackrel{~}{\varphi }_0^{\mathrm{𝑃𝑜𝑖𝑛𝑐𝑎𝑟𝑒}}(\omega )=𝑑Te^{i\omega T}\varphi _0^{\mathrm{𝑃𝑜𝑖𝑛𝑐𝑎𝑟𝑒}}(T)$$
vanishes if $`\omega <0`$. Thus it follows from the equivalence of the bulk Poincaré and global vacua that a positive-frequency global boundary field corresponds to a positive-frequency Poincaré boundary field. We used this in section 3.2 to go from (30) to (31). Note that the Jacobian appearing in (46) is crucial: the global boundary field (45) by itself, when expressed in terms of Poincaré time, is not positive-frequency.
Now let us turn to the equivalence between the global and Hartle-Hawking vacua. For these purposes it is convenient to introduce null Kruskal coordinates
$$u=\mathrm{tan}\frac{\tau +\rho }{2}v=\mathrm{tan}\frac{\tau \rho }{2}$$
(48)
in which
$$ds^2=\frac{4R^2dudv}{(1+uv)^2}.$$
(49)
These coordinates cover the region shown in Fig. 4; the restriction $`\pi /2<\rho <\pi /2`$ corresponds to $`uv>1`$. For $`u<0`$ note that points with $`uv=1`$ make up the boundary of the left Rindler wedge, while for $`u>0`$ points with $`uv=1`$ make up the boundary of the right Rindler wedge.
Working in the right Rindler wedge ($`u>0`$ and $`v<0`$) mode solutions with normalizable fall-off near the right boundary of AdS are
$$\varphi (u,v)=u^{i\omega }(1+uv)^\mathrm{\Delta }F(\mathrm{\Delta },\mathrm{\Delta }i\omega ,2\mathrm{\Delta },1+uv).$$
(50)
Here $`\mathrm{}<\omega <\mathrm{}`$ is a parameter which can be understood as the frequency measured with respect to Rindler time, since under the isometry corresponding to a Rindler time translation
$$ue^\lambda uve^\lambda v$$
these modes transform by a definite phase $`\varphi e^{i\omega \lambda }\varphi `$. Note that $`\omega `$ does not correspond to the frequency measured with respect to a Kruskal time coordinate.
We want to find a set of positive-frequency Kruskal modes<sup>14</sup><sup>14</sup>14Positive frequency only in the sense that they multiply annihilation operators in the mode expansion of the field. They do not transform by a definite phase under a Kruskal time translation. for which (by construction) the Kruskal vacuum will be equivalent to the global vacuum. This depends on making the correct analytic continuation from right to left. To do this we first write the modes (50) in a more symmetric form, using a transformation formula for the hypergeometric function
$`\varphi (u,v)`$ $`=`$ $`(1+uv)^\mathrm{\Delta }[{\displaystyle \frac{\mathrm{\Gamma }(2\mathrm{\Delta })\mathrm{\Gamma }(i\omega )}{\mathrm{\Gamma }(\mathrm{\Delta })\mathrm{\Gamma }(\mathrm{\Delta }+i\omega )}}u^{i\omega }F(\mathrm{\Delta },\mathrm{\Delta }i\omega ,1i\omega ,uv)+`$ (51)
$`{\displaystyle \frac{\mathrm{\Gamma }(2\mathrm{\Delta })\mathrm{\Gamma }(i\omega )}{\mathrm{\Gamma }(\mathrm{\Delta })\mathrm{\Gamma }(\mathrm{\Delta }i\omega )}}(v)^{i\omega }F(\mathrm{\Delta },\mathrm{\Delta }+i\omega ,1+i\omega ,uv)].`$
This makes it clear that, aside from the prescribed behavior near the boundary of AdS, the modes have branch points on the horizon (at $`u=0`$ or $`v=0`$). A positive frequency Kruskal mode is defined by analytically continuing from right to left going through the lower half of the complex $`u`$ and $`v`$ planes, while a negative frequency Kruskal mode is defined by analytically continuing through the upper half $`u`$ and $`v`$ planes.<sup>15</sup><sup>15</sup>15This is the same prescription used in Rindler space or, more generally, on a bifurcate Killing horizon .
It is straightforward to obtain the corresponding analyticity properties working in terms of boundary fields. In Kruskal coordinates we define the boundary field by
$$\varphi _0^{\mathrm{𝐾𝑟𝑢𝑠𝑘𝑎𝑙}}(u)=\underset{uv1}{lim}\frac{\varphi (u,v)}{(1+uv)^\mathrm{\Delta }}.$$
(52)
The relation between Kruskal and global boundary fields is then
$$\varphi _0^{Kruskal}(u)=\{\begin{array}{cc}\underset{uv1}{lim}\frac{\mathrm{cos}^\mathrm{\Delta }\rho }{(1+uv)^\mathrm{\Delta }}\varphi _0^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},R}(\tau )\hfill & \text{for }u>0\hfill \\ \underset{uv1}{lim}\frac{\mathrm{cos}^\mathrm{\Delta }\rho }{(1+uv)^\mathrm{\Delta }}\varphi _0^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},L}(\tau )\hfill & \text{for }u<0.\hfill \end{array}$$
(53)
A positive frequency global boundary field takes the form
$$\varphi _{0+}^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},R}(\tau )=\underset{n=0}{\overset{\mathrm{}}{}}c_ne^{i\omega _n\tau },\varphi _{0+}^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},L}(\tau )=\underset{n=0}{\overset{\mathrm{}}{}}c_n(1)^ne^{i\omega _n\tau }.$$
(54)
Note that
$$\underset{\tau +\pi /2}{lim}\varphi _{0+}^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},L}(\tau )=e^{i\pi \mathrm{\Delta }}\underset{\tau \pi /2}{lim}\varphi _{0+}^{\mathrm{𝑔𝑙𝑜𝑏𝑎𝑙},R}(\tau ).$$
(55)
Inserting (54) into (53) we find that a positive frequency global boundary field maps to a Kruskal boundary field given by
$$\varphi _0^{\mathrm{𝐾𝑟𝑢𝑠𝑘𝑎𝑙}}(u)=\underset{n=0}{\overset{\mathrm{}}{}}c_n\frac{(1)^{n/2}(iu)^\mathrm{\Delta }(u+i)^n}{(ui)^{2\mathrm{\Delta }+n}}.$$
(56)
In this expression the branch cut beginning at $`u=0`$ is taken to lie in the upper half complex $`u`$ plane in order to reproduce (55). Thus a positive frequency global mode corresponds to a Kruskal mode that is analytic and bounded in the lower half of the complex $`u`$ plane. We take this analyticity condition to define a positive frequency Kruskal boundary mode; this makes the global and Kruskal vacua equivalent on the boundary (just as they were equivalent in the bulk).
With this definition, from (51) a positive-frequency Kruskal mode can be characterized by the boundary behavior
$$\varphi _{0+}^{\mathrm{𝐾𝑟𝑢𝑠𝑘𝑎𝑙}}(u)=\{\begin{array}{cc}u^{i\omega }\hfill & \text{for }u>0\hfill \\ e^{\pi \omega }(u)^{i\omega }\hfill & \text{for }u<0\hfill \end{array}$$
while a negative-frequency Kruskal mode is characterized by
$$\varphi _0^{\mathrm{𝐾𝑟𝑢𝑠𝑘𝑎𝑙}}(u)=\{\begin{array}{cc}u^{i\omega }\hfill & \text{for }u>0\hfill \\ e^{+\pi \omega }(u)^{i\omega }\hfill & \text{for }u<0.\hfill \end{array}$$
At this point it is convenient to switch to Rindler coordinates. The relation is
$$u=\pm \left(\frac{r\sqrt{M}R}{r+\sqrt{M}R}\right)^{1/2}e^{\sqrt{M}t/R}v=\left(\frac{r\sqrt{M}R}{r+\sqrt{M}R}\right)^{1/2}e^{\sqrt{M}t/R}$$
where the upper (lower) sign holds in the right (left) Rindler wedge. The Jacobian for switching from Kruskal to Rindler is a constant,
$$\varphi _0^{\mathrm{𝑅𝑖𝑛𝑑𝑙𝑒𝑟}}=\underset{r\mathrm{}}{lim}(r(1+uv))^\mathrm{\Delta }\varphi _0^{\mathrm{𝐾𝑟𝑢𝑠𝑘𝑎𝑙}}=(2\sqrt{M}R)^\mathrm{\Delta }\varphi _0^{\mathrm{𝐾𝑟𝑢𝑠𝑘𝑎𝑙}},$$
so we can characterize Rindler modes which are positive or negative frequency with respect to Kruskal time as having the behavior
$$\varphi _{0\pm }^{\mathrm{𝑅𝑖𝑛𝑑𝑙𝑒𝑟},R}e^{i\omega t}\varphi _{0\pm }^{\mathrm{𝑅𝑖𝑛𝑑𝑙𝑒𝑟},L}e^{\beta \omega /2}e^{i\omega t}$$
where $`\beta =2\pi R/\sqrt{M}`$ is the inverse temperature of the black hole. Finally, this lets us define a projection operator which picks out the part of the Rindler boundary field which is positive or negative frequency with respect to Kruskal time. Let
$$\stackrel{~}{\varphi }_0^{\mathrm{𝑅𝑖𝑛𝑑𝑙𝑒𝑟},L/R}(\omega )=_{\mathrm{}}^{\mathrm{}}𝑑te^{i\omega t}\varphi _0^{\mathrm{𝑅𝑖𝑛𝑑𝑙𝑒𝑟},L/R}(t)$$
be the Fourier transform of the Rindler boundary field. Then Rindler boundary fields which are positive or negative frequency with respect to Kruskal time are given by
$`\varphi _{0\pm }^{\mathrm{𝑅𝑖𝑛𝑑𝑙𝑒𝑟},R}(t)={\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{d\omega }{2\pi }}f_\pm (\omega )e^{i\omega t}`$ (57)
$`\varphi _{0\pm }^{\mathrm{𝑅𝑖𝑛𝑑𝑙𝑒𝑟},L}(t)={\displaystyle _{\mathrm{}}^{\mathrm{}}}{\displaystyle \frac{d\omega }{2\pi }}f_\pm (\omega )e^{\beta \omega /2}e^{i\omega t}`$
where the conditions
$`f_++f_{}=\stackrel{~}{\varphi }_0^{\mathrm{𝑅𝑖𝑛𝑑𝑙𝑒𝑟},R}`$ (58)
$`f_+e^{\beta \omega /2}+f_{}e^{\beta \omega /2}=\stackrel{~}{\varphi }_0^{\mathrm{𝑅𝑖𝑛𝑑𝑙𝑒𝑟},L}`$
fix
$`f_+(\omega )={\displaystyle \frac{\stackrel{~}{\varphi }_0^{\mathrm{𝑅𝑖𝑛𝑑𝑙𝑒𝑟},R}(\omega )e^{\beta \omega /2}\stackrel{~}{\varphi }_0^{\mathrm{𝑅𝑖𝑛𝑑𝑙𝑒𝑟},L}(\omega )}{2\mathrm{sinh}\beta \omega /2}}`$ (59)
$`f_{}(\omega )={\displaystyle \frac{\stackrel{~}{\varphi }_0^{\mathrm{𝑅𝑖𝑛𝑑𝑙𝑒𝑟},L}(\omega )\stackrel{~}{\varphi }_0^{\mathrm{𝑅𝑖𝑛𝑑𝑙𝑒𝑟},R}(\omega )e^{\beta \omega /2}}{2\mathrm{sinh}\beta \omega /2}}.`$
Multiplying by the Rindler-to-global Jacobian
$$\underset{r\mathrm{}}{lim}\frac{1}{(r\mathrm{cos}\rho )^\mathrm{\Delta }}=\left(\frac{\mathrm{cosh}(\sqrt{M}t/R)}{\sqrt{M}R}\right)^\mathrm{\Delta }$$
(60)
the boundary fields (57) become positive or negative frequency with respect to global rather than Kruskal time. They can then be substituted in (40) to obtain the Rindler smearing function for a point inside the horizon.
|
warning/0506/hep-lat0506008.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
Lattice QCD is playing an important rôle in the interpretation of B-physics experiments \[?\]. However, as new physics is hiding behind the standard model, the required precision for lattice computations of B-meson transition amplitudes is becoming more and more demanding. The development of new algorithms \[?,?,?,?\] promises that light dynamical fermions at small lattice spacings, $`a`$, can be reached in the near future, which will reduce one of the most important systematic errors.
Still, the treatment of b-quarks on the lattice remains another difficult part of these computations, since it appears unlikely that lattice spacings small enough to satisfy $`a<1/m_\mathrm{b}`$ will soon come into reach. We refer the reader to reviews for an explanation of the various approaches to this problem \[?,?,?\].
A theoretically clean solution is provided by HQET. This effective theory starts from the static approximation describing the asymptotics as $`m_\mathrm{b}\mathrm{}`$. Corrections $`\mathrm{O}(1/m_\mathrm{b})`$ have to be computed by a $`1/m_\mathrm{b}`$ expansion, where the higher dimensional interaction terms in the effective Lagrangian are treated as insertions into static correlation functions. An attractive feature of this approach is that the continuum limit exists and results are independent of the regularization. The theory requires non-perturbative renormalization, but a concrete way to carry this out has been proposed and tested for a simple case \[?\].
Furthermore also just the leading order term (static) is of considerable interest, since it provides a limit of the theory, which in many cases is not expected to be far from results at the physical point. Other methods to treat heavy quarks can thus be tested by checking whether they smoothly approach the static limit $`1/m_\mathrm{b}0`$. In fact, results can also be obtained from interpolations in $`1/m_\mathrm{b}`$ between points below the b-quark mass and the static limit, see \[?\] for a precise demonstration.
As we will discuss in detail in section 2.1, the noise-to-signal ratio of static-light correlation functions grows exponentially in Euclidean time, $`\mathrm{exp}(\mu x_0)`$ and the rate $`\mu 1/a`$ diverges as one approaches the continuum limit. Good statistical precision is difficult to reach for $`x_0>1\mathrm{fm}`$ \[?,?\]. In the past, sophisticated wavefunction techniques have been used to obtain ground state properties \[?,?,?\]. The results of these attempts, obtained with the standard Eichten-Hill regularization \[?\], were not completely satisfactory in all cases \[?\]. In \[?\] we used the fact that $`\mu `$ is not universal. By changing the regularization, we were able to obtain results at large $`x_0`$, where the ground state can be isolated with good statistical and systematic precision. Here we will discuss these alternative actions in more details. Particular emphasis is put on the size of discretization errors, since their smallness should always be the first criterion for choosing an action. Let us mention right away that we will consider only actions, which share the symmetries, eqs. (2.18,2.19), with the Eichten-Hill action, since these guarantee that no $`\mathrm{O}(a)`$ terms are needed in the static action to have $`\mathrm{O}(a)`$ improvement. For all actions considered, we determine the coefficients of the improvement terms needed for the axial current with 1-loop precision. In particular we correct for a mistake in the perturbative computation of the improvement coefficients in \[?\], see the Erratum. Furthermore we determine the action-dependent parts of the renormalization of the static axial current and the b-quark mass.
Before coming to the description of the static actions, we list some preliminaries. As a probe of the theory, we will consider correlation functions defined by the Schrödinger functional (SF) \[?,?\]. For the introduction of static quarks in the SF as well as any unexplained notation we refer to \[?\]. We adopted the $`\mathrm{O}(a)`$-improved Wilson regularization \[?,?\] with non-perturbatively determined value of $`c_{\mathrm{SW}}`$ \[?\] for the light quarks. The gauge sector is discretized through the Wilson plaquette action.
Below we will report also on results of Monte Carlo computations. Our simulation parameters are summarized in appendix C. All the results have been obtained in the quenched approximation, a part of them already appeared in \[?\] and \[?\].
## 2 Static actions
Static quarks are introduced on the lattice through fermionic fields $`\psi _\mathrm{h}`$ and $`\overline{\psi }_\mathrm{h}`$, which live on the lattice sites and satisfy the projection properties
$$P_+\psi _\mathrm{h}=\psi _\mathrm{h},\overline{\psi }_\mathrm{h}P_+=\overline{\psi }_\mathrm{h},P_+=\frac{1+\gamma _0}{2}.$$
(2.1)
The action $`S_\mathrm{h}`$ for static quarks has been derived by Eichten and Hill in \[?\]. Here we adopt a notation, which allows us to provide a generalization of that action and write it in the form
$$S_\mathrm{h}^\mathrm{W}=a^4\frac{1}{1+a\delta m_\mathrm{W}}\underset{x}{}\overline{\psi }_\mathrm{h}(x)(D_0^\mathrm{W}+\delta m_\mathrm{W})\psi _\mathrm{h}(x),$$
(2.2)
with the covariant derivative
$$D_0^\mathrm{W}\psi _\mathrm{h}(x)=\frac{1}{a}\left[\psi _\mathrm{h}(x)W^{}(xa\widehat{0},0)\psi _\mathrm{h}(xa\widehat{0})\right],$$
(2.3)
where $`W(x,0)`$ is a gauge parallel transporter with the gauge transformation properties of the link $`U(x,0)`$. The Eichten-Hill action is given by setting $`W(x,0)=U(x,0)`$. The quark propagator $`G_\mathrm{h}^\mathrm{W}(x,y)`$ satisfying $`(D_0^\mathrm{W}+\delta m_\mathrm{W})G_\mathrm{h}^\mathrm{W}(x,y)=\delta (xy)P_+`$, reads
$`G_\mathrm{h}^\mathrm{W}(x,y)`$ $`=`$ $`\theta (x_0y_0)\delta (𝐱𝐲)(1+a\delta m_\mathrm{W})^{(x_0y_0)/a}𝒫^\mathrm{W}(y,x)^{}P_+,`$ (2.4)
$`𝒫^\mathrm{W}(x,x)`$ $`=`$ $`1,`$ (2.5)
$`𝒫^\mathrm{W}(x,x+R\widehat{\mu })`$ $`=`$ $`W(x,\mu )W(x+a\widehat{\mu },\mu )\mathrm{}W(x+a(R1)\widehat{\mu },\mu )\mathrm{for}\mathrm{R}>0,`$ (2.6)
where $`\delta m_\mathrm{W}`$ cancels the divergence in the self-energy of the static quark (see below).
### 2.1 Noise to signal ratio
In order to simplify the following discussion we start from the action in eq. (2.2) but set $`\delta m_\mathrm{W}=0`$. This is possible since the dependence of correlation functions on $`\delta m_\mathrm{W}`$ is known exactly from the form of the static quark propagator in eqs. (2.4-2.6) and can be restored easily. The noise to signal ratio discussed here is independent of $`\delta m_\mathrm{W}`$. As an example we consider a correlation function of heavy-light fields such as $`f_\mathrm{A}^{\mathrm{stat}}(x_0)`$, which describes the propagation of a static–light pseudoscalar meson. Its definition
$$f_\mathrm{A}^{\mathrm{stat}}(x_0)=\frac{a^6}{2}\underset{𝐲,𝐳}{}A_0^{\mathrm{stat}}(x)\overline{\zeta }_\mathrm{h}(𝐲)\gamma _5\zeta _\mathrm{l}(𝐳),$$
(2.7)
involves the boundary fields $`\overline{\zeta }_\mathrm{h}`$ and $`\zeta _\mathrm{l}`$ (see \[?\]) as well as the time component of the axial current
$$A_0^{\mathrm{stat}}(x)=\overline{\psi }_\mathrm{l}(x)\gamma _0\gamma _5\psi _\mathrm{h}(x),$$
(2.8)
where the subscript ‘l’ indicates the light quark fields. The correlation function is represented pictorially in figure 1.
From the quantum mechanical representation one expects the large $`x_0`$ asymptotic behavior
$$f_\mathrm{A}^{\mathrm{stat}}(x_0)e^{E_{\mathrm{stat}}x_0},$$
(2.9)
where $`E_{\mathrm{stat}}`$ is the binding energy of the static–light system. It depends on the choice for $`W`$ and diverges approximately linearly in the lattice spacing.
$$E_{\mathrm{stat}}E_{\mathrm{self}}+\mathrm{O}(a^0)\frac{1}{a}e^{(1)}g_0^2+\mathrm{}.$$
(2.10)
This divergence is to be canceled by $`\delta m_\mathrm{W}=E_{\mathrm{self}}+\mathrm{O}(a^0)`$. The finite part of $`\delta m_\mathrm{W}`$ is of course scheme dependent, but in perturbation theory one can always perform the double expansion
$`E_{\mathrm{self}}`$ $`=`$ $`E_{\mathrm{self}}^{(1)}g_0^2+\mathrm{O}(g_0^4),`$ (2.11)
$`E_{\mathrm{self}}^{(1)}`$ $`=`$ $`{\displaystyle \frac{1}{a}}e^{(1)}+\mathrm{O}(a^0).`$ (2.12)
For example, if we define $`E_{\mathrm{self}}`$ from a Schrödinger functional correlation function (see eq. (A.76)), the $`a`$-expansion appears as an expansion in terms of $`a/L`$. The coefficient $`e^{(1)}`$ does then depend on the action, but not on the correlation function used to define it.
We now want to discuss the noise to signal ratio of the Monte Carlo estimate of $`f_\mathrm{A}^{\mathrm{stat}}`$. The variance of a generic observable $`𝒪`$ in a QCD simulation is given by $`\stackrel{~}{𝒪}^2_\mathrm{G}\stackrel{~}{𝒪}_\mathrm{G}^2`$, where the expectation value $`\mathrm{}_\mathrm{G}`$ is an average over the gauge fields including the weight obtained after integrating out the fermion fields (the fermion determinant). $`\stackrel{~}{𝒪}`$ is derived from $`𝒪`$ by performing the corresponding Wick contractions on each gauge field background. In the Monte Carlo one takes advantage of translation invariance on the r.h.s. of eq. (2.7) and estimates $`f_\mathrm{A}^{\mathrm{stat}}=𝒪`$ with $`𝒪=\frac{a^9}{2L^3}_{𝐲,𝐳,𝐱}A_0^{\mathrm{stat}}(x)\overline{\zeta }_\mathrm{h}(𝐲)\gamma _5\zeta _\mathrm{l}(𝐳)`$. Its variance, $`\sigma _\mathrm{A}(x_0)`$, can be rewritten as
$`\sigma _\mathrm{A}(x_0)`$ $`=`$ $`{\displaystyle \frac{a^{18}}{4L^6}}{\displaystyle \underset{𝐱,𝐲,𝐳,𝐱^{},𝐲^{},𝐳^{}}{}}[A_0^{\mathrm{stat}}]_{\mathrm{hl}}(x_0,𝐱)\overline{\zeta }_\mathrm{h}(𝐲)\gamma _5\zeta _\mathrm{l}(𝐳)\times `$ (2.13)
$`\left\{[A_0^{\mathrm{stat}}]_{\mathrm{h}^{}\mathrm{l}^{}}(x_0,𝐱^{})\overline{\zeta }_\mathrm{h}^{}(𝐲^{})\gamma _5\zeta _\mathrm{l}^{}(𝐳^{})\right\}^{}[f_\mathrm{A}^{\mathrm{stat}}(x_0)]^2,`$
where $`\mathrm{h}^{}`$ and $`\mathrm{l}^{}`$ denote copies of $`\mathrm{h}`$ and $`\mathrm{l}`$ differing only by their flavor, which are introduced to be able to write the variance in the form of a standard expectation value. Saturating $`\sigma _\mathrm{A}(x_0)`$ with intermediate states shows that its large $`x_0`$ asymptotics is
$$\sigma _\mathrm{A}(x_0)\underset{𝐱,𝐱^{}}{}C(𝐱,𝐱^{})\mathrm{e}^{x_0E_{\mathrm{ll}^{}}(𝐱,𝐱^{})},$$
(2.14)
with $`E_{\mathrm{ll}^{}}(𝐱,𝐱^{})`$ the energy of a state with a static quark-antiquark pair at positions $`𝐱,𝐱^{}`$ and a light quark-antiquark pair with flavors $`\mathrm{l},\mathrm{l}^{}`$. At large $`x_0`$ the sum is dominated by the term with the smallest energy $`E`$. This is expected to be given by $`𝐱=𝐱^{}`$, where the color charges of the static quark and anti-quark compensate. The light quark-antiquark pair should then feel little of the static quarks and its lowest energy be given approximately by $`m_\pi `$. This argument suggests
$$\mathrm{Min}_{𝐱,𝐱^{}}E_{\mathrm{ll}^{}}(𝐱,𝐱^{})=V(0)+m_\pi .$$
(2.15)
Here $`V(0)`$ is the lattice static quark potential at zero distance, which for example is computable via the large $`x_0`$ asymptotics
$`\overline{\psi }_\mathrm{h}(x)\gamma _5\psi _\mathrm{h}^{}(x)\overline{\zeta }_\mathrm{h}^{}(𝐱)\gamma _5\zeta _\mathrm{h}(𝐱)\mathrm{e}^{x_0V(0)}.`$ (2.16)
The noise to signal ratio $`R_{\mathrm{NS}}`$ for $`f_\mathrm{A}^{\mathrm{stat}}(x_0)`$ should then approach
$$R_{\mathrm{NS}}e^{[E_{\mathrm{stat}}(m_\pi +V(0))/2]x_0}.$$
(2.17)
For the Eichten-Hill action (or more generally whenever $`W(x,0)`$ is unitary), the lattice potential vanishes at zero distance and one obtains the formula given earlier by Lepage \[?\]. We infer from eq. (2.17) that the linear divergence in $`E_{\mathrm{stat}}`$ is responsible for the exponential growth of the error with the Euclidean time $`x_0`$, which has been observed in many numerical investigations of correlation functions such as $`f_\mathrm{A}^{\mathrm{stat}}`$ \[?,?,?,?\]. It is then clear that the problem becomes more severe as the continuum limit is approached. In particular, for the Eichten-Hill action the coefficient $`e^{(1)}`$ was found to be rather large \[?\] rendering precise computations hopeless. Finally we remark again that the form of the static propagator shows that $`R_{\mathrm{NS}}`$ is (exactly) independent of $`\delta m_\mathrm{W}`$. In the final formula eq. (2.17) this is realized since $`E_{\mathrm{stat}}`$ and $`V(0)/2`$ are shifted by the same amount,
$`aE_{\mathrm{stat}}|_{\delta m_\mathrm{W}}`$ $`=`$ $`aE_{\mathrm{stat}}|_{\delta m_\mathrm{W}=0}+\mathrm{ln}(1+a\delta m_\mathrm{W})`$
$`aV(0)|_{\delta m_\mathrm{W}}`$ $`=`$ $`aV(0)|_{\delta m_\mathrm{W}=0}+2\mathrm{ln}(1+a\delta m_\mathrm{W}).`$
### 2.2 Statistically improved discretizations
A possible way to improve on the problem discussed in the previous section is to employ actions $`S_\mathrm{h}^W`$ inspired by variance reduction methods. Such methods led for example to the introduction of the one-link integral (or multihit) \[?\] in the pure gauge theory. This provides unbiased estimators for quantities such as Polyakov loop correlation functions, significantly reducing at the same time their variance. Here, a similar change will not lead to an unbiased estimator, but rather has to be considered a change of the static action. Alternatively (or simultaneously) one can try to enhance the signal by choosing the parallel transporter $`W`$ such that $`E_{\mathrm{stat}}E_{\mathrm{self}}`$ is comparatively small (assuming $`V(0)`$ to be numerically less relevant). In addition to this, we want to preserve on the lattice the following symmetries of the static theory
* Heavy quark spin symmetry:
$`\psi _\mathrm{h}𝒱\psi _\mathrm{h},\overline{\psi }_\mathrm{h}\overline{\psi }_\mathrm{h}𝒱^1,\mathrm{with}𝒱=\mathrm{exp}(i\varphi _iϵ_{ijk}\sigma _{jk}),`$ (2.18)
* Local conservation of heavy quark flavor number:
$$\psi _\mathrm{h}\mathrm{e}^{i\eta (𝐱)}\psi _\mathrm{h},\overline{\psi }_\mathrm{h}\overline{\psi }_\mathrm{h}\mathrm{e}^{i\eta (𝐱)}.$$
(2.19)
Together with gauge invariance, parity and cubic symmetry, this is enough to guarantee that the universality class and the O($`a`$) improvement are unchanged with respect to the Eichten-Hill action, as has been discussed in \[?\]. Furthermore we want to keep the action as local as possible. We therefore exclude constructions of $`W`$ which involve fields at a distance two lattice spacings or more away from the link $`U(x,0)`$. Taking these considerations into account, we propose the following regularized actions
$`S_\mathrm{h}^\mathrm{A}`$ $`:W^\mathrm{A}(x,0)=`$ $`V(x,0),`$ (2.20)
$`S_\mathrm{h}^\mathrm{s}`$ $`:W^\mathrm{s}(x,0)=`$ $`V(x,0)\left[{\displaystyle \frac{g_0^2}{5}}+\left({\displaystyle \frac{1}{3}}\text{tr}V^{}(x,0)V(x,0)\right)^{1/2}\right]^1,`$ (2.21)
$`S_\mathrm{h}^{\mathrm{HYP}}`$ $`:W^{\mathrm{HYP}}(x,0)=`$ $`V_{\mathrm{HYP}}(x,0),`$ (2.22)
where $`V(x,0)`$ is the average of the six staples around the link $`U(x,0)`$
$`V(x,0)`$ $`=`$ $`{\displaystyle \frac{1}{6}}{\displaystyle \underset{j=1}{\overset{3}{}}}[U(x,j)U(x+a\widehat{j},0)U^{}(x+a\widehat{0},j)`$ (2.23)
$`+U^{}(xa\widehat{j},j)U(xa\widehat{j},0)U(x+a\widehat{0}a\widehat{j},j)],`$
and $`V_{\mathrm{HYP}}`$ is the HYP link \[?\]. In the latter case three coefficients $`(\alpha _1,\alpha _2,\alpha _3)\stackrel{}{\alpha }`$ need to be specified in order to define the combination of differently smeared links in the construction of the HYP link. In the following we will only discuss the choices $`\stackrel{}{\alpha }=(0.75,0.6,0.3)`$, motivated in \[?\] and $`\stackrel{}{\alpha }=(1.0,1.0,0.5)`$ obtained by an approximate minimization of $`R_{\mathrm{NS}}`$. They define $`S_\mathrm{h}^{\mathrm{HYP1}}`$ and $`S_\mathrm{h}^{\mathrm{HYP2}}`$ respectively. The construction of HYP links involves projecting $`3\times 3`$ complex matrices onto SU(3). As we expect deviations from SU(3) to be small, we prefer approximating the projection by an analytic function defined by the steps
$$WW/\sqrt{\text{tr}(WW^{})/3},$$
(2.24)
followed by 4 iterations of
$$WX\left(1\frac{\mathrm{i}}{3}\mathrm{Im}\left(detX\right)\right),\mathrm{with}X=W\left(\frac{3}{2}\frac{1}{2}W^{}W\right).$$
(2.25)
To be precise, this function is to be taken as part of the definition of $`V_{\mathrm{HYP}}`$, when our numerical results for improvement coefficients and renormalization factors are used in subsequent computations.
For the actions $`S_\mathrm{h}^{\mathrm{HYP1}}`$ and $`S_\mathrm{h}^{\mathrm{HYP2}}`$ the value of $`e^{(1)}`$ has been computed in \[?\]. In the cases $`W/`$ SU(3) it is more convenient to look at $`E_{\mathrm{self}}V(0)/2`$ not only because this is the relevant quantity for the noise to signal ratio, but also because at 1-loop order the deviations of the links $`W`$ from unitarity cancel in the combination. We indicate by $`r^{(1)}`$ the 1-loop coefficient in the perturbative expansion
$$E_{\mathrm{self}}V(0)/2\left(\frac{1}{a}r^{(1)}+\mathrm{O}(a^0)\right)g_0^2+\mathrm{O}(g_0^4).$$
(2.26)
Of course $`r^{(1)}=e^{(1)}`$ for $`S_\mathrm{h}^{\mathrm{HYP}}`$ and $`S_\mathrm{h}^{\mathrm{EH}}`$. In appendix A we outline a computation of $`r^{(1)}`$ for $`S_\mathrm{h}^\mathrm{A}`$ and explain why it is the same for $`S_\mathrm{h}^\mathrm{s}`$ (at one-loop order).
Some results for $`r^{(1)}`$ are collected in table 1. From the table we see that perturbation theory already suggests $`S_\mathrm{h}^{\mathrm{HYP2}}`$ as the most favorable choice concerning signal enhancement.
The discretizations $`S_\mathrm{h}^\mathrm{A}`$ and $`S_\mathrm{h}^\mathrm{s}`$ are inspired by noise reduction methods, namely APE smearing and one-link integral respectively. APE smearing was introduced in \[?\] where it was shown to suppress fluctuations in gauge invariant quantities in the pure gauge theory. As shown by $`r^{(1)}`$, this can also be interpreted as a reduction of the static self energy. Finally $`W^\mathrm{s}`$ is an approximation of the SU(3) one-link integral. For completeness we describe briefly how we arrived at $`S_\mathrm{h}^\mathrm{s}`$ in appendix B. The reader may, however, take it just as another ansatz satisfying our criteria explained above.
The effectiveness of these regularizations in reducing the noise to signal ratio in static–light correlation functions has been checked non-perturbatively by a simulation on a $`16^3\times 32`$ lattice at $`\beta =6/g_0^2=6`$, where the lattice spacing is $`a\mathrm{\hspace{0.17em}0.1}\mathrm{fm}`$ (the hopping parameter $`\kappa `$ has been set to the strange quark mass value, $`\kappa =0.133929`$, as in \[?\]). Figure 2 shows the results for $`R_{\mathrm{NS}}`$ obtained using an ensemble of almost 5000 configurations.
The largest improvement is again given by the action $`S_\mathrm{h}^{\mathrm{HYP2}}`$, but we see from the figure that all the proposed discretizations produce a clear signal (for the considered statistics) at least up to a time separation of roughly $`2\mathrm{fm}`$. The dotted lines in figure 2 represent the predictions from eq. (2.17), where for $`aE_{\mathrm{stat}}`$, $`aV(0)`$ and $`am_\pi `$ we insert the estimates from the data. We summarize the obtained values of $`aE_{\mathrm{stat}}`$ and $`aV(0)`$ in table 1. The formula in eq. (2.17) turns out to be always quite accurate in describing the results. Furthermore, assuming $`E_{\mathrm{stat}}V(0)/2`$ to be dominated by its divergent term and approximating it by the leading perturbative estimate yields the correct order for $`R_{\mathrm{NS}}`$. It is thus to be expected that the ordering of $`R_{\mathrm{NS}}`$ observed in figure 2 will be preserved when going to smaller lattice spacings, where it is even more important to keep $`R_{\mathrm{NS}}`$ at a reasonable level if one wants to reach distances $`x_0`$ around $`12`$ fm. Indeed, numerical experience supports this expectation \[?\].
## 3 The size of discretization errors
In addition to the reduction of statistical errors, the size of scaling violations has to be taken into account when one chooses between different actions. For this reason we designed a number of scaling tests by which we could study the approach of a set of quantities defined in the SF to their continuum limit values. Before going into the details of this study we first need to discuss the computation of the O($`a`$) improvement coefficients of the static axial current for the different actions introduced.
### 3.1 Computation of the improvement coefficients $`c_\mathrm{A}^{\mathrm{stat}}`$ and $`b_\mathrm{A}^{\mathrm{stat}}`$
The O($`a`$) improvement programme \[?,?\] has been carried out for the EH action in \[?\]. In \[?\] the discussion has been extended to the actions we are considering here. As mentioned in the previous section, arguments based on the symmetries preserved on the lattice allow to show that the improvement pattern is the same in all cases. In particular the actions are improved once the light sector has been improved; no new improvement terms are needed for the static part of the actions.
Concerning the static–light axial density $`A_0^{\mathrm{stat}}(x)`$, the improved version reads
$$(A_\mathrm{I}^{\mathrm{stat}})_0=A_0^{\mathrm{stat}}(x)+ac_\mathrm{A}^{\mathrm{stat}}\delta A_0^{\mathrm{stat}}(x),\delta A_0^{\mathrm{stat}}(x)=\overline{\psi }_\mathrm{l}(x)\gamma _j\gamma _5\frac{\stackrel{}{}_j+\stackrel{}{}_j}{2}\psi _\mathrm{h}(x),$$
(3.27)
where the improvement coefficient $`c_\mathrm{A}^{\mathrm{stat}}`$ has been introduced. In a mass independent renormalization scheme the renormalized density can be written
$$(A_\mathrm{R}^{\mathrm{stat}})_0=Z_\mathrm{A}^{\mathrm{stat}}(g_0,a\mu )(1+b_\mathrm{A}^{\mathrm{stat}}am_\mathrm{q})(A_\mathrm{I}^{\mathrm{stat}})_0,$$
(3.28)
with $`m_\mathrm{q}`$ the bare subtracted light quark mass (cf. eq. (C.104)), $`Z_\mathrm{A}^{\mathrm{stat}}(g_0,a\mu )`$ the scale dependent renormalization factor of the axial current and $`b_\mathrm{A}^{\mathrm{stat}}`$ a second improvement coefficient. The values of $`c_\mathrm{A}^{\mathrm{stat}}`$ and $`b_\mathrm{A}^{\mathrm{stat}}`$, as well as $`Z_\mathrm{A}^{\mathrm{stat}}`$, depend on the choice for the static action. For the EH action the improvement coefficients $`c_\mathrm{A}^{\mathrm{stat}}`$ and $`b_\mathrm{A}^{\mathrm{stat}}`$ have been determined at 1-loop order of perturbation theory in \[?,?,?\], while the renormalization constant $`Z_\mathrm{A}^{\mathrm{stat}}`$ has been computed non-perturbatively in the SF scheme in \[?\]. In the remainder of this section we describe our computation of the improvement coefficients $`c_\mathrm{A}^{\mathrm{stat}}`$ and $`b_\mathrm{A}^{\mathrm{stat}}`$ for the actions in eqs. (2.20 \- 2.22), and we present a set of scaling studies. We will come back to the renormalization constant $`Z_\mathrm{A}^{\mathrm{stat}}`$ in section 4, where we also discuss the improvement the new actions can bring in the computation of the b-quark mass following the strategy in \[?,?\].
#### 3.1.1 Improvement conditions and results for $`c_\mathrm{A}^{\mathrm{stat}}`$ and $`b_\mathrm{A}^{\mathrm{stat}}`$
We want to compute the 1-loop coefficients $`c_\mathrm{A}^{\mathrm{stat},(1)}`$ and $`b_\mathrm{A}^{\mathrm{stat},(1)}`$ of the improvement constants $`c_\mathrm{A}^{\mathrm{stat}}`$ and $`b_\mathrm{A}^{\mathrm{stat}}`$. To this end we have adopted a mixed strategy. For $`S_\mathrm{h}^\mathrm{A}`$ and $`S_\mathrm{h}^\mathrm{s}`$ the computation has been carried out analytically<sup>1</sup><sup>1</sup>1At this order in perturbation theory the results for the improvement coefficients are the same for the two regularizations, see appendix A., while for $`S_\mathrm{h}^{\mathrm{HYP1}}`$ and $`S_\mathrm{h}^{\mathrm{HYP2}}`$ we have used Monte Carlo simulations to numerically estimate an effective 1-loop coefficient for the coupling range relevant here. In both cases we have exploited the same improvement conditions. We collect some details of the analytic computation in appendix A.
We introduce the correlation $`f_{\mathrm{A},\mathrm{I}}^{\mathrm{stat}}`$ defined as $`f_\mathrm{A}^{\mathrm{stat}}`$ in eq. (2.7) with the current $`A_0^{\mathrm{stat}}`$ replaced by the improved current in eq. (3.27). Assuming the knowledge of $`c_\mathrm{A}^{\mathrm{stat}}`$ for a discretized action $`S_\mathrm{h}^1`$, we have enforced the condition<sup>2</sup><sup>2</sup>2Where necessary we explicitly indicate the dependence of the correlation functions and the improvement coefficients on the discretization of the static action.
$$\frac{f_{\mathrm{A},\mathrm{I}}^{\mathrm{stat}}(T/2,S_\mathrm{h}^1)|_\theta }{f_{\mathrm{A},\mathrm{I}}^{\mathrm{stat}}(T/2,S_\mathrm{h}^1)|_\theta ^{}}=\frac{f_{\mathrm{A},\mathrm{I}}^{\mathrm{stat}}(T/2,S_\mathrm{h}^2)|_\theta }{f_{\mathrm{A},\mathrm{I}}^{\mathrm{stat}}(T/2,S_\mathrm{h}^2)|_\theta ^{}},\theta =0,\theta ^{}=1,m_\mathrm{q}=0,$$
(3.29)
and solved the implicit relation for $`c_\mathrm{A}^{\mathrm{stat}}`$ of $`S_\mathrm{h}^2`$. Expanding eq. (3.29) in powers of the coupling $`g_0^2`$ and setting $`S_\mathrm{h}^1=S_\mathrm{h}^{\mathrm{EH}}`$, for which $`c_\mathrm{A}^{\mathrm{stat},(1)}`$ is known from \[?\], we obtain
$$c_\mathrm{A}^{\mathrm{stat}}=c_\mathrm{A}^{\mathrm{stat},(1)}g_0^2+\mathrm{O}(g_0^4),c_\mathrm{A}^{\mathrm{stat},(1)}=0.0072(4),\mathrm{for}S_\mathrm{h}^\mathrm{s}\mathrm{and}S_\mathrm{h}^\mathrm{A}.$$
(3.30)
The ratios in eq. (3.29) have then been evaluated, with $`S_\mathrm{h}^1=S_\mathrm{h}^\mathrm{A}`$, on the configurations generated in runs I to IV of table 6 in order to obtain $`c_\mathrm{A}^{\mathrm{stat}}`$ for $`S_\mathrm{h}^2=S_\mathrm{h}^{\mathrm{HYP1}},S_\mathrm{h}^{\mathrm{HYP2}}`$. There the size $`L/a`$ and the $`\beta `$ values have been chosen such that $`L`$ is kept fixed to $`1.436r_0`$ with $`r_0`$ from \[?,?\]. Our results are shown in figure 3.
The result for $`S_\mathrm{h}^\mathrm{s}`$ provides a test of the numerical procedure since it is known analytically (eq. (3.30)). We see that by ascribing to the numerical result an error, which covers the spread of the points we find agreement with eq. (3.30). This suggests that here higher orders in $`g_0^2`$ contribute little to the cutoff effects of the ratios appearing in eq. (3.29). We estimate in the same way the errors for $`c_\mathrm{A}^{\mathrm{stat}}`$ of $`S_\mathrm{h}^{\mathrm{HYP1}}`$ and $`S_\mathrm{h}^{\mathrm{HYP2}}`$, quoting the result
$`c_\mathrm{A}^{\mathrm{stat}}0.039(4)g_0^2,\mathrm{for}S_\mathrm{h}^{\mathrm{HYP1}}`$ (3.31)
$`c_\mathrm{A}^{\mathrm{stat}}0.220(14)g_0^2,\mathrm{for}S_\mathrm{h}^{\mathrm{HYP2}}.`$ (3.32)
Here the statistical errors are not the relevant ones. Rather our dominant error is due to the assumption (tested to some extent as just explained) that the 1-loop value for $`c_\mathrm{A}^{\mathrm{stat}}`$ for action $`S_\mathrm{h}^1=S_\mathrm{h}^\mathrm{A}`$ is accurate for our values of $`g_0^2`$. Apart from this, our determinations do provide a non-perturbative computation of $`c_\mathrm{A}^{\mathrm{stat}}`$ for the other actions. Thus the errors quoted in eqs. (3.31,3.32) include a reasonable (but not necessarily a safe) estimate of the perturbative uncertainty.
Note that for $`S_\mathrm{h}^{\mathrm{HYP2}}`$ the improvement coefficient is considerably larger than for the other actions. This means that before improvement the linear $`a`$-effects are larger in this case and one might expect that this will be the case also for higher order $`a`$-effects. We will see below that this is however not born out of our non-perturbative results.
Sensitivity to the improvement coefficient $`b_\mathrm{A}^{\mathrm{stat}}`$ is achieved by exploiting the quark mass dependence of the correlation function $`f_{\mathrm{A},\mathrm{I}}^{\mathrm{stat}}(x_0)`$. Again, supposing $`b_\mathrm{A}^{\mathrm{stat}}`$ be known for the action $`S_\mathrm{h}^1`$, we consider the improvement condition
$$\frac{(1+ab_\mathrm{A}^{\mathrm{stat}}(S_\mathrm{h}^1)m_\mathrm{q}^{})f_{\mathrm{A},\mathrm{I}}^{\mathrm{stat}}(T/2,S_\mathrm{h}^1)|_{m_\mathrm{q}^{}}}{(1+ab_\mathrm{A}^{\mathrm{stat}}(S_\mathrm{h}^1)m_\mathrm{q})f_{\mathrm{A},\mathrm{I}}^{\mathrm{stat}}(T/2,S_\mathrm{h}^1)|_{m_\mathrm{q}}}=\frac{(1+ab_\mathrm{A}^{\mathrm{stat}}(S_\mathrm{h}^2)m_\mathrm{q}^{})f_{\mathrm{A},\mathrm{I}}^{\mathrm{stat}}(T/2,S_\mathrm{h}^2)|_{m_\mathrm{q}^{}}}{(1+ab_\mathrm{A}^{\mathrm{stat}}(S_\mathrm{h}^2)m_\mathrm{q})f_{\mathrm{A},\mathrm{I}}^{\mathrm{stat}}(T/2,S_\mathrm{h}^2)|_{m_\mathrm{q}}},$$
(3.33)
with $`\theta =1`$, and solve for $`b_\mathrm{A}^{\mathrm{stat}}`$ of $`S_\mathrm{h}^2`$.
In perturbation theory $`b_\mathrm{A}^{\mathrm{stat}}`$ has been computed to 1-loop order in \[?\] for the Eichten–Hill regularization. Similarly to what we did for $`c_\mathrm{A}^{\mathrm{stat}}`$, we have then analytically computed $`b_\mathrm{A}^{\mathrm{stat}}`$ for $`S_\mathrm{h}^\mathrm{A}`$, i.e. we have expanded eq. (3.33) in perturbation theory with $`S_\mathrm{h}^1=S_\mathrm{h}^{\mathrm{EH}}`$ and $`S_\mathrm{h}^2=S_\mathrm{h}^\mathrm{A}`$. In this computation we set $`m_\mathrm{q}=0`$ and, following \[?\], $`m_\mathrm{q}^{}`$ such that $`Lm_\mathrm{R}^{}=0.24`$, with $`m_\mathrm{R}^{}`$ the quark mass renormalized at scale $`\mu =1/L`$ in the minimal subtraction scheme on the lattice. The result reads
$$b_\mathrm{A}^{\mathrm{stat}}=1/2+b_\mathrm{A}^{\mathrm{stat},(1)}g_0^2+\mathrm{O}(g_0^4),b_\mathrm{A}^{\mathrm{stat},(1)}=0.033(7),\mathrm{for}S_\mathrm{h}^\mathrm{s}\mathrm{and}S_\mathrm{h}^\mathrm{A}.$$
(3.34)
Next we impose eq. (3.33) on the data sets I to IV in table 6. Here, both $`\kappa `$ values and $`\theta =1`$ are used, where the second $`\kappa `$ value was determined such that $`Lm_\mathrm{R}^{}=0.24`$ in the SF-scheme at scale $`\mu =1/(1.436r_0)`$. With $`S_\mathrm{h}^1=S_\mathrm{h}^\mathrm{A}`$, we obtain
$`b_\mathrm{A}^{\mathrm{stat}}1/2+0.078(8)g_0^2,\mathrm{for}S_\mathrm{h}^{\mathrm{HYP1}}`$ (3.35)
$`b_\mathrm{A}^{\mathrm{stat}}1/2+0.259(13)g_0^2,\mathrm{for}S_\mathrm{h}^{\mathrm{HYP2}}.`$ (3.36)
The errors have been estimated as in the case of $`c_\mathrm{A}^{\mathrm{stat}}`$. We show the numerical results in figure 4. Again the improvement coefficient turns out largest for $`S_\mathrm{h}^{\mathrm{HYP2}}`$.
One comment is in order here. In the condition in eq. (3.33), O$`(am_\mathrm{q})`$-terms with coefficients $`b_\mathrm{g}`$ and $`b_{\zeta _\mathrm{h}}`$ \[?\] are neglected, while terms proportional to $`b_\zeta `$ (see \[?\]) drop out. Of course, $`b_{\zeta _\mathrm{h}}`$ is O($`g_0^4N_\mathrm{f}`$) and therefore irrelevant in a 1-loop computation. While $`b_g`$ is O$`(g_0^2N_\mathrm{f})`$, it enters expectation values only with an additional factor $`g_0^2`$ (unless there is a non-zero background gauge field). In other words, eq. (3.33) is correct up to $`\mathrm{O}(g_0^4)`$ in full QCD, but in the quenched approximation it is valid also non-perturbatively.
### 3.2 Scaling tests
On the same set of configurations used for the determination of $`c_\mathrm{A}^{\mathrm{stat}}`$ and $`b_\mathrm{A}^{\mathrm{stat}}`$ (runs I–IV), and for $`\kappa =\kappa _\mathrm{c}`$, we have computed (for each regularization) the ratios
$$\xi _\mathrm{A}(\theta ,\theta ^{})=\frac{f_{\mathrm{A},\mathrm{I}}^{\mathrm{stat}}(T/2)|_\theta }{f_{\mathrm{A},\mathrm{I}}^{\mathrm{stat}}(T/2)|_\theta ^{}},\xi _1(\theta ,\theta ^{})=\frac{f_1^{\mathrm{stat}}|_\theta }{f_1^{\mathrm{stat}}|_\theta ^{}},h(d/L)=\frac{f_1^{\mathrm{hh}}(d)}{f_1^{\mathrm{hh}}(L/2)}.$$
(3.37)
Here and in the following $`f_{\mathrm{A},\mathrm{I}}^{\mathrm{stat}}(x_0)`$ refers to the 1-loop improved axial current in eq. (3.27). The correlation function $`f_1^{\mathrm{stat}}`$ is schematically represented in the right part of figure 1, it precisely reads
$$f_1^{\mathrm{stat}}=\frac{a^{12}}{2L^6}\underset{𝐮,𝐯,𝐲,𝐳}{}\overline{\zeta }_\mathrm{l}^{}(𝐮)\gamma _5\zeta _\mathrm{h}^{}(𝐯)\overline{\zeta }_\mathrm{h}(𝐲)\gamma _5\zeta _\mathrm{l}(𝐳),$$
(3.38)
with the primed fields living on the boundary at $`x_0=T`$. The quantity $`h(d/L)`$ was introduced in \[?\]. It is defined through the boundary to boundary SF correlator
$$f_1^{\mathrm{hh}}(x_3)=\frac{a^8}{2L^2}\underset{x_1,x_2,𝐲,𝐳}{}\overline{\zeta }_\mathrm{h}^{}(𝐱)\gamma _5\zeta _\mathrm{h}^{}(\mathrm{𝟎})\overline{\zeta }_\mathrm{h}(𝐲)\gamma _5\zeta _\mathrm{h}(𝐳),$$
(3.39)
note that the sum runs on $`x_1`$ and $`x_2`$ and therefore yields an $`x_3`$-dependent correlation function.
All the quantities in eq. (3.37) have a finite continuum limit in a fixed, finite, volume. The continuum result is universal, but the way it is approached depends on the details of the regularization. The scaling of these ratios can then provide an idea about the size of discretization effects for the static actions in eqs. (2.20 \- 2.22). Our results are shown in figure 5.
The cutoff effects in $`\xi _\mathrm{A}`$ are somewhat larger with the Eichten–Hill action than with any of our alternatives. Still, it is more relevant to note that all cutoff effects visible in the figure are very small and linear in $`a^2`$. This suggests that the O($`a`$) improvement programme has been implemented in a satisfactory way for the lattice spacings considered here – also for $`S_\mathrm{h}^{\mathrm{HYP2}}`$, where the improvement coefficients turn out to be not that small.
The gain in statistical precision brought by the new discretizations of the static action can also be seen in figure 5, especially for the quantity $`h(1/4)`$, which involves two static quarks propagating over the whole temporal extent.
Another interesting observable is the step scaling function introduced in \[?\] for the renormalization of the b-quark mass and further discussed in the following section. Here we simply define it as
$`\mathrm{\Sigma }_\mathrm{m}(u,a/L)`$ $`=`$ $`2L\left[\mathrm{\Gamma }_{\mathrm{stat}}(2L)\mathrm{\Gamma }_{\mathrm{stat}}(L)\right]_{u=\overline{g}^2(L)}`$ (3.40)
$`\mathrm{\Gamma }_{\mathrm{stat}}(L)`$ $`=`$ $`{\displaystyle \frac{1}{2a}}\mathrm{ln}\left[f_{\mathrm{A},\mathrm{I}}^{\mathrm{stat}}(x_0a)/f_{\mathrm{A},\mathrm{I}}^{\mathrm{stat}}(x_0+a)\right],\theta =1/2,T=L,`$ (3.41)
in terms of the correlation function $`f_{\mathrm{A},\mathrm{I}}^{\mathrm{stat}}`$ and the Schrödinger functional coupling $`\overline{g}(L)`$ \[?,?,?\], which fixes the length scale $`L`$. The continuum limit of $`\mathrm{\Sigma }_\mathrm{m}`$ is, for each value of the coupling $`u`$, a universal quantity, independent of the regularization used. The step scaling function can hence be used for a further scaling study, in particular since discretization errors with the EH action turn out to be clearly visible \[?\]. Figure 6 shows that they are rather linear in $`(a/L)^2`$ for all actions considered. Compared to the EH action, the alternative ones show smaller lattice artefacts. In fact they are the smallest for $`S_\mathrm{h}^{\mathrm{HYP2}}`$ which also showed the best noise to signal ratio. We will return to another scaling test in section 4.1.
## 4 Renormalization of the axial current and quark mass
In the effective theory the static–light axial current is not derived from a symmetry transformation of the action. The renormalized current is therefore scale dependent. This dependence has been studied non-perturbatively, over a wide energy range, in the SF scheme (see \[?\]). The main quantity considered in that study is the step scaling function $`\mathrm{\Sigma }_\mathrm{A}^{\mathrm{stat}}(u,a/L)`$ defined as
$$\mathrm{\Sigma }_\mathrm{A}^{\mathrm{stat}}(u,a/L)=\frac{Z_\mathrm{A}^{\mathrm{stat}}(g_0,2L/a)}{Z_\mathrm{A}^{\mathrm{stat}}(g_0,L/a)}|_{\overline{g}^2(L)=u,m_\mathrm{q}=0},$$
(4.42)
with
$$Z_\mathrm{A}^{\mathrm{stat}}=\frac{\mathrm{\Xi }^{(0)}}{\mathrm{\Xi }},\mathrm{and}\mathrm{\Xi }=\frac{f_{\mathrm{A},\mathrm{I}}^{\mathrm{stat}}(L/2)}{\left[f_1f_1^{\mathrm{hh}}(L/2)\right]^{1/4}}|_{\theta =0.5},$$
(4.43)
here $`\mathrm{\Xi }^{(0)}`$ is the tree-level value of $`\mathrm{\Xi }`$, as in \[?\]. In eq. (4.43) $`f_1`$ is the correlator between two light-quark pseudoscalar boundary sources
$$f_1=\frac{a^{12}}{2L^6}\underset{𝐮,𝐯,𝐲,𝐳}{}\overline{\zeta }_1^{}(𝐮)\gamma _5\zeta _2^{}(𝐯)\overline{\zeta }_2(𝐲)\gamma _5\zeta _1(𝐳),$$
(4.44)
and we remind the reader that $`\overline{g}(L)`$ is the Schrödinger functional coupling. In the first part of this section we report on a scaling study of $`\mathrm{\Sigma }_\mathrm{A}^{\mathrm{stat}}`$.
In addition, for a chosen reference scale $`L_{\mathrm{ref}}=1.436r_0`$, we are interested in the dependence of the renormalization factor $`Z_\mathrm{A}^{\mathrm{stat}}(g_0,L_{\mathrm{ref}}/a)`$ on the bare coupling $`g_0`$. This allows to match bare matrix elements of the axial current to continuum ones (up to cutoff effects). The renormalization factor itself clearly depends on the regularization, and needs to be recomputed for the actions in eqs. (2.20 \- 2.22).
In the quenched approximation the b-quark mass has been obtained at leading order in HQET through the matching of the effective theory to the full one \[?,?\]. The mass of the B-meson is used in the procedure as phenomenological input. Therefore, although the matching is performed in a small volume, the effective theory has then to be connected to large volumes. Improving on this part of the computation would reduce the error on the result for the b-quark mass. The use of the static actions introduced here (in place of the Eichten-Hill action used in \[?,?\]) has been proven to be very efficient in this respect (see \[?\], where $`S_\mathrm{h}^{\mathrm{HYP1}}`$ has been considered). In the last part of this section we briefly discuss the strategy which has been used to compute the b-quark mass, with particular emphasis on the quantities we recompute in the new discretizations.
### 4.1 Results for the step scaling function and $`Z_\mathrm{A}^{\mathrm{stat}}`$
The step scaling function in eq. (4.42) has been evaluated for $`u=3.48`$ and with lattice resolution $`L/a=6,8,12`$ on the configurations produced in the runs ZI–sZIII of table 6 (the same sets were already used in figure 6). In figure 7 the results are compared to those for the Eichten-Hill action, obtained in \[?\]. In this case a larger set of lattice spacings was used and the data have been extrapolated to the continuum limit.
Here again we see that the different regularizations yield very similar results already for finite lattice spacing, although in principle agreement needs to be found only after having taken the continuum limit.
At the scale $`L_{\mathrm{ref}}=1.436r_0`$ we have also computed the renormalization constant $`Z_\mathrm{A}^{\mathrm{stat}}`$ in the $`\beta `$-range relevant for large volume simulations (again data from runs I-IV have been used here). We summarize the results in table 2. For $`6\beta 6.5`$ we parameterize them in the form $`(x=\beta 6)`$
$`Z_\mathrm{A}^{\mathrm{stat}}(g_0,L_{\mathrm{ref}}/a)`$ $`=`$ $`0.78890.1290x+0.1197x^2,\mathrm{for}S_\mathrm{h}^\mathrm{A}`$ (4.45)
$`Z_\mathrm{A}^{\mathrm{stat}}(g_0,L_{\mathrm{ref}}/a)`$ $`=`$ $`0.78950.1291x+0.1189x^2,\mathrm{for}S_\mathrm{h}^\mathrm{s}`$ (4.46)
$`Z_\mathrm{A}^{\mathrm{stat}}(g_0,L_{\mathrm{ref}}/a)`$ $`=`$ $`0.79620.1368x+0.1207x^2,\mathrm{for}S_\mathrm{h}^{\mathrm{HYP1}}`$ (4.47)
$`Z_\mathrm{A}^{\mathrm{stat}}(g_0,L_{\mathrm{ref}}/a)`$ $`=`$ $`0.84430.1740x+0.1351x^2,\mathrm{for}S_\mathrm{h}^{\mathrm{HYP2}}`$ (4.48)
when $`c_\mathrm{A}^{\mathrm{stat}}`$ is set to its 1-loop value, and
$`Z_\mathrm{A}^{\mathrm{stat}}(g_0,L_{\mathrm{ref}}/a)`$ $`=`$ $`0.79000.1286x+0.1231x^2,\mathrm{for}S_\mathrm{h}^\mathrm{A}`$ (4.49)
$`Z_\mathrm{A}^{\mathrm{stat}}(g_0,L_{\mathrm{ref}}/a)`$ $`=`$ $`0.79070.1287x+0.1224x^2,\mathrm{for}S_\mathrm{h}^\mathrm{s}`$ (4.50)
$`Z_\mathrm{A}^{\mathrm{stat}}(g_0,L_{\mathrm{ref}}/a)`$ $`=`$ $`0.80440.1404x+0.1244x^2,\mathrm{for}S_\mathrm{h}^{\mathrm{HYP1}}`$ (4.51)
$`Z_\mathrm{A}^{\mathrm{stat}}(g_0,L_{\mathrm{ref}}/a)`$ $`=`$ $`0.88660.1995x+0.1423x^2,\mathrm{for}S_\mathrm{h}^{\mathrm{HYP2}}`$ (4.52)
for $`c_\mathrm{A}^{\mathrm{stat}}=0`$. The formulae reproduce the numbers in table 2 within their errors. They may be assigned an uncertainty of about 2‰.
Given now a bare matrix element $`\mathrm{\Phi }_{\mathrm{bare}}`$ of the static–light axial current, the corresponding matrix element $`\mathrm{\Phi }(\mu )`$ renormalized in the SF scheme at the scale $`\mu ^1=1.436r_0`$ can be written
$$\mathrm{\Phi }(\mu )=Z_\mathrm{A}^{\mathrm{stat}}(g_0,L_{\mathrm{ref}}/a)\mathrm{\Phi }_{\mathrm{bare}}(g_0),\mu ^1=L_{\mathrm{ref}},$$
(4.53)
with $`Z_\mathrm{A}^{\mathrm{stat}}(g_0,L_{\mathrm{ref}}/a)`$ from eqs.(4.45-4.52) depending on the regularization used to compute the bare matrix element. Finally the Renormalization Group Invariant (RGI) matrix element $`\mathrm{\Phi }_{\mathrm{RGI}}`$ reads
$$\mathrm{\Phi }_{\mathrm{RGI}}=\frac{\mathrm{\Phi }_{\mathrm{RGI}}}{\mathrm{\Phi }(\mu )}\mathrm{\Phi }(\mu ),$$
(4.54)
where we have factorized out the universal ratio $`\mathrm{\Phi }_{\mathrm{RGI}}/\mathrm{\Phi }(\mu )`$, which has been computed in the continuum limit in \[?\]. There one finds the result
$$\frac{\mathrm{\Phi }(\mu )}{\mathrm{\Phi }_{\mathrm{RGI}}}=1.088(10),\mathrm{at}\mu =(1.436r_0)^1,$$
(4.55)
which, as discussed, applies also to the discretizations introduced here. Moreover, as the error on the ratio in eq. (4.55) refers to a continuum result, it propagates into $`\mathrm{\Phi }_{\mathrm{RGI}}`$ only once $`\mathrm{\Phi }(\mu )`$ has been extrapolated to the continuum limit.
### 4.2 Renormalization of the quark mass
In \[?\] a strategy for the non-perturbative computation of the b-quark mass in the static approximation has been introduced and discussed. We refer to that publication for all the details of the method and we only remind the reader of the basic formula, from which the RGI b-quark mass $`M_\mathrm{b}`$ can be implicitly derived:
$$L_0m_\mathrm{B}=L_0\mathrm{\Gamma }(L_0,M_\mathrm{b})+\underset{k=0}{\overset{K1}{}}2^{(k+1)}\sigma _\mathrm{m}(u_k)+L_0\mathrm{\Delta }E,$$
(4.56)
where the use of a finite volume scheme like the Schrödinger functional is assumed. In eq. (4.56)
* $`L_0`$ is the linear extent of the small volume where the matching between HQET and QCD is performed.
* $`m_\mathrm{B}`$ is the physical (spin-averaged) $`\mathrm{B}_{(\mathrm{s})}`$-meson mass.
* $`\mathrm{\Gamma }`$ is an energy defined in terms of heavy-light correlators (with a heavy quark of mass $`M`$) in a finite volume $`L^3`$. A relevant example is
$`\mathrm{\Gamma }(L,M)`$ $`=`$ $`\frac{1}{4}(\mathrm{\Gamma }_\mathrm{P}+3\mathrm{\Gamma }_\mathrm{V}),\mathrm{\Gamma }_\mathrm{P}={\displaystyle \frac{1}{2a}}\mathrm{ln}\left[f_\mathrm{A}(x_0a)/f_\mathrm{A}(x_0+a)\right],`$ (4.57)
($`x_0/T=1/2\mathrm{with}T/L\mathrm{fixed}`$) where $`\mathrm{\Gamma }_\mathrm{V}`$ is defined analogously in the vector channel. Their static version $`\mathrm{\Gamma }_{\mathrm{stat}}(L)`$ was defined in eq. (3.41). Its infinite volume limit gives the static binding energy $`E_{\mathrm{stat}}`$ in eq. (2.9).
* $`\sigma _\mathrm{m}`$ is a step scaling function
$$\sigma _\mathrm{m}(\overline{g}^2(L))=2L\left[\mathrm{\Gamma }_{\mathrm{stat}}(2L)\mathrm{\Gamma }_{\mathrm{stat}}(L)\right].$$
(4.58)
Notice that in the difference the dependence of $`\mathrm{\Gamma }_{\mathrm{stat}}`$ on $`\delta m_\mathrm{W}`$ cancels. Once the effective theory and the full one are matched in small volume the step scaling function provides, within the effective theory, the connection to large volumes, where contact with phenomenology can be made. This large scale difference is covered in several steps, such that at each stage scales differing only by a factor two appear. In this sense the sum in eq. (4.56) connects the size $`L_0`$ to $`L_K=2^KL_0`$, therefore $`u_k=\overline{g}^2(2^kL_0)`$ in the equation.
* $`\mathrm{\Delta }E`$ is an energy difference
$$\mathrm{\Delta }E=E_{\mathrm{stat}}\mathrm{\Gamma }_{\mathrm{stat}}(L_K).$$
(4.59)
It is important to remark that in the way eq. (4.56) has been written all the quantities on the r.h.s. are independent of $`\delta m_\mathrm{W}`$ and can be computed in the continuum limit.
In real applications of the method (see \[?,?\]) the binding energy $`E_{\mathrm{stat}}`$ has been computed on a lattice of size 1.6 fm while for $`L_K`$ it sufficed to have almost half of it, more precisely $`L_K=L_{\mathrm{ref}}=1.436r_0`$. In view of combining this strategy with the use of the new static actions we have calculated the quantity $`a\mathrm{\Gamma }_{\mathrm{stat}}(g_0,L_{\mathrm{ref}}/a)`$, on the configurations generated in runs I-IV of table 6. The results are collected in table 3.
The data can be described within one standard deviation by polynomial expressions. Explicitly, for the 1-loop value of $`c_\mathrm{A}^{\mathrm{stat}}`$ and for $`6\beta =6/g_0^26.5`$ we provide the parameterizations $`(x=\beta 6)`$
$`a\mathrm{\Gamma }_{\mathrm{stat}}(g_0,L_{\mathrm{ref}}/a)`$ $`=`$ $`0.59160.2140x+0.0418x^2,\mathrm{for}S_\mathrm{h}^\mathrm{A}`$ (4.60)
$`a\mathrm{\Gamma }_{\mathrm{stat}}(g_0,L_{\mathrm{ref}}/a)`$ $`=`$ $`0.50400.1726x+0.0285x^2,\mathrm{for}S_\mathrm{h}^\mathrm{s}`$ (4.61)
$`a\mathrm{\Gamma }_{\mathrm{stat}}(g_0,L_{\mathrm{ref}}/a)`$ $`=`$ $`0.20680.0890x+0.0106x^2,\mathrm{for}S_\mathrm{h}^{\mathrm{HYP1}}`$ (4.62)
$`a\mathrm{\Gamma }_{\mathrm{stat}}(g_0,L_{\mathrm{ref}}/a)`$ $`=`$ $`0.17870.0915x+0.0255x^2,\mathrm{for}S_\mathrm{h}^{\mathrm{HYP2}}`$ (4.63)
with an uncertainty of about 2.5‰. In addition we see from the table that $`c_\mathrm{A}^{\mathrm{stat}}`$ has very little impact on this quantity.
## 5 Summary and conclusions
We have proposed and investigated alternative discretizations for static quarks on the lattice. All of them have automatic $`\mathrm{O}(a)`$ improvement, i.e. energies approach the continuum limit with correction $`\mathrm{O}(a^2)`$ if the light-quark sector is improved. The purpose of introducing these actions was to achieve a better noise to signal ratio at large Euclidean times. This goal could be reached, since the exponential growth of the noise to signal ratio is non-universal and can be reduced by choosing a discretization with a small self energy for the static quark (relative to the lattice potential at the origin). The two actions with HYP links are best in this respect.
Of course a prime criterion for choosing an action is to have small lattice artifacts. We have therefore investigated several quantities (figures 5,6,7). It turns out that lattice artifacts are mostly indistinguishable for the different actions except for the case of $`\mathrm{\Sigma }_\mathrm{m}`$, figure 6 (and to a smaller extent $`\xi _\mathrm{A}`$), where the lattice artifacts with the two actions with HYP links are comfortably small compared to the ones with the standard action. Since an implementation of two or three static actions is no problem in any practical computation, it is advisable to compute with both $`S_\mathrm{h}^{\mathrm{HYP2}}`$ and $`S_\mathrm{h}^{\mathrm{HYP1}}`$ and to perform continuum extrapolations constrained by universality. This promises to stabilize continuum extrapolations in a non-trivial manner, since e.g. figure 6 provides an example where the $`a`$-effects are different for the two actions.
In view of future applications we have computed all coefficients needed to improve and renormalize the static-light axial current for all actions considered. Here, the renormalization factors are now known non-perturbatively (in the quenched approximation) and the improvement coefficients are given up to uncertainties $`\mathrm{O}(g_0^4)`$. In the course of these computations we again observed that the $`a`$-effects (now the ones before $`\mathrm{O}(a)`$ improvement) are different for the different actions.
We finally emphasize that in \[?\] it was shown that the force between static quarks, computed e.g. through Polyakov loop correlation functions, is automatically $`\mathrm{O}(a)`$ improved. This is derived from the corresponding property of the Eichten Hill action for static quarks. This proof is also valid when the alternative actions are employed. Of course a significant gain in the noise to signal ratio can be achieved in this way as has first been seen in \[?\]. Such a gain in precision is especially relevant in the theory with dynamical quarks, where the noise-reduction methods of \[?\] are not applicable. Indeed, with the action $`S_\mathrm{h}^\mathrm{A}`$, it has been possible very recently to observe string breaking in $`N_\mathrm{f}=2`$ QCD \[?\].
In addition, these discretizations can be used to efficiently calculate the $`1/m_\mathrm{b}`$ corrections to the static limit. In this context a preliminary study has been presented in \[?\].
Acknowledgements. We are grateful to H. Simma and F. Knechtli for useful discussions. Many thanks go to R. Hoffmann and F. Knechtli for performing the computation of $`e^{(1)}`$ for the $`S_\mathrm{h}^{\mathrm{HYP}}`$ actions. We thank DESY for allocating computer time on the APEmille computers at DESY Zeuthen to this project and the APE group for its help. This work is also supported by the EU IHP Network on Hadron Phenomenology from Lattice QCD under grant HPRN-CT-2000-00145 and by the Deutsche Forschungsgemeinschaft in the SFB/TR 09.
## Appendix A Perturbation theory for the APE action
### A.1 Feynman rules
We use the conventions of \[?,?\]. Here we generalize the Feynman rules given for the EH action in appendix B.1 of \[?\] (again we use the notation of that reference), for the action
$$S_\mathrm{h}^{\mathrm{A},\alpha }:W^{\mathrm{A},\alpha }(x,0)=(1\alpha )U(x,0)+\alpha V(x,0),$$
(A.64)
which reduces to eq. (2.20), when $`\alpha =1`$. We set $`\delta m_W=0`$. The propagation of a static quark from the boundary of the SF to the bulk is given by the matrix $`H_\mathrm{h}`$ with a perturbative expansion
$$H_\mathrm{h}(x)=\underset{k=0}{\overset{\mathrm{}}{}}g_0^kH_\mathrm{h}^{(k)}(x).$$
(A.65)
The terms proportional to $`g_0^k`$ with $`k=0,1,2`$ are given by
$`H_\mathrm{h}^{(0)}(x)`$ $`=`$ $`P_+,`$ (A.66)
$`H_\mathrm{h}^{(1)}(x)`$ $`=`$ $`\frac{a}{L^3}{\displaystyle \underset{𝐩}{}}{\displaystyle \underset{u_0=a}{\overset{x_0}{}}}\mathrm{e}^{\mathrm{i}\mathrm{𝐩𝐱}}{\displaystyle \underset{i=1}{\overset{6}{}}}h_i^{(1)}(𝐩)\stackrel{~}{q}_{\mu (i)}^a(u_0a+as(i);𝐩)T^aP_+,`$ (A.67)
$`H_\mathrm{h}^{(2)}(x)`$ $`=`$ $`H_\mathrm{h}^{(2)}(x)_1+H_\mathrm{h}^{(2)}(x)_2,`$ (A.68)
$`H_\mathrm{h}^{(2)}(x)_1`$ $`=`$ $`\frac{a^2}{L^6}{\displaystyle \underset{𝐩,𝐪}{}}{\displaystyle \underset{u_0=a}{\overset{x_0}{}}}{\displaystyle \underset{v_0=a}{\overset{u_0a}{}}}\mathrm{e}^{\mathrm{i}(𝐩+𝐪)𝐱}{\displaystyle \underset{i=0}{\overset{6}{}}}h_i^{(1)}(𝐩)\stackrel{~}{q}_{\mu (i)}^a(u_0a+as(i);𝐩)`$ (A.70)
$`{\displaystyle \underset{j=0}{\overset{6}{}}}h_j^{(1)}(𝐪)\stackrel{~}{q}_{\mu (j)}^b(v_0a+as(j);𝐪)T^aT^bP_+,`$
$`H_\mathrm{h}^{(2)}(x)_2`$ $`=`$ $`\frac{a^2}{2L^6}{\displaystyle \underset{𝐩,𝐪}{}}{\displaystyle \underset{u_0=a}{\overset{x_0}{}}}\mathrm{e}^{\mathrm{i}(𝐩+𝐪)𝐱}{\displaystyle \underset{i=0}{\overset{15}{}}}h_i^{(2)}(𝐩,𝐪)\stackrel{~}{q}_{\mu (i)}^a(u_0a+as(i);𝐩)`$ (A.72)
$`\stackrel{~}{q}_{\nu (i)}^b(u_0a+at(i);𝐪)T^aT^bP_+,`$
where the gluon field in the time momentum representation is defined by
$`q_0(x)`$ $`=`$ $`\frac{1}{L^3}{\displaystyle \underset{𝐩}{}}\mathrm{e}^{i\mathrm{𝐩𝐱}}\stackrel{~}{q}_0(x_0,𝐩),`$ (A.73)
$`q_k(x)`$ $`=`$ $`\frac{1}{L^3}{\displaystyle \underset{𝐩}{}}\mathrm{e}^{i(\mathrm{𝐩𝐱}+ap_k/2)}\stackrel{~}{q}_k(x_0,𝐩)\mathrm{with}k=1,2,3.`$ (A.74)
The vertex functions $`h_i^{(1)}`$ and $`h_i^{(2)}`$ are listed in tables 4 and 5. They are linear in $`\alpha `$ and reduce to the ones for the EH action for $`\alpha =0`$. It is thus sufficient to give results for $`\alpha =1`$, to which we specialize from now on.
### A.2 Self energy
Here we compute, at one loop order, the linearly divergent contributions to the static propagator and the potential at zero distance for the action $`S_\mathrm{h}^\mathrm{A}`$. We expand the correlation function $`f_1^{\mathrm{hh}}(x_3)`$ defined in eq. (3.39)
$$\frac{1}{N_c}f_1^{\mathrm{hh}}(x_3)=1+g_0^2f_1^{\mathrm{hh},(1)}(x_3,\frac{a}{L})+\mathrm{}.$$
(A.75)
The Feynman diagrams that contribute to $`f_1^{\mathrm{hh},(1)}`$ are given in fig. B.1 of \[?\]. We then define
$$e^{(1)}=\underset{\frac{a}{L}0}{lim}f_1^{\mathrm{hh},(1)}(x_3,\frac{a}{L})\frac{a}{2L}.$$
(A.76)
As explained in sec. 2.1 the relevant quantity for the signal to noise ratio is $`E_{\mathrm{self}}V(0)/2`$. In order to compute the divergent contribution of $`V(0)`$ we introduce
$$\widehat{f}_1^{\mathrm{hh}}(𝐱)=\frac{a^6}{2}\underset{𝐲,𝐳}{}\overline{\zeta }_\mathrm{h}^{}(𝐱)\gamma _5\zeta _\mathrm{h}^{}(\mathrm{𝟎})\overline{\zeta }_\mathrm{h}(𝐲)\gamma _5\zeta _\mathrm{h}(𝐳),$$
(A.77)
which is related to $`f_1^{\mathrm{hh}}`$ via $`f_1^{\mathrm{hh}}(x_3)=(a/L)^2_{x_1,x_2}\widehat{f}_1^{\mathrm{hh}}(𝐱)`$. With the by now familiar notation for the one-loop coefficients, we then define
$$\delta V^{(1)}(0)=\underset{\frac{a}{L}0}{lim}\widehat{f}_1^{\mathrm{hh},(1)}\left(𝐱=\mathrm{𝟎},\frac{a}{L}\right)\frac{a}{L}.$$
(A.78)
The combination
$$r^{(1)}=e^{(1)}\frac{\delta V^{(1)}(0)}{2}$$
(A.79)
is listed in table 1.
While the Feynman rules given above are valid for $`S_\mathrm{h}^\mathrm{A}`$, the result for $`r^{(1)}`$ is the same for the action $`S_\mathrm{h}^\mathrm{s}`$. These two actions differ only by the normalization term
$$t=\left[\frac{g_0^2}{5}+\left(\frac{1}{3}\text{tr}V^{}(x,0)V(x,0)\right)^{1/2}\right]^1,$$
(A.80)
multiplying the links $`W`$. This normalization factor is gauge invariant. Consequently, at one-loop order, the gluon fields which are obtained from the expansion of $`t`$, are connected only to themselves. There is no connected graph from $`t`$ to the rest of the variables. In other words, these are tadpole graphs, which factorize (at one-loop!) in the sense that
$`𝒪_\mathrm{h}_t=𝒪_\mathrm{h}_{t=1}\times \left[1+{\displaystyle \underset{x_0}{}}t_1(x_0)g_0^2\right].`$ (A.81)
Here we have written down the expression for $`𝒪_\mathrm{h}`$ which contains a single heavy quark, otherwise several terms $`t_1`$ would appear. The function $`t_1`$ is the result of the sum over the loop momentum of the tadpole graphs. Due to translation invariance in space it depends only on the time coordinate $`x_0`$. The sum over $`x_0`$ in eq. (A.81) extends over all timeslices over which the heavy quark propagates. If, for the simplicity of the argument, we assume that we also have translation invariance in time, $`t_1`$ does not depend on $`x_0`$ and is a pure (dimensionless) number (note that we do not have a factor $`a`$ in front of the sum). It is then evident that $`e^{(1)}`$ is shifted by an amount $`t_1`$, while $`\delta V^{(1)}(0)`$ is shifted by $`2t_1`$ and $`r^{(1)}`$ is unchanged as claimed.
As (at one-loop order and with periodic boundary conditions) the effect of $`t`$ can entirely be absorbed into a change of $`e^{(1)}`$, it is also clear that the improvement coefficients $`c_\mathrm{A}^{\mathrm{stat}},b_\mathrm{A}^{\mathrm{stat}}`$ are independent of $`t`$ and in particular they are identical for $`S_\mathrm{h}^\mathrm{A}`$ and $`S_\mathrm{h}^\mathrm{s}`$.
### A.3 Improvement coefficients
To compute the improvement coefficients of the static-light axial current we make a one loop perturbative expansion of a correlator involving the static-light O($`a`$) improved axial current
$$f_{\mathrm{A},\mathrm{I}}^{\mathrm{stat}}(x_0)=f_\mathrm{A}^{\mathrm{stat}}(x_0)+ac_\mathrm{A}^{\mathrm{stat}}f_{\delta \mathrm{A}}^{\mathrm{stat}}(x_0).$$
(A.82)
The perturbative expansion reads
$$f_\mathrm{A}^{\mathrm{stat}}(x_0)=\underset{k=0}{\overset{\mathrm{}}{}}g_0^{2k}f_\mathrm{A}^{\mathrm{stat},(k)}(x_0),$$
(A.83)
and the Feynman diagrams contributing to the one loop term $`f_\mathrm{A}^{\mathrm{stat},(1)}`$ are summarized in figure B.1 of \[?\].
In the following, if not explicitly indicated, we will insert the local operators always at $`x_0=T/2`$ and supress that argument.
### A.4 Determination of $`c_\mathrm{A}^{\mathrm{stat}}`$
The universality of the continuum limit gives us a handle to compute the improvement coefficient for a certain static action, once the improvement coefficient is known for another static action. This remark, given two actions with the same continuum limit, is valid in general and not restricted to the static case. It can be translated into the fact that the ratio
$$\rho _\mathrm{I}(S_\mathrm{h};\theta ,\theta ^{})=\frac{f_{\mathrm{A},\mathrm{I}}^{\mathrm{stat}}(S_h)|_\theta }{f_{\mathrm{A},\mathrm{I}}^{\mathrm{stat}}(S_h)|_\theta ^{}},$$
(A.84)
is independent of the action used up to O($`a^2`$), once suitable values of the improvement coefficients are chosen. The basic idea is then to use the already known value of $`c_\mathrm{A}^{\mathrm{stat},(1)}`$ for the EH action \[?,?\] and to enforce
$$\rho _\mathrm{I}(S_\mathrm{h}^{\mathrm{EH}};\theta ,\theta ^{})=\rho _\mathrm{I}(S_\mathrm{h}^\mathrm{A};\theta ,\theta ^{}),$$
(A.85)
in order to compute $`c_\mathrm{A}^{\mathrm{stat},(1)}`$ for $`S_\mathrm{h}^\mathrm{A}`$. Different choices for $`\theta `$ and $`\theta ^{}`$ will give equivalent definitions for the improvement coefficient, differing only by cutoff effects $`\mathrm{O}(a/L)`$. The perturbative expansion of eq. (A.84) reads
$$\rho _\mathrm{I}(S_\mathrm{h};\theta ,\theta ^{})=\underset{k=0}{\overset{\mathrm{}}{}}g_0^{2k}\rho _\mathrm{I}^{(k)}(S_\mathrm{h};\theta ,\theta ^{}).$$
(A.86)
After some trivial algebra we obtain
$$\rho _\mathrm{I}^{(1)}(S_\mathrm{h};\theta ,\theta ^{})=\rho ^{(1)}(S_\mathrm{h};\theta ,\theta ^{})+ac_\mathrm{A}^{\mathrm{stat},(1)}(S_\mathrm{h})\left[\mathrm{\Delta }(\theta )\mathrm{\Delta }(\theta ^{})\right],$$
(A.87)
and
$$\rho ^{(1)}(S_\mathrm{h};\theta ,\theta ^{})=\frac{f_\mathrm{A}^{\mathrm{stat},(1)}(S_\mathrm{h})|_\theta }{f_\mathrm{A}^{\mathrm{stat},(0)}|_\theta }\frac{f_\mathrm{A}^{\mathrm{stat},(1)}(S_\mathrm{h})|_\theta ^{}}{f_\mathrm{A}^{\mathrm{stat},(0)}|_\theta ^{}},$$
(A.88)
where we use the fact that the correlator at tree-level does not depend on the static action, and
$$\mathrm{\Delta }(\theta )=\frac{f_{\delta \mathrm{A}}^{\mathrm{stat},(0)}|_\theta }{f_\mathrm{A}^{\mathrm{stat},(0)}|_\theta }.$$
(A.89)
An explicit formula for the desired improvement coefficient is
$`c_\mathrm{A}^{\mathrm{stat},(1)}(S_\mathrm{h}^\mathrm{A})`$ $`=`$ $`c_\mathrm{A}^{\mathrm{stat},(1)}(S_\mathrm{h}^{\mathrm{EH}})\underset{a/L0}{lim}\mathrm{\Delta }c`$
$`\mathrm{\Delta }c`$ $`=`$ $`{\displaystyle \frac{\rho ^{(1)}(S_\mathrm{h}^\mathrm{A};\theta ,\theta ^{})\rho ^{(1)}(S_\mathrm{h}^{\mathrm{EH}};\theta ,\theta ^{})}{\mathrm{\Delta }(\theta )\mathrm{\Delta }(\theta ^{})}}.`$
We have performed a one loop computation of $`\mathrm{\Delta }c`$ for $`\theta ,\theta ^{}\{0.0,0.5,1.0\}`$, $`T=L`$, and for resolutions $`L/a=4,6,\mathrm{},48`$. The bare light quark mass was set to the critical mass
$$m_\mathrm{c}=m_\mathrm{c}^{(1)}g_0^2+O(g_0^4),$$
(A.91)
with \[?\]
$$am_\mathrm{c}^{(1)}=0.2025565\times C_\mathrm{F}.$$
(A.92)
The input value
$$c_\mathrm{A}^{\mathrm{stat},(1)}(S_\mathrm{h}^{\mathrm{EH}})=\frac{1}{4\pi }\times 1.0351(1),$$
(A.93)
was taken from \[?\].
In figure 8 we plot $`\mathrm{\Delta }c`$ as a function of $`a/L`$ for the $`3`$ possible choices of ($`\theta `$,$`\theta ^{}`$). It is clear that the limit $`a/L0`$ can be taken with reasonable precision. The plot indicates also that defining the improvement coefficient without performing the continuum limit will give a perturbative O($`a/L`$) uncertainty of the order of $`0.02g_0^2`$. To obtain the final number eq. (3.30) we used both the extrapolation procedures of \[?\] and \[?\], finding consistent results.
### A.5 Determination of $`b_\mathrm{A}^{\mathrm{stat}}`$
To determine $`b_\mathrm{A}^{\mathrm{stat}}`$ we follow the strategy just applied to compute $`c_\mathrm{A}^{\mathrm{stat}}`$. The ratio
$$\stackrel{~}{\rho }_\mathrm{I}(S_\mathrm{h};m_\mathrm{q}^{},m_\mathrm{q})=\frac{(1+ab_\mathrm{A}^{\mathrm{stat}}(S_\mathrm{h})m_\mathrm{q}^{})f_{\mathrm{A},\mathrm{I}}^{\mathrm{stat}}(S_\mathrm{h})|_{m_\mathrm{q}^{}}}{(1+ab_\mathrm{A}^{\mathrm{stat}}(S_\mathrm{h})m_\mathrm{q})f_{\mathrm{A},\mathrm{I}}^{\mathrm{stat}}(S_\mathrm{h})|_{m_\mathrm{q}}},$$
(A.94)
has a well defined continuum limit. Once suitable values of the improvement coefficients are chosen, the continuum limit is approached with a rate of O($`a^2`$), as long as one considers (i) the quenched approximation or (ii) is interested in one-loop accuracy, only. These restrictions are necessary, since for full QCD and starting at order $`g_0^4`$, additional terms proportional to $`am_\mathrm{q}`$ with coefficients denoted by $`b_{\zeta _\mathrm{h}},b_g`$ (see \[?\]) are needed to cancel all $`\mathrm{O}(a)`$ effects (the term proportional to $`b_\zeta `$ cancels trivially in the above ratio).
We can hence require
$$\stackrel{~}{\rho }_\mathrm{I}(S_\mathrm{h}^{\mathrm{EH}};m_\mathrm{q}^{},m_\mathrm{q})=\stackrel{~}{\rho }_\mathrm{I}(S_\mathrm{h}^\mathrm{A};m_\mathrm{q}^{},m_\mathrm{q})+\mathrm{O}(g_0^4),$$
(A.95)
which yields immediately
$`b_\mathrm{A}^{\mathrm{stat},(1)}(S_\mathrm{h}^\mathrm{A})`$ $`=`$ $`b_\mathrm{A}^{\mathrm{stat},(1)}(S_\mathrm{h}^{\mathrm{EH}})+\underset{a/L0}{lim}\mathrm{\Delta }b,`$ (A.96)
$`\mathrm{\Delta }b`$ $`=`$ $`{\displaystyle \frac{1}{a(m_\mathrm{q}^{}m_\mathrm{q})}}\{[{\displaystyle \frac{f_{\mathrm{A},\mathrm{I}}^{\mathrm{stat},(1)}(S_\mathrm{h}^{\mathrm{EH}})|_{m_\mathrm{q}^{}}}{f_\mathrm{A}^{\mathrm{stat},(0)}(S_\mathrm{h}^{\mathrm{EH}})|_{m_\mathrm{q}^{}}}}{\displaystyle \frac{f_{\mathrm{A},\mathrm{I}}^{\mathrm{stat},(1)}(Sh^{\mathrm{EH}})|_{m_\mathrm{q}}}{f_\mathrm{A}^{\mathrm{stat},(0)}(S_\mathrm{h}^{\mathrm{EH}})|_{m_\mathrm{q}}}}]`$ (A.97)
$`[{\displaystyle \frac{f_{\mathrm{A},\mathrm{I}}^{\mathrm{stat},(1)}(S_\mathrm{h}^\mathrm{A})|_{m_\mathrm{q}^{}}}{f_\mathrm{A}^{\mathrm{stat},(0)}(S_\mathrm{h}^\mathrm{A})|_{m_\mathrm{q}^{}}}}{\displaystyle \frac{f_{\mathrm{A},\mathrm{I}}^{\mathrm{stat},(1)}(S_\mathrm{h}^\mathrm{A})|_{m_\mathrm{q}}}{f_\mathrm{A}^{\mathrm{stat},(0)}(S_\mathrm{h}^\mathrm{A})|_{m_\mathrm{q}}}}]\}.`$
In the numerical evaluation, we set $`m_\mathrm{q}=0`$ and choose $`m_\mathrm{q}^{}`$ such that
$$Lm_\mathrm{R}^{}=L(Z_\mathrm{m}m_\mathrm{q}^{}(1+b_\mathrm{m}am_\mathrm{q}^{}))=0.24.$$
(A.98)
Here, in contrast to the MC evaluation, $`m_\mathrm{R}^{}`$ has been defined in the lattice minimal subtraction scheme, where $`Z_\mathrm{m}=1+\frac{1}{2\pi ^2}\mathrm{ln}(L/a)g_0^2`$. The improvement constant $`b_\mathrm{m}=\frac{1}{2}0.07217C_\mathrm{F}g_0^2`$ is known from \[?\] and the input value of $`b_\mathrm{A}^{\mathrm{stat},(1)}`$ for the EH action
$$b_\mathrm{A}^{\mathrm{stat},(1)}(S_\mathrm{h}^{\mathrm{EH}})=0.056(7),$$
(A.99)
was computed in \[?\].
In figure 9 we plot $`\mathrm{\Delta }b`$ as a function of $`a/L`$ for the 3 possible choices of $`\theta `$. Again the O($`a/L`$) effects are well under control and almost independent of the choice of $`\theta `$. Extrapolating $`a/L0`$ as above yields the result eq. (3.34).
## Appendix B Approximate one-link integral
The basic idea of the one-link integral is that (in the pure gauge theory) correlation functions of Polyakov loops remain unchanged if the link variable $`U(x,0)`$ is replaced by
$$\overline{U}(x,0)=\frac{dUU\mathrm{exp}\left(12/g_0^2\mathrm{Re}\text{tr}[UV^{}(x,0)]\right)}{dU\mathrm{exp}\left(12/g_0^2\mathrm{Re}\text{tr}[UV^{}(x,0)]\right)},$$
(B.100)
whereas the variance (and therefore the error) is substantially reduced. While $`\overline{U}(x,0)`$ would thus be a possible choice for $`W(x,0)`$, its expression is rather unhandy for gauge groups SU($`N`$) with $`N>2`$, \[?,?\]. A more convenient approximation can be inferred form the SU($`2`$) case, where the group integral is proportional to $`V(x,0)`$ and takes the form
$$\overline{U}(x,0)=\frac{V}{v}\frac{I_2(24v/g_0^2)}{I_1(24v/g_0^2)},v=\left\{\frac{1}{N}\text{tr}VV^{}\right\}^{1/2},VV(x,0),$$
(B.101)
in terms of the Bessel functions $`I_1`$ and $`I_2`$. For SU($`3`$) we make a similar ansatz
$$W(x,0)=g_0^2V(x,0)s(g_0^2v).$$
(B.102)
Although this can not reproduce the SU($`3`$) one-link integral exactly, since that is a more complicated function of $`V`$ and $`V^{}`$, it turns out that the simple form
$$s(y)=k/(y+1/5),$$
(B.103)
with $`k=1.092`$ comes very close and thus is a good candidate to define an action with reduced noise and roughly unchanged signal with respect to the Eichten-Hill action. The form of $`s(y)`$ has been obtained numerically by minimizing the difference between the norm of the links $`\overline{U}(x,0)`$ (constructed by a multihit procedure) and the norm of the links in eq. (B.102) with $`s(y)`$ replaced by some trial functions of $`y`$. For this tuning we used a set of quenched configurations with SF boundary conditions on an $`8^4`$ lattice at $`\beta =6`$ and $`\beta =6.5`$. <sup>3</sup><sup>3</sup>3The actual value of $`k`$ is irrelevant for physics as it only influences the self energy. We chose $`k=1`$ in eq. (2.21).
## Appendix C Simulation parameters
We list in table 6 the lattice sizes and the bare parameters of our simulations. The angle $`\theta `$ defines the periodicity of the fermionic fields in all the spatial directions. We set $`T/L=1`$ and choose the background field to vanish.
The first $`\kappa `$ value always corresponds to $`\kappa _\mathrm{c}`$, where the PCAC mass vanishes \[?,?\], while the second one for the runs I to IV has been chosen such that $`Lm_\mathrm{R}=0.24`$ with $`m_\mathrm{R}`$ the running quark mass in the SF scheme at the scale $`\mu ^1=2L_{\mathrm{max}}=1.436r_0`$:
$$m_\mathrm{R}=Z_m\stackrel{~}{m}_\mathrm{q},\stackrel{~}{m}_\mathrm{q}=m_\mathrm{q}(1+b_\mathrm{m}am_\mathrm{q}),am_\mathrm{q}=\frac{1}{2}\left(\frac{1}{\kappa }\frac{1}{\kappa _\mathrm{c}}\right).$$
(C.104)
The low energy scale $`r_0`$ ($`0.5`$ fm) has been introduced in \[?\] and in \[?\] the ratio $`r_0/a`$ has been computed to a precision of about $`0.5\%`$ in a range of $`\beta `$ values which includes runs I to IV in table 6. For the coefficients $`Z_m`$ and $`b_\mathrm{m}`$ in eq. (C.104) we have used the results in \[?\] and the improvement coefficients $`c_\mathrm{t},\stackrel{~}{c}_\mathrm{t}`$ were set to their two-loop and one-loop values respectively \[?,?,?\].
|
warning/0506/quant-ph0506198.html
|
ar5iv
|
text
|
# Spin transport and quasi 2D architectures for donor-based quantum computing
## Abstract
Through the introduction of a new electron spin transport mechanism, a 2D donor electron spin quantum computer architecture is proposed. This design addresses major technical issues in the original Kane design, including spatial oscillations in the exchange coupling strength and cross-talk in gate control. It is also expected that the introduction of a degree of non-locality in qubit gates will significantly improve the scaling fault-tolerant threshold over the nearest-neighbour linear array.
The Kane paradigm of donor nuclear spin quantum computing in silicon Kane , based on single atom placement fabrication techniques Schofield ; Jamieson , is an important realization of Feynman’s original concept of nanotechnology in the solid-state. Variations on this theme include electron spin qubits Vrijen ; DeSousa ; Hill and charge qubits charge\_qubit . There are significant advantages of the donor spin as a qubit, including uniformity of the confinement potential and high number of gate operations possible within the electron spin coherence time, measured to be in excess of 60ms Tyryshkin . Consequently, there is great interest in donor-based architectures and progress towards their fabrication Clark Review ; Schenkel ; Buehler .
It is often assumed that solid-state designs should be inherently scalable given the capabilities of semi-conductor device fabrication. In reality this weak-scalability argument should be replaced with a stronger version as scalability of a given architecture is considerably more complex than fabricating many interacting qubits. Fault-tolerant scale-up requires quantum error correction over concatenated logical qubits with all the attendant ancillas, syndrome measurements, and classical feed-forward processing. Both parallelism and communication must be optimised Steane . Only by considering such systems-level issues in conjunction with the underlying qubit physics will the requirements of quantum computation in a given implementation be understood, and new concepts generated. In this paper we introduce a new mechanism for coherent donor electron spin state transport, and in a similar design path to the QCCD ion trap proposal Kielpinsky , we construct a 2D donor architecture based on distinct qubit storage and interaction regions.
The significant interest in scaling up the donor-based solid-state designs, has led to a number of works considering these scalability issues. As a result, several serious problems have been identified, including: sensitivity of the exchange interaction and control to qubit placement (at the 2-3 lattice site level) KHD ; Wellard\_Josc ; Kane2 , qubit control and fabrication limitations associated with high gate densities Oskin ; Copsey1 , spin readout based on spin-charge transduction Kane ; LHFIR , and the communication bottlenecks for linear nearest neighbour (LNN) qubit arrays Oskin ; Metodiev .
The issue of local versus non-local fault-tolerant operation is non-trivial Gottesman ; Svore . A recent surprising result is that Shor’s algorithm can be implemented on a LNN circuit for the minimal qubit case with no increase at leading order in the circuit gate count or depth FowlerLNNShor ; Devitt . However, at the systems level one expects a linear nearest neighbour qubit array to suffer from swap gate overheads, particularly when concatenated qubit encoding is employed. The general analysis in Svore shows that locality forces the threshold down inversely with the physical encoding scale. Recently, the extent of the LNN penalty has been estimated to bring the threshold down by two orders of magnitude compared to the non-local case Szkopek .
For the Kane, or related donor based architectures, all of the above implies the imperative of finding ways of traversing the linear array constraints, as the most effective way to improve the threshold and tackle the technical problems listed. An important step in this direction is the proposal for sub-interfacial transport of electrons in a one dimensional array Skinner . This design has many desirable features, digitising the single and two qubit gate problems in an elegant way, but also has problems with scalability due to the relative closeness of gates Oskin .
The 2D architecture introduced here requires relatively low gate densities and specifically address the problems listed above. In Fig. 1 the geometry is shown for the specific case of the exchange-interaction based Kane architecture. We note that the transport ideas presented here allow for a similar, but non-trivial development for the digital-Kane case. A buried array of ionised donors provide pathways for coherent transport of electron spins for in-plane horizontal and vertical shuttling (dashed-border sections) of qubit states into and out of the interaction zone. The overall gate density is low compared to the Kane case, and can be further reduced by increasing the transport pathway length (Fig. 2). Initially all gates inhibit tunnelling along any given channel. Coherent spin transport along one segment is achieved by adiabatically lowering the barriers in a well defined sequence to effect coherent transfer by adiabatic passage (CTAP) without populating the intervening channel donors GreentreeCTAP . We show that with appropriate donor separations, the shuttling time can be in the nanosecond range for one section. In Fig. 1 the coherent transport scheme is defined for the minimum number of donors. Higher order schemes with more donors reduces the gate density (see Fig. 2).
Logic gates are carried out in interaction zones distinct from qubit storage regions – shown in Fig. 1 are the cannonical A and J gates for electron spin based qubit control at the microsecond level Hill . After mandatory precision characterisation SchirmerOi ; Cole , interaction regions with unacceptably low couplings can be identified and bypassed in the circuit flow, thereby avoiding bottleneck issues arising from the sensitivity of the exchange interaction to donor placement. This design allows for new variations on the theme, e.g. digitisation of hyperfine control Skinner , or introduction of local buried B-field antennae structures LidarBwires , and space for SET readout techniques Kane ; LHFIR ; GreentreeTriple .
A schematic of the minimal three donor transport pathway is given in Fig. 2. The triple-well system $`|1\sigma `$, $`|2\sigma `$, $`|3\sigma `$ ($`\sigma =,`$) facilitates coherent state transport from $`\alpha |1+\beta |1`$ to $`\alpha |3+\beta |3`$ without populating the $`|2\sigma `$ states. Techniques for coherent transfer by adiabatic passage are well known Vitanov , and for the donor system was proposed in GreentreeCTAP for the case of charge transfer. A superconducting version of the three state case has also been proposed Siewert . The system is controlled by shift gates, $`S`$, which can modify the energy levels of the end donors, and barrier gates, $`B_{i,i+1}`$ which control the tunnelling rate $`\mathrm{\Omega }_{i,i+1}`$ between donors $`i`$ and $`i+1`$.
Although the scheme we introduce here necessarily includes spin, we first consider the zero field case and ignore spin degrees of freedom GreentreeCTAP to illustrate the principles of CTAP in the one-electron three-donor system, 3D<sup>2+</sup>. The effective Hamiltonian for the 3D<sup>2+</sup> system is:
$$=\mathrm{\Delta }|22|\mathrm{}(\mathrm{\Omega }_{12}|12|+\mathrm{\Omega }_{23}|23|+\mathrm{h}.\mathrm{c}.),$$
(1)
where $`\mathrm{\Omega }_{ij}=\mathrm{\Omega }_{ij}(t)`$ is the coherent tunnelling rate between donors $`|i`$ and $`|j`$ and $`\mathrm{\Delta }=E_2E_1=E_2E_3`$. The eigenstates of $``$ (with energies $`_\pm `$ and $`_0`$) are
$`|𝒟_+`$ $`=`$ $`\mathrm{sin}\mathrm{\Theta }_1\mathrm{sin}\mathrm{\Theta }_2|1+\mathrm{cos}\mathrm{\Theta }_2|2+\mathrm{cos}\mathrm{\Theta }_1\mathrm{sin}\mathrm{\Theta }_2|3,`$
$`|𝒟_{}`$ $`=`$ $`\mathrm{sin}\mathrm{\Theta }_1\mathrm{cos}\mathrm{\Theta }_2|1\mathrm{sin}\mathrm{\Theta }_2|2+\mathrm{cos}\mathrm{\Theta }_1\mathrm{cos}\mathrm{\Theta }_2|3,`$
$`|𝒟_0`$ $`=`$ $`\mathrm{cos}\mathrm{\Theta }_1|1\mathrm{sin}\mathrm{\Theta }_1|3,`$ (2)
where we have introduced $`\mathrm{\Theta }_1=\mathrm{arctan}\left(\mathrm{\Omega }_{12}/\mathrm{\Omega }_{23}\right)`$ and $`\mathrm{\Theta }_2=\mathrm{arctan}[2\mathrm{}\sqrt{(\mathrm{\Omega }_{12})^2+(\mathrm{\Omega }_{23})^2}/\mathrm{\Delta }]/2`$. Transfer from state $`|1`$ to $`|3`$ is achieved by maintaining the system in state $`|𝒟_0`$ and changing the characteristics of $`|𝒟_0`$ adiabatically ($`|_0_\pm ||\dot{𝒟}_0|𝒟_\pm |`$) from $`|1`$ at $`t=0`$ to $`|3`$ at $`t=t_{\mathrm{max}}`$ by appropriate control of the tunnelling rates, without population leakage into the other eigenstates.
For the case of coherent spin transport we write the 3D<sup>2+</sup> Hamiltonian in terms of spin/site operators as:
$$=\underset{i=1}{\overset{3}{}}\underset{\sigma =,}{}E_{i\sigma }c_{i\sigma }^{}c_{i\sigma }+\underset{<ij>}{}\underset{\sigma =,}{}\mathrm{\Omega }_{ij}(t)c_{j\sigma }^{}c_{i\sigma }$$
(3)
and numerically solve for the density matrix, $`\rho (t)`$, in the presence of a (dominant) charge dephasing rate $`\mathrm{\Gamma }`$, assumed to act equally on all coherences. Without attempting to fully optimize control we apply Gaussian pulses of the form $`\mathrm{\Omega }_{ij}(t)=\mathrm{\Omega }_{ij}^{\mathrm{max}}\mathrm{exp}\left[(tt_{ij})^2/(2w_{ij}^2)\right]`$, where $`t_{ij}`$ and $`w_{ij}`$ are the peak time and width of the control pulse modulating the tunnelling rate between position states $`|i`$ and $`|j`$. To simplify matters for initial simulations we set the maximum tunnelling rates and standard deviations for each transition to be equal, i.e. $`\mathrm{\Omega }_{ij}^{\mathrm{max}}=\mathrm{\Omega }^{\mathrm{max}}`$ and $`w_{ij}=w`$, and set $`\mathrm{\Delta }=0`$ (these conditions can be relaxed with no effect on the conclusions of this paper). Transfer is then optimized when the width of the pulses equals the time delay between the pulses Gaubatz . With total pulse time $`t_{\mathrm{max}}`$, we choose $`w=t_{max}/8`$ so that $`t_{12}=(t_{\mathrm{max}}+w)/2`$ and $`t_{23}=(t_{max}w)/2`$. This ordering, where $`\mathrm{\Omega }_{23}`$ is applied before $`\mathrm{\Omega }_{12}`$ is known as the counter-intuitive pulse sequence and has significant advantages in improving transfer fidelity over other pulse sequences GreentreeCTAP . In Fig. 3 we present results showing transport using the counter-intuitive pulse ordering for a spin superposition (phases relative to the untransported state).
Generally, when the adiabaticity criterion is satisfied and the transport time is at least an order of magnitude faster than charge dephasing, the transport fidelity is high. These results are consistent with those of Ivanov et al Ivanov who considered the role of dephasing in three-state Stimulated Raman Adiabatic Passage (STIRAP). Although these competing timescales are essentially unmeasured at present, estimates Barrett ; Fedichkin for the P-P<sup>+</sup> charge dephasing time are of order 10ns and a value of 220ns was reported recently for a Si:P double-dot Gorman , whereas sub-nanosecond tunnelling times are possible due to the strong confining potential of donor nuclei. The CTAP transport time will be defined primarily by the gate-assisted tunnelling rate, which we calculate as follows. Using the TCAD package we compute the potential due to a surface B-gate bias and determine the donor electron wave function in an effective mass basis, e.g. $`F_{\pm z}^{n,l,m}(𝐫)=\phi _{n,l,m}(x,y,\gamma z)`$, about the six band minima where the $`\phi _{n,l,m}`$ are hydrogenic orbitals with Bohr radius $`a_{}`$, and $`\gamma =a_{}/a_{}`$. Diagonalising the total Hamiltonian of the system, using pseudopotentials to describe the silicon bandstructure, we obtain a generalised Kohn-Luttinger wave function:
$$\psi (𝐫,V)=\underset{n,l,m}{}c_{n,l,m}(V)\underset{\mu =1}{\overset{6}{}}F_\mu ^{n,l,m}(𝐫)\mathrm{e}^{i𝐤_\mu .𝐫}u_{𝐤_\mu }(𝐫),$$
(4)
where the Bloch states are $`u_{𝐤_\mu }(𝐫)=_𝐆A_{𝐤_\mu }(𝐆)e^{i𝐆𝐤_\mu }`$. We form bonding and anti-bonding states $`\mathrm{\Psi }_\pm (𝐫,V)=𝒩(\psi _L(𝐫,V)\pm \psi _R(𝐫,V))`$, normalised by $`𝒩`$, and compute the gap as shown in Fig. 4 for basis sizes 55 and 140 ($`n_{\mathrm{max}}`$ = 5 and 7). Comparison of the non-linear regions indicates that the range of validity is $`|V_b|`$ 200 mV.
In contrast to what one expects for an isolated P-P<sup>+</sup> system in vacuum where the nodal structure of the bonding and anti-bonding states is simple, the non-trivial nodal properties of the donor electron wave function and the proximity of the oxide interface complicates the tunnelling control. These calculations directly extend similar effects noted in the ungated P-P<sup>+</sup> system Hu\_Charge\_Qubit . From Fig. 4 we see that for this configuration the tunnelling rate can be varied from zero at +100mV to $``$ 10 GHz at -200mV, giving a gate assisted tunnelling time of 60 ps.
Based on this value, CTAP simulations for 5, 7 and 9 donor chains are presented in Fig. 5. The adiabatic nature of the transport scheme provides an inherent robustness, as evidenced in Fig. 5, which shows a remarkable uniformity in the response to charge dephasing for the different path lengths once the adiabatic regime is reached. Another consequence is that inevitable variations in tunnelling rates due to donor placement Hu\_Charge\_Qubit will not affect the viability of the scheme, as further simulations have explicitly verified. The extent to which $`\mathrm{\Gamma }`$ controls the transport fidelity is also clear, although we note that there is room for improvement through optimisation of control pulses and minimisation of charge fluctuations through fabrication development. Non-zero transport errors may require monitoring mechanisms for heralding successful transport, or an error correction protocol for transport loss. As intrinsic spin-orbit coupling for donor states in silicon is very low, dephasing of donor electron spin is dominated by spectral diffusion due to spin impurities and is mitigated by isotopic purification DeSousa2 . For the bound state spin-orbit coupling, at $`V_B200`$ mV we calculate from Eqn(4) the non-S components to be $`_{n,l>0,m}|c_{n,l,m}(V)|^2<10^4`$ indicating that the deviation from the S sector is minimal. Together with the zero occupation of channel states, this suggests that charge dephasing will have a negligible second order effect on the spin coherence during transport. Decoupling of orbital and spin sectors has already given rise to demonstrations of coherent transport of electron spins over 100$`\mu `$m Kikkawa .
The basic layout of 2D donor arrays, with storage regions, vertical and horizontal transport pathways and interaction zones, allows us to explicitly consider designs for fault-tolerant operation. For example, we can arrange the logical qubit groups and ancillas so that the transport rails allow for non-local intralogical qubit interactions (qubit – ancillas) and LNN interlogical interactions. With inherent parallelism of operation, interlogical gates can then be applied transversally as required to implement fault-tolerant gates. Another possibility with less stringent fabrication requirements is a linear qubit storage with transport and interaction rails either side.
The optimum arrangement for fault-tolerant operation requires sophisticated systems level simulations Copsey2 to determine the best use of this medium range quantum transport capability, and the corresponding improvements on the LNN threshold. In any case, it is clear that the introduction of coherent spin transport to donor quantum computing allows us to address many problems in the Kane concept, and consider scalable fault-tolerant architectures with low gate densities, room for SET structures and control, and a bypass mechanism for low value exchange gates. One expects the realities of the silicon crystaline environment will necessitate the characterisation of transport pathways, however, the precision requirements of the adiabatic CTAP mechanism would be far less than the quantum gate threshold.
We thank S. Das Sarma, G. Milburn, F. Wilhelm and J. Cole for comments, and support of the Australian Research Council, the U.S. National Security Agency, Advanced Research and Development Activity and Army Research Office under contract DAAD19-01-1-0653.
|
warning/0506/hep-ph0506207.html
|
ar5iv
|
text
|
# Spin transfer and polarization of antihyperons in lepton induced reactions
## I Introduction
The polarizations of hyperons have been widely used to study various aspects of spin effects in high energy reactions, in particular, the spin-dependent fragmentation functions for their self spin-analyzing parity violating decay tdlee . One of the important aspects in this connection is the spin transfer in high energy fragmentation processes. Here, it is of particular interest to know whether SU(6) wave-function or the results drawn from polarized deeply inelastic lepton-nucleon scattering and related data should be used in connecting the polarization of the fragmenting quark and that of the produced hadrons. Clearly such study can provide us with useful information on hadronization mechanism and the spin structure of hadron.
Theoretical calculations of hyperon polarizations in different reactions have been carried out using different models GH93 ; BL98 ; Kot98 ; Florian98 ; LL00 ; LXL01 ; XLL02 ; LL02 ; Ma:1999wp ; Ma:1998pd ; BQMa01 ; Ma:2000cg ; Ma:2000uu ; Ellis2002 . The results show, in particular, that it is possible to differentiate different pictures by measuring the polarizations of different hyperons in $`e^+e^{}`$ annihilations, polarized deeply inelastic lepton-nucleon scattering at high energies, and in the high transverse momentum regions in polarized $`pp`$ collisions. They show also that the decay contributions to $`\mathrm{\Lambda }`$ polarization are usually high and should be taken into account in the calculations, and that the contamination from the fragmentation of remnant of target in deeply inelastic scattering at relatively low energies such as the CERN NOMAD or DESY HERMES energies are very important. Both of them have to be taken into account in comparing theoretical predictions with the experimental results.
Presently, data in this connection are already available for $`\mathrm{\Lambda }`$ polarization in $`e^+e^{}`$ annihilation at the $`Z^0`$ pole ALEPH96 ; OPAL98 , in deeply inelastic scattering using neutrino beam NOMAD00 and that using electron or muon beam HERMES ; E665 ; COMPASS . But the amount of data and the statistics of them are still not high enough to judge which picture is better. More complementary measurements are necessary and some of them are underway. In this connection, it is interesting to note that the NOMAD and COMPASS Collaboration at CERN has measured not only the polarization of $`\mathrm{\Lambda }`$ but also that of $`\overline{\mathrm{\Lambda }}`$ NOMAD00 ; COMPASS . Since the polarization of antihyperon in semi-inclusive deeply inelastic lepton-nucleon scattering is definitely more sensitive to the nucleon sea, it is thus instructive to make more detailed study in this direction not only to get more information on spin transfer in fragmentation but also on nucleon spin structure.
In this paper, we make calculations of the polarizations of antihyperons in lepton induced reactions such as $`e^+e^{}Z^0\overline{H}+X`$, $`e^{}+pe^{}+\overline{H}+X`$ and $`\nu _\mu +p\mu ^{}+\overline{H}+X`$. In Sec. II, we present the general formulas used in these calculations and summarize the key points of different models for spin transfer in fragmentation. We present the results for antihyperon polarizations in Sec. III and compare them with those for hyperons and available data. We give a short summary and an outlook in Sec. IV.
## II The calculation formulas
As have been shown in LL00 ; LXL01 ; XLL02 ; LL02 , to calculate hyperon polarization in high energy reactions, it is necessary to take also the decay contribution into account. In this section, we present the general formulas for such calculations with the contribution from decay. We first present the formulas for a pure quark fragmentation in Sec. II A, then the formulas for hyperon and antihyperon polarization in Sec. II B. We summarize the key points of a few models for spin transfer in fragmentation processes in Sec. II C.
### II.1 Calculation formulas for a pure quark fragmentation
We first present the formulas for a pure quark fragmentation which are the basis for different reactions. We present the formulas for $`q_fH+X`$, $`q_f\overline{H}+X`$, $`\overline{q}_fH+X`$ and $`\overline{q}_f\overline{H}+X`$, respectively, and the relations between them obtained from charge conjugation symmetry.
#### II.1.1 $`q_fH+X`$
Now we consider the fragmentation process $`q_fH+X`$, where we use the subscript $`f`$ to denote the flavor of the quark, $`H`$ to denote hyperon. We use $`D_f^H(z)`$ to denote the fragmentation function, which is defined as the number density of $`H`$ produced in the fragmentation of $`q_f`$, where $`z`$ is the fraction of momentum of $`q_f`$ carried away by $`H`$. We do not study the transverse momentum dependence in this paper so an integration over transverse momentum is understood. In general, if the contribution from decay is taken into account, $`D_f^H(z)`$ can be written as two parts, i.e.,
$$D_f^H(z)=D_f^H(z;\text{dir})+D_f^H(z;\text{dec}),$$
(1)
where $`D_f^H(z;\text{dir})`$ and $`D_f^H(z;\text{dec})`$ denote the directly produced part and the decay contribution respectively.
It is clear that the decay contribution can be calculated from the following convolution,
$$D_f^H(z;\text{dec})=\underset{j}{}𝑑z^{}K_{H,H_j}(z,z^{})D_f^{H_j}(z^{}),$$
(2)
where the kernel function $`K_{H,H_j}(z,z^{})`$ is the probability for $`H_j`$ with a fractional momentum $`z^{}`$ to decay into an $`H`$ with $`z`$ and anything. If we consider, as usual, only the $`J^P=(1/2)^+`$ octet and $`J^P=(3/2)^+`$ decuplet baryon production, most of the decay processes are two body decay. For an unpolarized two body decay $`H_jH+M`$, the kernel function $`K_{H,H_j}(z,z^{})`$ can be calculated easily. In this case, the magnitude of the momentum of the decay product in the rest frame of $`H_j`$ is fixed and it has to be isotropically distributed. By making a Lorentz transformation of this isotropic distribution to the moving frame of $`H_j`$, we obtain the result for $`K_{H,H_j}(z,z^{})`$ as given by
$$K_{H,H_j}(z,\stackrel{}{p}_i;z^{},\stackrel{}{p}_{,i}^{})=\frac{N}{E_j}Br(H_jH_iM)\delta (p_ip_jm_jE_i^{}),$$
(3)
where $`Br(H_jHM)`$ is the corresponding decay branching ratio, $`N`$ is a normalization constant, $`E_i^{}`$ is the energy of $`H_i`$ in the rest frame of $`H_j`$ which is a function of the masses $`m_j`$, $`m_i`$ and $`m_M`$ of $`H_j`$, $`H_i`$ and $`M`$.
Similarly, in polarized case, we have
$$\mathrm{\Delta }D_f^H(z)=\mathrm{\Delta }D_f^H(z;\text{dir})+\mathrm{\Delta }D_f^H(z;\text{dec}),$$
(4)
where $`\mathrm{\Delta }D_f^H(z)=D_f^H(z,+)\mathrm{\Delta }D_f^H(z,)`$, and the $`+`$ or $``$ denotes that the produced $`H`$ is polarized in the same or opposite direction as the initial quark $`q_f`$. $`\mathrm{\Delta }D_f^H(z;\text{dir})`$ and $`\mathrm{\Delta }D_f^H(z;\text{dec})`$ are the corresponding quantities for directly produced $`H`$ and decay contribution. For decay contribution, we have ft
$$\mathrm{\Delta }D_f^H(z;\text{dec})=\underset{j}{}𝑑z^{}t_{H,H_j}^DK_{H,H_j}(z,z^{})\mathrm{\Delta }D_f^{H_j}(z^{}),$$
(5)
where $`t_{H,H_j}^D`$ is the spin transfer factor for the decay process $`H_jH+M`$. $`t_{H,H_j}^D`$ is a constant which is independent of the process where $`H_j`$ is produced. It is completely determined by the decay process. For different decay processes, $`t_{H,H_j}^D`$ can be found, e.g., in Table II of LL00 .
The unknowns left now are $`D_f^H(z;\text{dir})`$ and $`\mathrm{\Delta }D_f^H(z;\text{dir})`$. They are determined by the hadronization mechanism and the structure of hadrons. Presently, we can calculate $`D_f^H(z;\text{dir})`$ using a hadronization model or using a parametrization of fragmentation functions. For $`\mathrm{\Delta }D_f^H(z;\text{dir})`$, there are also different models in literature GH93 ; BL98 ; Kot98 ; Florian98 ; LL00 ; LXL01 ; XLL02 ; LL02 ; Ma:1999wp ; Ma:1998pd ; BQMa01 ; Ma:2000cg ; Ma:2000uu . We will briefly summarize a few of them in Sec. II C.
#### II.1.2 $`\overline{q}_fH+X`$
For $`\overline{q}_fH+X`$, we have
$$D_{\overline{f}}^H(z)=D_{\overline{f}}^H(z;\text{dir})+D_{\overline{f}}^H(z;\text{dec}),$$
(6)
$$D_{\overline{f}}^H(z;\text{dec})=\underset{j}{}𝑑z^{}K_{H,H_j}(z,z^{})D_{\overline{f}}^{H_j}(z^{}),$$
(7)
$$\mathrm{\Delta }D_{\overline{f}}^H(z)=\mathrm{\Delta }D_{\overline{f}}^H(z;\text{dir})+\mathrm{\Delta }D_{\overline{f}}^H(z;\text{dec}),$$
(8)
$$\mathrm{\Delta }D_{\overline{f}}^H(z;\text{dec})=\underset{j}{}𝑑z^{}t_{H,H_j}^DK_{H,H_j}(z,z^{})\mathrm{\Delta }D_{\overline{f}}^{H_j}(z^{}).$$
(9)
#### II.1.3 $`q_f\overline{H}+X`$
For $`q_f\overline{H}+X`$, we have
$$D_f^{\overline{H}}(z)=D_f^{\overline{H}}(z;\text{dir})+D_f^{\overline{H}}(z;\text{dec}),$$
(10)
$$D_f^{\overline{H}}(z;\text{dec})=\underset{j}{}𝑑z^{}K_{H,H_j}(z,z^{})D_f^{\overline{H}_j}(z^{}),$$
(11)
$$\mathrm{\Delta }D_f^{\overline{H}}(z)=\mathrm{\Delta }D_f^{\overline{H}}(z;\text{dir})+\mathrm{\Delta }D_f^{\overline{H}}(z;\text{dec}),$$
(12)
$$\mathrm{\Delta }D_f^{\overline{H}}(z;\text{dec})=\underset{j}{}𝑑z^{}t_{H,H_j}^DK_{H,H_j}(z,z^{})\mathrm{\Delta }D_f^{\overline{H}_j}(z^{}).$$
(13)
Here we assume that the charge conjugation symmetry is applicable to the decay process so that the relations $`K_{\overline{H},\overline{H}_j}(z,z^{})=K_{H,H_j}(z,z^{})`$ and $`t_{\overline{H},\overline{H}_j}^D=t_{H,H_j}^D`$ are valid. We assume also the validity of the charge conjugation symmetry for the fragmentation functions and obtain further the following relations:
$$D_f^{\overline{H}}(z;i)=D_{\overline{f}}^H(z;i),$$
(14)
$$\mathrm{\Delta }D_f^{\overline{H}}(z;i)=\mathrm{\Delta }D_{\overline{f}}^H(z;i),$$
(15)
for total, the direct ($`i=dir`$) and decay ($`i=dec`$) parts respectively.
#### II.1.4 $`\overline{q}_f\overline{H}+X`$
For $`\overline{q}_f\overline{H}+X`$, we use charge conjugation symmetry and obtain that
$$D_{\overline{f}}^{\overline{H}}(z;i)=D_f^H(z;i),$$
(16)
$$\mathrm{\Delta }D_{\overline{f}}^{\overline{H}}(z;i)=\mathrm{\Delta }D_f^H(z;i),$$
(17)
for total, the direct ($`i=dir`$) and decay ($`i=dec`$) parts respectively. Hence, in the following, we need only to write out the formulas for $`q_fH+X`$ and $`\overline{q}_fH+X`$. Those for the other two cases are obtained from charge conjugation symmetry.
Independent of the models, we expect the following qualitative features for $`D_f^H(z;\text{dir})`$ and $`\mathrm{\Delta }D_f^H(z;\text{dir})`$. Since hadrons containing the initial quark, i.e. the first rank hadrons in Feynman-Field type of cascade fragmentation models, usually carry a large fraction of momentum of the initial quark, we expect that, for large $`z`$
$$D_{q_f}^H(z;\text{dir})D_{\overline{q}_f}^H(z;\text{dir}),$$
(18)
$$|\mathrm{\Delta }D_{q_f}^H(z;\text{dir})||\mathrm{\Delta }D_{\overline{q}_f}^H(z;\text{dir})|.$$
(19)
But for small $`z`$, they can be comparable. Since $`P_H`$ due to spin transfer is expected to be significant for large $`z`$, we should see that
$$|P_H^{q_fHX}(z)||P_H^{\overline{q}_fHX}(z)|.$$
(20)
This implies that the qualitative behavior of $`P_H`$ in a given reaction is determined mainly by quark fragmentation and that of $`P_{\overline{H}}`$ is determined mainly by anti-quark fragmentation.
### II.2 Polarization of hyperon or antihyperon in lepton induced reactions
To calculate the polarization of hyperon or antihyperon in a given reaction, we need to sum over the contributions from the fragmentation of quarks and anti-quarks of different flavors. For example, for $`A+BH+X`$, we have
$$\frac{d\sigma }{dx_F}(ABHX)=\underset{f}{}[\frac{d\widehat{\sigma }}{d\alpha }(ABq_fX)D_f^H(z)+\frac{d\widehat{\sigma }}{d\alpha }(AB\overline{q}_fX)D_{\overline{f}}^H(z)],$$
(21)
$$\frac{d\mathrm{\Delta }\sigma }{dx_F}(ABHX)=\underset{f}{}[\frac{d\widehat{\sigma }}{d\alpha }(ABq_fX)P_f(\alpha )\mathrm{\Delta }D_f^H(z)+\frac{d\widehat{\sigma }}{d\alpha }(AB\overline{q}_fX)P_{\overline{f}}(\alpha )\mathrm{\Delta }D_{\overline{f}}^H(z)],$$
(22)
where $`x_F`$ is the fractional momentum carried by the produced $`H`$; $`\widehat{\sigma }`$ denotes the cross section for the production of $`q_f`$ or $`\overline{q}_f`$ in $`A+B`$ collisions and $`\alpha `$ denotes the kinematic variables describing cross section; $`P_f(\alpha )(\mathrm{\Delta }d\widehat{\sigma }/d\alpha )/(d\widehat{\sigma }/d\alpha )`$ is the polarization of $`q_f`$, and similar for $`P_{\overline{f}}`$. In general, they can be dependent on some kinematic variables hence we use $``$ to denote convolutions. Or equivalently, we have,
$$N(x_F,H)=\underset{f}{}[R_f(\alpha )D_f^H(z)+R_{\overline{f}}(\alpha )D_{\overline{f}}^H(z)],$$
(23)
$$\mathrm{\Delta }N(x_F,H)=\underset{f}{}[R_f(\alpha )P_f(\alpha )\mathrm{\Delta }D_f^H(z)+R_{\overline{f}}(\alpha )P_{\overline{f}}(\alpha )\mathrm{\Delta }D_{\overline{f}}^H(z)],$$
(24)
$$N(x_F,\overline{H})=\underset{f}{}[R_f(\alpha )D_{\overline{f}}^H(z)+R_{\overline{f}}(\alpha )D_f^H(z)],$$
(25)
$$\mathrm{\Delta }N(x_F,\overline{H})=\underset{f}{}R_f(\alpha )P_f(\alpha )\mathrm{\Delta }D_{\overline{f}}^H(z)+R_{\overline{f}}(\alpha )P_{\overline{f}}(\alpha )\mathrm{\Delta }D_f^H(z)],$$
(26)
where we use $`N(x_F,H)`$ and $`N(x_F,\overline{H})`$ to denote the number density of $`H`$ and that of $`\overline{H}`$ at a given momentum fraction $`x_F`$ produced in the reaction respectively, $`\mathrm{\Delta }N(x_F,H)`$ and $`\mathrm{\Delta }N(x_F,\overline{H})`$ to denote the corresponding differences in the polarized case. $`P_f`$ and $`P_{\overline{f}}`$ are respectively the polarization of $`q_f`$ and that of $`\overline{q}_f`$; $`R_f`$ and $`R_{\overline{f}}`$ are the fractional contributions of $`q_fh+X`$ and $`\overline{q}_fh+X`$ to the whole hadronic events. They are related to $`d\widehat{\sigma }`$ by
$$R_f(\alpha )=\frac{1}{\sigma _{\text{inel}}}\frac{d\widehat{\sigma }}{d\alpha }(ABq_fX),$$
(27)
where $`\sigma _{\text{inel}}`$ is the total inelastic cross section for hadron production in $`A+B`$ collisions. The final result for the polarization is given by,
$$P_H(x_F)=\frac{\mathrm{\Delta }N(x_F,H)}{N(x_F,H)}.$$
(28)
$$P_{\overline{H}}(x_F)=\frac{\mathrm{\Delta }N(x_F,\overline{H})}{N(x_F,\overline{H})}.$$
(29)
For different reactions, the fragmentation functions $`D`$’s and $`\mathrm{\Delta }D`$’s are assumed to be universal, but the results of $`R_f`$, $`R_{\overline{f}}`$, $`P_f`$ and $`P_{\overline{f}}`$ are different. These differences lead to different results for the polarization of hyperons and/or antihyperons. In the following, we summarize the formulas in different lepton induced reactions, respectively, and discuss the qualitative features of these results. The numerical results are given in Sec. III.
#### II.2.1 $`e^+e^{}H(\mathrm{or}\overline{H})+X`$
At high energies, due to the contribution through weak-interaction at the $`e^+e^{}`$ annihilation vertex such as $`e^+e^{}Z^0q_f\overline{q}_f`$, the initial $`q_f`$ and $`\overline{q}_f`$ are longitudinally polarized. The magnitude of the polarization of $`q_f`$ and that of $`\overline{q}_f`$ are the same but the sign are different. They are constants at a given center of mass (c.m.) energy of $`e^+e^{}`$ system. Also the relative weights for the contributions of different flavors are constants when averaging over the different jet (or initial quark) directions. They are the same for quarks and for anti-quarks. Namely, we have
$$P_f^{e^+e^{}}=P_{\overline{f}}^{e^+e^{}},$$
(30)
$$R_f^{e^+e^{}}=R_{\overline{f}}^{e^+e^{}}.$$
(31)
and $`R_f^{e^+e^{}}`$ is given by
$$R_f^{e^+e^{}}=\frac{\sigma (e^+e^{}q_f\overline{q}_f)}{_f\sigma (e^+e^{}q_f\overline{q}_f)}.$$
(32)
For example, for reactions at the $`Z^0`$-pole, we neglect the contribution from the annihilation via virtual photon. We have $`P_f=0.67`$ for $`f=u`$ or $`c`$ and $`P_f=0.94`$ for $`f=d,s`$, or $`b`$. Hence, we have
$$N^{e^+e^{}}(z,H)=\underset{f}{}R_f^{e^+e^{}}[D_f^H(z)+D_{\overline{f}}^H(z)],$$
(33)
$$\mathrm{\Delta }N^{e^+e^{}}(z,H)=\underset{f}{}R_f^{e^+e^{}}P_f^{e^+e^{}}[\mathrm{\Delta }D_f^H(z)\mathrm{\Delta }D_{\overline{f}}^H(z)].$$
(34)
For antihyperons, we have
$$N^{e^+e^{}}(z,\overline{H})=\underset{f}{}R_f^{e^+e^{}}[D_f^H(z)+D_{\overline{f}}^H(z)],$$
(35)
$$\mathrm{\Delta }N^{e^+e^{}}(z,\overline{H})=\underset{f}{}R_f^{e^+e^{}}P_f^{e^+e^{}}[\mathrm{\Delta }D_{\overline{f}}^H(z)\mathrm{\Delta }D_f^H(z)].$$
(36)
We see that,
$$N^{e^+e^{}}(z,\overline{H})=N^{e^+e^{}}(z,H),$$
(37)
$$\mathrm{\Delta }N^{e^+e^{}}(z,\overline{H})=\mathrm{\Delta }N^{e^+e^{}}(z,H).$$
(38)
This means that, under the charge conjugation symmetries for the fragmentation functions and decay processes, we have
$$P_{\overline{H}}^{e^+e^{}}(z)=P_H^{e^+e^{}}(z).$$
(39)
This result is true independent of model for fragmentation functions. It is a direct consequence of the charge conjugation symmetries for fragmentation and decay processes. Since we do not have the contaminations from the initial state such as the structure of nucleon in this reaction, this is an ideal place to test such charge conjugation symmetries.
#### II.2.2 $`e^{}+Ne^{}+H(\mathrm{or}\overline{H})+X`$
At sufficiently large energies and momentum transfer $`Q^2`$, hadron produced in the current fragmentation region of semi-inclusive deeply inelastic lepton-nucleon scattering such as $`e^{}+Ne^{}+H(\mathrm{or}\overline{H})+X`$ can be considered as a pure product of fragmentation of the struck quark. In this case, the cross sections are given by
$$\frac{d^3\sigma }{dxdydz}(eNeHX)=\frac{d\widehat{\sigma }_{\text{Mott}}}{dy}\underset{f}{}e_f^2[q_f(x,Q^2)D_f^H(z)+\overline{q}_f(x,Q^2)D_{\overline{f}}^H(z)],$$
(40)
$$\frac{\mathrm{\Delta }d^3\sigma }{dxdydz}(eNeHX)=\frac{d\widehat{\sigma }_{\text{Mott}}}{dy}\underset{f}{}e_f^2[P_f^{eN}(x,y)q_f(x,Q^2)\mathrm{\Delta }D_f^H(z)+P_{\overline{f}}^{eN}(x,y)\overline{q}_f(x,Q^2)\mathrm{\Delta }D_{\overline{f}}^H(z)],$$
(41)
where $`\sigma _{\text{Mott}}`$ is the Mott cross section; $`x`$ is the usual Bjorken $`x`$; $`y`$ is the fractional energy transfer in the rest frame of the nucleon; and $`z`$ is the fraction of momentum of struck $`q`$ carried by $`H`$. The polarization of the struck quark $`q_f`$ and that of anti-quark $`\overline{q}_f`$ are given by
$$P_f^{eN}(x,y)=\frac{P_L^{(e)}D_L(y)q_f(x)+P_L^{(N)}\mathrm{\Delta }q_f(x)}{q_f(x)+P_L^{(e)}D_L(y)P_L^{(N)}\mathrm{\Delta }q_f(x)},$$
(42)
$$P_{\overline{f}}^{eN}(x,y)=\frac{P_L^{(e)}D_L(y)\overline{q}_f(x)+P_L^{(N)}\mathrm{\Delta }\overline{q}_f(x)}{\overline{q}_f(x)+P_L^{(e)}D_L(y)P_L^{(N)}\mathrm{\Delta }\overline{q}_f(x)},$$
(43)
for longitudinally polarized case, where $`P_L^{(e)}`$ and $`P_L^{(N)}`$ are polarizations of the incident electron and nucleon, respectively; $`D_L(y)`$ is the longitudinal spin transfer factor in $`eqeq`$. It is given by
$$D_L(y)=\frac{1(1y)^2}{1+(1y)^2}.$$
(44)
In the transversely polarized case, we have
$$P_{fT}^{eN}(x,y)=P_T^{(N)}\frac{\delta q_f(x)}{q_f(x)}D_T(y).$$
(45)
$$P_{\overline{f}T}^{eN}(x,y)=P_T^{(N)}\frac{\delta \overline{q}_f(x)}{\overline{q}_f(x)}D_T(y).$$
(46)
where the transversal spin transfer factor $`D_T(y)`$ in $`eqeq`$ is given by
$$D_T(y)=\frac{2(1y)}{1+(1y)^2}.$$
(47)
We see that the relative contributions of different flavors are given by
$$R_f^{eN}(x,y)=\frac{1}{\sigma _{\text{inel}}}\frac{d\widehat{\sigma }_{\text{Mott}}}{dy}e_f^2q_f(x,Q^2),$$
(48)
$$R_{\overline{f}}^{eN}(x,y)=\frac{1}{\sigma _{\text{inel}}}\frac{d\widehat{\sigma }_{\text{Mott}}}{dy}e_f^2\overline{q}_f(x,Q^2).$$
(49)
From Eqs.(42)–(49), we see that both $`P_f`$ and $`R_f`$ depend on $`x`$ and $`y`$. We see also that, in general, $`P_f^{eN}(x,y)P_{\overline{f}}^{eN}(x,y)`$, $`R_f^{eN}(x,y)R_{\overline{f}}^{eN}(x,y)`$, they are equal only if $`q_f(x,Q^2)=\overline{q}_f(x,Q^2)`$ and $`\mathrm{\Delta }q_f(x,Q^2)=\mathrm{\Delta }\overline{q}_f(x,Q^2)`$.
For the corresponding number densities of $`H`$ or $`\overline{H}`$ in $`e+Ne+H(\mathrm{or}\overline{H})+X`$, we have
$$N^{eN}(z,H)=\underset{f}{}𝑑x𝑑y\left[R_f^{eN}(x,y)D_f^H(z)+R_{\overline{f}}^{eN}(x,y)D_{\overline{f}}^H(z)\right],$$
(50)
$$N^{eN}(z,\overline{H})=\underset{f}{}𝑑x𝑑y\left[R_f^{eN}(x,y)D_{\overline{f}}^H(z)+R_{\overline{f}}^{eN}(x,y)D_f^H(z)\right],$$
(51)
$$\mathrm{\Delta }N^{eN}(z,H)=\underset{f}{}𝑑x𝑑y\left[R_f^{eN}(x,y)P_f^{eN}(x,y)\mathrm{\Delta }D_f^H(z)+R_{\overline{f}}^{eN}(x,y)P_{\overline{f}}^{eN}(x,y)\mathrm{\Delta }D_{\overline{f}}^H(z)\right],$$
(52)
$$\mathrm{\Delta }N^{eN}(z,\overline{H})=\underset{f}{}𝑑x𝑑y\left[R_f^{eN}(x,y)P_f^{eN}(x,y)\mathrm{\Delta }D_{\overline{f}}^H(z)+R_{\overline{f}}^{eN}(x,y)P_{\overline{f}}^{eN}(x,y)\mathrm{\Delta }D_f^H(z)\right],$$
(53)
Now, we compare the results for $`\overline{H}`$ with those for $`H`$, and we see the following qualitative features in the two different cases.
In the first case, we consider reactions at very high energies so that small $`x`$ contribution dominates. In this case, we can neglect the valence-quark contributions with good accuracy, i.e., $`q_f(x,Q^2)q_{f,s}(x,Q^2)`$ and $`\mathrm{\Delta }q_f(x,Q^2)\mathrm{\Delta }q_{f,s}(x,Q^2)`$, where the subscript $`s`$ denote “sea.” If we assume the charge conjugation symmetry in nucleon sea, i.e., $`q_{f,s}(x,Q^2)=\overline{q}_{f,s}(x,Q^2)=\overline{q}_f(x,Q^2)`$, and $`\mathrm{\Delta }q_{f,s}(x,Q^2)=\mathrm{\Delta }\overline{q}_{f,s}(x,Q^2)=\mathrm{\Delta }\overline{q}_f(x,Q^2)`$, we expect only a very small difference between quark and anti-quark distributions, i.e., $`q_f(x,Q^2)\overline{q}_f(x,Q^2)`$, and $`\mathrm{\Delta }q_f(x,Q^2)\mathrm{\Delta }\overline{q}_f(x,Q^2)`$. Hence, we expect that, $`P_f^{eN}(x,y)P_{\overline{f}}^{eN}(x,y)`$, $`R_f^{eN}(x,y)R_{\overline{f}}^{eN}(x,y)`$, and finally $`P_H(z)P_{\overline{H}}(z)`$.
In the second case, we consider reactions where large $`x`$ contributions dominate. In this case, valence-quark contribution plays an important role and there should be significant differences between quark and anti-quark distributions — thus significant differences between $`P_H`$ and $`P_{\overline{H}}`$. To get some feeling of these differences, we make the following rough estimations.
We denote $`R_f^{eN}(x,y)=R_{f,v}^{eN}(x,y)+R_{f,s}^{eN}(x,y)`$, (where the $`v`$ and $`s`$ in the subscripts denote valence or sea contribution), and expect the following approximate relations, if we neglect the flavor-dependence in the quark distribution functions:
$$R_{u,v}^{ep}(x,y)8R_{d,v}^{ep}(x,y),$$
(54)
$$R_{u,s}^{eN}(x,y)4R_{d,s}^{eN}(x,y)4R_{s,s}^{eN}(x,y),$$
(55)
$$R_{\overline{u},s}^{eN}(x,y)4R_{\overline{d},s}^{eN}(x,y)4R_{\overline{s},s}^{eN}(x,y).$$
(56)
These approximate relations can be used to give us some guidances of the qualitative features of hyperon and antihyperon polarizations in the reaction. For example, for $`e^{}+pe^{}+\mathrm{\Lambda }+X`$, we note further that, $`D_u^\mathrm{\Lambda }(z)=D_d^\mathrm{\Lambda }(z)`$ and for large $`z`$ both of them have the strangeness suppressions factor $`\lambda 0.3`$ relative to $`D_s^\mathrm{\Lambda }(z)`$ and that both $`u`$ and $`d`$ have, if any, small and negative contributions to the spin of $`\mathrm{\Lambda }`$ but $`s`$ has large and positive contribution. Hence, for reactions where the valence-quark contributions dominate for hyperon production, we expect $`u\mathrm{\Lambda }+X`$ dominates in large $`z`$ region, and $`P_\mathrm{\Lambda }`$ should be small. But for $`\overline{\mathrm{\Lambda }}`$, there is no contribution from valence-quark to the first rank particles; we expect that, for large $`z`$, $`\overline{u}\overline{\mathrm{\Lambda }}+X`$ and $`\overline{s}\overline{\mathrm{\Lambda }}+X`$ give comparable contributions to $`\overline{\mathrm{\Lambda }}`$ ptroduction and $`P_{\overline{\mathrm{\Lambda }}}`$ should be mainly determined by $`\overline{s}\overline{\mathrm{\Lambda }}+X`$. This implies that, in this case, $`P_{\overline{\mathrm{\Lambda }}}`$ should be positive and the magnitude is much larger than $`P_\mathrm{\Lambda }`$ at large $`z`$ values.
We can see that the qualitative features are independent of the models for spin transfer but depend strongly on the kinematic region.
#### II.2.3 $`\nu _\mu +N\mu ^{}+H(\mathrm{or}\overline{H})+X`$
For charged current neutrino reaction, we have
$`{\displaystyle \frac{d^3\sigma }{dxdydz}}(\nu _\mu N\mu ^{}HX)`$ $`=`$ $`{\displaystyle \underset{f,f^{}}{}}{\displaystyle \frac{d\widehat{\sigma }_0}{dy}}(\nu _\mu q_f\mu ^{}q_f^{})q_f(x,Q^2)D_f^{}^H(z)`$
$`+`$ $`{\displaystyle \underset{f,f^{}}{}}{\displaystyle \frac{d\widehat{\sigma }_0}{dy}}(\nu _\mu \overline{q}_f\mu ^{}\overline{q}_f^{})\overline{q}_f(x,Q^2)D_{\overline{f}^{}}^H(z).`$
We do not consider top production. In this case we have two Cabbibo favored elementary processes for quarks. The differential cross sections are given by
$$\frac{d\widehat{\sigma }_0}{dy}(\nu _\mu d\mu ^{}u)=\frac{d\widehat{\sigma }_0}{dy}(\nu _\mu s\mu ^{}c)=\frac{G^2xs}{\pi }\mathrm{cos}^2\theta _c,$$
(57)
and two Cabbibo suppressed elementary processes with cross sections,
$$\frac{d\widehat{\sigma }_0}{dy}(\nu _\mu d\mu ^{}c)=\frac{d\widehat{\sigma }_0}{dy}(\nu _\mu s\mu ^{}u)=\frac{G^2xs}{\pi }\mathrm{sin}^2\theta _c,$$
(58)
where $`\theta _c`$ is the Cabbibo angle. Similarly for anti-quarks, we have two Cabbibo favored elementary processes,
$$\frac{d\widehat{\sigma }_0}{dy}(\nu _\mu \overline{u}\mu ^{}\overline{d})=\frac{d\widehat{\sigma }_0}{dy}(\nu _\mu \overline{c}\mu ^{}\overline{s})=\frac{G^2xs}{\pi }(1y)^2\mathrm{cos}^2\theta _c,$$
(59)
and two Cabbibo suppressed processes,
$$\frac{d\widehat{\sigma }_0}{dy}(\nu _\mu \overline{u}\mu ^{}\overline{s})=\frac{d\widehat{\sigma }_0}{dy}(\nu _\mu \overline{c}\mu ^{}\overline{d})=\frac{G^2xs}{\pi }(1y)^2\mathrm{sin}^2\theta _c,$$
(60)
This means that
$`{\displaystyle \frac{d^3\sigma }{dxdydz}}(\nu _\mu N\mu ^{}HX)`$ $`=`$ $`{\displaystyle \frac{G^2xs}{\pi }}\{[d(x,Q^2)\mathrm{cos}^2\theta _c+s(x,Q^2)\mathrm{sin}^2\theta _c]D_u^H(z)`$ (61)
$`+`$ $`[d(x,Q^2)\mathrm{sin}^2\theta _c+s(x,Q^2)\mathrm{cos}^2\theta _c]D_c^H(z)`$
$`+`$ $`(1y)^2[\overline{u}(x,Q^2)\mathrm{cos}^2\theta _c+\overline{c}(x,Q^2)\mathrm{sin}^2\theta _c]D_{\overline{d}}^H(z)`$
$`+`$ $`(1y)^2[\overline{u}(x,Q^2)\mathrm{sin}^2\theta _c+\overline{c}(x,Q^2)\mathrm{cos}^2\theta _c]D_{\overline{s}}^H(z)]\},`$
We note further that the quarks in the final state of $`\nu _\mu q\mu ^{}q^{}`$ are longitudinally polarized with polarization $`P_q^{}=1`$, while the anti-quarks in the final state of $`\nu _\mu \overline{q}\mu ^{}\overline{q}^{}`$ are longitudinally polarized with $`P_{\overline{q}^{}}=1`$. We obtain that
$`{\displaystyle \frac{d^3\mathrm{\Delta }\sigma }{dxdydz}}(\nu _\mu N\mu ^{}HX)`$ $`=`$ $`{\displaystyle \frac{G^2xs}{\pi }}\{[d(x,Q^2)\mathrm{cos}^2\theta _c+s(x,Q^2)\mathrm{sin}^2\theta _c]\mathrm{\Delta }D_u^H(z)`$ (62)
$``$ $`[d(x,Q^2)\mathrm{sin}^2\theta _c+s(x,Q^2)\mathrm{cos}^2\theta _c]\mathrm{\Delta }D_c^H(z)`$
$`+`$ $`(1y)^2[\overline{u}(x,Q^2)\mathrm{cos}^2\theta _c+\overline{c}(x,Q^2)\mathrm{sin}^2\theta _c]\mathrm{\Delta }D_{\overline{d}}^H(z)`$
$`+`$ $`(1y)^2[\overline{u}(x,Q^2)\mathrm{sin}^2\theta _c+\overline{c}(x,Q^2)\mathrm{cos}^2\theta _c]\mathrm{\Delta }D_{\overline{s}}^H(z)\},`$
For antihyperon, we have
$`{\displaystyle \frac{d^3\sigma }{dxdydz}}(\nu _\mu N\mu ^{}\overline{H}X)`$ $`=`$ $`{\displaystyle \frac{G^2xs}{\pi }}\{[d(x,Q^2)\mathrm{cos}^2\theta _c+s(x,Q^2)\mathrm{sin}^2\theta _c]D_{\overline{u}}^H(z)`$ (63)
$`+`$ $`[d(x,Q^2)\mathrm{sin}^2\theta _c+s(x,Q^2)\mathrm{cos}^2\theta _c]D_{\overline{c}}^H(z)`$
$`+`$ $`(1y)^2[\overline{u}(x,Q^2)\mathrm{cos}^2\theta _c+\overline{c}(x,Q^2)\mathrm{sin}^2\theta _c]D_d^H(z)+`$
$`+`$ $`(1y)^2[\overline{u}(x,Q^2)\mathrm{sin}^2\theta _c+\overline{c}(x,Q^2)\mathrm{cos}^2\theta _c]D_s^H(z)]\},`$
$`{\displaystyle \frac{d^3\mathrm{\Delta }\sigma }{dxdydz}}(\nu _\mu N\mu ^{}\overline{H}X)`$ $`=`$ $`{\displaystyle \frac{G^2xs}{\pi }}\{[d(x,Q^2)\mathrm{cos}^2\theta _c+s(x,Q^2)\mathrm{sin}^2\theta _c]\mathrm{\Delta }D_{\overline{u}}^H(z)`$ (64)
$``$ $`[d(x,Q^2)\mathrm{sin}^2\theta _c+s(x,Q^2)\mathrm{cos}^2\theta _c]\mathrm{\Delta }D_{\overline{c}}^H(z)`$
$`+`$ $`(1y)^2[\overline{u}(x,Q^2)\mathrm{cos}^2\theta _c+\overline{c}(x,Q^2)\mathrm{sin}^2\theta _c]\mathrm{\Delta }D_d^H(z)`$
$`+`$ $`(1y)^2[\overline{u}(x,Q^2)\mathrm{sin}^2\theta _c+\overline{c}(x,Q^2)\mathrm{cos}^2\theta _c]\mathrm{\Delta }D_s^H(z)]\}.`$
From these equations, we expect that, unlike that in $`e^{}N`$ scattering, there should be a significant difference for the polarization of hyperon and that for the corresponding antihyperon. This is because, in $`\nu _\mu +N\mu ^{}+H(\mathrm{or}\overline{H})+X`$, (i) hyperons in the current fragmentation region are mainly from the fragmentation of $`u`$ and $`c`$ quarks but the antihyperons are from $`\overline{d}`$ and $`\overline{s}`$; and (ii) the helicities of the struck quarks are opposite to those of the anti-quarks.
To see the qualitative features more explicitly, we now make a qualitative analysis by taking the following approximations. We keep only Cabbibo favored processes and neglect $`\mathrm{\Delta }D_{\overline{f}}^H(z)`$ compared to $`\mathrm{\Delta }D_f^H(z)`$. Under these approximations, we have
$$\frac{d^3\mathrm{\Delta }\sigma }{dxdydz}(\nu _\mu N\mu ^{}HX)\frac{G^2xs}{\pi }\mathrm{cos}^2\theta _c[d(x,Q^2)\mathrm{\Delta }D_u^H(z)+s(x,Q^2)\mathrm{\Delta }D_c^H(z)],$$
(65)
$$\frac{d^3\mathrm{\Delta }\sigma }{dxdydz}(\nu _\mu N\mu ^{}\overline{H}X)\frac{G^2xs}{\pi }(1y)^2\mathrm{cos}^2\theta _c[\overline{u}(x,Q^2)\mathrm{\Delta }D_d^H(z)+\overline{c}(x,Q^2)\mathrm{\Delta }D_s^H(z)].$$
(66)
We see that the polarization of hyperons in this reaction is mainly determined by $`\mathrm{\Delta }D_u^H(z)`$ and $`\mathrm{\Delta }D_c^H(z)`$ while those for antihyperons are determined by $`\mathrm{\Delta }D_d^H(z)`$ and $`\mathrm{\Delta }D_s^H(z)`$. We thus expect the following qualitative features:
(1) For $`\mathrm{\Lambda }`$, we note $`\mathrm{\Delta }D_d^\mathrm{\Lambda }(z)=\mathrm{\Delta }D_u^\mathrm{\Lambda }(z)<0`$ and the magnitude is very small, while $`\mathrm{\Delta }D_s^\mathrm{\Lambda }(z)>0`$ and the magnitude is large. There is a significant contribution from $`\mathrm{\Lambda }_c`$ decay to $`\mathrm{\Lambda }`$, but the decay spin transfer in $`\mathrm{\Lambda }_c\mathrm{\Lambda }X`$ is unclear. Hence it is very difficult to make any estimate on $`P_\mathrm{\Lambda }`$. But for $`\overline{\mathrm{\Lambda }}`$, we expect that the qualitative feature of $`P_{\overline{\mathrm{\Lambda }}}`$ in $`\nu _\mu +N\mu ^{}+\overline{\mathrm{\Lambda }}+X`$ is mainly determined by $`\mathrm{\Delta }D_s^\mathrm{\Lambda }(z)`$. The results should be positive and the magnitude is large for large $`z`$.
(2) For $`\mathrm{\Sigma }`$ production, we recall that, from isospin symmetry, $`\mathrm{\Delta }D_u^{\mathrm{\Sigma }^+}=\mathrm{\Delta }D_d^\mathrm{\Sigma }^{}`$ is positive and large, while $`\mathrm{\Delta }D_d^{\mathrm{\Sigma }^+}=\mathrm{\Delta }D_u^\mathrm{\Sigma }^{}`$ is very small. The spin transfer from charmed baryon decay to $`\mathrm{\Sigma }`$ is unknown but such decay contribution is relatively small compared to $`\mathrm{\Lambda }`$. We thus expect that $`P_{\mathrm{\Sigma }^+}`$ is negative and the magnitude is large for large $`z`$. But the magnitude of $`P_\mathrm{\Sigma }^{}`$ is very small.
We note further that $`\mathrm{\Delta }D_s^{\mathrm{\Sigma }^\pm }`$ is negative and the magnitude is smaller than $`\mathrm{\Delta }D_d^\mathrm{\Sigma }^{}`$ \[half of it in SU(6) and similar in DIS picture\]. But there is a strange suppression for $`D_d^\mathrm{\Sigma }`$ compared to $`D_s^\mathrm{\Sigma }`$. We expected the contributions from the two terms in Eq.(66) to $`P_{\overline{\mathrm{\Sigma }}^+}`$ are opposite in sign with similar magnitudes. These two contributions partly cancel each other and the final results for $`P_{\overline{\mathrm{\Sigma }}^+}`$ should be small in magnitude and very sensitive to the quark distribution functions. For $`P_{\overline{\mathrm{\Sigma }}^{}}`$, the contribution from $`\mathrm{\Delta }D_{\overline{d}}^{\overline{\mathrm{\Sigma }}^{}}=\mathrm{\Delta }D_d^{\mathrm{\Sigma }^+}`$ is very small and the results are mainly determined by $`\mathrm{\Delta }D_s^{\mathrm{\Sigma }^+}`$, which is negative and larger at large $`z`$.
(3) For $`\mathrm{\Xi }`$ production, charmed baryon decay contribution can be neglected. $`P_\mathrm{\Xi }`$ is mainly determined by $`\mathrm{\Delta }D_u^\mathrm{\Xi }`$. Since $`\mathrm{\Delta }D_u^{\mathrm{\Xi }^0}(z,\text{dir})`$ is negative but the decay contribution from $`\mathrm{\Xi }^0`$, i.e. $`\mathrm{\Delta }D_u^{\mathrm{\Xi }^0}(z,\text{dec})`$, is positive, and the magnitude of polarization of the latter is larger, the final results for $`P_{\mathrm{\Xi }^0}`$ can be quite sensitive to the hadronization model. But in gerenal, we expect both the magnitude of $`P_{\mathrm{\Xi }^0}`$ and that of $`P_\mathrm{\Xi }^{}`$ to be small.
For $`\overline{\mathrm{\Xi }}^0`$, the contribution from the first term of Eq.(66), i.e. that from $`\mathrm{\Delta }D_d^{\mathrm{\Xi }^0}`$, is much smaller than that from the second term since $`\mathrm{\Xi }^0`$ does not contain $`d`$ as a valence-quark. For $`\overline{\mathrm{\Xi }}^+`$, there is a contribution from the first term of Eq.(66), i.e. that from $`\mathrm{\Delta }D_d^\mathrm{\Xi }^{}`$, but the magnitude is smaller than $`\mathrm{\Delta }D_s^\mathrm{\Xi }^{}`$ \[half of it in SU(6)\]. Furthermore, because of the strangeness suppression, $`D_d^\mathrm{\Xi }`$ also is suppressed compared to $`D_s^\mathrm{\Xi }^{}`$. We thus expect that $`|\mathrm{\Delta }D_d^\mathrm{\Xi }^{}|\mathrm{\Delta }D_s^\mathrm{\Xi }^{}`$ too. Hence for both $`\overline{\mathrm{\Xi }}^0`$ and $`\overline{\mathrm{\Xi }}^+`$, we expect that $`P_{\overline{\mathrm{\Xi }}}`$ is mainly determined by $`\mathrm{\Delta }D_s^\mathrm{\Xi }`$ and results should be positive and large for large $`z`$.
These qualitative features are independent of models for spin transfer and can be checked by experiments.
#### II.2.4 $`\overline{\nu }_\mu +N\mu ^++H(\mathrm{or}\overline{H})+X`$
For charged current reactions with anti-neutrino beam such as $`\overline{\nu }_\mu +N\mu ^++H(\mathrm{or}\overline{H})+X`$, the contributing elementary processes are the following. We have two Cabbibo favored elementary processes for quarks with the differential cross sections as given by
$$\frac{d\widehat{\sigma }_0}{dy}(\overline{\nu }_\mu u\mu ^+d)=\frac{d\widehat{\sigma }_0}{dy}(\overline{\nu }_\mu c\mu ^+s)=\frac{G^2xs}{\pi }(1y)^2\mathrm{cos}^2\theta _c,$$
(67)
and two Cabbibo suppressed elementary processes with cross sections,
$$\frac{d\widehat{\sigma }_0}{dy}(\overline{\nu }_\mu u\mu ^+s)=\frac{d\widehat{\sigma }_0}{dy}(\overline{\nu }_\mu c\mu ^+d)=\frac{G^2xs}{\pi }(1y)^2\mathrm{sin}^2\theta _c.$$
(68)
Similarly for anti-quarks, we have two Cabbibo favored elementary processes,
$$\frac{d\widehat{\sigma }_0}{dy}(\overline{\nu }_\mu \overline{d}\mu ^+\overline{u})=\frac{d\widehat{\sigma }_0}{dy}(\overline{\nu }_\mu \overline{s}\mu ^+\overline{c})=\frac{G^2xs}{\pi }\mathrm{cos}^2\theta _c,$$
(69)
and two Cabbibo suppressed processes,
$$\frac{d\widehat{\sigma }_0}{dy}(\overline{\nu }_\mu \overline{d}\mu ^+\overline{c})=\frac{d\widehat{\sigma }_0}{dy}(\overline{\nu }_\mu \overline{s}\mu ^+\overline{u})=\frac{G^2xs}{\pi }\mathrm{sin}^2\theta _c.$$
(70)
Hence, we obtain that
$`{\displaystyle \frac{d^3\sigma }{dxdydz}}(\overline{\nu }_\mu N\mu ^+HX)`$ $`=`$ $`{\displaystyle \frac{G^2xs}{\pi }}\{[\overline{d}(x,Q^2)\mathrm{cos}^2\theta _c+\overline{s}(x,Q^2)\mathrm{sin}^2\theta _c]D_{\overline{u}}^H(z)`$ (71)
$`+`$ $`[\overline{d}(x,Q^2)\mathrm{sin}^2\theta _c+\overline{s}(x,Q^2)\mathrm{cos}^2\theta _c]D_{\overline{c}}^H(z)`$
$`+`$ $`(1y)^2[u(x,Q^2)\mathrm{cos}^2\theta _c+c(x,Q^2)\mathrm{sin}^2\theta _c]D_d^H(z)`$
$`+`$ $`(1y)^2[u(x,Q^2)\mathrm{sin}^2\theta _c+c(x,Q^2)\mathrm{cos}^2\theta _c]D_s^H(z)\},`$
$`{\displaystyle \frac{d^3\sigma }{dxdydz}}(\overline{\nu }_\mu N\mu ^+\overline{H}X)`$ $`=`$ $`{\displaystyle \frac{G^2xs}{\pi }}\{[\overline{d}(x,Q^2)\mathrm{cos}^2\theta _c+\overline{s}(x,Q^2)\mathrm{sin}^2\theta _c]D_u^H(z)`$ (72)
$`+`$ $`[\overline{d}(x,Q^2)\mathrm{sin}^2\theta _c+\overline{s}(x,Q^2)\mathrm{cos}^2\theta _c]D_c^H(z)`$
$`+`$ $`(1y)^2[u(x,Q^2)\mathrm{cos}^2\theta _c+c(x,Q^2)\mathrm{sin}^2\theta _c]D_{\overline{d}}^H(z)`$
$`+`$ $`(1y)^2[u(x,Q^2)\mathrm{sin}^2\theta _c+c(x,Q^2)\mathrm{cos}^2\theta _c]D_{\overline{s}}^H(z)]\},`$
We note further that the quarks in the final state of $`\overline{\nu }_\mu q\mu ^+q^{}`$ are longitudinally polarized with polarization $`P_q^{}=1`$, while the anti-quarks in the final state of $`\overline{\nu }_\mu \overline{q}\mu ^+\overline{q}^{}`$ are longitudinally polarized with $`P_{\overline{q}^{}}=1`$. We obtain that,
$`{\displaystyle \frac{d^3\mathrm{\Delta }\sigma }{dxdydz}}(\overline{\nu }_\mu N\mu ^+HX)`$ $`=`$ $`{\displaystyle \frac{G^2xs}{\pi }}\{[\overline{d}(x,Q^2)\mathrm{cos}^2\theta _c+\overline{s}(x,Q^2)\mathrm{sin}^2\theta _c]\mathrm{\Delta }D_{\overline{u}}^H(z)`$ (73)
$`+`$ $`[\overline{d}(x,Q^2)\mathrm{sin}^2\theta _c+\overline{s}(x,Q^2)\mathrm{cos}^2\theta _c]\mathrm{\Delta }D_{\overline{c}}^H(z)`$
$``$ $`(1y)^2[u(x,Q^2)\mathrm{cos}^2\theta _c+c(x,Q^2)\mathrm{sin}^2\theta _c]\mathrm{\Delta }D_d^H(z)`$
$``$ $`(1y)^2[u(x,Q^2)\mathrm{sin}^2\theta _c+c(x,Q^2)\mathrm{cos}^2\theta _c]\mathrm{\Delta }D_s^H(z)\},`$
$`{\displaystyle \frac{d^3\mathrm{\Delta }\sigma }{dxdydz}}(\overline{\nu }_\mu N\mu ^+\overline{H}X)`$ $`=`$ $`{\displaystyle \frac{G^2xs}{\pi }}\{[\overline{d}(x,Q^2)\mathrm{cos}^2\theta _c+\overline{s}(x,Q^2)\mathrm{sin}^2\theta _c]\mathrm{\Delta }D_u^H(z)`$ (74)
$`+`$ $`[\overline{d}(x,Q^2)\mathrm{sin}^2\theta _c+\overline{s}(x,Q^2)\mathrm{cos}^2\theta _c]\mathrm{\Delta }D_c^H(z)`$
$``$ $`(1y)^2[u(x,Q^2)\mathrm{cos}^2\theta _c+c(x,Q^2)\mathrm{sin}^2\theta _c]\mathrm{\Delta }D_{\overline{d}}^H(z)`$
$``$ $`(1y)^2[u(x,Q^2)\mathrm{sin}^2\theta _c+c(x,Q^2)\mathrm{cos}^2\theta _c]\mathrm{\Delta }D_{\overline{s}}^H(z)]\},`$
We compare these results with those for $`\nu _\mu +N\mu ^{}+H(\mathrm{or}\overline{H})+X`$ presented in last subsection. We see that the results for $`\overline{\nu }_\mu +N\mu ^++H(\mathrm{or}\overline{H})+X`$ are the same as the corresponding results $`\nu _\mu +N\mu ^{}+\overline{H}(\mathrm{or}H)+X`$ under the exchange $`q_f(x,Q^2)\overline{q}_f(x,Q^2)`$. Hence, if we consider the reactions at very high energies where small $`x`$ contribution dominates so that $`q_f(x,Q^2)\overline{q}_f(x,Q^2)`$, we have
$$P_H^{\nu N}(z)P_{\overline{H}}^{\overline{\nu }N}(z),$$
(75)
$$P_{\overline{H}}^{\nu N}(z)P_H^{\overline{\nu }N}(z).$$
(76)
As we have emphasized before, to test different models for spin transfer in fragmentation, it is important to have high energy so that the results in the current fragmentation region can be considered as purely from the struck quark (anti-quark) fragmentation. In this case, there is no new result for anti-neutrino charged current interactions compared with those for the corresponding neutrino reactions. The differences come from the valence-quark contributions which are small at very high energies where small $`x`$ dominate. In view of this and the difficulties in performing such experiments in the near future, we will not discuss this reaction in the next section.
### II.3 Models for $`\mathrm{\Delta }D_f^H(z)`$
There exist many different approaches for $`\mathrm{\Delta }D_f^H(z)`$ in literature GH93 ; BL98 ; Kot98 ; Florian98 ; LL00 ; LXL01 ; XLL02 ; LL02 ; Ma:1999wp ; Ma:1998pd ; BQMa01 ; Ma:2000cg ; Ma:2000uu . We summarize the key points of some of them in the following.
#### II.3.1 Calculation of $`\mathrm{\Delta }D_f^H(z)`$ according to the origin of $`H`$
In GH93 ; BL98 ; LL00 ; LXL01 ; XLL02 ; LL02 , $`\mathrm{\Delta }D_f^H(z)`$ has been calculated according to the origins of $`H`$. The produced $`H`$’s are divided into the following four categories: (A) those are directly produced and contain $`q_f`$; (B) decay products of polarized heavy hyperons; (C) those are directly produced and do not contain $`q_f`$; (D) decay products of unpolarized heavy hyperons. This is to say that we divide further
$$D_f^H(z;\text{dir})=D_f^{H(A)}(z)+D_f^{H(C)}(z);$$
(77)
$$D_f^H(z;\text{dec})=D_f^{H(B)}(z)+D_f^{H(D)}(z).$$
(78)
For the polarized case,
$$\mathrm{\Delta }D_f^H(z;\text{dir})=\mathrm{\Delta }D_f^{H(A)}(z)+\mathrm{\Delta }D_f^{H(C)}(z);$$
(79)
$$\mathrm{\Delta }D_f^H(z;\text{dec})=\mathrm{\Delta }D_f^{H(B)}(z)+\mathrm{\Delta }D_f^{H(D)}(z).$$
(80)
It is assumed that,
$$\mathrm{\Delta }D_f^{H(A)}(z)=t_{H,f}^FD_f^{H(A)}(z),$$
(81)
$$\mathrm{\Delta }D_f^{H(C)}(z)=\mathrm{\Delta }D_f^{H(D)}(z)=0.$$
(82)
Here, $`t_{H,f}^F`$ is a constant and is taken as,
$$t_{H,f}^F=\mathrm{\Delta }Q_f/n_f$$
(83)
where $`\mathrm{\Delta }Q_f`$ and $`n_f`$ are the fractional contribution of spin of quark with flavor $`f`$ to the spin of $`H`$ and the number of valence-quarks of flavor $`f`$ in $`H`$. Clearly, $`Q_f`$ is different in the SU(6) picture from those drawn from polarized deeply inelastic scattering (DIS) data. They are given in e.g. Table I of LL00 .
The decay contribution part $`\mathrm{\Delta }D_f^{H(B)}(z)`$ is calculated using Eq.(5), i.e.,
$$\mathrm{\Delta }D_f^{H(B)}(z)=\underset{j}{}𝑑z^{}t_{H,H_j}^DK_{H,H_j}(z,z^{})[\mathrm{\Delta }D_f^{H_j(A)}(z^{})+\mathrm{\Delta }D_f^{H_j(B)}(z^{})],$$
(84)
We should note that, in the Feynman-Field type of recursive cascade fragmentation model, $`D_f^{H(A)}(z)`$ is nothing else but the probability to produce a first rank $`H`$ with $`z`$. It is usually denoted by $`f_{q_f}^H(z)`$ in such fragmentation models, i.e.,
$$D_f^{H(A)}(z)=f_{q_f}^H(z).$$
(85)
It follows that,
$$D_f^{H(C)}(z)=D_f^H(z;\text{dir})f_{q_f}^H(z).$$
(86)
In this case, we have,
$$\mathrm{\Delta }D_f^{H(A)}(z)=t_{H,f}^Ff_{q_f}^H(z),$$
(87)
$$\mathrm{\Delta }D_f^{H(B)}(z)=\underset{j}{}𝑑z^{}t_{H,H_j}^DK_{H,H_j}(z,z^{})[t_{H,f}^Ff_{q_f}^H(z)+\mathrm{\Delta }D_f^{H_j(B)}(z^{})].$$
(88)
In this model, for $`\overline{q}_fHX`$, we should have,
$$D_{\overline{f}}^{H(A)}(z)=0,$$
(89)
$$\mathrm{\Delta }D_{\overline{f}}^H(z)=0.$$
(90)
We recall that $`f_{q_f}^H(z)`$ is a quite essential input in such recursive cascade models. In most cases, there exist explicit expression for it. Hence, the $`z`$ dependence of $`\mathrm{\Delta }D`$ in this model is determined by the fragmentation model for unpolarized case, which are empirically known from the experimental facts for unpolarized reactions. The only unknown is the spin transfer constant $`t_{H,f}^F`$ for type (A) hyperon which influences mainly the magnitudes of the polarizations of the hyperons. So the ingredients of this model can be tested separately by testing the $`z`$ dependence and magnitudes of the polarizations.
In GH93 ; BL98 ; LL00 ; LXL01 ; XLL02 ; LL02 , calculations have been carried out using a Monte Carlo event generator PYTHIA jetset for $`e^+e^{}`$ annihilation and pythia for $`lp`$ or $`pp`$ collisions based on Lund fragmentation model lund . A set of formulas are given there which are more convenient for Monte Carlo calculations.
#### II.3.2 Calculation of $`\mathrm{\Delta }D_f^H(z)`$ using Gribov relation
In Ma:1999wp ; Ma:1998pd ; BQMa01 ; Ma:2000cg ; Ma:2000uu , $`\mathrm{\Delta }D_f^H(z)`$ has been calculated using the following proportionality relation, i.e., $`D_f^H(z)q_f^H(z)`$ and $`\mathrm{\Delta }D_f^H(z)\mathrm{\Delta }q_f^H(z)`$, which they referred as “Gribov relation” since it was first shown in Gribov by Gribov and collaborator in 1971. In terms of the language given above, this relation, if true, should be valid for the directly produced part, i.e.
$$D_f^H(z;\text{dir})q_f^H(z),$$
(91)
$$\mathrm{\Delta }D_f^H(z;\text{dir})\mathrm{\Delta }q_f^H(z).$$
(92)
The decay contribution parts can be calculated using Eq.(5).
For $`\overline{q}_fHX`$, we have similar results, i.e.
$$D_{\overline{f}}^H(z;\text{dir})\overline{q}_f^H(z),$$
(93)
$$\mathrm{\Delta }D_{\overline{f}}^H(z;\text{dir})\mathrm{\Delta }\overline{q}_f^H(z).$$
(94)
In Ma:1999wp ; Ma:1998pd ; BQMa01 ; Ma:2000cg ; Ma:2000uu , as an approximation, decay contributions are not considered.
#### II.3.3 Other models
Other models for $`\mathrm{\Delta }D_f^H(z)`$ have also discussed in the literature Kot98 ; Florian98 . They are essentially different combinations of the two models discussed above. We will not go to the details of these models but refer the interested readers to the references.
## III Results and discussions
By applying the formulas presented in the last section to different reactions, we obtain the results for the polarization of hyperons and antihyperons in these reactions. We use the model for $`\mathrm{\Delta }D_f^H(z)`$ as described in the first subsection of Sec. II C, i.e. to calculate $`\mathrm{\Delta }D_f^H(z)`$ according to the origin of $`H`$. The different contributions to $`D_f^H(z)`$ are calculated using a Monte Carlo event generator pythia PYTHIA based on Lund string fragmentation model lund . The results for hyperon polarization are given in BL98 ; LL00 ; LXL01 ; XLL02 ; LL02 . We present the results for antihyperons and compare them with those for the corresponding hyperon in the following. As we have shown in Section IIB, the results for antihyperons in $`e^+e^{}`$ annihilations differ from those for the corresponding hyperons only by a minus sign. We will not repeat them here. We will present the results for $`e^{}+Ne^{}+\overline{H}+X`$ and those for $`\nu _\mu +N\mu ^{}+\overline{H}+X`$, respectively, in the following.
### III.1 $`e^{}+Ne^{}+\overline{H}+X`$
In Fig. 1, we show the results for antihyperons in $`e^{}+Ne^{}+\overline{H}+X`$ with polarized beam or target. Clearly the situation is completely the same for $`e^++Ne^++\overline{H}+X`$ or $`\mu ^\pm +N\mu ^\pm +\overline{H}+X`$, where one virtual photon exchange plays the dominate role. The incident energy of electron is taken as $`E=500`$ GeV off a fixed target. We take this electron energy to gurantee that the energy is already high enough that we need only take the struck quark or anti-quark fragmentation into account when we look at the current fragmentation region. The results are shown as functions of $`z`$, where $`0.2<y<0.9`$, $`Q^2>1\text{GeV}^2`$ and $`W^2>1\text{GeV}^2`$. The results depend not much on the energy so long as $`E`$ is considerably large (say, larger than $`200`$GeV).
From Fig. 1, we see that there is indeed little difference between the results for hyperons and those for antihyperons at such energies for reactions with polarized beams. This confirms the qualitative expectations presented in Sec.IIB. We also see that the polarizations are much larger in reactions with polarized beams than those for reactions where only the nucleon target is polarized. The results in the latter case are very sensitive to the parametrization of polarized quark (anti-quark) distribution functions used. The relatively large difference between the results for hyperons and those for the corresponding antihyperons reflects the differences in the polarized quark (anti-quark) distributions.
To compare the results with the results COMPASS from COMPASS at CERN, we lowered the energy to $`E=160`$GeV and obtained the results as shown in the left column of Fig. 2. Our results show that, at this energy, the difference between the results for $`\mathrm{\Lambda }`$ and those for $`\overline{\mathrm{\Lambda }}`$ polarization is also quite small, in particular, in the $`x_F0`$ region. The difference can be a little bit significant only for larger $`x_F`$. According to this result, we should not see a large difference between $`P_\mathrm{\Lambda }`$ and $`P_{\overline{\mathrm{\Lambda }}}`$ as indicated by the COMPASS data COMPASS .
As we have discussed in last section, in the theoretical framework described in this paper, the following symmetries are supposed: (i) Charge conjugation symmetry in the fragmentation functions and decay processes \[see Eqs.(14)–(17)\]; and (ii) charge conjugation symmetry in nucleon sea in particular $`s(x)=\overline{s}(x)`$ and $`\mathrm{\Delta }s(x)=\mathrm{\Delta }\overline{s}(x)`$. In this case, the origins of the difference between the polarizations of hyperons and those of antihyperons can be only the contributions from the valence-quarks of the initial nucleons. Our results show that the valence-quark contributions are already quite small at the COMPASS energy. To see this explicitly, we have examined the $`x`$ values of the struck quark or antiquark for events with the kinematic constraints as imposed by COMPASS. (Here, $`x`$ is the Bjorken $`x`$ used in describing deeply inelastic lepton-nucleon scattering.) The results are shown in Fig.3. Here, in Fig.3(a) and 3(b), we show the $`x`$-distribution of the struck quark and anti-quark that lead to the production of the hyperon and antihyperon, respectively; and in 3(c) and 3(d), we show the average values of $`x`$ of such struck quarks and anti-quarks as functions of $`x_F`$ of the produced hyperons or antihyperons. We see that the $`x`$ value of the struck quark or anti-quark can be as small as $`0.004`$. We see also that $`x`$ in this case is in general of the order of 0.01 for most of the $`x_F`$ values. In this $`x`$ region, $`q(x,Q^2)`$ is already dominated by the sea quark distribution $`q_s(x,Q^2)`$. There exists a valence-quark contribution $`q_v(x,Q^2)`$, but it is much smaller than $`q_s(x,Q^2)`$. This is why the obtained $`P_{\overline{H}}`$ differs little from $`P_H`$.
Another effect that relates to the valence-quark contribution and may cause a difference between $`P_H`$ and $`P_{\overline{H}}`$ in $`e^{}+Ne^{}+H(\mathrm{or}\overline{H})+X`$ is the contribution from the hadronization of the remnant of target nucleon. It has been pointed out first in LL02 that contribution of the hadronization of target remnant is important to hyperon production even for reasonably large $`x_F`$ at lower energies. The effect has been confirmed by the calculations presented in Ellis2002 . It has been shown that LL02 at the CERN NOMAD energies, contribution from the hadronization of nucleon target remnant dominates hyperon production at $`z`$ around zero. It is impossible to separate the contribution of the struck quark fragmentation from those of the target remnant fragmentation. Clearly, this effect can be different for hyperon and antihyperon production since the target remnant contribution comes mainly from the fragmentation of the valence diquark. It contributes quite differently for hyperons and antihyperons. To see whether this effect plays an important role at COMPASS energy, we have also calculated the contributions of target remnant fragmentation. We found out that this contribution is already very small for $`z>0`$ at the COMPASS energy. To see the influence on hyperon polarizations, we have also included the contributions in the results shown in Fig. 2(a) by using a valence-quark model for the quark polarization in the remnant of the target as described in LL02 . We see that there is indeed some influence at $`x_F0`$ for the polarization of the hyperons but the effect is already very small at the COMPASS energy.
After having made these checks, we are quite confident that the valence-quark contributions are quite small at the COMPASS energy. Hence, under the conjugation symmetries in fragmentation function, decay processes and in nucleon sea, there should not be a large difference between $`P_\mathrm{\Lambda }`$ and $`P_{\overline{\mathrm{\Lambda }}}`$ at COMPASS energy. A significant difference could be a signature for the violation of such conjugation symmetries. In view of the recent discussions on the asymmetric strange and antistrange sea BM96 , it would be very significant to check whether similar asymmetry is also possible for the polarized case.
We also made the calculations at HERMES energy, i.e. at $`E=27.6`$GeV. The results are shown in the right column of Fig. 2. We found out that at this energy, the major contribution comes mainly from the $`x`$-region of $`0.02<x<0.8`$. In this $`x`$ region, valence-quark plays an important role. Hence, we obtained a significant difference between the polarizations of antihyperons and those of hyperons in the current fragmentation region. We see, in particular, that the difference for $`\mathrm{\Lambda }`$ and $`\overline{\mathrm{\Lambda }}`$ is indeed as expected in the qualitative analysis given in Sec. II B 2.
We also calculated the contribution from target remnant fragmentation and found that it is important at HERMES energy. This can be seen clearly in Fig. 4, where different contributions to $`\mathrm{\Lambda }`$ at this energy are shown. We see clearly that the contribution from target remnant fragmentation is much higher than that from the struck quarks in the region near $`x_F=0`$. To show the influence of this effect on the polarizations, we calculate these contributions from target remnant fragmentation using a valence-quark model for the polarizations of the quarks in the remnant of the nucleon as described in LL02 . The results are added to the struck quark fragmentation contribution and are shown in Fig. 2(b). We see that the contribution is very important in the $`x_F0`$ region. It is therefore meaningless to factorize the $`\mathrm{\Lambda }`$ production cross section as one usually does at very high energies. On the other hand, the influence from target remnant fragmentation to antihyperon polarization is negligibly small. This is another reason for the differences between the polarizations of hyperons and those of antihyperons shown in Fig. 2(b) in particular at the $`x_F0`$ region.
### III.2 $`\nu _\mu +N\mu ^{}+\overline{H}+X`$
In Fig. 5, we show the results obtained for $`\nu _\mu +N\mu ^{}+\overline{H}\text{(or }H\text{)}+X`$ at an incident muon-neutrino beam energy $`E=500\text{GeV}`$ off a fixed target. We calculated for proton and neutron target and different hyperons and antihyperons. To obtain the results for $`\mathrm{\Lambda }`$, we simply take $`t_{\mathrm{\Lambda },\mathrm{\Lambda }_c}^D=1`$ as we did in . From these figures, we see that the results for hyperons and those for antihyperons are indeed significantly different from each other. These features are quite different from that in electron nucleon collision where hyperon and antihyperon polarizations are essentially the same at very high energies. The qualitative features are the same as we obtained in the qualitative analysis using Eqs.(65) and (66) in the last section.
We also calculated the polarizations for antihyperons at the CERN NOMAD energies. The results are shown in Fig. 6. As it has been pointed out in LL02 , at this energy, contribution from the fragmentation of the remnant of target nucleon to hyperon production is very important. We also included this contribution from a rough estimation by using a valence-quark model for the polarization of the quarks in the target remnant as in LL02 . For this reason, the polarizations obtained for hyperons at this energy differ significantly from those obtained at very high energis as shown in Fig. 5. But for antihyperons, this contribution is small thus the difference between the results and those shown in Fig. 5 is small. This is also one of the reasons for the differences between the results for hyperons and those for antihyperons, in particular, in the $`x_F0`$ region. The qualitative features are consistent with the NOMAD data NOMAD00 and it will be interesting to make high statistics check in the future.
## IV Summary and outlook
In summary, we have calculated the polarizations of antihyperons in lepton induced reactions, in particular $`l+pl^{^{}}+\overline{H}+X`$ at different energies with polarized beams, using different models for spin transfer for fragmentation. The results show little difference between the polarization of antihyperon and that of the corresponding hyperon in $`e^{}+Ne^{}+\overline{H}(\mathrm{or}\mathrm{H})+\mathrm{X}`$ at COMPASS or higher energies when the charge conjugation symmetries for fragmentation, decay and nucleon sea are assumed. But there are in general large differences between those for antihyperons and the corresponding hyperons in neutrino induced charged current reactions since the flavors of dominant fragmenting quarks and those of the anti-quarks and their polarizations are different. A detailed discussion is given and the results can be tested by future experiments.
## Acknowledgments
We thank M.G. Sapozhnikov of the COMPASS Collaboration for suggesting that we make such calculations and for stimulating discussions. This work was supported in part by the National Science Foundation of China (NSFC) with grant Nos. 10135030 and 10405016.
|
warning/0506/cond-mat0506799.html
|
ar5iv
|
text
|
# Deformation of Silica Aerogel During Fluid Adsorption
## I Introduction
Porous media have large surface areas, and consequently interfacial energy contributes significantly to their behavior. The energetic cost of the solid-vacuum interface in an empty porous medium produces a stress on the matrix; when fluid is adsorbed the interfacial energy may decrease, reducing the stress on the matrix and allowing it to expand. As more fluid is adsorbed, liquid begins to capillary condense, creating a large number of curved liquid-vapor interfaces throughout the sample. The energy of these liquid-vapor interfaces creates another stress which may result in contraction of the porous medium. The expansion of a porous medium upon adsorption of fluids has been observed in a number of systems over the past centuryYates (1954); Beaume et al. (1980); Takamori and Tomozawa (1982); Scherer (1986); Thibault et al. (1995); contraction has also been observedDolino et al. (1996).
Aerogels present a somewhat different system than the denser porous media since they are composed of a very low density network of silica strands, often with total porosities over 95%. As such they have very low elastic constants, with most elastic strain being accounted for by the bending and twisting of the strandsMa et al. (2002) rather than compression of the silica making up those strands. Therefore the effective elastic constants for aerogels can be orders of magnitude smaller than bulk silica, and aerogels are very sensitive to small changes in stress caused by interfacial energy.
A porous medium exposed to vapor at low pressure collects a thin film on the surface of the pores. As vapor pressure is increased this film thickens slowly until the fluid suddenly capillary condenses at some pressure below bulk saturation. Liquid invades the pores over a narrow pressure range, driven by the pressure difference across the curved liquid-vapor interface within the pores. While this pressure difference may be small compared to the elastic constants of typical solids, it can be comparable to the bulk moduli of low density aerogels. In such a situation the forces generated by the liquid-vapor interface during capillary condensation can deform the medium measurably. The capillary forces generated by water in aerogel are sufficient to completely crush the gel — a fact known to anyone who has accidentally gotten a drop of water on a hydrophilic aerogel.
Deformation of aerogel during liquid nitrogen adsorption and desorption has been measured by Reichenauer and SchererReichenauer and Scherer (2001, 2000) for some aerogels with densities between $`150240\frac{kg}{m^3}`$ (porosities between 88% and 93%) with a view to incorporating the distortion of the aerogel into existent methods of determining pore sizes from N<sub>2</sub> adsorption isotherms. The aerogels in that study had Young’s moduli of $`3.8`$MPa (implying a bulk modulus of about 2MPa) and greater, much larger than the aerogels in this study. Their aerogels exhibited large changes in volume during capillary condensation, and were permanently damaged by the process. They also observed that information about the elastic properties of their samples could be extracted from the isotherms.
The high surface tension and large contact angle of mercury prevents it from entering the pores of aerogels when placed in a mercury porosimeter. In this case the surface tension is so great that when the aerogel is subjected to large pressures it plastically deformsBeurroies et al. (1998); Woignier et al. (1997), down to a fraction of its original volume.
Shen and Monson performed a Monte Carlo study of fluid adsorption in a flexible porous network which resembled a high porosity aerogelShen and Monson (2002). Their simulated adsorption isotherms showed that the flexibility of the network had a large effect on the adsorption isotherms, and that the network exhibited a large volumetric change, especially during desorption.
Aerogels are widely used as a method to introduce a “quenched impurity” into a fluid, often with the goal of exploring the effect of fixed disorder on fluid order and phase transitions. Experiments with quantum fluids have included <sup>3</sup>HeMatsumoto et al. (1997); Porto and Parpia (1999), <sup>4</sup>HeChan et al. (1988); Yoon et al. (1998), and <sup>3</sup>He-<sup>4</sup>He mixturesChan et al. (1996); Kim et al. (1993); Mulders and Chan (1995). Liquid crystals in aerogelsBellini et al. (2001) have also been widely studied. Implicit in all of these studies is an assumption that the structure of the aerogel is not affected by the fluid within its pores, nor is the aerogel altered during filling or emptying. While studying the temperature and porosity dependence of adsorption isotherms in aerogelsHerman et al. (2005), we noticed that the isotherm shapes might be showing effects from the deformation of the aerogel by the adsorbed fluid. This study is aimed at quantifying the degree of aerogel deformation during helium adsorption at a variety of temperatures.
We have measured the macroscopic linear strain in two different low density aerogels during adsorption and desorption of low surface tension fluids. Our measurements included an isotherm of neon adsorbed in a $`51\frac{kg}{m^3}`$ ($`98\%`$ porosity) silica aerogel at 43K and several isotherms of helium adsorbed in a $`110\frac{kg}{m^3}`$ (95% porosity) silica aerogel at temperatures from 2.4K to 5.0K. At these temperatures the fluid surface tensions were small, but the low elastic constants of the aerogels allowed a significant deformation nevertheless. The compression of the aerogel was greatest during desorption. From the isotherms we show that the degree of compression during capillary condensation within the aerogel is directly related to the surface tension of the adsorbate. Furthermore, the bulk modulus of the aerogel was extracted from the adsorption and desorption isotherms.
The compression due to capillary condensation has not previously been studied in such compliant materials. By using high porosity aerogels, we were able to directly observe, for the first time, the deformation associated with low surface tension fluids like helium. By making measurements very slowly, we avoided rate dependent effects and were able to study the hysteresis between compression during filling and emptying of aerogels. Our measurements extended close to helium’s critical point, which allowed us to vary the surface tension over a wide range compared to previous measurements, which used nitrogen and much more rigid samples. From our results, it is clear that the deformation associated with surface tension must be taken into account when interpreting adsorption isotherms and to avoid damage when filling high porosity aerogels.
## II Experimental Method
The aerogel samples were synthesized in our lab from tetramethyl orthosilicate (TMOS) using a standard one-step base catalyzed method followed by supercritical extraction of the methanol solventPoelz and Riethmüller (1982). The two aerogels studied had densities of $`110\frac{kg}{m^3}`$ and $`51\frac{kg}{m^3}`$, corresponding to porosities of 95% and slightly less than 98% respectively. They are referred to as aerogels “B110” and “B51” throughout this paper. Both aerogel samples were cut into cylinders about 1cm long. B51, used in the first experiment, was about 1.2cm in diameter — it equilibrated so slowly that it took weeks to measure a single isotherm. To reach thermal equilibrium more quickly, a smaller sample of B110 (4mm in diameter) was used for the second experiment.
Our linear variable differential transformer (LVDT) allowed us to make high precision, non-contact measurements of the relative position of a cylindrical ferromagnetic core and a set of primary and secondary coils<sup>1</sup><sup>1</sup>1The neon data were collected using hand wound coils; the helium data were collected using a Schevitz HR050 sensor, available from Measurement Specialties, Inc.. The LVDT output was measured using an LR400 mutual inductance bridge, from Linear Research Inc.. The response of the LVDT was highly temperature dependent, and room temperature calibrations were not used to interpret the low temperature data. In the liquid helium cryostat it was possible to make a direct calibration of the LVDT, but for measurements with neon an approximate calibration had to be computed from the adsorption isotherm itself (as will be explained later).
Two experimental cells were used over the course of this experiment to accommodate the two different aerogel samples. Both cells were made of copper and had the same general layout, with the LVDT core supported about 2cm above the sample by a thin brass rod. The support was kept in contact with the sample by gravity, and the cells kept upright once assembled. The initial position of the external LVDT coils relative to the core was controlled through set screws.
In our initial experiment neon was admitted to the experimental cell containing B51 in volumetric shots from a room temperature gas handling system, and pressure measurements were made with a room temperature gauge<sup>2</sup><sup>2</sup>2Mensor 1000psi gauge, Model 4040. Cooling was provided by a Gifford-McMahon closed cycle refrigerator; the temperature was controlled to $`\pm 1`$mK using a platinum resistance thermometer and a thick-film heater mounted directly on the cell.
In the second set of measurements, on helium in B110, the fluid condensation into the system was controlled by stepping the pressure in the cell through the use of a low temperature pressure regulation ballastHerman et al. (2005). This second method did not provide information on the absolute quantity of fluid adsorbed by the aerogel but ensured long term pressure stability. A Straty-AdamsStraty and Adams (1969) type capacitive pressure gauge was used for low temperature *in situ* pressure measurement. The cell was mounted on a liquid helium cryostat; the temperature was controlled to $`\pm 50\mu `$K using a germanium resistive thermometer and thick-film heater mounted directly on the cell.
## III Neon at 43K in B51
The adsorption and desorption isotherms for neon at 43K in B51 are shown in Fig. 1a.
Note that neon at 43K is very close to its critical point (T$`{}_{c}{}^{}44.5`$K, P$`{}_{c}{}^{}2.7`$MPa, $`\rho _c485\frac{kg}{m^3}`$). As such, it has a very low surface tension (at 43K neon has the same surface tension as liquid helium does at 3.7K) and its liquid and vapor densities differ by only a factor of three ($`\rho _l=740\frac{kg}{m^3}`$ and $`\rho _v=250\frac{kg}{m^3}`$). We could not perform a direct calibration of the LVDT at 43K because there was no direct access to the LVDT within the cryostat; however, as shown by Reichenauer and SchererReichenauer and Scherer (2001, 2000); Reichenauer (2004), information can be extracted about the sample shrinkage from the adsorption isotherm itself.
Since capillary condensation occurs over a narrow pressure range ($`0.998<\frac{P}{P_0}<1`$), the liquid can be treated as incompressible during capillary condensation. The high density corners of the adsorption and desorption isotherms (points“B” and “D” in Fig. 1a respectively) correspond to the aerogel being full of liquid, but compressed because of capillary forces at the surface pores. After capillary condensation was complete in this sample, the aerogel continued to adsorb fluid up until bulk saturation was reached (just below point “C”). The significant slope of the isotherm between capillary condensation and bulk condensation (i.e. $`0.999<\frac{P}{P_0}<1`$) indicates that the aerogel was still adsorbing liquid even though there was no longer any vapor within the pores. By measuring how much fluid was adsorbed after the completion of capillary condensation it is possible to calculate how much swelling occurred in the aerogel over this pressure range. The degree of swelling can then be used to form a calibration for the LVDT.
The bulk neon vapor at 43K has a high density and the cell included a large bulk volume, so the total amount of neon admitted into the cell does not correspond to the amount of neon adsorbed by the aerogel. However, since we are dealing with such a small pressure range during capillary condensation we can assume that all neon admitted over the pressure range $`0.999<\frac{P}{P_0}<1`$ is adsorbed by the aerogel – the bulk vapor density changes very little over this range.
Assuming the deformation of the aerogel is isotropic, the change in aerogel volume ($`\mathrm{\Delta }V`$) can be related to the change in its length ($`\mathrm{\Delta }L`$) and the quantity of fluid ($`\mathrm{\Delta }n`$) adsorbed during swelling (or shrinking) by:
$$\mathrm{\Delta }V=\frac{\mathrm{\Delta }n}{\rho _l\rho _v}=V_0\left[1\left(1\frac{\mathrm{\Delta }L}{L_0}\right)^3\right]$$
(1)
where “$`L_0`$” and “$`V_0`$” are the initial length and volume of the aerogel sample and “$`\rho _l`$” and “$`\rho _v`$” are the density of liquid and gaseous neon respectively. While this technique is not very precise, it does give an estimate of the maximum linear aerogel compression which can then be used as an LVDT calibration.
Using this calibration the aerogel length change has been plotted as a function of the amount of neon adsorbed in Fig. 1b and as a function of pressure in Fig. 1c. Zero has been set to be the fully relaxed aerogel filled with liquid. Within the resolution of the LVDT no deformation was seen for the first 0.06 moles of neon admitted to the cell — in this regime, a thin film was collecting on the aerogel strands in equilibrium with bulk vapor. Then, as neon began to capillary condense near point A, the aerogel contracted reaching a maximum linear compression of almost 2% at point B. At this point the gel was full of liquid, but compressed from its original volume. As more neon was added to the cell it was adsorbed by the aerogel, allowing the gel to relax and expand. Once the gel was full and bulk liquid began collecting (at point C, $`n_{neon}0.095`$moles), there were no longer any curved liquid-vapor interfaces causing stress within the aerogel and it had re-expanded to its original size. Upon desorption of the neon, the process occurred in reverse; however during desorption the aerogel was linearly compressed by almost 3.5% at point D. By point E the aerogel had re-expanded — in this pressure regime only a film of neon remains on the silica strands. However, the gel did not fully return to its original size; there was a slight permanent deformation (about 0.2% linear compression).
Hysteresis is generally observed for adsorption of fluids in porous media — the adsorption and desorption of the fluid taking place at different partial pressures. The precise mechanism of this hysteresis is an open question for aerogels, but it may indicate a different liquid-vapor interface shape for adsorption and desorption. The lower partial pressure for desorption means that there exists a greater pressure difference across the curved liquid-vapor interface, and results in a greater maximum stress on the aerogel during desorption.
The portion of the isotherms closest to $`P_0`$ are identical for the emptying and filling branches of the isotherm; in this region the gel is full of liquid and the aerogel matrix is swelling or shrinking in response to the pressure difference across the liquid-vapor interface present at the aerogel surface rather than interfaces within the pores.
## IV Helium in Aerogel B110
The second experimental cell, used to investigate helium adsorption in aerogel, allowed more precise compression and pressure measurements. It also allowed a direct low temperature calibration for the LVDT. The LVDT calibration was reproducible to within about 1% upon thermal cycling. Since the sample had a smaller diameter than the B51 sample, the pressure exerted by the weight of the ferromagnetic core was larger. To avoid overloading the sample we used a denser aerogel, B110, in this cell; the gel was compressed about 0.3% by the weight of the core ($`3`$g).
An isotherm for helium in B110 at 4.200K is shown as Fig. 2.
Since the pressure in the cell is controlled rather than the quantity of helium, the isotherm is only plotted with length change as a function of cell pressure, as in Fig. 1c. During the low pressure, $`P/P_0<0.95`$, formation of a thin film of helium the aerogel expanded by about 0.08%. Equilibration took many hours for these films, so equilibrium data of the initial dilation of aerogel upon helium adsorption were not collected. Similar long equilibration times for helium films in aerogels have been observed before, as has the dilationThibault et al. (1995). Equilibration of points within the hysteretic region was also very slowHerman et al. (2005). The maximum contraction occurred during desorption and was equal to about 0.6% of total length.
Figure 3
shows three desorption isotherms taken at 2.4K, 4.2K and 5.0K respectively. As the temperature was raised, the surface tension of the helium decreased, causing less deformation of the aerogel. The maximum deformation decreased by more than a factor of twenty between 2.4K and 5.0K. No permanent compression was observed for this sample. Above the critical temperature of helium (T<sub>c</sub>=5.195K) no contraction was observed, as would be expected without a liquid-vapor interface.
## V Analysis/Discussion
### V.1 Dilation at low vapor pressure
Consistent with an earlier studyThibault et al. (1995), dilation of the aerogel during the initial stages of helium adsorption was observed at all temperatures studied. The maximum dilation depended much less on temperature than did the deformation during capillary condensation — in all cases it was just under 0.1% of the total sample length. This is consistent with the dilation being governed by the energetics of adsorption sites on the silica strands, which can involve binding energies much larger the liquid-vapor interfacial energy or the thermal energy at liquid helium temperatures.
### V.2 Aerogel bulk modulus
During fluid adsorption and desorption, curved liquid-vapor interfaces abound within the open aerogel pore structure. A Laplace pressure exists across each of these interfaces, which can place stress on the aerogel. The Laplace pressure is a purely mechanical construct, with the surface tension ($`\sigma _{lv}`$) across the interface responsible for the pressure difference between the liquid ($`P_l`$) and vapor ($`P_v`$) phases:
$$\mathrm{\Delta }P=P_lP_v=\frac{2\sigma _{lv}}{r}$$
(2)
assuming a hemispherical interface or radius $`r`$. Usually the porous medium is assumed to be rigid, but aerogels are extraordinarily compliant, and any stresses must be balanced by deformation of the aerogel. Once the aerogel is completely filled with liquid, the liquid-vapor interface exists only at the aerogel surface and the Laplace pressure is felt as a macroscopic stress.
The Laplace pressure also results in undersaturation — the condensation of fluids below bulk saturation pressure ($`P_0`$). This behavior is described by the Kelvin equation, which can be expressed in many forms. The simplest form for our purposesLandau and Lifshitz (1958) relates the stress caused by the fluid interface to the vapor pressure in the cell:
$$P_lP_v=\left(\frac{V_vV_l}{V_l}\right)\left(P_vP_0\right)$$
(3)
Here $`P_l`$ and $`P_v`$ are the pressures, and $`V_l`$ and $`V_v`$ the molar volumes, of the liquid and vapor phases respectively. This derivation assumes that the liquid and vapor are incompressible, which is a fairly accurate approximation given the small pressure ranges over which capillary condensation occurs in aerogels.
When the Laplace pressure acts as a macroscopic stress, then the Kelvin equation can be used to relate the stress on the aerogel to the vapor pressure in the cell. Thus Eq. 3 can be combined with the measured slope of the isotherms after capillary condensation has occurred (e.g. for $`0.995<\frac{P}{P_0}<1`$ for helium in B110 at 4.2K) to calculate the elastic properties of the aerogel. In this regime the gel was full of liquid and the pressure difference between the liquid and vapor phases was balanced by the elastic stress of the aerogel. If the deformation of the aerogels is isotropic then the bulk modulus, K<sub>gel</sub>, can be used to characterize its response to the stress exerted by the liquid-vapor interface.
For high porosity aerogels and small deformations (i.e. $`\frac{\mathrm{\Delta }L}{L_0}\frac{\mathrm{\Delta }V}{3V_0}`$) the sample experiences this pressure difference as a compressive stress, which gives:
$$\frac{\mathrm{\Delta }L}{L_0}\frac{1}{3}\frac{1}{K_{gel}}\left(\frac{V_vV_l}{V_l}\right)\left(P_vP_0\right)$$
(4)
Thus we can extract a value for K<sub>gel</sub> from the slope of each isotherm when plotted as $`\mathrm{\Delta }L/L_0vsP_v`$, without needing to know either the fluid’s surface tension or the aerogel’s pore size. In fact, measurements on this portion of our isotherm, where the aerogel is full but partially compressed, cannot tell us about the effective pore size. That information (R<sub>cap</sub>) comes from the “breakthrough radius,” the point at which the meniscus curvature $`r`$ becomes equal to an effective pore size during desorption and the interface becomes unstable so that the pores suddenly empty.
If we know the liquid’s surface tension and the pressure, $`P_v`$, at which pores begin to empty, we can determine a pore size, $`R_{cap}`$. In fact, this is a standard way of determining pore size from isotherms — there is no need to measure sample deformation $`\mathrm{\Delta }`$L, the desorption pressure is all that is necessary. Assuming the simplest (hemispherical) form for the meniscus, then at breakthrough:
$$\left(P_0P_v\right)=\left(\frac{V_l}{V_vV_l}\right)\frac{2\sigma _{lv}}{R_{cap}}$$
(5)
A similar calculation can define a radius during adsorption, although its meaning is less clear.
This derivation assumes that the liquid and vapor are incompressible, which is a fairly accurate approximation given the small pressure ranges over which capillary condensation occurs in aerogels. More commonly one assumes the vapor behaves like an ideal gas and the liquid molar volume is much smaller and can be neglected. This gives the more familiar form of the Kelvin equation:
$$RT\mathrm{ln}\frac{P_v}{P_0}=V_l\frac{2\sigma _{lv}}{r}+V_l\left(P_0P_v\right)$$
(6)
Neither of these assumptions is appropriate near the liquid-vapor critical point, so we use the form in Eq. 5 rather than Eq. 6.
What our aerogel pore size analysis (from the “breakthrough” pressure, $`P_v`$) does is show that this rather macroscopic, classical treatment is valid over a wide range of $`\sigma _{lv}`$ near T<sub>c</sub> (i.e. the value for $`R_{cap}`$ is essentially the same, even though $`\sigma _{lv}`$ varies by nearly twenty times). This also shows that, despite their unique tenuous structure, quite unlike an array of uniform pores, a description of capillary condensation in terms of a single effective pore size seems adequate.
The bulk moduli calculated from several helium adsorption isotherms in B110 are plotted in Fig. 4.
Calculations were made using all adsorption and desorption isotherms separately over a temperature range where $`\sigma _{lv}`$ changes by a factor of eighteen, confirming the validity of Eq. 4. The values extracted for K<sub>gel</sub> are roughly constant, and the mean value (K$`{}_{gel}{}^{}0.43`$MPa) has been included on the plot as a solid line.
The bulk modulus for silica aerogel depends sensitively on aerogel density, and can be extracted from measurements of the Young’s modulus (E) Woignier and Phalippou (1989) or shear modulus (G) Daughton et al. (2003) using:
$$K=\frac{E}{3(12\nu )}=\frac{2(1\nu )G}{3(12\nu )}$$
(7)
These equations require knowledge of the Poisson’s ratio ($`\nu `$) for aerogel, which is about 0.2 Gross et al. (1988). Pure silica has a bulk modulus of roughly $`3.5\times 10^4`$MPa, much larger than aerogel. The moduli for base catalyzed silica aerogels with densities used in this study ($`51\frac{kg}{m^3}`$ and $`110\frac{kg}{m^3}`$) should be about 0.08MPa and 0.8MPa respectively, although these values can vary greatly between samples. Our value for the bulk modulus of B110 of 0.43MPa is consistent with these values, although it is a little lower than expected. A similar calculation for the data in Fig. 1 yields a bulk modulus of K=0.065MPa for aerogel B51.
### V.3 Maximum deformation
Since helium adsorption isotherms in B110 were collected at several temperatures, the temperature dependence of the maximum compression during capillary condensation could be analyzed; Fig. 5
shows the degree of compression during capillary condensation, as well as the bulk surface tension at each temperature. While the degree of contraction depends sensitively on temperature, it scales roughly with surface tension (plotted as a solid line in Fig. 5.
The scaling of maximum deformation with surface tension is consistent with a characteristic “breakthrough radius” describing the curvature of the liquid-vapor interface at the surface of the aerogel just prior to the percolation of the vapor phase into the sample (invasion of the helium vapor phase into aerogel during desorption has been observedLambert et al. (2004) in optical experiments). While it is not clear how this breakthrough radius relates to the aerogel structure, the magnitude of this breakthrough radius can be estimated from the desorption isotherm using the Kelvin Equation. A similar calculation can be performed using the pressure at which capillary condensation is complete along the adsorption isotherms, although the form of the liquid-vapor interface during filling is even less clear. Table 1 summarizes the values of the interface radii calculated from our isotherms, as well as the effective capillary pressure and predicted compression of the aerogels.
The capillary pressure generated by the curved liquid-vapor interface during desorption is significant compared to the bulk moduli of the aerogels. It is easy to see how higher surface tension fluids can easily damage aerogels — assuming a breakthrough radius of 20nm, water (295K) and liquid nitrogen (77K) would generate capillary pressures of 7MPa and 0.9MPa respectively, much larger than the bulk moduli of these aerogels.
### V.4 Damage to aerogels
No permanent densification was seen in our denser aerogel (B110) over the range of this experiment. The maximum volumetric compression at any time was about 6%, which is generally within the elastic regime for aerogels. However, B51 appeared to exhibit damage after desorption of neon. It had experienced a 10% volumetric compression, enough to cause permanent structural changes in the aerogel. The manner in which the maximum compression scales with surface tension allows us to predict the compression for a given aerogel, fluid, and temperature. This is especially important for very low density gels, with their associated low bulk moduli — for such samples permanent damage could result from the surface tension of liquid helium. At common filling temperatures, such as 4.2K, gels with porosities or 98% or more will likely be damaged.
Compression of aerogels at room temperature leads to bulk densification and important microstructural changes in the sample. Such changes have been investigatedBeurroies et al. (1998); Woignier et al. (1997) during the room temperature densification of silica aerogels in a mercury porosimeter. Three characteristics of the aerogel structure were measured — the fractal dimension, the solid particle size, and the cluster size (or correlation length). As the aerogel was compressed the particle size remained unchanged, consistent with the silica particles (that compose the aerogel strands) being relatively unaffected by small pressures. However, the fractal dimension of the aerogels increased slightly, and the correlation length decreased significantly, during compression. These three factors are all very important in determining how fluids are affected by the presence of aerogel, and they are usually assumed to remain constant throughout an experiment. However, if the aerogel is compressed during the experiment these factors may change.
The increase in aerogel density is generally assumed to be due to the preferential collapse of the largest “pores,” with relatively little damage to the smallest scale structures. However, the presence of large, open, pores is what distinguishes aerogel from other porous media and any damage to this structure may have dramatic effects on the behavior of fluids within the aerogels. To avoid such damage, one must be careful to avoid capillary stresses that could deform the gel beyond its elastic limit. The safest way to fill or empty an aerogel is above the liquid-vapor critical point, where no liquid-vapor interface exists. However, using calculations such as those shown in Table 1 allows one to estimate the degree of deformation induced during filling and emptying at lower temperatures, and a filling temperature can be chosen that does not allow the sample to experience plastic deformation.
### V.5 Shape of adsorption isotherms
Note that the amount of neon adsorbed or desorbed at 43K during capillary condensation — the steepest portion of the curve at $`P/P_00.9990`$(adsorption) or $`0.9983`$(desorption) — is about 65% of the total neon within the gel. The volumetric compression of the aerogel by about 10% during desorption is a significant portion of this, so that compression is a very important part of the capillary condensation behavior. Qualitatively the effect of aerogel compliance is to make the capillary condensation/desorption portions smaller (i.e. the fluid density change from point A to point B in Fig. 1 would be spread out over a wider range, from point A to point C if the aerogel was rigid). Simultaneously, the isotherms become steeper in a compliant material like aerogels since effective pore radius decreases during compression (by up to 3.5%).
## VI Summary
We have investigated the effect of small capillary pressures on the deformation of low density silica aerogels during adsorption and desorption of fluids. For most conditions in this study, this deformation was found to be completely elastic, but in the lower density aerogel (B51) permanent damage was seen. The compression of the aerogel during adsorption and desorption can be used to compute the bulk modulus of the aerogel, with no knowledge of the aerogel pore structure being necessary.
The highly compliant nature of low density aerogels also has important implications for the shape of their adsorption isotherms. This effect becomes less important as the degree of compression is reduced — it is not very significant in the helium adsorption/desorption isotherms in B110. This effect may be further reduced if the aerogel is not free to deform, such as when grown within the small pores of a metal sinter.
Finally, it should be pointed out that when low density silica aerogels are used as a method to introduce disorder into fluid systems such as helium or liquid crystals the aerogels may be damaged by any fluid interfaces present. Even the low surface tensions of <sup>3</sup>He and <sup>4</sup>He are capable of damaging low density aerogels during filling and emptying.
|
warning/0506/quant-ph0506030.html
|
ar5iv
|
text
|
# Cooling trapped atoms in optical resonators
## Abstract
We derive an equation for the cooling dynamics of the quantum motion of an atom trapped by an external potential inside an optical resonator. This equation has broad validity and allows us to identify novel regimes where the motion can be efficiently cooled to the potential ground state. Our result shows that the motion is critically affected by quantum correlations induced by the mechanical coupling with the resonator, which may lead to selective suppression of certain transitions for the appropriate parameters regimes, thereby increasing the cooling efficiency.
Cavity cooling is a recent expression, which stresses the role of the mechanical effects of a resonator on the atomic and molecular center-of-mass dynamics. Indeed, the coupling between resonator and atom gives rise to complex dynamics, one aspect of which is the substantial modification of the atom spectroscopic properties CQED94 . This property allows one to change the atom scattering cross section, thereby affecting, and eventually tailor, the mechanical dynamics of the atomic center of mass Domokos03 . Moreover, the motion of the atom changes the medium density, thereby affecting the resonator field itself. Several recent experiments have reported relevant features of these complex dynamics Rempe04 ; KimbleFORT03 ; Vuletic03 ; Ring03 ; Zimmerman04 ; Buschev04 , Experimental demonstrations of atom cooling in resonators Rempe04 ; KimbleFORT03 ; Vuletic03 ; Ring03 ; Zimmerman04 have shown, among others, that cavities are a promising tool for preparing and controlling cold atomic samples of scalable dimensions, which may find relevant applications, for instance in quantum information processing QuInfo .
In this letter we present a study of the quantum dynamics of the center of mass motion of an atomic dipole, which couples to a resonator and to a laser field driving it from the side. The system is sketched in Fig. 1. Differently from recent theoretical works on cavity cooling of atomic clouds Domokos02 , here the center of mass is confined by a tight trap, in a configuration that may correspond to the experimental situations reported for instance in KimbleFORT03 ; IonCQED ; Sauer03 . Starting from a master equation for the quantum variables of dipole, cavity and center-of-mass, we derive a closed equation for the center-of-mass dynamics. This equation generalizes previous theoretical studies Cirac95 ; Vuletic01 and allows us to identify novel parameter regimes, where cooling can be efficient. Moreover, its form permits us to identify the individual scattering contributions, thereby gaining insight into the role of the various physical parameters. We show that quantum correlations between atom and resonator may lead to the suppression of scattering transitions, thereby enhancing the cooling efficiency.
The starting point is the master equation for an atom of mass $`M`$ whose dipole transition between the ground and excited states $`|g`$ and $`|e`$ couples (quasi-)resonantly to a laser and the mode of an optical resonator with wave vector $`k_L`$ and $`k_c`$, respectively ($`|k_L||k_c|=k`$). In the reference frame of the laser the state $`|e`$ is at frequency $`\mathrm{\Delta }`$ and the cavity mode at $`\delta _c`$. The atomic center-of-mass is trapped by a harmonic oscillator of frequency $`\nu `$ along the $`x`$-axis, while the motion in the orthogonal plane is frozen out. Later on we discuss how the treatment can be generalized to three dimensional motion. The density matrix $`\rho `$ of cavity mode and atom internal and external degrees of freedom evolves according to the master equation $`\rho /t=\rho `$ where the Liouvillian $``$ describes the coherent and incoherent dynamics. External and internal degrees of freedom are coupled through the spatial gradient of the interaction with the electromagnetic field, which scales with the Lamb-Dicke parameter $`\eta =\sqrt{\mathrm{}k^2/2M\nu }`$ Stenholm86 . We assume tight confinement, such that the gradient of the e.m.-field over the atomic wave packet is small and the Lamb-Dicke regime holds, $`\eta 1`$, and treat the coupling of the center-of-mass with dipole and cavity variables in perturbation theory. At second order in $`\eta `$ we decompose $`=_0+\eta _1+\eta ^2_2+\mathrm{o}(\eta ^3)`$ where the subscripts indicate the corresponding order in the expansion. The term $`_0=_{0E}+_{0I}`$ is the sum of the Liouvillian for the external ($`E`$) and internal and cavity degrees of freedom ($`I`$). Here, $`_{0E}\rho =[H_{\mathrm{mec}},\rho ]/\mathrm{i}\mathrm{}`$ and $`H_{\mathrm{mec}}=\mathrm{}\nu (\widehat{n}+1/2)`$ center-of-mass harmonic oscillator with $`\widehat{n}`$ number operator for its phononic excitations; the term $`_{0I}\rho =[H_0,\rho ]/\mathrm{i}\mathrm{}+(_s+𝒦)\rho `$, where $`𝒦`$ and $`_s`$ describe cavity decay at rate $`\kappa `$ and dipole spontaneous emission at rate $`\gamma `$, respectively Carmichael93 , and $`H_0=\mathrm{}\mathrm{\Delta }\sigma ^{}\sigma \mathrm{}\delta _ca^{}a+\mathrm{}[\sigma ^{}(\mathrm{\Omega }+\stackrel{~}{g}a)+\mathrm{H}.\mathrm{c}.]`$, with $`\sigma =|ge|`$, $`a,a^{}`$ annihilation and creation operators of a cavity photon, $`\stackrel{~}{g}`$ coupling constant between cavity and dipole at the center of the trap, and $`\mathrm{\Omega }`$ laser Rabi frequency. The first-order correction $`_1\rho =[V,\rho ]/\mathrm{i}\mathrm{}`$ contains the operator $`V=(V_L+V_c)(b+b^{})`$, which gives the mechanical coupling induced by laser and resonator, whereby $`b`$, $`b^{}`$ are the creation and annihilation operators of a vibrational quantum, the operators
$`V_L`$ $`=`$ $`\mathrm{i}\mathrm{}\phi _L\mathrm{\Omega }(\sigma ^{}\sigma )`$ (1)
$`V_c`$ $`=`$ $`\mathrm{}\phi _c\stackrel{~}{g}(a\sigma ^{}+a^{}\sigma )`$ (2)
and the coefficients $`\phi _L`$, $`\phi _c`$ scale the corresponding recoil and depend on the geometry of the setup Footnote1 . The term $`_2`$ describes mechanical effects associated with the spontaneous emission of a photon, which determine diffusion, and state-dependent energy shifts Javanainen84 ; Cirac92 . The modifications induced by these shifts in the cavity field dynamics have been discussed in Rice04 . In the present treatment, where $`\eta 1`$, they are high-order corrections to the external potential and hence do not appreciably affect the center-of-mass dynamics.
At second order in the Lamb-Dicke expansion we derive a rate equation for the occupations of the number states $`|n`$ of the center-of-mass harmonic oscillator Javanainen84 ; ZippilliNew . The rate equation is expressed in terms of the heating rate $`\mathrm{\Gamma }_{nn+1}=\eta ^2(n+1)A_+`$ and the cooling rate $`\mathrm{\Gamma }_{nn1}=\eta ^2nA_{}`$, which describe transitions that change $`|n`$ by one excitation. Here,
$`A_\pm =\mathrm{Re}\{2\mathrm{T}\mathrm{r}_I\{V(_{0I}+\mathrm{i}\nu )^1V\rho _0\}+\alpha \gamma \mathrm{Tr}\{|ee|\rho _0\}\}`$
with $`\mathrm{Tr}_I`$ trace over the dipole and cavity degrees of freedom and $`\alpha `$ geometric factor giving the angular dispersion of atomic momentum due to the random recoil of spontaneous emission. Their explicit evaluation requires the knowledge of the spectrum of the correlation of the operator $`V`$ and of the steady state $`\rho _0`$ of dipole and cavity mode at zero order in $`\eta `$ Javanainen84 ; Cirac92 ; Morigi03 . When the laser weakly perturbs the atom-cavity ground state $`|\varphi _0=|g,0_c`$, we find ZippilliNew
$`A_\pm `$ $`=`$ $`\gamma \alpha |𝒯_S|^2+\gamma |\phi _L𝒯_L^{\gamma ,\pm }+\phi _c𝒯_c^{\gamma ,\pm }|^2`$
$`+\kappa |\phi _L𝒯_L^{\kappa ,\pm }+\phi _c𝒯_c^{\kappa ,\pm }|^2`$
where
$`𝒯_S=\mathrm{\Omega }{\displaystyle \frac{\delta _c+\mathrm{i}\kappa /2}{f(0)}}`$ (4)
$`𝒯_L^{\gamma ,\pm }=\mathrm{i}\mathrm{\Omega }{\displaystyle \frac{(\delta _c\nu +\mathrm{i}\kappa /2)}{f(\nu )}}`$ (5)
$`𝒯_L^{\kappa ,\pm }=\mathrm{i}\mathrm{\Omega }{\displaystyle \frac{\stackrel{~}{g}}{f(\nu )}}`$ (6)
$`𝒯_c^{\gamma ,\pm }=\mathrm{\Omega }{\displaystyle \frac{\stackrel{~}{g}^2(2\delta _c\nu +\mathrm{i}\kappa )}{f(0)f(\nu )}}`$ (7)
$`𝒯_c^{\kappa ,\pm }=\mathrm{\Omega }{\displaystyle \frac{\stackrel{~}{g}\left[(\mathrm{\Delta }\nu +\mathrm{i}\gamma /2)(\delta _c+\mathrm{i}\kappa /2)+\stackrel{~}{g}^2\right]}{f(0)f(\nu )}}`$ (8)
with
$`f(x)=(x+\delta _c+\mathrm{i}\kappa /2)(x+\mathrm{\Delta }+\mathrm{i}\gamma /2)\stackrel{~}{g}^2`$ (9)
Equations (Cooling trapped atoms in optical resonators) and (4)-(8) are the main result of this letter. They have been derived for one-dimensional motion. However, since at second order in $`\eta `$ the three directions of oscillation decouple in an anisotropic trap, they can be generalized to three-dimensional motion as they hold for any geometry of the setup. Their analytic form allows one for insight into these complex dynamics, which we now discuss.
The rates are the incoherent sum of three contributions. The first term on the right-hand side (rhs) of Eq. (Cooling trapped atoms in optical resonators), $`\gamma \alpha |𝒯_S|^2`$, describes absorption of a laser photon and spontaneous emission, whereby the change in the center-of-mass state is due to the recoil induced by the spontaneously emitted photon. This process is shown in Fig. 2(a), where the excitation paths are depicted in the dressed states basis $`|\pm ;n`$ of cavity and atom, which are superposition of the states $`|g,1_c`$ and $`|e,0_c`$ Domokos03 at fixed number of phonons $`|n`$. The scattering rate is scaled by the geometric factor $`\alpha `$ and is found after averaging over the solid angle of photon emission into free space. This contribution is diffusive, as the motion can be scattered into a higher or lower vibrational state with probabilities depending on the overlap integrals Cirac92 . Diffusion can be suppressed, namely term (4) can become very small, in good resonators, $`\kappa \stackrel{~}{g}`$, and for $`\delta _c=0`$. In this limit the corresponding dynamics are characterized by destructive interference between the two excitation paths $`|\varphi _0;n|\pm ;n`$ Alsing92 ; Zippilli04 ; Rice96 . This situation appears at zero order in the mechanical effects and is reminiscent of analogous phenomena encountered in the cooling dynamics of trapped multi-level atoms Morigi00 .
The second term on the rhs of Eq. (Cooling trapped atoms in optical resonators), $`\gamma |\phi _L𝒯_L^{\gamma ,\pm }+\phi _c𝒯_c^{\gamma ,\pm }|^2`$, describes scattering of a laser photon by spontaneous emission where the motion is changed by mechanical coupling to the laser ($`𝒯_L^{\gamma ,\pm }`$) and to the cavity ($`𝒯_c^{\gamma ,\pm }`$) field. Process $`𝒯_L^{\gamma ,\pm }`$ scales with the geometric factor $`\phi _L`$, which accounts for the recoil due to absorption of a laser photon, and is depicted in Fig. 2(b). The transition amplitude $`𝒯_c^{\gamma ,\pm }`$, shown in Fig. 2(c), describes the mechanical coupling between the dressed states $`|\pm `$ due to the resonator. It thus scales with the geometric factor $`\phi _c`$ which accounts for the recoil due to interaction with the cavity mode. Since the final state of the two scattering processes is the same, they interfere Footnote2 . In addition, each term is composed by multiple excitation paths, and can vanish in some parameter regimes. A representative case is found in good resonators, $`\kappa \stackrel{~}{g}`$, when the laser is orthogonal to the motion axis ($`\phi _L=0`$). Then, the transition amplitude, $`𝒯_c^{\gamma ,+}`$ is suppressed for $`\delta _c=\nu /2`$, as one can see in Eq. (7). This suppression arises from destructive interference between the multiple paths of mechanical coupling via the dressed states $`|\pm ;n`$ and $`|\pm ;n+1`$, thereby giving a vanishing heating rate. Similarly, the cooling transition can be suppressed by choosing $`\delta _c=\nu /2`$.
The third term in the rhs of Eq. (Cooling trapped atoms in optical resonators), $`\kappa |\phi _L𝒯_L^{\kappa ,\pm }+\phi _c𝒯_c^{\kappa ,\pm }|^2`$, describes scattering of a laser photon by cavity decay, where the motion is changed by mechanical coupling to the laser ($`𝒯_L^{\kappa ,\pm }`$) and to the cavity ($`𝒯_c^{\kappa ,\pm }`$) field. The scattered photon is transmitted through the cavity mirrors into the external modes, and therefore these two processes do not interfere with the ones above discussed. They are depicted in Fig. 2(d) and (e) and add up coherently with one another. An interesting situation has been discussed in Cirac95 , where it was shown that interference between these two terms can lead to suppression of the heating transition in bad resonators. The result of Cirac95 is recovered from Eqs. (Cooling trapped atoms in optical resonators)-(8) in the corresponding parameter regime ZippilliNew .
The general dynamics are a competition of all these processes. By inspection of the form of Eqs. (Cooling trapped atoms in optical resonators-9), one can identify a strategy for obtaining efficient cooling by searching for the parameters such that $`\text{Re}\{f(\nu )\}=0,`$ thereby minimizing the denominator of $`A_{}`$, and therefore enhancing the cooling rate over the heating rate. This condition leads to the equation
$`\mathrm{\Delta }_{\mathrm{opt}}(\delta _c){\displaystyle \frac{\stackrel{~}{g}^2+\gamma \kappa /4}{\delta _c+\nu }}\nu `$ (10)
which relates the cavity detuning $`\delta _c`$ to the atom detuning $`\mathrm{\Delta }`$ for fixed couplings and decay rates, such that, when $`\mathrm{\Delta }=\mathrm{\Delta }_{\mathrm{opt}}(\delta _c)`$, the cooling rate $`A_{}`$ is maximum. In general, Eq. (10) sets the best parameters for ground state cooling, provided that either $`\gamma \nu `$ or $`\kappa \nu `$. Below we discuss cooling in the good cavity limit. The bad cavity limit will be discussed elsewhere. Figures 3(a) and (b) are contour plots, displaying the average number of phonon at steady state, $`n_{\mathrm{}}=A_+/(A_{}A_+)`$, and the cooling rate, $`W=\eta ^2(A_{}A_+)`$, as a function of $`\mathrm{\Delta }`$ and $`\delta _c`$ for a good cavity and $`\kappa \nu \gamma `$, namely when processes in Fig. 2(a)-(c) are dominant. These curves have been obtained from Eq. (Cooling trapped atoms in optical resonators), and their validity has been checked by comparison with full numerical simulations ZippilliNew . The dashed line indicates the curve $`\mathrm{\Delta }=\mathrm{\Delta }_{\mathrm{opt}}(\delta _c)`$. It is evident that this curve corresponds to the parameter region where the lowest temperature is achieved. The final limit is set by the ratio $`\kappa /\nu `$, since in this regime $`\kappa `$ is the narrowest linewidth one can achieve for the scattering processes Footnote3 .
Good parameter regimes are found for relatively small values of $`\mathrm{\Delta }`$ and $`\delta _c`$, in the range $`0<\delta _c<\nu `$, $`\mathrm{\Delta }>0`$, where one has low temperatures and relatively large cooling rates for a wide range of values. These regions correspond to novel cooling dynamics, which are characterized by suppression of heating and diffusive transitions by quantum correlations induced by the resonator. Here, for instance, the potential ground-state occupation can reach $`99\%`$ in a time $`\tau 2`$ msec for an atom in a trap with frequency $`\nu =2\pi \times 500`$ KHz and $`\eta =0.1`$, whose dipole has line width $`\gamma =2\pi \times 5`$ MHz and couple to a resonator with $`g=2\pi \times 5`$ MHz and $`\kappa =2\pi \times 100`$ KHz. Larger values of $`\kappa `$ lead to smaller occupations, which scale like $`n_{\mathrm{}}\kappa ^2/\nu ^2`$ in good resonators for $`\kappa <\nu `$. Figure 3 shows also that the dynamics are relatively insensitive to fluctuations of the parameters in the range $`0<\delta _c<\nu `$, $`\mathrm{\Delta }>0`$. This implies that several oscillation modes, like the ones of an ion chain Morigi04 , could be simultaneously cooled. We remark that good cooling regions are found in Fig. 3 also for large $`|\mathrm{\Delta }|`$ and $`\delta _c\nu `$. This is the so–called cavity sideband cooling regime, discussed in Vuletic01 . Here, enhancement of the cooling rate is achieved by means of resonance with the very narrow transition, at large $`|\mathrm{\Delta }|`$, between the dressed states of the cavity-atom coupling. This regime exhibits a critical sensitivity to variations of the value of $`\delta _c`$ with respect to $`\mathrm{\Delta }`$.
To conclude, we have presented the explicit form of cooling and heating rates of the center-of-mass motion of an atom, trapped in an optical resonator and driven by a laser. This result permits us to identify the basic processes determining the center-of-mass dynamics and the parameter regions where cooling can be efficiently obtained. Novel regimes have been identified, where the motion is cooled by exploiting destructive interference in the multiple paths of the mechanical excitations. The resulting interference bases itself on the discreteness of the vibrational spectrum, which is the same for the dipolar ground and the excited state. Some analogies with interference processes in cooling of trapped multilevel atoms Marzoli94 ; Morigi00 are due to a similar dressed state structure at low excitation, nevertheless the mechanical effects induced by the resonator give rise to a wider wealth of phenomena. Dynamics will be substantially modified when the external potential depends appreciably on the internal state Rice04 ; Morigi03 . In future works we will investigate how the dynamics are modified by collective effects, when many atoms are trapped inside the cavity.
Beyond applications to cooling, these results allow one for gaining insight in the complex dynamics of the mechanical effects of optical resonators on atoms, and their wealth of phenomena could be exploited for implementing coherent control of this kind of systems.
We thank Helmut Ritsch and Axel Kuhn for discussions. Support from the IST-network QGATES and the spanish Ministerio de Educación y Ciencia (Ramon-y-Cajal fellowship 129170) are acknowledged.
|
warning/0506/astro-ph0506647.html
|
ar5iv
|
text
|
# Pair winds in Schwarzschild space-time with application to hot bare strange stars
## 1 Introduction
Compact astronomical objects identified with neutron stars, strange stars, and black holes are believed to be sources of electron-positron ($`e^\pm `$) pairs that form in their vicinity by different mechanisms and flow away. Among these objects are radio pulsars (Sturrock 1971, Ruderman & Sutherland 1975, Arons 1981; Usov & Melrose 1996; Baring & Harding 2001), accretion-disc coronas of the Galactic X-ray binaries (White & Lightman 1989; Sunyaev et al. 1992), soft $`\gamma `$-ray repeaters (Thompson & Duncan 1995; Usov 2001b), cosmological $`\gamma `$-ray bursters (Eichler et al. 1989; Paczyński 1990; Usov 1992), etc.. The estimated luminosity in $`e^\pm `$ pairs varies greatly depending on the object and the specific conditions: from $`10^{31}10^{35}`$ ergs s<sup>-1</sup> or less for radio pulsars up to $`10^{50}10^{52}`$ ergs s<sup>-1</sup> in cosmological $`\gamma `$-ray bursters.
For a wind out-flowing spherically from a surface of radius $`R`$ there is a maximum (isotropic, unbeamed) pair luminosity beyond which the pairs annihilate significantly before they escape (see Beloborodov 1999 and references therein). This is given by
$$L_\pm ^{\mathrm{max}}=\frac{4\pi m_ec^3R\mathrm{\Gamma }^2}{\sigma _\mathrm{T}}10^{36}(R/10^6\mathrm{cm})\mathrm{\Gamma }^2\mathrm{ergs}\mathrm{s}^1,$$
(1)
where $`\mathrm{\Gamma }`$ is the pair bulk Lorentz factor, and $`\sigma _\mathrm{T}`$ the Thomson cross section. When the injected pair luminosity, $`\stackrel{~}{L}_\pm `$, greatly exceeds this value the emerging pair luminosity , $`L_\pm `$ , cannot significantly exceed $`L_\pm ^{\mathrm{max}}`$; in this case photons strongly dominate in the emerging emission: $`L_\pm L_\pm ^{\mathrm{max}}\stackrel{~}{L}_\pm L_\gamma `$. Injected pair luminosities typical of cosmological $`\gamma `$-ray bursts (e.g., Piran 2000), $`\stackrel{~}{L}_\pm 10^{50}10^{52}`$ ergs s<sup>-1</sup>, greatly exceed $`L_\pm ^{\mathrm{max}}`$ for their estimated $`\mathrm{\Gamma }_\pm 10^2`$. For such a powerful wind the pair density near the source is very high, and the out-flowing pairs and photons are nearly in thermal equilibrium almost up to the wind photosphere (e.g., Paczyński 1990). The outflow process of such a wind may be described fairly well by relativistic hydrodynamics (Paczyński 1986, 1990; Goodman 1986; Grimsrud & Wasserman 1998; Iwamoto & Takahara 2002).
In contrast if $`\stackrel{~}{L}_\pm L_\pm ^{\mathrm{max}}`$ annihilation of the outflowing pairs is negligible. It is now commonly accepted that the magnetospheres of radio pulsars contain such a very rarefied ultra-relativistic ($`\mathrm{\Gamma }_\pm 1010^2`$) pair plasma that is practically collisionless (see Melrose 1995 for a review).
Recently we described a numerical code and the results of calculations of spherically out-flowing, non-relativistic ($`\mathrm{\Gamma }_\pm 1`$) pair winds with total luminosity in the range $`10^{34}10^{42}`$ ergs s<sup>-1</sup>, that is $`L(10^210^6)L_\pm ^{\mathrm{max}}`$ (Aksenov, Milgrom, & Usov 2004, hereafter Paper I). (A brief account of the emerging emission from such a pair wind has been given by Aksenov Milgrom & Usov 2003.) While our numerical code can be more generally employed, the results we presented in Paper I were for a hot, bare, strange star as the a wind injection source. Such stars are thought to be powerful sources of pairs created by the Coulomb barrier at the quark surface (Usov 1998, 2001a, see Fig. 1). For such luminosities it is not justified to assume thermal equilibrium, and so we have calculated the detailed microphysics including all two-body processes ($`eeee`$, $`\gamma e\gamma e`$, $`e^+e^{}\gamma \gamma `$) as well as their radiative three-body variants ($`eeee\gamma `$, $`\gamma e\gamma e\gamma `$, $`e^+e^{}\gamma \gamma \gamma `$). The relativistic kinetic Boltzmann equations for pairs and photons were solved numerically to describe the time evolution and structure of pair winds and to find the emergent emission (we only presented results for stationary situations). Gravity of the star was neglected, and it was assumed that only $`e^\pm `$ pairs are emitted from the stellar surface. Thermal emission of photons from a bare quark surface, which has been neglected in Paper I, is strongly suppressed relative to black-body emission, but not completely, (e.g., Cheng & Harko 2003; Jaikumar et al. 2004, see Fig. 1). Still, whatever little is emitted may significantly affect the wind properties (see below).
In the present paper we extend our previous studies to also include the effects of gravity by solving the Boltzmann equations for pairs and photons in a Schwarzschild space-time appropriate for the strange star. We also include thermal emission of photons from the stellar surface.
In § 2 we formulate the equations that describe the pair wind and the boundary conditions. In § 3 we describe the computational method used to solve these equations. In § 4 we present the results of our numerical simulations. Finally, in § 5, we discuss our results and some potential astrophysical applications.
## 2 Formulation of the problem
We consider an $`e^\pm `$ pair wind that flows away from a hot, bare, unmagnetized, non-rotating, strange star. Space-time outside the star is described by Schwarzschild’s metric with the line element
$$ds^2=e^{2\varphi }c^2dt^2+e^{2\varphi }dr^2+r^2(d\theta ^2+\mathrm{sin}^2\theta d\phi ^2),$$
(2)
where
$$e^\varphi =(1r_g/r)^{1/2}$$
(3)
is the lapse or redshift factor induced by gravity at a distance $`r`$ from the stellar center,
$$r_g=2GM/c^22.95\times 10^5M/M_{}\mathrm{cm}$$
(4)
is the gravitational radius, with $`M`$ the gravitational mass of the star.
Following Page & Usov (2002) we consider, as a representative case, a strange star with $`M=1.4M_{}`$ which is constructed by solving the Tolman-Oppenheimer- Volkoff equation of hydrostatic equilibrium using an equation of state for SQM as described in Alcock, Farhi, & Olinto (1986a) and in Haensel, Zdunik, & Schaefer (1986) \[with a bag constant $`B=(140\mathrm{M}\mathrm{e}\mathrm{V})^4`$, QCD coupling constant $`\alpha _cg^2/4\pi =0.3`$, and the mass of strange quarks $`m_s=150`$ MeV\]. The circumferential radius of the star is $`R=1.1\times 10^6`$ cm.
We assume that the temperature is constant over the surface of the star. The thermal emission in pairs and photons from the surface of strange quark matter (SQM) depends on the surface temperature, $`T_\mathrm{S}`$, alone (Usov 1998, 2001a; Cheng & Harko 2003; Jaikumar et al. 2004). The state of the plasma in the wind may be described by the distribution functions $`f_\pm (𝐩,r,t)`$ and $`f_\gamma (𝐩,r,t)`$ for positrons $`(+)`$, electrons $`()`$, and photons, respectively, where $`𝐩`$ is the momentum of particles. There is no emission of nuclei from the stellar surface, so the distribution functions of positrons and electrons are identical, $`f_+=f_{}`$.
Since we fix the mass and radius of the central star the only remaining parameters that determine the wind are the surface temperature, which determines the pair injection rate and spectrum, and the photon surface radiation, which for default of better knowledge we take to be thermal with a suppression factor, $`\eta `$, relative to black body; $`\eta `$ is our second free parameter. Because of the very steep dependence of the pair injection rate on the surface temperature, we actually use the former as a free parameter: in our calculations here $`\stackrel{~}{L}_\pm =10^{34},\mathrm{\hspace{0.17em}10}^{35},\mathrm{},\mathrm{\hspace{0.17em}10}^{39}`$ ergs s<sup>-1</sup>.
### 2.1 General Relativistic Boltzmann Equations
The Boltzmann equations in a Schwarzschild background for either massless or massive particles can be obtained as a special case from the results of Harleston & Holcomb (1991) and Harleston & Vishniac (1992). These are given in a conservative form, which makes them particularly amenable to numerical treatment:
$`{\displaystyle \frac{e^\varphi }{c}}{\displaystyle \frac{f_i}{t}}+{\displaystyle \frac{1}{r^2}}{\displaystyle \frac{}{r}}(r^2\mu e^\varphi \beta _if_i){\displaystyle \frac{e^\varphi }{p^2}}{\displaystyle \frac{}{p}}\left(p^3\mu {\displaystyle \frac{\varphi ^{}}{\beta _i}}f_i\right)`$
$`{\displaystyle \frac{}{\mu }}\left[(1\mu ^2)e^\varphi \left({\displaystyle \frac{\varphi ^{}}{\beta _i}}{\displaystyle \frac{\beta _i}{r}}\right)f_i\right]={\displaystyle \underset{q}{}}(\overline{\eta }_i^q\chi _i^qf_i).`$ (5)
Here, $`i`$ is the species type ($`i=e`$ for $`e^\pm `$ pairs and $`i=\gamma `$ for photons), $`\mu =\mathrm{cos}\theta `$, $`\theta `$ is the angle between the radius-vector $`𝐫`$ from the stellar center and the particle momentum $`𝐩`$, $`p=|𝐩|`$, $`\beta _e=v_e/c=[1(m_ec^2/ϵ_e)^2]^{1/2}`$, $`\beta _\gamma =1`$, and $`ϵ_e=c[p^2+(m_ec)^2]^{1/2}`$ is the energy of electrons and positrons (for photons $`ϵ_\gamma =pc`$). Also, $`\overline{\eta }_i^q`$ is the emission coefficient for the production of a particle of type $`i`$ via the physical process labelled by $`q`$, and $`\chi _i^q`$ is the corresponding absorption coefficient. The summation runs over physical processes that involve a particle of type $`i`$.
For convenience of numerical simulations we use instead of $`f_i`$ the quantities
$$E_i(ϵ,\mu ,r,t)=\frac{2\pi ϵ^3\beta _if_i}{c^3},$$
(6)
standardly used in the “conservative” numerical method. This can provide exact conservation of energy on a finite computational grid (see below). $`E_i`$ is the energy density in the $`\{𝐫,\mu ,ϵ\}`$ phase space. Since all equations are the same for all particle species from now on we omit the index $`i`$.
From equations (5) and (6) the Boltzmann equations can be written in terms of $`E`$ as
$`{\displaystyle \frac{e^\varphi }{c}}{\displaystyle \frac{E}{t}}+\mu e^\varphi {\displaystyle \frac{}{r^2r}}(r^2\beta e^{2\varphi }E)\mu e^\varphi \varphi ^{}{\displaystyle \frac{}{ϵ}}(ϵ\beta E)`$
$`e^\varphi {\displaystyle \frac{}{\mu }}\left[(1\mu ^2)\left({\displaystyle \frac{\varphi ^{}}{\beta }}{\displaystyle \frac{\beta }{r}}\right)E\right]={\displaystyle \underset{q}{}}(\eta ^q\chi ^qE),`$ (7)
where
$$\eta ^q=\frac{2\pi ϵ^3\beta \overline{\eta }^q}{c^3}.$$
(8)
### 2.2 Boundary conditions
The computational boundaries are
$$R<r<r_{\mathrm{ext}},\mathrm{\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}0}<t<t_{\mathrm{st}},$$
(9)
where $`r_{\mathrm{ext}}=1.66\times 10^8`$ cm, and $`t_{\mathrm{st}}`$ is the time when the wind approaches stationarity close enough, since here we concentrate on steady-state winds.
Considering the interior of the quark star, it was shown that a thin ($`10^{10}`$ cm) layer of electrons with an extremely strong electric field–the electrosphere–forms at the surface of strange quark matter (SQM) (Alcock et al. 1986a; Kettner et al. 1995), which prevents the electrons from escaping to infinity. The electric field in the elecrosphere is a few ten times higher than the critical field, $`E_{\mathrm{cr}}1.3\times 10^{16}`$ V cm<sup>-1</sup>, at which the vacuum is unstable to creation of $`e^+e^{}`$ pairs Schwinger (1951). Therefore, a hot, bare strange star is potentially a powerful source of $`e^+e^{}`$ pairs created in the electrosphere and flowing away from the star (Usov 1998). We use in our simulations a pair injection rate (Usov 2001a):
$$\dot{N}_\pm ^{\mathrm{in}}=4\pi R^2F_\pm ,$$
(10)
where
$`F_\pm =3\times 10^{42}\mathrm{exp}(0.593\zeta )`$
$`\times \left[{\displaystyle \frac{\mathrm{ln}(1+2\zeta ^1)}{(1+0.074\zeta )^3}}+{\displaystyle \frac{\pi ^5\zeta }{2(13.9+\zeta )^4}}\right]\mathrm{cm}^2\mathrm{s}^1,`$ (11)
and $`\zeta =20(T_\mathrm{S}/10^9\mathrm{K})^1`$. The energy spectrum of injected pairs is thermal with the surface temperature $`T_\mathrm{S}`$, and their angular distribution is isotropic for $`0\mu 1`$.
Thus, the pair injection luminosity, our first free parameter, is given by
$$\stackrel{~}{L}_\pm =\dot{N}_\pm ^{\mathrm{in}}[m_ec^2+(3/2)k_\mathrm{B}T_\mathrm{S}],$$
(12)
where $`k_\mathrm{B}`$ is the Boltzmann constant (Usov 2001a). For the range we consider, $`\stackrel{~}{L}_\pm =10^{34}10^{39}`$ ergs s<sup>-1</sup>, the surface temperature is in a rather narrow range, $`T_\mathrm{S}(46)\times 10^8`$ K (see Fig. 1).
Thermal emission of photons from the surface of a bare strange star is strongly suppressed if the surface temperature is not very high, $`T_\mathrm{S}10^{11}`$ K (Alcock et al. 1986a). The reason is that the plasma frequency of quarks in SQM is very large, $`\mathrm{}\omega _\mathrm{p}2025`$ MeV, and only hard photons with energies $`ϵ_\gamma >\mathrm{}\omega _\mathrm{p}`$ can propagate in the SQM. The luminosity in such hard photons, which are in thermodynamic equilibrium with quarks, decreases very fast for $`T_\mathrm{S}\mathrm{}\omega _\mathrm{p}/k_\mathrm{B}`$ (Chmaj, Haensel, & Slomiński 1991; Usov 2001a) and in our case, where $`T_\mathrm{S}10^9`$ K, it is negligible (see Fig. 1). However, low-energy ($`ϵ_\gamma <\mathrm{}\omega _\mathrm{p}`$) photons may still leave the SQM if they are produced by nonequilibrium processes in the surface layer of thickness $`c/\omega _\mathrm{p}10^{12}`$ cm (Chmaj et al. 1991). The emissivity of SQM in nonequilibrium quark-quark bremsstrahlung radiation has been estimated by Cheng & Harko (2003) who find that it is suppressed at least by a factor of $`10^6`$ in comparison with black-body emission, $`\stackrel{~}{L}_\gamma 10^6L_{\mathrm{bb}}`$. Usov (2004) and Usov, Harko, & Cheng (2005) have recently considered a modified model of the electrosphere taking into account surface effects for SQM . The modification of the electrosphere results in an increase in the density of electrons by a factor of $`2030`$ in comparison with the case when the surface effects are ignored. The electron density increase can additionally suppress the outgoing radiation from SQM in nonequilibrium quark-quark bremsstrahlung photons. In our simulations we take
$$\stackrel{~}{L}_\gamma =\eta L_{\mathrm{bb}}=\eta \mathrm{\hspace{0.17em}4}\pi \sigma R^2T_\mathrm{S}^4,$$
(13)
where $`\eta `$ is a dimensionless free parameter. We present results with $`\eta =0,3\times 10^8,10^6`$. In default of better knowledge the photon spectrum is taken as black-body with the surface temperature $`T_\mathrm{S}`$ corresponding to $`\stackrel{~}{L}_\pm `$.
An important parameter characterizing the affect of photon emission from the surface on the out-flowing wind is the Eddington luminosity $`L_{\mathrm{Edd}}^\pm `$ for the pair plasma, above which the radiation pressure force dominates over gravity,
$$L_{\mathrm{Edd}}^\pm 10^{35}(M/1.4M_{})\mathrm{ergs}\mathrm{s}^1.$$
(14)
Within the narrow range of surface temperatures studied here, $`T_\mathrm{S}(46)\times 10^8`$ K, we get $`\stackrel{~}{L}_\gamma `$ that is $`(110)L_{\mathrm{Edd}}^\pm `$ for $`\eta =3\times 10^8`$, and $`(10^210^3)L_{\mathrm{Edd}}^\pm `$ for $`\eta =10^6`$ (see Fig. 1).
The stellar surface is assumed to be a perfect mirror for both $`e^\pm `$ pairs and photons. At the external boundary ($`r=r_{\mathrm{ext}}`$) pairs and photons escape freely from the studied region, i.e., the inward ($`\mu <0`$) fluxes of both $`e^\pm `$ pairs and photons vanish there.
### 2.3 Physical processes in the pair plasma
As the plasma moves outwards photons are produced by pair annihilation ($`e^+e^{}\gamma \gamma `$). Other two-body processes that occur in the outflowing plasma, and are included in the simulations, are Møller ($`e^+e^+e^+e^+`$ and $`e^{}e^{}e^{}e^{}`$) and Bhaba ($`e^+e^{}e^+e^{}`$) scattering, Compton scattering ($`\gamma e\gamma e`$), and photon-photon pair production ($`\gamma \gamma e^+e^{}`$).
Two-body processes do not change the total number of particles in the system, and thus cannot in themselves lead to thermal equilibrium. So, we also include radiative processes (bremsstrahlung, double Compton scattering, and three-photon annihilation with their inverse processes), even though their cross-sections are at least $`\alpha ^110^2`$ times smaller than those of the two-body processes ($`\alpha =e^2/\mathrm{}c=1/137`$ is the fine structure constant).
## 3 The computational method
Our grid in the $`\{𝐫,\mu ,ϵ\}`$ phase-space is defined as follows. The $`r`$ domain ($`R<r<r_{\mathrm{ext}}`$) is divided into $`j_{\mathrm{max}}`$ spherical shells whose boundaries are designated with half integer indices. The $`j`$ shell ($`1jj_{\mathrm{max}}`$) is between $`r_{j1/2}`$ and $`r_{j+1/2}`$, with $`\mathrm{\Delta }r_j=r_{j+1/2}r_{j1/2}`$ ($`r_{1/2}=R`$ and $`r_{j_{\mathrm{max}}+1/2}=r_{\mathrm{ext}}`$).
The $`\mu `$-grid is made of $`k_{\mathrm{max}}`$ intervals $`\mathrm{\Delta }\mu _k=\mu _{k+1/2}\mu _{k1/2}`$: $`1kk_{\mathrm{max}}`$.
The energy grids for photons and electrons are both made of $`\omega _{\mathrm{max}}`$ energy intervals $`\mathrm{\Delta }ϵ_\omega =ϵ_{\omega +1/2}ϵ_{\omega 1/2}`$: $`1\omega \omega _{\mathrm{max}}`$, but the lowest energy for photons is 0, while that for pairs is $`m_ec^2`$.
The quantities we compute are the energy densities averaged over phase-space cells
$$E_{\omega ,k,j}(t)=\frac{1}{\mathrm{\Delta }X}_{\mathrm{\Delta }ϵ_\omega ,\mathrm{\Delta }\mu _k,\mathrm{\Delta }r_j}E𝑑ϵ𝑑\mu r^2𝑑r.$$
(15)
where $`\mathrm{\Delta }X=\mathrm{\Delta }ϵ_\omega \mathrm{\Delta }\mu _k\mathrm{\Delta }(r_j^3)/3`$ and $`\mathrm{\Delta }(r_j^3)=r_{j+1/2}^3r_{j1/2}^3`$.
Replacing the space, angle, and energy derivatives in the Boltzmann equations (7) by finite differences (e.g., Mezzacappa & Bruenn 1993) we have the following set of ordinary differential equations (ODEs) for $`E_{\omega ,k,j}`$ specified on the computational grid:
$`{\displaystyle \frac{e^{\varphi _j}}{c}}{\displaystyle \frac{dE_{\omega ,k,j}}{dt}}+e^{\varphi _j}\beta _\omega {\displaystyle \frac{\mathrm{\Delta }(r^2\mu _ke^{2\varphi }E_{\omega ,k})_j}{\mathrm{\Delta }(r_j^3)/3}}`$
$`\mu _ke^{\varphi _j}\varphi _j^{}{\displaystyle \frac{\mathrm{\Delta }(ϵ\beta E_{k,j})_\omega }{\mathrm{\Delta }ϵ_\omega }}e^{\varphi _j}\left({\displaystyle \frac{\varphi _j^{}}{\beta _\omega }}{\displaystyle \frac{1}{r}}_j\beta _\omega \right)`$
$`\times {\displaystyle \frac{\mathrm{\Delta }\left[(1\mu ^2)E_{\omega ,j}\right]_k}{\mathrm{\Delta }\mu _k}}={\displaystyle \underset{q}{}}[\eta _{\omega ,k,j}^q(\chi E)_{\omega ,k,j}^q],`$ (16)
where
$$\beta _\omega =\{\begin{array}{cc}1\hfill & \text{for photons,}\hfill \\ [1(m_ec^2/ϵ_\omega )^2]^{1/2}\hfill & \text{for electrons,}\hfill \end{array}$$
(17)
$$ϵ_\omega =\frac{ϵ_{\omega 1/2}+ϵ_{\omega +1/2}}{2},$$
(18)
$$\mu _k=\frac{\mu _{k1/2}+\mu _{k+1/2}}{2},$$
(19)
$$r_j=\frac{r_{11/2}+r_{j+1/2}}{2}.$$
(20)
$$\varphi _j=\varphi (r_j),\varphi _j^{}=\varphi ^{}(r_j)$$
(21)
$$\frac{1}{r}_j=\frac{(r_{j+1/2}^2r_{j1/2}^2)/2}{(r_{j+1/2}^3r_{j1/2}^3)/3},$$
(22)
$$E_{\omega ,k}(r)=\frac{1}{\mathrm{\Delta }ϵ_\omega \mathrm{\Delta }\mu _k}_{\mathrm{\Delta }ϵ_\omega \mathrm{\Delta }\mu _k}E(ϵ,\mu ,r)𝑑ϵ𝑑\mu ,$$
(23)
$$E_{\omega ,j}(\mu )=\frac{3}{\mathrm{\Delta }ϵ_\omega \mathrm{\Delta }r_j^3}_{\mathrm{\Delta }ϵ_\omega \mathrm{\Delta }r_j}E(ϵ,\mu ,r)𝑑r𝑑ϵ,$$
(24)
$$E_{k,j}(ϵ)=\frac{3}{\mathrm{\Delta }\mu _k\mathrm{\Delta }r_j^3}_{\mathrm{\Delta }\mu _k\mathrm{\Delta }r_j^3}E(ϵ,\mu ,r)𝑑r𝑑\mu ,$$
(25)
$`\mathrm{\Delta }(r^2\mu _ke^{2\varphi }E_{\omega ,k})_j=r_{j+1/2}^2e^{2\varphi _{j+1/2}}(\mu _kE_{\omega ,k})_{r=r_{j+1/2}}`$
$`r_{j1/2}^2e^{2\varphi _{j1/2}}(\mu _kE_{\omega ,k})_{r=r_{j1/2}},`$ (26)
$`\mathrm{\Delta }\left(ϵ\beta E_{k,j}\right)_\omega =ϵ_{\omega +1/2}\beta _{\omega +1/2}(E_{k,j})_{\omega +1/2}`$
$`ϵ_{\omega 1/2}\beta _{\omega 1/2}(E_{k,j})_{\omega 1/2},`$ (27)
$`\mathrm{\Delta }\left[(1\mu ^2)E_{\omega ,j}\right]_k=(1\mu _{k+1/2}^2)(E_{\omega ,j})_{\mu =\mu _{k+1/2}}`$
$`(1\mu _{k1/2}^2)(E_{\omega ,j})_{\mu =\mu _{k1/2}},`$ (28)
$`(E_{k,j})_{\omega +1/2}=\{\begin{array}{cc}E_{\omega +1,k,j}\hfill & \\ +\frac{(E_{\omega +2,k,j}E_{\omega +1,k,j})(ϵ_{\omega +1/2}ϵ_{\omega +1})}{ϵ_{\omega +2}ϵ_{\omega +1}},\hfill & \\ \mu 0\hfill & \\ E_{\omega ,k,j}\hfill & \\ +\frac{(E_{\omega ,k,j}E_{\omega 1,k,j})(ϵ_{\omega +1/2}ϵ_\omega )}{ϵ_\omega ϵ_{\omega 1}},\hfill & \\ \mu <0,\hfill & \end{array}`$ (29)
$`(E_{\omega ,j})_{\mu =\mu _{k+1/2}}=E_{\omega ,k,j}`$
$`+{\displaystyle \frac{\mathrm{\Delta }\mu _k(E_{\omega ,k,j}E_{\omega ,k1,j})}{\mathrm{\Delta }\mu _{k1}+\mathrm{\Delta }\mu _k}},`$ (30)
$`(\mu _kE_{\omega ,k})_{r=r_{j+1/2}}=(1\stackrel{~}{\chi }_{\omega ,k,j+1/2})`$
$`\times \left({\displaystyle \frac{\mu _k+|\mu _k|}{2}}E_{\omega ,k,j}+{\displaystyle \frac{\mu _k|\mu _k|}{2}}E_{\omega ,k,j+1}\right)`$
$`+\stackrel{~}{\chi }_{\omega ,k,j+1/2}\mu _k{\displaystyle \frac{E_{\omega ,k,j}+E_{\omega ,k,j+1}}{2}},`$ (31)
$`\stackrel{~}{\chi }_{\omega ,k,j+1/2}^1=1+{\displaystyle \frac{1}{\chi _{\omega ,k,j}\mathrm{\Delta }r_j}}+{\displaystyle \frac{1}{\chi _{\omega ,k,j}\mathrm{\Delta }r_{j+1}}}.`$ (32)
The dimensionless coefficient $`\stackrel{~}{\chi }`$ is introduced to describe correctly both optically thin and optically thick computational cells by means of a compromise between the high order method and the monotonic transport scheme without the artificial viscosity (e.g., Richtmeyer & Morton 1967; Mezzacappa & Bruenn 1993; Aksenov 1998). It is worth noting that if the values of $`(E_{k,j})_{\omega +1/2}`$ and $`(E_{\omega ,j})_{\mu =\mu _{k+1/2}}`$ calculated by equations (29) and (30), respectively, are negative they are taken to be zero.
The right side terms of equation (16) are
$$\eta _{\omega ,k,j}^q=\frac{1}{\mathrm{\Delta }X}_{\mathrm{\Delta }ϵ_\omega ,\mathrm{\Delta }\mu _k,\mathrm{\Delta }r_j}\eta ^q𝑑ϵ𝑑\mu r^2𝑑r.$$
(33)
$$(\chi E)_{\omega ,k,j}^q=\frac{1}{\mathrm{\Delta }X}_{\mathrm{\Delta }ϵ_\omega ,\mathrm{\Delta }\mu _k,\mathrm{\Delta }r_j}\chi E^q𝑑ϵ𝑑\mu r^2𝑑r.$$
(34)
For the physical processes included in our simulations expressions $`\eta _{\omega ,k,j}^q`$ and $`(\chi E)_{\omega ,k,j}^q`$ are calculated in Paper I.
In paper I, where gravity is neglected, particle number is explicitly conserved when only two-body processes are taken into account; particle production occurs only via three-body processes. In the general relativistic scheme developed in this paper we were not able to achieve exact number conservation. However, in all runs for a stationary solution the change of the particle flux due to computational effects is at most a few percents. Energy conservation is still exactly satisfied.
There are several characteristic times in the system. Some are related to particle reaction times, some to the time to reach steady state. These times may greatly differ from each other, especially at high luminosities ($`10^{38}`$ ergs s<sup>-1</sup>) when the pair wind is optically thick. The set of equations (16) is then stiff: at least some eigenvalues of the Jacobi matrix differ significantly from each other, and the real parts of the eigenvalues are negative. In contrast to Mezzacappa & Bruenn (1993) we use Gear’s method (Hall & Watt 1976) to solve the ODEs finite differences analog (16) of the Boltzmann equations. This high-order, implicit method was developed especially to find a solution of stiff sets of ODE. To solve a system of linear algebraic equations at any time step of Gear’s method we use the cyclic reduction method (Mezzacappa & Bruenn 1993). The number of operations per time step is $`(\omega _{\mathrm{max}}k_{\mathrm{max}})^3j_{\mathrm{max}}`$, which increases rapidly with the increase of $`\omega _{\mathrm{max}}`$ and $`k_{\mathrm{max}}`$. Therefore, the numbers of energy and angle intervals have to be rather limited in our simulations.
Here we use an $`(ϵ,\mu ,r)`$-grid with $`\omega _{\mathrm{max}}=13`$, $`k_{\mathrm{max}}=8`$, and $`j_{\mathrm{max}}=100`$. The discrete energies (in keV) of the $`ϵ`$-grid minus the rest mass of the particles are 0, 2, 27, 111, 255, 353, 436, 491, 511, 530, 585, 669, 766, and $`\mathrm{}`$. This gives a denser grid at low energies and near the threshold of pair production, $`ϵ=m_ec^2`$. The $`\mu `$-grid is uniform, with $`\mathrm{\Delta }\mu _k=2/k_{\mathrm{max}}=1/4`$. The shell thicknesses are geometrically spaced: $`\mathrm{\Delta }r_1=2\times 10^4\text{ cm}`$, and $`\mathrm{\Delta }r_j=1.3\mathrm{\Delta }r_{j1}`$.
To check the effects of grid coarseness we also performed test computations with $`k_{\mathrm{max}}=4`$ and 6, and separately with $`\omega _{\mathrm{max}}=10`$ and 11. We did not observe changes in the results.
## 4 Numerical results
In this section, we give the results for the properties of spherically symmetric winds consisting of $`e^\pm `$ pairs and photons. The pair injection luminosity $`\stackrel{~}{L}_\pm `$ and $`\eta `$ are the parameters in our simulations. The photon injection luminosity $`\stackrel{~}{L}_\gamma `$ is determined by equation (13) where the surface temperature $`T_\mathrm{S}`$ relates to the value of $`\stackrel{~}{L}_\pm `$ \[see equations (10) - (12)\]. We start from an empty wind, injecting both pairs at a rate $`10^{34}`$ ergs s<sup>-1</sup> and photons at a corresponding rate. After a steady state is reached we start a new run with this steady state as initial condition, increase the energy injection rate in pairs by a factor of 10, wait for steady state, and so on.
We next present the results for the structure of the stationary winds and their emergent emission at the external boundary.
Figure 2 shows the mean optical depth $`\tau _\gamma (r)`$ for photons, from $`r`$ to $`r_{\mathrm{ext}}`$. The wind as a whole is optically thick $`[\tau _\gamma (R)>1]`$ for $`\stackrel{~}{L}_\pm 10^{37}`$ ergs s<sup>-1</sup> irrespective of $`\eta `$. For $`\stackrel{~}{L}_\pm =10^{39}`$ ergs s<sup>-1</sup> the radius of the wind photosphere $`r_{\mathrm{ph}}`$, determined by condition $`\tau (r_{\mathrm{ph}})=1`$, is $`4\times 10^6`$ cm (see Fig. 2). The wind photosphere is always deep inside our chosen external boundary ($`r_{\mathrm{ph}}r_{\mathrm{ext}}`$), justifying our neglect of the inward ($`\mu <0`$) fluxes at $`r=r_{\mathrm{ext}}`$.
Figures 3 and 4 show the pair number density ($`n_e`$) and the bulk velocity of the pair plasma ouflow ($`v_e^{\mathrm{out}}`$), respectively, as functions of the distance from the stellar surface. For high luminosities ($`\stackrel{~}{L}_\pm 10^{37}`$ ergs s<sup>-1</sup>) both these quantities hardly depend on $`\eta `$, i.e., the pair wind structure is determined only by $`\stackrel{~}{L}_\pm `$. For low luminosities ($`\stackrel{~}{L}_\pm 10^{35}`$ ergs s<sup>-1</sup>), $`n_e`$ and $`v_e^{\mathrm{out}}`$ depend significantly on $`\eta `$. In particular, for $`\stackrel{~}{L}_\pm =10^{34}`$ ergs s<sup>-1</sup> near the stellar surface ($`rR10^5`$ cm) the pair density for $`\eta =0`$ is $`6`$ times higher than for $`\eta =10^6`$, while the velocity of the pair plasma is $`6`$ times smaller. This is because for a surface temperature $`T_\mathrm{S}4\times 10^8`$ K, which corresponds to $`\stackrel{~}{L}_\pm =10^{34}`$ ergs s<sup>-1</sup>, the mean kinetic energy of pairs near the surface, $`(3/2)k_\mathrm{B}T_\mathrm{S}8.3\times 10^8`$ ergs is about half of the gravitational binding energy, so an atmosphere may form. In addition, for low surface photon luminosities ($`\eta \mathrm{a}\mathrm{few}\times 10^8`$) radiation pressure is too weak to suppress the plasma atmosphere that results from the presence of gravity. So, an atmosphere does form, and, in turn, is conducive to pair annihilation. In contradistinction for $`\eta =10^6`$ radiation pressure does suppress the formation of an atmosphere, and the pair wind structure, i.e., the runs of $`n_e(r)`$ and $`v_e^{\mathrm{out}}(r)`$, is very similar to the case with no gravity and $`\eta =0`$.
Figure 5 shows the rates of outflow of $`e^\pm `$ pairs ($`\dot{N}_\pm `$) and photons ($`\dot{N}_\gamma `$) through the surface as functions of $`r`$. For high luminosities ($`\stackrel{~}{L}_\pm 10^{37}`$ ergs s<sup>-1</sup>) the rate of pair outflow $`\dot{N}_\pm `$ almost doesn’t depend on $`\eta `$ and decreases significantly at the distance
$$l_{\mathrm{ann}}10\left(\frac{\stackrel{~}{L}_\pm }{10^{39}\mathrm{ergs}\mathrm{s}^1}\right)^1\mathrm{cm}$$
(35)
from the stellar surface because of pair annihilation (see the estimate of $`l_{\mathrm{ann}}`$ in Paper I). For low luminosities ($`\stackrel{~}{L}_\pm 10^{35}`$ ergs s<sup>-1</sup>) the dependence of $`\dot{N}_\pm `$ on $`r`$ is different for different values of $`\eta `$. For $`\eta =0`$ the value of $`\dot{N}_\pm `$ decreases outwards at least by a factor of 2 even when $`\stackrel{~}{L}_\pm `$ is as small as $`10^{34}`$ ergs s<sup>-1</sup>, while for $`\eta =10^6`$ the decrease of $`\dot{N}_\pm `$ due to pair annihilation is very small for low luminosities, again reflecting the degree to which a pair atmosphere is formed.
The photon outflow rate is seen to increase with increasing $`r`$ not only because of pair annihilation, but also because of accumulative production of photons by radiative three-body processes. These processes are important for high luminosities ($`\stackrel{~}{L}_\pm 10^{38}`$ ergs s<sup>-1</sup>) and are responsible for the increase of $`\dot{N}_\gamma `$ at $`rR10^310^5`$ cm for such high luminosities (see Fig. 5).
The rates of energy outflow in $`e^\pm `$ pairs ($`\dot{E}_\pm `$) and photons ($`\dot{E}_\gamma `$) vary with radius more or less similarly to the particle outflow rates, except that the total energy rate is explicitly conserved in all processes (see Fig. 6). The total energy rate decreases with increase of $`r`$ because of the gravity effects,
$$\dot{E}(r)=\dot{E}_\pm (r)+\dot{E}_\gamma (r)=e^{2(\stackrel{~}{\varphi }\varphi )}\stackrel{~}{L}=\frac{1r_g/R}{1r_g/r}\stackrel{~}{L},$$
(36)
and varies from $`\stackrel{~}{L}=\stackrel{~}{L}_\pm +\stackrel{~}{L}_\gamma `$ at $`r=R`$ to the total emerging luminosity
$$L=L_\pm +L_\gamma =\dot{E}(r_{\mathrm{ext}})=\frac{1r_g/R}{1r_g/r_{\mathrm{ext}}}\stackrel{~}{L}$$
(37)
at $`r=r_{\mathrm{ext}}`$. This differs qualitatively from Paper I where the gravity effects are ignored, and $`\dot{E}(r)`$ is constant.
The number rates of emerging pairs $`(\dot{N}_\pm )`$ as functions of $`\stackrel{~}{L}_\pm `$ for different values of $`\eta `$ are shown in Figure 7. For $`\stackrel{~}{L}_\pm =10^{34}`$ ergs s<sup>-1</sup> the value of $`\dot{N}_\pm `$ for $`\eta =0`$ is $`3`$ times smaller than the same for $`\eta =10^6`$. The results of our new simulations for $`\eta =10^6`$ and our old simulations with $`\eta =0`$ and no gravity practically coincide. This is consistent with our results on the wind structure. For $`\stackrel{~}{L}_\pm 10^{37}`$ ergs s<sup>-1</sup> $`\dot{N}_\pm `$ hardly depends on $`\eta `$, and is $`1.52`$ times smaller than the result of Paper I (see Fig. 7). This is due to partial suppression of pair creation as the photon energies are reduced by gravitational redshift.
Figure 8 shows the emerging total luminosities in $`e^\pm `$ pairs ($`L_\pm `$, including the rest mass) and photons ($`L_\gamma `$)–given as fractions of the total emerging luminosity $`L=L_\pm +L_\gamma `$–as functions of the injection luminosity, and for different values of $`\eta `$. For the luminosity range we consider photons dominate in the emerging emission ($`L_\gamma >L_\pm `$), especially at high luminosities where $`L_\gamma L_\pm `$.
Figure 9 presents the energy spectra of the emerging photons for different values of $`\stackrel{~}{L}_\pm `$ and $`\eta `$. For low luminosities ($`\stackrel{~}{L}_\pm 10^{36}`$ ergs s<sup>-1</sup>) and $`\eta =0`$, photons, which form by pair annihilation, escape more or less freely from the star’s vicinity, and the photon spectra represent a very wide annihilation line that is redshifted in the gravitational field. The mean energy of the emerging photons varies from $`430`$ keV for $`\stackrel{~}{L}_\pm =10^{34}`$ ergs s<sup>-1</sup> to 400 keV for $`\stackrel{~}{L}_\pm =3\times 10^{36}`$ ergs s<sup>-1</sup> (see Fig. 10). If $`\eta `$ is near its maximum value ($`10^6`$) there is no annihilation line in the spectra of emerging photons. Annihilation photons only modify these spectra by producing high-energy tails (see Fig. 9c). For high luminosities ($`\stackrel{~}{L}_\pm 10^{38}`$ ergs s<sup>-1</sup>) the energy spectra of emerging photons practically don’t depend on $`\eta `$ and coincide with the results of Paper I.
## 5 Discussion
We have identified certain characteristics of pair winds outflowing from hot, bare, strange stars and their emerging emission. For the energy injection rate in pairs we consider ($`\stackrel{~}{L}_\pm =10^{34}10^{39}`$ ergs s<sup>-1</sup>) photons dominate in the emerging emission (see Fig. 8.). The spectrum of the emerging photons, we find, is rather hard (see Figs. 9 and 10) and differs qualitatively from the spectrum of the thermal emission from a neutron star with the same luminosity (e.g., Romani 1987; Shibanov et al. 1992; Rajagopal & Romani 1996; Page et al. 2004). In particular, for a bare strange star the mean energy of the emerging photons (see Fig. 10) is at least an order of magnitude larger than the same for a neutron star. This opens observational possibilities to distinguish strange stars from neutron stars. Hard spectra of bare strange stars are amenable to detection and study by sensitive, hard X-ray and soft $`\gamma `$-ray instruments, such as INTEGRAL (e.g., Winkler et al. 2003).
In this study we take into account both the thermal emission of photons from the strange star surface and gravity that have been neglected in Paper I. We have shown that for low values of $`\eta \mathrm{a}\mathrm{few}\times 10^8`$ and $`\stackrel{~}{L}_\pm 10^{35}`$ ergs s<sup>-1</sup> pairs emitted by the stellar surface are mainly captured by the gravitational field, and a pair atmosphere forms. The probability of pair annihilation increases because of the increase of the pair number density in the atmosphere, and this results in decrease of the fraction of pairs in the emerging emission in comparison with the case of Paper I (see Figs. 7 and 8). However, if the surface emission in photons is as high as its upper limit ($`\eta 10^6`$) radiation pressure forces dominate over gravity, and the pair wind structure practically coincides with the results for no gravity, irrespective of $`\stackrel{~}{L}_\pm `$.
The dominant physical processes in pair winds change significantly depending on $`\stackrel{~}{L}_\pm `$. For $`\stackrel{~}{L}_\pm <L_{}10^{37}`$ ergs s<sup>-1</sup> the optical depth for photons is smaller than unity, $`\tau _\gamma <1`$ (see Fig. 2). In this case some part of $`e^\pm `$ pairs ejected from the stellar surface annihilates into photons in the process of their outflow, while photo-creation of pairs is rather rare. Above $`\stackrel{~}{L}_\pm L_{}`$ pair creation by photons becomes important, and for $`\stackrel{~}{L}_\pm L_{}`$ the rates of pair creation and pair annihilation are more or less comparable at $`l_{\mathrm{ann}}rRr_{\mathrm{ph}}`$. At such distances from the surface the number density of pairs is nearly constant (see Fig. 3 and Fig. 8 in Paper I). In addition, for high luminosities ($`\stackrel{~}{L}_\pm 10^{38}`$ ergs s<sup>-1</sup>) radiative three-body processes are important, and the total rate of the particle outflow increases with radius (see Fig. 5). These processes favour thermalization of pairs and photons in the outflow.
For low luminosities ($`\stackrel{~}{L}_\pm <L_{}`$) the spectrum of emerging photons significantly depends on the photo-emission from the surface, especially at low energies, $`ϵ_\gamma m_ec^2`$ (see Fig. 9). If surface photo-emission is negligible ($`\eta =0`$), this spectrum resembles a very wide annihilation line. The fractional emerging luminosity in the annihilation line decreases with the increase of $`\eta `$. For $`\eta =10^6`$ the annihilation line is completely washed out, and a high-energy part of this line is observed as a high-energy tail (see Fig. 9c). For high luminosities ($`\stackrel{~}{L}_\pm 10^{38}`$ ergs s<sup>-1</sup>) the spectrum of emerging photons is practically independent of the photon emission from the stellar surface.
Soft $`\gamma `$-ray repeaters (SGRs), which are the sources of short bursts of hard X-rays with super-Eddington luminosities (up to $`10^{42}10^{47}`$ ergs s<sup>-1</sup>), are potential candidates for strange stars (e.g., Alcock et al. 1986b; Cheng & Dai 1998; Usov 2001c; Ouyed et al. 2005). The bursting activity of SGRs may be explained by fast heating of the stellar surface up to a temperature of $`(13)\times 10^9`$ K and its subsequent thermal emission (Usov 2001b,c). The heating mechanism may be either impacts of comets onto bare strange stars (Zhang et al. 2000; Usov 2001b) or fast decay of superstrong ($`10^{14}10^{15}`$ G) magnetic fields (Thompson & Duncan 1995; Heyl & Kulkarni 1998). Since for high luminosities ($`10^{38}`$ ergs s<sup>-1</sup>) the spectrum of emerging photons practically doesn’t depend on the surface photo-emission, which is poorly known, it is possible to calculate more securely many properties (light curves, energy spectra, etc.) of powerful ($`L10^{38}`$ ergs s<sup>-1</sup>) X-ray bursts expected from the stellar surface heating neglecting surface photo-emission. We plan to apply the tools developed here to study these problems.
We thank members of the Israeli Center for High Energy Astrophysics, D. Eichler, A. Levinson, T. Piran, and E. Waxman, for helpful discussions. The research was supported by the Israel Science Foundation of the Israel Academy of Sciences and Humanities.
|
warning/0506/cond-mat0506095.html
|
ar5iv
|
text
|
# Rectification in one–dimensional electronic systems
## I Introduction
Current rectification, also known as the diode effect, occurs whenever the transport characteristics in electronic conductors are asymmetric under the application of the driving voltage. This asymmetry can have various origins. Best known perhaps is the mismatch of band structures in macroscopic diodes, which leads to a potential wall that blocks the motion of the particles in one direction, while it is not seen by the particles moving in the opposite direction.
In the last years, rectification in mesoscopic and single–molecule systems has attracted much attention. Even though rectification by single asymmetric molecules has been suggested 30 years agoAviram and Ratner (1974), it has been realized experimentally in the 1990s onlyGeddes et al. (1990); Martin et al. (1993); Joachim et al. (2000). Asymmetric wave guides were constructed in the inversion layer of semiconductor heterostructuresLinke et al. (1999); Löfgren et al. (2003). Transport asymmetries have been observed in carbon nanotubesPostma et al. (2001); Papadopoulos et al. (2004) and in tunneling in quantum Hall edge statesRoddaro et al. (2003). The experimental progress has been accompanied by much theoretical activityChristen and Büttiker (1996); Reimann et al. (1997); Lehmann et al. (2002); Scheidl and Vinokur (2002); Komnik and Gogolin (2003); Sánchez and Büttiker (2004); Spivak and Zyuzin (2004); Feldman et al. (2005) with the main focus on Fermi–liquid systems.
An interesting source for rectification arises when the particle interaction is strong enough that a single–particle description becomes invalid. This naturally occurs in one–dimensional systems, for which a Luttinger liquid behavior is expected. It has been known for some time nowKane and Fisher (1992) that in such systems the current is strongly affected by the presence of impurities, even when they are weak, due to their renormalization by the electron–electron interactions. One can thus expect that asymmetries in the impurity distribution leads to strongly asymmetric current–voltage curves. This issue was addressed very recently in the framework of Luttinger liquids of spinless particlesFeldman et al. (2005). The *rectification current* $`I_r(V)=[I(V)+I(V)]/2`$ can be measured as the dc response to a low-frequency square voltage wave of amplitude $`V`$ and expresses the asymmetry of the $`IV`$ curve for forward and reverse bias. It was shown that a single weak asymmetric impurity is sufficient for a pronounced rectification effect (see Fig. 1 for a sketch of the system), leading to a large rectification current, $`I_r(V)`$, at low voltages $`V`$. Moreover an unusual behavior of the current was revealed for systems with strong repulsive interactions: A power–law dependence of $`I_r`$ on the voltage with a negative exponent, $`I_rV^{|z|}`$, within a range $`V^{}<V<V^{}`$ \[with $`V^{}`$ and $`V^{}`$ being expressed by some powers of the impurity potential $`U`$; see also below\], i.e. the rectification current *increases* as the voltage is lowered. At $`V<V^{}`$ the increase crosses over into a regular decrease such that the equilibrium condition $`I=0`$ at $`V=0`$ is met. The qualitative behavior is shown in Fig. 2.
In this paper we pursue this work for real electrons carrying a spin. We show that we can provide lower and upper limits for the voltage $`V`$, within which perturbation theory can be applied and determine the leading power–law behavior of the rectification current, $`I_rV^z`$, as a function of the charge and spin interaction strengths, $`g_c`$ and $`g_s`$. This leads to a phase diagram for the leading power–law dependence of $`I_r`$ with a much richer structure than in the spinless case (Fig. 3). We show that, similar to the spinless case, a negative exponent $`z`$, $`I_rV^{|z|}`$, appears for strong repulsive electron–electron interactions within a voltage range that is determined by the bare impurity scattering strength. Fig. 3 shows our main result, the leading power–law behavior of the rectification current $`I_r`$. The hatched region in the figure marks the range of $`(g_c,g_s)`$, in which the exponent $`z`$ becomes negative and in which the rectification current shows the behavior sketched in Fig. 2.
The paper is organized as follows: In the next section, we introduce the model of electrons in a one–dimensional system, and argue how scattering on a single impurity can lead to rectification. We then quantitatively address this problem within the bosonization approach, and show that the most relevant power–law behavior of the rectification current can be obtained within second and third order perturbation theory. The results are listed in Table 1 and in Fig. 3. In the appendix we show that higher order perturbative contributions cannot exceed the leading second and third order expressions.
## II Model and Origin of Rectification
Let us consider a one–dimensional conductor of length $`L`$ with electron–electron interactions that are effectively short–ranged due to screening by gates. Such a system allows a description by the Tomonaga–Luttinger model, given by the Hamiltonian
$`H`$ $`={\displaystyle }\mathrm{d}x{\displaystyle \underset{\sigma }{}}\{\mathrm{}v_F[\psi _{R\sigma }^{}(x)i_x\psi _{R\sigma }(x)\psi _{L\sigma }^{}(x)i_x\psi _{L\sigma }(x)]`$
$`+U(x)\psi _\sigma ^{}(x)\psi _\sigma (x)\}`$
$`+{\displaystyle dxdy\underset{\sigma \sigma ^{}}{}K_{\sigma \sigma ^{}}(xy)\psi _\sigma ^{}(x)\psi _\sigma ^{}^{}(y)\psi _\sigma ^{}(y)\psi _\sigma (x)},`$ (1)
where $`\psi _{R\sigma }`$ and $`\psi _{L\sigma }`$ are the operators for right– and left–moving electrons with spin $`\sigma =,`$, and $`\psi _\sigma =\psi _{R\sigma }+\psi _{L\sigma }`$ is the conventional electron operator. $`U_\sigma (x)`$ is the asymmetric potential (i.e. $`U(x)U(x)`$) localized in a small region about $`x=0`$. $`K(xy)`$ describes the electron–electron interaction. For the following discussion it is important to assume that the long–ranged Coulomb interaction is screened by the gates, so that $`K(xy)`$ becomes a short–ranged, rapidly decreasing function of $`(xy)`$.
On its two ends, at $`x=\pm L/2`$, the system is adiabatically coupled to electrodes that serve as reservoirs of particles, and whose chemical potentials $`\mu _{1,2}`$ are controlled experimentally: We assume that one electrode is fixed at the ground, $`\mu _2=\mu `$, while the other one is connected to the voltage source, $`\mu _1=\mu +eV`$. For such situations, it is possible to distinguish between two effects, addressed more quantitatively below, that lead to rectificationFeldman et al. (2005):
(1) *The “injected-density driven” rectification* as the result of the dependence of the charge density on the voltage: For simplicity, let us consider noninteracting particles. For the voltages $`\pm V`$, only electrons in the energy ranges between $`[\mu ,\mu +eV]`$ and $`[\mu eV,\mu ]`$ (at zero temperature) can contribute to the transport. The presence of a scatterer $`U`$ with an energy–dependent transmission coefficient $`R(E)`$ in the system leads to different transmission coefficients for $`\pm V`$ and thus to different currents. This is seen as follows: For noninteracting particles, the reflection coefficient $`R(E)`$ is independent of the propagation directionLandau and Lifshitz (1964), and the current is $`I(V)_{\mu _2}^{\mu _1}[1R(E)]dE`$. If the bandwidth $`E_F`$ is the only relevant scale for the energy dependence of $`R(E)`$, then, for small $`U`$ and $`V`$, we have $`R(E)U^2/E_F^2`$ and the rectification current $`I_r_0^{eV}[R(E_FE)R(E_F+E)]dE2R^{}(E_F)_0^{eV}EdER(E_F)(eV)^2/E_FU^2(eV)^2/E_F^3`$. The rectification current is nonzero. For noninteracting particles it is proportional to $`V^2`$. As shown below, a modified power–law dependence on $`V`$ is obtained in the presence of electron–electron interactions.
We remark that this argument does not require a spatially asymmetric impurity potential since the asymmetry is introduced through the injected charge densities.
(2) *The “asymmetric-impurity driven” rectification effect* is independent of injected densities. It results from the renormalization of the asymmetric potential $`U(x)`$ by the electron–electron interactions, which leads to the asymmetric current–voltage curves. For the Luttinger liquid, this naturally involves multi–particle processes, so that all such possible terms have to be taken into account in the modeling. As shown in Ref. Feldman et al., 2005, this effect is absent in the first two orders in the scattering potential $`U`$, and we must address it perturbatively at the order $`U^3`$.
The “asymmetry-driven” rectification effect can be qualitatively visualized (in a mean–field picture) as follows: In an interacting system, electrons are backscattered by a combined potential $`\stackrel{~}{U}(x)=U(x)+W(x)`$, where $`W(x)`$ is the self–consistent electrostatic potential created by the average local (nonuniform) charge density. Depending on the voltage sign, the density decreases or increases as a function of the position $`x`$ and the magnitude of $`U(x)`$ (see Fig. 1). Asymmetric impurities create different density profiles for opposite voltages, and lead to different combined backscattering potentials $`\stackrel{~}{U}(x)`$. The rectification effect is a consequence of the modification of the current by the backscattering potential.
## III Bosonization
For the quantitative treatment of the rectification effect we use the bosonization technique, which has become a standard tool for one–dimensional problemsGiamarchi (2004). In the bosonization language, the creation of a right– or left–moving electron at the coordinate $`x`$ is expressed by $`\psi _{\nu \sigma }^{}(x)\eta _{\nu \sigma }^{}\mathrm{exp}(+i[\pm k_{F\nu }x+\varphi _{\nu \sigma }(x)])`$, for $`\nu =R,L`$, where $`k_{F\nu }/\pi `$ are the densities of right– and left–moving particles, where $`\eta _{\nu \sigma }^{}`$, the Klein factors, raise the total number of right– or left–moving electrons with spin $`\sigma `$ by one, and where $`\varphi _{\nu \sigma }(x)`$ is a boson field that describes the dressing of this particle by a chain of particle–hole excitations. In correlation functions, the Klein factors keep track of particle conservation and can contribute with signs to the expressions, as they anticommute for different $`(\nu ,\sigma )`$. These signs are all equal in the present calculations since we assume conservation of the $`S_z`$ component of the spin by the scattering process. This allows us to drop the Klein factors in the following expressions.
In one dimension, the charge and spin degrees of freedom of the Hamiltonian (1) decouple, and it is convenient to set $`\varphi _c=_{\nu =L,R}[\varphi _\nu +\varphi _\nu ]/2`$ and $`\varphi _s=_{\nu =L,R}[\varphi _\nu \varphi _\nu ]/2`$, which are bosonic fields related to the charge and spin densities as $`\rho _c=e[_x\varphi _c+2k_F]/\pi `$ and $`\rho _s=(\mathrm{}/2)_x\varphi _s/\pi `$, where $`k_F=[k_{FR}+k_{FL}]/2`$.
For $`U=0`$ the bosonization leads to a quadratic action for the fields $`\varphi _{c,s}`$, which can be written in the form
$$𝒮_0=dxdt\underset{a=c,s}{}\frac{1}{8\pi g_a}\left[(_t\varphi _a)^2(_x\varphi _a)^2\right].$$
(2)
The quantities $`g_c`$ and $`g_s`$ are the charge and spin interaction strengths resulting from the screened electron–electron interaction $`K(xy)`$. A noninteracting system is characterized by $`g_c=g_s=1/2`$. Repulsive (attractive) interactions are expressed by $`g_c<1/2`$ ($`g_c>1/2`$). Interactions with conserved $`SU(2)`$ spin symmetry have $`g_s=1/2`$. For $`g_s<1/2`$, a neglected sine–Gordon term of $`\varphi _s`$ in $`𝒮_0`$, describing spin–flip backscattering, would become relevant. The physics of such systems are considerably different to the conductors described here (see e.g. Ref. Giamarchi, 2004), since the fields $`\varphi _s`$ would order and become massive. We exclude such situations explicitly by assuming $`g_s1/2`$, such that all backscattering terms in $`𝒮_0`$ are scaled to zero.
If we assume that $`U(x)`$ is concentrated in a small region $`a_UL`$ about $`x=0`$, the backscattering can be described by the values of the fields at $`x=0`$ only, $`\mathrm{\Phi }_{c,s}(t)\varphi _{c,s}(x=0,t)`$. This “single–point” description of $`U(x)`$, restricts the model to wavelengths longer than $`a_U`$, i.e. to energy scales smaller than $`E_U\mathrm{}v_F/a_U`$ (with $`v_F`$ the Fermi velocity). Since the characteristic energy scale is set by the voltage, $`E_U`$ provides an upper limit for the applied voltage, $`eVE_U`$, and the validity of the model. Under this assumption, the impurity term has the formKane and Fisher (1992),
$$𝒮_{\text{imp}}=dt\underset{n_c,n_s}{^{}}U(n_c,n_s)\mathrm{e}^{i\alpha (n_c,n_s)}\mathrm{e}^{in_c\mathrm{\Phi }_c+in_s\mathrm{\Phi }_s}.$$
(3)
The integers $`n_{c,s}`$ physically express the transfer of $`n_c`$ charges and $`n_s`$ spins between the right– and the left–moving electrons (where, for instance, $`n_c=n_s=1`$ corresponds to the backscattering of an up–spin electron, and $`n_c=0,n_s=2`$ to the backscattering of an up–spin electron and a down–spin electron with opposite incident directions). $`U(n_c,n_s)`$ and $`\alpha (n_c,n_s)`$ are modulus and phase of the effective multi–particle backscattering potential (for symmetric potentials, $`U(x)=U(x)`$, the $`\alpha (n_c,n_s)`$ would vanish). Forward scattering can be absorbed in $`𝒮_0`$ by a shift of $`\varphi _c`$. Since charges and spins are bound to physical electrons, the sum (indicated by the next to it) runs over $`n_{c,s}`$ such that $`n_c+n_s`$ is even. $`|n_c|=|n_s|=1`$ corresponds to the usual $`2k_F`$ backscattering of electrons. The most relevant contribution to the currents arises from the backscattering processes with $`n_c+n_s=0,\pm 2`$. These terms dominate in the “density-driven” rectification effect, but higher orders in $`n_c+n_s`$ are required for the “asymmetry-driven” rectification effect, and the full sum over $`n_{c,s}`$, therefore, must be kept.
Throughout the following analysis, $`U(x)`$ is assumed to be weak, $`|U|E_F`$ (with $`E_F`$ bandwidth). Due to the strong renormalization of the impurity strength through the electron–electron interactions, the precise nature of $`U`$ is of no major importance; $`U(n_c,n_s)`$ is an effective many–particle potential, dressed by short–time electron processes with frequencies well above $`E_U`$. Its magnitude can be roughly estimated asKane and Fisher (1992)
$$U(n_c,n_s)k_Fdx\mathrm{e}^{2ik_Fn_cx}U(x),$$
(4)
which means that $`U(n_c,n_s)U`$. Its further renormalization by low–energy processes, leading to a power–law correction in $`V`$, is considered below. The coefficients $`U(n_c,n_s)`$ depend further on the applied voltage through the densities $`k_{FL}`$ and $`k_{FR}`$.
Since the Hamiltonian is Hermitian, $`U(n_c,n_s)=U(n_c,n_s)`$ and $`\alpha (n_c,n_s)=\alpha (n_c,n_s)`$. Because of the spin symmetry, $`U(n_c,n_s)=U(n_c,n_s)`$ and $`\alpha (n_c,n_s)=\alpha (n_c,n_s)`$. For spatially asymmetric potentials, as considered here, $`\alpha (n_c,n_s)\alpha (n_c,n_s)`$.
If $`U=0`$, the system is characterized by right–moving particles injected from the left reservoir with the chemical potential $`\mu _1`$, and left–moving particles injected from the right reservoir with chemical potential $`\mu _2`$. Due to the absence of backscattering in Eq. (2), the right– and left–movers are in equilibrium with the corresponding reservoirs. In the presence of an impurity $`U`$, the equilibrium notion of chemical potentials becomes meaningless in the system; the quantities $`\mu _{1,2}`$ enter only through the voltage dependence of the injected carrier densities in the description and appear in the backscattering term that couples right– and left–moving particles. Explicitly, this can be seen by switching to the interaction representation $`H\widehat{H}=H\mu _1N_R\mu _2N_L`$ for $`U=0`$. This shifts the single particle energies of the injected particles, and the electron operators become $`\psi _R(x,t)\widehat{\psi }_R(x,t)=\mathrm{e}^{i\mu _1t/\mathrm{}}\psi _R(x,t)`$ and $`\psi _L(x,t)\widehat{\psi }_L(x,t)=\mathrm{e}^{i\mu _2t/\mathrm{}}\psi _L(x,t)`$. Hence, for $`U0`$, the backscattering of every charge described by $`𝒮_{\text{imp}}`$ acquires a time–dependence on the difference of chemical potentials, $`\mu _1\mu _2=eV`$, as $`\mathrm{e}^{\pm ieVt/\mathrm{}}`$. For the $`n_c`$ backscattered charges, this becomes
$`𝒮_{\text{imp}}={\displaystyle dt}`$ $`\underset{n_c,n_s}{{\displaystyle ^{}}}U(n_c,n_s)\mathrm{e}^{i\alpha (n_c,n_s)}`$ (5)
$`\times \mathrm{e}^{in_c\mathrm{\Phi }_c+in_s\mathrm{\Phi }_s}\mathrm{e}^{in_ceVt/\mathrm{}}.`$
The time–dependence of the backscattering operator reflects the nonequilibrium constraints. In order to calculate expectation values, we have to consider an appropriate nonequilibrium technique, such as provided by the Keldysh methodKeldysh (1965); Rammer and Smith (1986): We assume that in the remote past, $`t\mathrm{}`$, the system is fully described by $`𝒮_0`$ and the ground state $`|0`$ of the shifted Hamiltonian $`H\mu _1N_R\mu _2N_L`$. Averages are taken over this well–defined ground state $`|0`$ only, while the impurity term $`𝒮_{\text{imp}}`$ is taken into account through the time evolution operator $`S(\mathrm{},t)`$, given by
$$S(t^{},t^{\prime \prime })=T\mathrm{exp}\left(\frac{i}{\mathrm{}}_t^{}^{t^{\prime \prime }}dtH_{\text{imp}}(t)\right),$$
(6)
with $`H_{\text{imp}}`$ the impurity Hamiltonian in the interaction representation, and $`T`$ the time order operator. The expectation value of an operator $`O(t)`$ is expressed by
$$O(t)=0|S(\mathrm{},t)O(t)S(t,\mathrm{})|0,$$
(7)
with $`S(t,\mathrm{})=S(\mathrm{},t)^{}`$.
## IV Currents
In a clean system, the current flow is proportional to the voltage, and is given by the Landauer formulaMaslov and Stone (1995); Ponomarenko (1995); Safi and Schulz (1995), $`I_0(V)=2Ve^2/h`$ (the factor 2 accounts for the two spin channels). In the presence of the impurity scatterer $`U`$, the backscattering corrects the current as $`I(V)=I_0(V)+I_{bs}(V)`$. This shows that the rectification current depends only on the backscattering currents, $`I_r(V)=[I_{bs}(V)+I_{bs}(V)]/2`$. The backscattering current operator can be obtained, for instance, by the time variation of the number $`N_R`$ of right–moving particles, $`I_{bs}=\frac{\mathrm{d}}{\mathrm{d}t}N_R=i[H,N_R]/\mathrm{}`$. If we set $`e=\mathrm{}=v_F=1`$ for the ease of notation, this yields, in the interaction representation,
$$I_{bs}(V,t)=i\underset{n_c,n_s}{^{}}n_cU(n_c,n_s)\mathrm{e}^{i\alpha (n_c,n_s)}\mathrm{e}^{in_c\mathrm{\Phi }_c+in_s\mathrm{\Phi }_s}\mathrm{e}^{in_cVt},$$
(8)
and we need to calculate
$$I_{bs}(V)=0|S(\mathrm{},0)I_{bs}(V,0)S(0,\mathrm{})|0,$$
(9)
where $`S(t,t^{})`$ is the time evolution operator, and $`|0`$ the ground state of the system described by $`𝒮_0`$.
## V Results from weak coupling theory
For a weak impurity potential, the currents can be calculated within perturbation theory.
As shown by Kane and FisherKane and Fisher (1992), the backscattering potential is renormalized by the electron–electron interactions and scales as a power–law with the characteristic energy (here set by $`V`$) as $`U_{\text{eff}}(n_c,n_s)U(n_c,n_s)(V/E_F)^{g_cn_c^2+g_sn_s^21}`$. Perturbation theory is applicable when $`U_{\text{eff}}E_F`$.
For $`n_c^2g_c+n_s^2g_s>1`$, this condition is always met for $`V,UE_F`$, while for $`n_c^2g_c+n_s^2g_s<1`$, we must have
$$VV_{n_c,n_s}^{}(U/E_F)^{1/(1n_c^2g_cn_s^2g_s)}E_F,$$
(10)
for any admissible values of $`n_c,n_s`$. We set $`V_{n_c,n_s}^{}=0`$ whenever $`n_c^2g_c+n_s^2g_s>1`$. Since $`g_s1/2`$, only $`n_s=0,1`$ have to be considered. We always have $`V_{2n+2,0}^{}<V_{2n,0}^{}`$ and $`V_{2n+1,1}^{}<V_{2n1,1}^{}`$, so that the only important quantities are $`V_{1,1}^{}`$ and $`V_{2,0}^{}`$. For $`g_c+g_s<1`$ and $`g_c<1/4`$ both $`V_{1,1}^{}`$ and $`V_{2,0}^{}`$ are nonzero, and we have $`V_{1,1}^{}>V_{2,0}^{}`$ for $`g_s<3g_c`$. The lower cutoff energy is given by
$$V^{}=\mathrm{max}[V_{1,1}^{},V_{2,0}^{}].$$
(11)
Interestingly, if $`V^{}>0`$, perturbation theory requires a voltage $`V`$ that is not too small, $`V>V^{}`$, which is much in contrast to the usual perturbation theory, in which $`V`$ would be required to be a small parameter close to $`V=0`$.
The upper energy limit is set by the energy $`E_U`$, at which the corrections to the model become important. The perturbation theory is valid in the range $`eV^{}eVE_U`$.
In this case, the rectification current is dominated by the second and third order perturbative expressions from an expansion in powers of $`U`$. Due to the renormalization of the potentials, and due to the constraints on particle conservation and even sums of $`n_c+n_s`$, third order expressions can become larger than the second order ones and have to be taken into account. In the appendix we further show that, for $`V>V^{}`$, higher orders cannot exceed the second and third order expressions, and can be safely neglected.
### V.1 Second order
As mentioned, the rectification effect arises from the asymmetry of the charge/spin density profiles. The “injected-density-driven” rectification effect is due to the dependence of the density on the voltage, and hence depends on how the coefficients $`U(n_c,n_s)`$ change upon the variation of the density (see Eqs. (3) and (4)). The dominant contribution appears at second order in $`U`$, which reads after expanding Eq. (9)
$$\begin{array}{c}I_{bs}^{(2)}(V)=_{C_K}dt\underset{n,m}{^{}}in_cU(n)U(m)\hfill \\ \hfill \times \mathrm{e}^{i\alpha (n)+i\alpha (m)}\mathrm{e}^{im_cVt}T_c\mathrm{e}^{in\mathrm{\Phi }(0)}\mathrm{e}^{im\mathrm{\Phi }(t)}_0,\end{array}$$
(12)
where we have used the notations $`n=(n_c,n_s)`$ and $`n\mathrm{\Phi }(t)=n_c\mathrm{\Phi }_c(t)+n_s\mathrm{\Phi }_s(t)`$. $`\mathrm{}_0`$ is the average over the ground state of $`𝒮_0`$, $`C_K`$ is the Keldysh contour $`\mathrm{}0\mathrm{}`$, and $`T_c`$ is the chronological order operator on $`C_K`$. Particle conservation imposes $`m=n`$. Since $`\alpha (n)=\alpha (n)`$ the phases $`\alpha `$ cancel each other. Ignoring the voltage dependence of $`U`$, this expression, therefore, changes the sign upon $`VV`$, and would naively not contribute to the rectification effect (which is a consequence of the invariance of the action $`𝒮_0`$ under the change $`\varphi _{c,s}\varphi _{c,s}`$). However, since $`U`$ depends on the voltage through the density $`k_F`$ (see Eq. (4)), we can expand it in powers of $`V`$ as $`U=U_0+VU_1+\mathrm{}`$. At small voltages, $`VE_F`$, the correction is linear, and if the bandwidth $`E_F`$ is the only relevant scale for the energy dependence of $`U`$, we have $`U_1\frac{\mathrm{d}U_0}{\mathrm{d}E}U_0/E_F`$. The rectification current can be written as (choosing $`V>0`$)
$$\begin{array}{c}I_r^{(2)}(V)=2V_{C_K}dt\underset{n}{^{}}in_cU_0(n)U_1(n)\mathrm{e}^{in_cVt}\hfill \\ \hfill \times T_c\mathrm{e}^{in\mathrm{\Phi }(0)}\mathrm{e}^{in\mathrm{\Phi }(t)}_0.\end{array}$$
(13)
The propagators in the latter equation have the usual form known from the Luttinger liquid theoryGiamarchi (2004):
$$\begin{array}{c}T_c\mathrm{e}^{in_c\mathrm{\Phi }_c(0)+in_s\mathrm{\Phi }_s(0)}\mathrm{e}^{in_c\mathrm{\Phi }_c(t)in_s\mathrm{\Phi }_s(t)}_0\hfill \\ \hfill =\mathrm{e}^{n_c^2T_c\mathrm{\Phi }_c(0)\mathrm{\Phi }_c(t)_0+n_s^2T_c\mathrm{\Phi }_s(0)\mathrm{\Phi }_s(t)_0},\end{array}$$
(14)
where the free propagators are given by
$$\mathrm{\Phi }_{c,s}(t_1)\mathrm{\Phi }_{c,s}(t_2)_0=2g_{c,s}\mathrm{ln}\left(i(t_1t_2)+\delta \right),$$
(15)
with $`\delta >0`$ an infinitesimal constant. We can now extract the voltage dependence of the current by changing to the time variable $`s=Vt`$. The propagators become
$$\begin{array}{c}T_c\mathrm{e}^{in_c\mathrm{\Phi }_c(0)+in_s\mathrm{\Phi }_s(0)}\mathrm{e}^{in_c\mathrm{\Phi }_c(t)in_s\mathrm{\Phi }_s(t)}_0\hfill \\ \hfill V^{2n_c^2g_c+2n_s^2g_s}.\end{array}$$
(16)
From the integration measure, we obtain another factor $`V^1`$, which cancels the $`V`$ from the expansion of the potential. Hence the rectification current has the form
$$I_r^{(2)}(V)=\underset{n_c,n_s}{^{}}n_cC(n)V^{z^{(2)}(n)}$$
(17)
with $`C`$ constants of the order of $`CU_0U_11`$, and
$$z^{(2)}(n)=z^{(2)}(n_c,n_s)=2n_c^2g_c+2n_s^2g_s.$$
(18)
The leading order is obtained by minimizing the exponent, which is achieved by one of the processes $`(n_c,n_s)=(2,0),(0,2),(1,1)`$ (and the combinations obtained by changing signs) only, as backscattering of more particles leads to larger $`z^{(2)}`$. Terms with $`n_c=0`$ do not couple to the voltage ($`e^{in_cVt}1`$) and coincide with the equilibrium value at $`V=0`$. Therefore, their amplitude vanishes. The comparison of the remaining two processes shows the dominance of $`(n_c,n_s)=(1,1)`$ \[labeled as (A) in Fig. 3\] for $`g_s<3g_c`$ and $`(n_c,n_s)=(2,0)`$ for $`g_s>3g_c`$ \[(B) in Fig. 3\]. The noninteracting system is characterized by $`g_c=g_s=1/2`$, leading to $`z=2`$. This result certainly agrees with the result from the application of the Landauer formalism to a noninteracting problem.
### V.2 Third order
The “asymmetry-driven” rectification effect appears at third order. The contribution to the backscattering currents becomes
$$\begin{array}{c}I_{bs}^{(3)}(V)=\frac{1}{2!}_{C_K}dt_1dt_2\underset{n,m,l}{^{}}in_c\mathrm{e}^{im_cVt_1+il_cVt_2}\hfill \\ \hfill \times U(n)U(m)U(l)\mathrm{e}^{i\alpha (n)+i\alpha (m)+i\alpha (l)}\\ \hfill \times T_c\mathrm{e}^{in\mathrm{\Phi }(0)}\mathrm{e}^{im\mathrm{\Phi }(t_1)}\mathrm{e}^{il\mathrm{\Phi }(t_2)}_0,\end{array}$$
(19)
which has to be evaluated with the constraints $`n_c+m_c+l_c=n_s+m_s+l_s=0`$, together with $`n_c+n_s,m_c+m_s`$, and $`l_c+l_s`$ being even. Again, diagrams in which $`n_c=m_c=l_c=0`$, do not couple to the voltage and vanish. The third order term is no longer invariant under the transformation $`(V,\mathrm{\Phi }_c)(V,\mathrm{\Phi }_c)`$ because, generally, $`\alpha (n)+\alpha (m)+\alpha (l)0`$ (i.e. $`\alpha (n_c,n_s)+\alpha (m_c,m_s)\alpha (n_c+m_c,n_s+m_s)`$); the invariance would require the additional transformation $`\alpha \alpha `$, corresponding to a mirror reflection $`xx`$ of $`U`$. Since the sum of the exponents $`\alpha `$ does not vanish, $`I_r0`$, and the same analysis as before leads to
$$I_r^{(3)}(V)=\underset{n,m,l}{^{}}C(n,m,l)V^{z^{(3)}(n,m,l)},$$
(20)
with
$$z^{(3)}(n,m,l)=g_c(n_c^2+m_c^2+l_c^2)+g_s(n_s^2+m_s^2+l_s^2)2,$$
(21)
and the amplitudes $`CU^3`$. Table 1 lists the smallest exponents $`z`$ at this order, characterized by the letters (C) and (D), for which the rectification current is nonzero. The exponent $`z_D`$ is smaller than $`z_C`$ for $`g_s>9g_c`$.
To obtain the leading power–law behavior for given $`(g_c,g_s)`$, we compare the amplitudes from second order, $`U^2V^{z^{(2)}}`$, and third order, $`U^3V^{z^{(3)}}`$, for voltages $`VV^{}`$ using Eq. (10). If $`V^{}=0`$, the mere comparison of exponents is sufficient. The result of comparison is represented in Fig. 3. The diagram is not a phase diagram in the proper sense because the crossover lines shift with the voltage. Close to the crossover lines, all contributions of the neighboring phases are large, whereas far from the boundaries, a single power–law dominates. Finally, as shown in the appendix, higher order perturbative corrections lead to less relevant power–laws and can be neglected.
## VI Conclusions
The above results show that the inclusion of spin degrees of freedom leads to a more diverse behavior of the rectification current for different interaction strengths than in the spinless case of Ref. Feldman et al., 2005. The leading power–laws for the rectification current as a function of the interaction strengths $`g_c`$ and $`g_s`$ are represented in Fig. 3.
The most interesting region is characterized by a negative exponent, $`z<0`$. It leads to the unusual behavior of a decreasing rectification current as the voltage $`V>V^{}`$ is raised. Since negative exponents appear only from the third order contributions, they are a realization of the “asymmetry-driven” rectification effect, and due to the presence of an asymmetric impurity in the system. The decrease stops at the upper voltage $`V^{}U^{1/(|z^{(3)}|+z^{(2)})}`$, where the amplitude of the second order contribution exceeds that of the third order. The qualitative behavior is sketched in Fig. 2.
The rectification effect is strong if the magnitude of $`I_r`$ becomes comparable to that of the total current $`I(V)V`$ and to that of the most relevant contribution to the backscattering currentKane and Fisher (1992), $`I_{bs}(V)V^{2n_c^2g_c+2n_s^2g_s1}`$ with $`(n_c,n_s)=(1,1)`$ or $`(n_c,n_s)=(2,0)`$. For interaction strengths $`g_{c,s}`$ such that $`V^{}=0`$, the corrections are weak and generally $`I_r(V)I_{bs}(V)I(V)`$. For $`V^{}>0`$, however, $`I_{bs}(V)`$ becomes comparable to $`I(V)`$ at $`VV^{}`$, i.e. close to the limit of validity of perturbation theory. For such voltages, in regions (C) and (D) the ratio of rectification current to total current can roughly be estimated as $`I_r(V)/I(V)U^3(V^{})^{z^{(3)}1}`$. In phase (C), when $`V^{}=V_{1,1}^{}`$ (i.e. $`3g_cg_s>0`$), this leads to the ratio $`I_r(V)/I(V)(V^{})^{3g_cg_s}=U^{(3g_cg_s)/(1g_cg_s)}`$. The rectification current, therefore, can become comparable to $`I(V)`$ (as well as to $`I_{bs}(V)`$) when $`3g_cg_s`$ is small. The comparison in the region where $`V^{}=V_{2,0}^{}`$ (i.e. $`3g_cg_s<0`$) yields a similar condition for $`2g_s6g_c`$ being small. In phase (D), we find $`I_r(V)/I(V)(V_{2,0}^{})^{12g_c}`$, which may lead to an enhancement of the rectification effect for weak impurities as $`g_c`$ becomes small.
To conclude, we have shown that one–dimensional electronic conductors with screened electron–electron interactions can exhibit a strong rectification effect in the presence of spatially asymmetric impurity scatterers. The rectification current shows a power–law behavior, $`I_rV^z`$, with an exponent $`z`$ that depends on the electron interaction strengths $`g_c`$ and $`g_s`$ only, and which can be determined from second and third order perturbation theory. For small values of $`g_c`$ the exponent becomes negative, leading to the upturn of $`I_r`$ about the voltage $`V^{}`$ (Fig. 2). We note that this effect was obtained in the framework of weak coupling theory. For voltages $`V<V^{}`$, the effective impurity scattering strength exceeds $`E_F`$, marking a crossover into the strong backscattering regimeKane and Fisher (1992); Feldman et al. (2005). All currents eventually vanish at $`V=0`$.
###### Acknowledgements.
This work is supported in part by the NSF under grant number DMR–0213818 and the Swiss National Fonds.
## Appendix A Estimate of higher perturbative orders
In the preceding study we have shown that third order contributions can exceed the second order ones for small values of $`g_c,g_s`$. To complete this, we have to give an additional proof that higher perturbative orders $`N4`$ cannot exceed these values in the physical range of $`g_c>0`$ and $`g_s1/2`$.
Since the calculation of the rectification current involves contributions that are lesser relevant than the leading power–laws for the backscattering current, we have also to show that the neglected backscattering term in $`𝒮_0`$, proportional to $`dx\mathrm{cos}(2\varphi _s)`$, cannot generate additional important corrections to the rectification current. Under renormalization, this term has to be completed by allowing the general interaction $`dx\mathrm{cos}(2\widehat{n}_s\varphi _s)`$ with a summation over integer $`\widehat{n}_s0`$ (the notation $`\widehat{n}_s`$ is used to distinguish these indices from those $`n_s`$ appearing in $`U(n_c,n_s)`$). Under the renormalization group, when integrating over high energies down to the characteristic energy $`V`$, such terms are renormalized asGiamarchi (2004) $`(V/E_F)^{(2\widehat{n}_s)^2g_s2}`$.
We choose the following strategy: Let us assume to be in the perturbative region with voltages $`V>V^{}`$, such that $`U,VE_F`$ and $`U(V/E_F)^{g_cn_c^2+g_sn_s^21}E_F`$. A general higher order correction to the current has the form
$$I^{(N,M)}=V\underset{\begin{array}{c}\{n_c,n_s\}\\ N\text{factors}\end{array}}{}\left(UV^{g_cn_c^2+g_sn_s^21}\right)\underset{\begin{array}{c}\{\widehat{n}_s\}\\ M\text{factors}\end{array}}{}V^{(2\widehat{n}_s)^22},$$
(22)
where for simplicity we set $`E_F=1`$ and neglect prefactors of order 1. Charge and spin conservation here requires that $`n_c=0`$ and $`n_s+(2\widehat{n}_s)=0`$. Since $`g_s1/2`$, the product over the $`V^{(2\widehat{n}_s)^22}`$ is always $`1`$, and if we denote by $`I^{(N)}`$ the expression obtained from Eq. (22) by suppressing these $`M`$ factors, we always have $`I^{(N,M)}I^{(N)}`$. Therefore, the corrections arising from $`dx\mathrm{cos}(2\varphi _s)`$ are always small.
Let us choose a subset of factors of $`I^{(N)}`$, and denote this quantity by $`J`$,
$$J=V\underset{\begin{array}{c}\text{selection of }\{n_c,n_s\}\end{array}}{}\left(UV^{g_cn_c^2+g_sn_s^21}\right).$$
(23)
Since all factors are smaller than 1, we have
$$I^{(N,M)}I^{(N)}<J.$$
(24)
We will now show that, for $`N4`$, it is always possible to choose a $`J`$ that is smaller or equal to the dominating second or third order contribution. The larger the $`n_{c,s}`$, the smaller become the factors in $`I^{(N)}`$ and $`J`$. Much in the analysis, therefore, depends on the maximal value of $`|n_c|`$, which we denote by $`\overline{n}_c`$.
This method of comparison does not hold for the comparison between the second and third order expressions themselves. Since these expressions are products of two or three factors only, we cannot choose a suitable subset for $`J`$ but have to deal with the full expressions. The exponents are therefore largely determined by the constraints of particle conservation plus having even sums of charge and spin numbers. The result of comparison was discussed above. For $`N4`$, however, the larger freedom of choice of $`J`$ allows us to show that these higher order expressions are small compared to the second and third order ones. The following estimates contain the proof of this statement.
For any choice of maximal $`\overline{n}_c`$ we choose a suitable $`J`$ consisting of two or three factors $`UV^{n_c^2g_c+n_s^2g_s1}`$ and the prefactor $`V`$ or, if required, the prefactor $`V^2`$ with the additional $`V`$ arising from the expansion of a $`U`$ (similarly to the second order calculation). We then show that this $`J`$ is smaller or equal to one of the expressions
$`I_A`$ $`=U^2V^{2g_c+2g_s},`$ (25)
$`I_B`$ $`=U^2V^{8g_c},`$ (26)
$`I_C`$ $`=U^3V^{6g_c+2g_s2},`$ (27)
$`I_D`$ $`=U^3V^{24g_c2},`$ (28)
for any $`g_c0`$ and $`g_s1/2`$. This will prove that $`I^{(N)}<J`$ cannot exceed the dominant second or third order expression, given by the maximum of $`I_{A,B,C,D}`$.
Since $`\overline{n}_c1`$, the particle conservation imposes that another or several numbers $`n_c`$ are different from zero. These are denoted by $`\stackrel{~}{n}_c`$ and $`\stackrel{~}{m}_c`$ below. If not further specified we only assume that $`|\stackrel{~}{n}_c|,|\stackrel{~}{m}_c|1`$. If $`\overline{n}_c`$ is odd there is at least another odd $`|\stackrel{~}{n}_c|1`$ and, due to the requirement of even $`n_c+n_s`$, there exist at least two odd $`|n_s|1`$, denoted by $`\stackrel{~}{n}_s`$ and $`\stackrel{~}{m}_s`$.
We distinguish between the following five cases:
1. *$`\overline{n}_c2`$ is even; all nonzero $`|n_c|=\overline{n}_c`$*: If in the couples $`(n_c,n_s)`$ all $`n_c0`$ have $`n_s=0`$, $`I^{(N)}`$ (and $`I^{(N,M)}`$) is symmetric under the change of sign $`VV`$, and the rectification current vanishes (note that the $`n_s`$ for which $`n_c=0`$ and the $`\widehat{n}_s`$ can be nonzero though). This situation is similar to the second order case discussed above. A rectification current exists since the potential $`U`$ depends on the voltage through the density $`k_F`$. An expansion of a $`U`$ to linear order in $`V`$ allows us to choose a $`J`$ of the form
$$J=V^2(UV^{\overline{n}_c^2g_c1})(UV^{\overline{n}_c^2g_c1})U^2V^{(2^2+2^2)g_c}=I_B.$$
(29)
On the other hand, if not all $`n_s`$ are zero, there must be an even $`n_s=\stackrel{~}{n}_s`$ with $`|\stackrel{~}{n}_s|2`$, and we can choose for $`J`$
$$\begin{array}{c}J=V(UV^{\overline{n}_c^2g_c+\stackrel{~}{n}_s^2g_s1})(UV^{\overline{n}_c^2g_c1})\hfill \\ \hfill U^2V^{(2^2+2^2)g_c+2^2g_s1}U^2V^{8g_c+2g_s}I_{A,B},\end{array}$$
(30)
where we have used $`g_s1/2`$.
2. *$`\overline{n}_c2`$ is even; all nonzero $`|n_c|`$ are even but are not all equal*: This condition excludes $`\overline{n}_c=2`$ as it coincides with the previous case. For $`\overline{n}_c4`$ there exist two other nonzero and even $`|\stackrel{~}{n}_c|,|\stackrel{~}{m}_c|2`$ or another $`|\stackrel{~}{n}_c|4`$ and an arbitrary $`|\stackrel{~}{m}_c|0`$. We then choose
$$J=U^3V^{(\overline{n}^2+\stackrel{~}{n}_c^2+\stackrel{~}{m}_c^2)g_c2}U^3V^{(4^2+8)g_c2}=I_D.$$
(31)
3. *$`\overline{n}_c2`$ is even; there is an odd $`|n_c|`$*: Since $`n_c=0`$ there exist at least two odd $`|\stackrel{~}{n}_c|,|\stackrel{~}{m}_c|1`$, and since $`n_c+n_s`$ must be even, there must be two odd $`|\stackrel{~}{n}_s|,|\stackrel{~}{m}_s|1`$. This leads to the bound
$$\begin{array}{c}J=U^3V^{(\overline{n}_c^2+\stackrel{~}{n}_c^2+\stackrel{~}{m}_c^2)g_c+(\stackrel{~}{n}_s^2+\stackrel{~}{m}_s^2)g_s2}\hfill \\ \hfill U^3V^{(2^2+1+1)g_c+(1+1)g_s2}=I_C.\end{array}$$
(32)
4. *$`\overline{n}_c=1`$*: In this case all $`n_c`$ are $`|n_c|=1`$ or zero. If all nonzero $`n_c`$ have $`n_s=\pm 1`$, $`I^{(N,M)}`$ is symmetric under $`VV`$, and we have to expand a $`U`$ to linear order in $`V`$. If we choose a $`|\stackrel{~}{n}_c|=1`$ and two $`|\stackrel{~}{n}_s|,|\stackrel{~}{m}_s|=1`$, we can set
$$\begin{array}{c}J=VU^2V^{(\overline{n}_c^2+\stackrel{~}{n}_c^2)g_c+(\stackrel{~}{n}_s^2+\stackrel{~}{m}_s^2)g_s1}\hfill \\ \hfill U^2V^{(1+1)g_c+(1+1)g_s}=I_A.\end{array}$$
(33)
On the other hand, if there is a couple $`(\stackrel{~}{n}_c,\stackrel{~}{n}_s)`$ with $`\stackrel{~}{n}_c0`$ and $`|\stackrel{~}{n}_s|3`$, we have the estimate
$$\begin{array}{c}J=U^2V^{(\overline{n}_c^2+\stackrel{~}{n}_c^2)g_c+\stackrel{~}{n}_s^2g_s1}U^2V^{(1+1)g_c+3^2g_s1}\hfill \\ \hfill U^2V^{2g_c+7g_s}I_A.\end{array}$$
(34)
5. *$`\overline{n}_c3`$ is odd*: Since $`\overline{n}_c`$ is odd there must be at least another odd $`|\stackrel{~}{n}_c|1`$, and, to fulfill that $`n_c+n_s`$ is even, there must be two odd $`|\stackrel{~}{n}_s|,|\stackrel{~}{m}_s|1`$, such that
$$\begin{array}{c}J=U^2V^{(\overline{n}_c^2+\stackrel{~}{n}_c^2)g_c+(\stackrel{~}{n}_s^2+\stackrel{~}{m}_s^2)g_s1}\hfill \\ \hfill U^2V^{(3^2+1)g_c+(1+1)g_s1}U^2V^{10g_c}I_B.\end{array}$$
(35)
We conclude that higher order terms cannot exceed the second and third order expressions for $`VV^{}`$.
|
warning/0506/cond-mat0506305.html
|
ar5iv
|
text
|
# Magnetocapacitance effect in perovskite -superlattice based multiferroics
## Abstract
We report the structural and magnetoelectrical properties of La<sub>0.7</sub>Ca<sub>0.3</sub>MnO<sub>3</sub>/BaTiO<sub>3</sub> perovskite superlattices grown on (001)-oriented SrTiO<sub>3</sub> by the pulsed laser deposition technique. Magnetic hysteresis loops together with temperature dependent magnetic properties exhibit well-defined coercivity and magnetic transition temperature (T<sub>C</sub>) ~140 K. $`DC`$ electrical studies of films show that the magnetoresistance (MR) is dependent on the BaTiO<sub>3</sub> thickness and negative $`MR`$ as high as 30% at 100K are observed. The $`AC`$ electrical studies reveal that the impedance and capacitance in these films vary with the applied magnetic field due to the magnetoelectrical coupling in these structures - a key feature of multiferroics. A negative magnetocapacitance value in the film as high as $`3\%`$ per tesla at 1kHz and 100K is demonstrated, opening the route for designing novel functional materials.
Superlattices, which are composed of thin layers of two or more different structural counterparts that stacked in a well-defined sequence, may exhibit some remarkable properties that do not exist in either of their parent forms. For example, a (LaFeO<sub>3</sub>)/(LaCrO<sub>3</sub>) superlattice stacked on (111)-SrTiO<sub>3</sub> exhibits a ferromagnetic behavior, whereas each parent material is antiferromagnetic. Similarly, many perovskite superlattices can exhibit new properties such as high temperature superconductivity. To grow these superlattices, various thin films techniques such as molecular beam epitaxy, chemical vapor deposition, pulsed laser deposition (PLD) technique etc. have been employed. In particular, the PLD process is one of the most suitable and frequently used techniques to grow the superlattice of multi-component perovskite oxides in a moderate oxygen pressure.
A multiferroic is a material in which ferromagnetism and ferroelectricity coexist . As a consequence, the magnetic domains can be tuned by the application of an external electric field, and likewise electric domains are switched by magnetic field. Thus, these materials offer an additional degree of freedom in designing the various devices, e.g. transducers, actuators, storage devices, which is unachievable separately in either ferroelectric or magnetic materials. Hitherto, a very few materials, e.g. perovskite-type BiFeO<sub>3</sub>, hexagonal REMnO<sub>3</sub> (RE=rare earths), and the rare-earth molybdates, exist in nature or synthesized in laboratory which exhibit multiferroism . Note that most of these compounds display an antiferromagnetic behavior. These non-trivial spin-lattice coupling in the multiferroics has been manifested through various forms, such as linear and bilinear magnetoelectric effects, polarization change through field-induced phase transition, magneto-dielectric effect, and dielectric anomalies at magnetic transition temperatures . Why and under what circumstances a large coupling should come about is a major open question, but this problem has proved difficult to tackle owing to the lack of materials that show such large coupling. Also the absence of multiferroics with large coupling at moderate conditions is one of the big hurdles in the realization of multiferroic devices. Thus, it is essential to design novel multiferroic materials with essential properties. To synthesize these novel multiferroics, various efforts have been made by mixing the ferroelectric and magnetic materials in forms of either composites or multilayers in view of possible applications.
In the present work, we have utilized the versatility of the PLD technique to create a multilayer structure in a superlattice form composed of a piezoelectric and ferroelectric, namely BaTiO<sub>3</sub> (BTO); and a ferromagnet, namely La<sub>0.7</sub>Ca<sub>0.3</sub>MnO<sub>3</sub> (LCMO). The superlattices were characterized by the various techniques for their structural, magnetic, electrical, and magneto-electrical properties, and our results are reported in this article.
Superlattices of BTO/LCMO were grown on (001)-oriented SrTiO<sub>3</sub> (STO) by the pulsed laser deposition technique, using stoichiometric targets, at 720C in a flowing 100 mTorr oxygen atmosphere. Superlattices with individual BTO layer thickness of 1 to 25 unit cells (u.c.) by keeping the LCMO layer thickness as 5 u.c. were realized. The choice of 5 u.c. came because thin layer of LCMO behaves as a ferromagnetic insulator . The superlattice is composed of 25 repeated units of BTO/LCMO bilayers with LCMO as the bottom layer.
The samples were characterized by X-ray diffraction (XRD) using Seifert 3000P diffractometer (Cu K$`\alpha `$ , $`\lambda `$= 0.15406 nm) and a Philips X’Pert for the in-plane measurements. Magnetization (M) was measured as a function of temperature (T) and magnetic field (H) using a superconducting quantum interference device magnetometer (SQUID). $`DC`$ electrical resistivity ($`\rho `$) of films were measured in four probe configuration. $`AC`$ electrical properties of the films were measured by a lock-in amplifier (Stanford Research 850) in the frequency range of 1-10<sup>6</sup> Hz, where the sample was held in PPMS system. To measure the electrical properties of the films in current-perpendicular-to-the-plane (CPP) geometry, a LaNiO<sub>3</sub> electrode was fabricated through a shadow mask .
In Fig. 1a, we show the $`\mathrm{\Theta }2\mathrm{\Theta }`$ XRD scan around the (002) fundamental peak (40-52 in 2$`\mathrm{\Theta }`$ ) of 5 u.c. LCMO/ 10 u.c. BTO (denoted hereafter as $`5/10`$) superlattice. The denoted number $`i`$ indicates the $`i^{th}`$ satellite peak. The presence of higher order satellite peaks adjacent to the main peak, arising from chemical modulation of multilayer structure, indicates that the films were indeed coherent heterostructurally grown. The periodic chemical modulation for the ($`5/10`$) superlattice, as extracted from $`\mathrm{\Theta }2\mathrm{\Theta }`$ XRD, is $`\mathrm{\Lambda }=5.98`$nm which is in agreement with theoretical values ($`5.94`$ nm) based on the lattice parameters of each constituent ($`0.386`$nm for LCMO and $`0.4006`$nm for BTO). The full-width-at-half-maximum (FWHM) of the rocking curve, recorded around the fundamental ($`002`$) diffraction peak of the same superlattice, is very close to the instrumental broadening ($`<0.2^{}`$), indicating a well crystallinity and a good coherency (see inset of Fig1a). Further, to examine the in-plane coherence, $`\mathrm{\Phi }`$-scan was recorded around the $`103`$ reflection of the cubic unit cell. Different $`\mathrm{\Phi }`$scans (not shown here) were recorded by at various tilted angle leading to a pole figure. Four peaks are clearly observed at $`90^{}`$ from each other, indicating a four-fold symmetry as expected for the perovskite structures LCMO and BTO. This well defined pattern is an evidence for the in-plane texture of the superlattice. Similar scans recorded on other films confirm that the superlattices grow epitaxially on STO. The film morphology, examined by atomic force microscopy, gave a roughness in the range of 3-6A (close to one perovskite u.c.) for all films showing that they have a very smooth surface.
Magnetization vs. applied magnetic field ($`MH`$) loop and vs. temperature $`M(T)`$ measurements were performed on all samples and Fig.1b shows an example in-plane hysteresis loop for a ($`5/10`$) superlattice, recorded at 10K. The curve clearly shows a well-defined coercivity confirming the ferromagnetic nature of the film. Furthermore, the temperature-dependance of the magnetization $`M(T)`$ recorded under 3000Oe applied magnetic field, shows the magnetic transition from ferromagnetic to paramagnetic at T<sub>C</sub> (Curie temperature) around 140K. This value is lower than the observed bulk LCMO (250K), but results from both the substrate-induced strain and the thickness of the layer (5 u.c.). Surprisingly, the Curie temperature of the superlattices is almost independent of the BTO thickness layer (in the range 135-142K for all films). Despite the similarity in the shape of the hysteresis loop, the magnetic data reveal some differences. For example, the magnetization of the films is dependent on the thickness of BTO spacer layer (see inset of Fig.1b). A detailed study shows that the total magnetization of the films is increasing with increase in the BTO thickness up to 15 u.c. This is surprising because the LCMO thickness is constant throughout all the samples. Thus, it is possible to increase the magnetization only when the LCMO is inducing the magnetization in BTO layer via magnetoelectric coupling. With further increase in BTO thickness (i.e. above 15 u.c.) magnetization is decreasing, which is mostly attributed to the variation in the strains, as previously reported for superlattices
Figure 2a displays the $`DC`$ resistivity of ($`5/5`$), ($`5/10`$) and ($`5/15`$) superlattices which were measured as a function of temperature without magnetic field. The inset of Fig.2a displays the magnetic field-dependance of the magnetoresistance ($`MR`$) of a ($`5/15`$) superlattice taken at 100K. $`MR`$ is defined as $`MR`$ $`(\%)=100\times [R(H)R(0)]/R(0)`$, where $`R(H)`$ and $`R(0)`$ are the resistance measured with and without magnetic field, respectively. Fig.2a clearly reveals that the resistivity is increasing with BTO thickness, which is consistent with the BTO insulating behavior. With increasing in the BTO layer thickness above 15 u.c., the resistance of the samples is indeed getting too large to measure above 100K, as it should be in BTO. This shows that the ”magnetic” BTO is also more conducting than that of classical BTO.
In order to understand the coherent spin transport in these films, MR were measured for different samples. Fig. 3b shows the evolution of MR as a function of temperature for a ($`5/15`$) superlattice and inset of Fig. 3b shows the $`MR`$ for several BTO thickness (from 5 to 25) measured at 100K. Fig.3b clearly indicates that with increasing temperature, the $`MR`$ is increasing and vanishes above the transition temperature which is consistent with the LCMO property. Further, with increase in the BTO thickness up to 15 u.c. (Inset of Fig.3b), there is enhancement in $`MR`$. This is surprising because, the $`MR`$ in a multilayer with an insulating barrier basically arises from the tunneling of coherent magnetic carriers and spin polarization of either side of the magnetic layer, i.e. magnetic layer thickness and structure, whereas in the present case the number of interfaces and magnetic layer thickness is constant for all superlattices. . However, in principle the $`MR`$ does not increase with increase of tunnel barrier thickness. The enhancement in MR may be understood as follows. The total resistivity ( $`\rho _{total})`$ of the samples can be expressed as $`\rho _{total}`$ = $`x`$ $`\rho _{BTO}`$ \+ ($`1x`$) $`\rho _{LCMO}`$ where ’$`x`$’ is the BTO compositional resistivity coefficient; $`\rho _{BTO}and\rho _{LCMO}`$ is bulk resistivity of BTO and LCMO, respectively . Since the thickness of the LCMO as well as the number of interfaces are constant in all samples, the increase in the enhancement MR will be arising from the BTO layer. Thus, it is evident that, up to 15 u.c., the BTO layer is magnetically polarized, which corroborates the ($`MH`$) findings. Henceforth, the observed enhancement in magnetoresistance with the increase of BTO thickness may be attributed to the possible magnetoelectric coupling in these superlattices as previously seen for (Pr<sub>0.85</sub>Ca<sub>0.15</sub>MnO<sub>3</sub>)/(Ba<sub>0.4</sub>Sr<sub>0.6</sub>TiO<sub>3</sub>) superlattices. Furthermore, $`MR`$ was diminishing with further increase in BTO thickness (i.e. above 15 u.c.), which is attributed to the loss of spin coherence due to the large tunnel barrier thickness. Thus, the above results (Fig.1b and Fig.2) clearly exhibit that the maximum magnetoelectric coupling can be expected in ($`5/15`$) superlattice.
In order to study the magnetoelectric coupling in these films, $`AC`$ electrical properties were measured. Fig.3 shows the impedance ($`Z`$) vs frequency at different temperatures under 0T and 5T applied magnetic field, for a ($`5/15`$) superlattice. Two regimes are observed. Below 150K, the impedance of the sample decreases with applied magnetic field, whereas above 150K it vanishes and no significant variation is seen with magnetic field at higher temperatures (see Fig.3d). Further, we have extracted the capacitance of the film based on an equivalent parallel $`RC`$ circuit, which has been chosen based on the Cole-Cole plot (not shown) of the observed impedance data. The estimated capacitance up to 150K is reported in the Fig 4. The capacitances vs. frequency curves exhibit the two important features. First, the capacitance of the sample is constant at low frequency and decreases with the progressive increase in frequency. In a given device dimension, the capacitance of the device dependence on the total electrical polarizability of the materials. There are the four electrical polarizability components, namely electronic, ionic, dipolar and space charge, and their contribution strongly depend on the frequencies. At low frequency (~1Hz -1MHz), the contribution comes from all these components, whereas above 1kHz frequency it starts decreasing and the dipolar contribution becomes zero around 100MHz. Thus, the decrease of the capacitance at higher frequency indicates presence of electric dipoles in the film. Hence, it indirectly provides an evidence of the presence ferroelectricity in these films. Second, there is a decrease of the capacitance under applied magnetic fields. In the previous reports, usually the magnetocapacitance effects have been shown near the Curie temperature. In the present case it occurs well below the Curie temperature and is significant large in magnitude. The magnetocapacitance $`MC`$ is defined, by analogy to the $`MR`$ as, $`MC(T)`$ $`(\%)=100\times [C(H,T)C(0,T)]/C(0,T)`$, where $`C(H,T)`$ represents the capacitance at a magnetic field $`H`$ and a temperature $`T`$. The negative $`MC`$ value observed at 1kHz and 100K was ~$`17\%`$. The detail study reveals that the $`MC`$ was less than 1% at 5K whereas it increases with increasing the temperature and diminishes above 150K. The $`MC`$ effect, at low frequency (1kHz) is interesting since it shows that the superlattice can be considered as a new compound with novel properties that are not observed in both the parent compounds.
It clearly evidenced that the artificial structure is behaving as a multiferroic. Such effect was not reported on artificial superlattices. This effect is a direct evidence of magnetoelectric coupling in the film, as alluded above, and as reported in other multiferroics. However, it is worth to note that the magnetocapacitance effects in some ceramics of the hexagonal rare earth manganites are of the order of 1% per tesla near transition temperature (~40K), whereas in the present case it is of the order of 3% per tesla at 100K. It should also be noticed that the $`MC`$ effect appears only at low frequency. This can be understood if one uses the model already proposed in Ref. where the BTO layer is replaced by a $`R`$ and $`C`$ in parallel, the LCMO as a single $`R`$. In this model, the $`MR`$ of the BTO layer is in agreement with the $`MC`$ at low frequency in the whole material.
To summarize, we have successfully grown high quality (BaTiO<sub>3</sub>/La<sub>0.7</sub>Ca<sub>0.3</sub>MnO<sub>3</sub>) superlattices on STO by PLD process. Despite the lattice mismatch between substrate and LCMO (-1.17%), and BTO (+2.2%), the films were grown heteroepitaxial. Magnetoelectrical measurements revealed that the films have at 100K, a negative magnetoresistance close to $`4\%`$ per tesla and a negative magnetocapacitance effect of the order of $`3\%`$ per tesla at 1kHz. The presence of the coupling between dielectric and magnetic orders have thusly been demonstrated. This high magnetocapacitance as well as the large magnetoresistance open a path in designing novel multiferroic thin films.
We would like to thank Prof. James N. Eckstein, Dr. L. Méchin, Prof. B. Mercey and Ms. N. Bellido for helpful discussions.
This work has been carried out in the frame of the European Network of Excellence ”Functionalized Advanced Materials Engineering of Hybrids and Ceramics” FAME (FP6-500159-1) supported by the European Community, and by Centre National de la Recherche Scientifique.
Figure 1: (a) $`\mathrm{\Theta }`$-2$`\mathrm{\Theta }`$ XRD pattern of a ($`5/10`$) superlattice. Inset of Fig.1a shows the ($`002`$) rocking curve of ($`5/10`$) superlattice, (b) $`M(H)`$ curve of ($`5/10`$) superlattice measured at 10 K. The field is applied along the direction. Inset of Fig.1b shows magnetic moment as a function of BaTiO<sub>3</sub> thickness at 10 K (Dots are experimental data point and solid line is just for guiding eyes). The magnetic moment of a LCMO (5 u.c.) is 3.10<sup>-3</sup> emu/cm<sup>2</sup>.
Figure 2: (a) $`DC`$ $`\rho (T)`$ for different superlattices at zero magnetic field, (b) $`MR(T)`$ of ($`5/15`$) superlattice extracted from 0T and 5T $`\rho `$ data. Inset of Fig.2a shows $`MR`$ of ($`5/15`$) superlattice at 100K and Inset of Fig.2b is the $`MR`$ measured at 100 K as a function of the BaTiO<sub>3</sub> spacer layer.
Figure 3: Modulus of complex impedance ($`Z`$) vs frequency with applied magnetic field at different temperatures of ($`5/15`$) superlattice. Full dots : 0T, open dots : 5T.
Figure 4: Capacitance vs frequency with applied magnetic field of a ($`5/15`$) superlattice at (a) 10 K, (b) 50 K, (c) 100K and (d) 150 K . Full dots : 0T, open dots : 5T.
|
warning/0506/cond-mat0506319.html
|
ar5iv
|
text
|
# Formation of Vector Solitary waves with Mixed Dispersion in Bose-Einstein Condensates.
## I Introduction
Dispersion is a fundamental process in the dynamics of wavepackets. Its effect is revealed dramatically as the expansion of the wavepacket’s size in time. For practical applications this extension becomes a serious limitation: degrades communication bit rates, and prevents efficient methods for storage and transport of classical or quantum information. Fortunately, this situation changes radically in non-linear systems, where the non-linear potential can counteract dispersion and may result in a preservation of the wavepacket shape in time. The characterization of situations where these soliton solutions emerge constitutes a fundamental challenge for research. Besides its mathematical interest, especially relevant is the description of situations in which solitons or soliton-like structures appear in Nature, opening the door for technological applications (as happened with electromagnetic waves in Kerr media molle80A ).
The non-linear Schrödinger equation (NLSE) constitutes one paradigmatic case for the study of solitons zakha71A ; haseg73A . It describes different physical systems, among them the mean field dynamics of Bose-Einstein condensates (BEC). Bright solitons, or more generally solitary waves, in BEC have been demonstrated experimentally for the case of attractive interactions khayk02A ; strec02A and dark solitons in the case of repulsive interactions burge99A ; densc00A . In addition, bright solitons can be also obtained for repulsive interactions in condensates near the edge of the Brillouin zone of an optical lattice (gap solitons) . In this case, a condensate in the lowest band will experience a negative effective mass, thus inverting the sign of dispersion, which now can be counterbalanced by the repulsive non-linear potential. The first experimental observation of this type of solitary waves has been reported recently eierm04A .
Vector solitons may appear in situations where two or more different wave modes can coexist. In this case, the soliton is composed by a set of wavepackets, one for each mode that propagate jointly. Every mode, therefore, contributes to the overall non-linear potential that allows for self-trapping. Although the integrability of the resulting system of equations is not ensured for the general two-component case, special cases as biased photorefractive crystals can be demonstrated to enclose Manakov solitons as solutions manak74A . Other examples of two-component vector solitary waves can be found in waveguide arrays, where the two radiation modes correspond to different energy bands mande03A ; mande03B . In the general case, each equation of the set describing the vector soliton includes the counter-balance between dispersion and non-linearity. Therefore each vector component, even evolving uncoupled, has the dynamical conditions to reach a soliton wave, which grants in great manner the stability of the coupled system. However, this condition is quite restrictive, since one may think in situations in which not all the vector components are described by equations in which the non-linearity and the dispersion balances. A particular case is the formation of vector gap solitons in which the geometry of the bands at the edge of the Brillouin zone only leads to a negative sign in the dispersion of the lowest band. In the uncoupled case, therefore, only the vector component corresponding to the lowest band will evolve to a soliton, while the other components (belonging to the upper bands) will rapidly disperse. In this case, the coupling is a fundamental aspect for the formation of a localized vector wavepacket. The aim of this letter is, first, to propose for the first time a set of two coupled NSE with opposite signs in the dispersion term which sustains localized self-trapped vector solutions, and, second, to demonstrate that such solitary solutions can be observed in physical systems as BEC in oscillating optical lattices.
To begin, let us consider the following system of coupled NLSE equations:
$`i{\displaystyle \frac{u_1}{\tau }}`$ $`=`$ $`{\displaystyle \frac{\alpha }{2}}\left[{\displaystyle \frac{1}{2}}\mathrm{\Delta }k+i{\displaystyle \frac{}{\eta }}\right]^2u_1+\beta u_2+\left(\gamma _1|u_1|^2+|u_2|^2\right)u_1`$
$`i{\displaystyle \frac{u_2}{\tau }}`$ $`=`$ $`\delta u_2+{\displaystyle \frac{\alpha ^1}{2}}\left[{\displaystyle \frac{1}{2}}\mathrm{\Delta }ki{\displaystyle \frac{}{\eta }}\right]^2u_2+\beta ^{}u_1+`$
$`\left(|u_1|^2+\gamma _2|u_2|^2\right)u_2`$
Where $`\alpha `$, $`\gamma _1`$ and $`\gamma _2`$ are positive. $`\alpha `$ accounts for the mass difference between the two vector components, $`\mathrm{\Delta }\kappa `$ accounts for the difference between the group velocities of the vector components evolving freely (i.e. without coupling), and $`\delta `$ accounts for the gap between the energy dispersion curves of each component. These two later parameters are of no fundamental importance, as we have demonstrated selftrapping also for the case $`\mathrm{\Delta }\kappa =\delta =0`$, but should be taken into account to describe particular physical systems as BEC in optical lattices, for which each component belongs to a different energy band. In many aspects, Eqs. (LABEL:eq:syssol1-LABEL:eq:syssol2) resemble to those describing the formation of solitary waves in light passing through a periodically twisted birrefringent fiber wabni91A . In both cases, the vector components interact through linear and non-linear couplings, and velocity and energy mismatch between the vector components are taken into account. However, in contrast with Eqs. (LABEL:eq:syssol1-LABEL:eq:syssol2), the dispersion terms in wabni91A have the same signs and the non-linear potentials are attractive, i. e. dispersion balances the non-linearity in each equation. Up to our knowledge, this characteristic is shared by the generality of models for vector solitons. Therefore, the novelty of the system (LABEL:eq:syssol1-LABEL:eq:syssol2) is the opposite sign in the dispersion terms for the two components. As discussed above, in this case the coupling is fundamental for the stabilization of the second component of the vector solitary wave. Fig. 1 shows the possibility of self-trapping through the numerical integration of Eqs. (LABEL:eq:syssol1-LABEL:eq:syssol2). The initial wavefunction is a Gaussian in one of the vector components and zero in the other, therefore a situation quite different from the two-component solitary wave in which it finally evolves. The parameters are taken as $`\alpha =0.775`$, $`\delta =0.0114`$, $`\beta =0.014`$, $`\gamma _1=1.43`$, $`\gamma _2=0.56`$ and the norm of the initial wavefunction is $`0.69`$, this particular values come from estimations from the Bose-Einstein experiment computed below. Figure 1(a) and (b) correspond to the time evolution of the first and second component of the vector function, respectively. In both cases a transient time is observed before the system reaches a stable configuration. During this transient, the population oscillates between both vector components while radiation is emitted at both sides of the wavefunction. Matter radiation is associated to the transient and may have a fundamental role as a dissipation process that allows the system to converge to the stable solution plaja04A ; yulin03A . The final population enclosed in the solitary wave depends on this transient, which is more pronounced the greater the difference between the initial wavepacket and the final solitary wave is.
Let us now focus to the experimental realization of this kind of vector solitary waves. To this end we will first demonstrate that Eqs. (LABEL:eq:syssol1-LABEL:eq:syssol2) can describe the evolution of a Bose-Einstein condensate in an oscillating optical lattice. Afterwards, as a further proof, we will compute directly the time evolution of the condensate, as given by Gross-Pitaevskii equation. These ab initio computations will demonstrate that, for conditions close to the experiment eierm04A , vector solitary waves with mixed dispersion appear spontaneously.
Following the philosophy of eierm04A , we will consider a Bose-Einstein condensate prepared near the edge of the Brillouin zone of an optical lattice. The condensate is tightly confined in the transversal direction by a trapping potential. Under these circumstances, the condensate dynamics can be considered one dimensional, provided $`(\mu /\mathrm{}\omega _{}1)/2<<1`$ plaja02A , $`\mu `$ being the chemical potential and $`\omega _{}`$ the transversal frequency of the trap. The dynamical equation is then
$$i\mathrm{}\frac{}{t}\mathrm{\Phi }(z,t)=\left(H_0+H_{nl}\right)\mathrm{\Phi }(z,t)$$
(3)
where $`H_0=\widehat{p}^2/2m+U(z,t)`$, $`H_{nl}=g_{1D}\left|\xi (z,t)\right|^2`$, with $`g_{1D}=N\mathrm{}a_s/ma_{}`$, $`N`$ being the number of atoms, $`a_s`$ the scattering length of the condensed element (in our case $`{}_{}{}^{87}Rb`$) and $`a_{}`$ the transversal size of the wavefunction (assumed to be in the lowest state of the radial harmonic potential). The potential $`U(z,t)`$ corresponds to a time-dependent axial lattice. In our case this potential includes an oscillatory motion of frequency $`\omega `$, and amplitude $`\zeta _0`$. Therefore $`U(z,t)=U_0[z\zeta (t)]`$ with $`\zeta (t)=\zeta _0\mathrm{sin}(\omega t)`$, and $`U_0(\varsigma )=V_0\mathrm{cos}^2(\pi \varsigma /a_0)`$ the stationary lattice potential.
Let us now define a new wavefunction $`\varphi `$ according to the transformation
$$\varphi (z,t)=\mathrm{exp}[i\frac{\widehat{p}}{\mathrm{}}\zeta (t)]\mathrm{\Phi }(z,t)=\mathrm{\Phi }(z+\zeta (t))$$
(4)
Eq. (3) may now be written as
$`i\mathrm{}{\displaystyle \frac{}{t}}\varphi (z,t)`$ $`=`$ $`\left[{\displaystyle \frac{\widehat{p}^2}{2m}}\dot{\zeta }(t)\widehat{p}\right]\varphi (z,t)+U_0(z)\varphi (z,t)+`$ (5)
$`g_{1D}|\varphi (z,t)|^2\varphi (z,t)`$
in which the time dependence is removed from the potential term and translated to a linear coupling term. Note that this equation is formally identical to the NLSE of a 1D system in interaction with a semiclassical electromagnetic field in dipole approximation, the effective vector potential being $`A(t)=(mc/q)\dot{\zeta }(t)`$. This optical analogy is fruitful in understanding lattice oscillations as direct interband transitions, in the same fashion as electromagnetic waves produce optical transitions of electrons in crystals. This type of coherent transfer between bands has been experimentally characterized in hecke02A .
Now we express the wavefunction $`\varphi (z,t)`$ as a superposition of states of the first and second band. Using the Wannier basis ziman ,
$$\varphi (z,t)=\underset{\mathrm{}}{}e^{ik_0\mathrm{}}\left[f_1(\mathrm{},t)a_1(z\mathrm{})+f_2(\mathrm{},t)a_2(z\mathrm{})\right]$$
(6)
where $`k_0=mv_0/\mathrm{}`$ corresponds to the initial momentum of the wavefunction, located near the edge of the Brillouin zone. We may now follow the standard procedure for the effective mass approach in the form used, for instance, in plaja04A . Substituting (6) in (5), and assuming negligible overlap between Wannier functions of neighbor sites (tight-binding approx.), we find the dynamical equations for the wavefunction envelopes
$`i\mathrm{}`$ $`{\displaystyle \frac{}{t}}`$ $`f_p(\xi ,t)=\widehat{T}_pf_p(\xi ,t)`$ (7)
$`{\displaystyle \frac{2a_0}{\mathrm{}}}\dot{\zeta }{\displaystyle \underset{qp}{}}\pi _{p,q}\left(i\mathrm{}{\displaystyle \frac{}{\xi }}\right)f_q(\xi ,t)+`$
$`g_{1D}{\displaystyle \underset{q,n,m}{}}\mathrm{\Gamma }_{p,q,n,m}f_q^{}(\xi ,t)f_n(\xi ,t)f_m(\xi ,t)`$
with band labels $`\{p,q,n,m\}`$ valued $`1`$ or $`2`$, $`\xi `$ defined as the continuous limit of $`\mathrm{}`$, and with $`\pi _{p,q}=i𝑑za_p^{}(z\mathrm{})(i\mathrm{})a_q(z\mathrm{}a_0)/z`$ and $`\mathrm{\Gamma }_{p,q,n,m}𝑑za_p^{}(z\mathrm{})a_q^{}(z\mathrm{})a_n(z\mathrm{})a_m(z\mathrm{})`$ constant parameters. The kinetic operator is defined as
$$\widehat{T}_p=ϵ_p(k_0)+\left(v_{g,p}\frac{2a_0\pi _{p,p}}{\mathrm{}}\dot{\zeta }\right)\left(i\mathrm{}\frac{}{\xi }\right)\frac{\mathrm{}^2}{2m_p^{}}\frac{^2}{\xi ^2}$$
(8)
with $`ϵ_p(k_0)`$ the energy of the $`p`$ band evaluated at the point $`k_0`$ of the Brillouin zone and $`v_{g,p}(1/\mathrm{})(/k)ϵ_p|_{k_0}`$ and $`1/m_p^{}=(1/\mathrm{}^2)(^2/^2k)ϵ_p|_{k_0}`$ the group velocity and the effective mass. The linear coupling in Eq. (7) describes the direct (optical) interband transitions induced by the oscillating lattice. Its derivation follows from the evaluation of the linear coupling in Eq. (5) projected on the Wannier function $`\mathrm{exp}(ik_0\mathrm{})a_p(z\mathrm{})`$ which leads to sum of the type $`_{n,\mathrm{}^{}}\mathrm{exp}\left[ik_0(\mathrm{}^{}\mathrm{})\right]f_n(\mathrm{}^{},t)𝑑za_p^{}(z\mathrm{})\widehat{p}a_n(z\mathrm{}^{})`$. Assuming that the leading term in the sum corresponds to the coupling between nearest neighbors, and since $`k_0a_0\pi `$, we can approximate this sum to $`i_n\pi _{p,n}\left[f_n(\mathrm{}+a_0,t)f_n(\mathrm{}a_0,t)\right]`$ which, in the continuous limit, can be written as $`i2a_0_n\pi _{p,n}\frac{}{\xi }f_n(\xi ,t)`$.
We shall assume that the lattice oscillation is tuned near the resonance of the interband transition: $`\mathrm{}\omega =ϵ_2(k_0)m_2^{}v_{g,2}^2/2ϵ_1(k_0)+m_1^{}v_{g,1}^2/2\mathrm{}\mathrm{\Delta }`$, $`\mathrm{\Delta }`$ being a small quantity. Following the standard approach to resonance allen87A , we define new rotating system wavefunctions $`g_p`$ according to
$`f_1(\xi ,t)`$ $``$ $`\mathrm{exp}\left[i\left(ϵ_1{\displaystyle \frac{m_1^{}v_{g,1}^2}{2}}\right){\displaystyle \frac{t}{\mathrm{}}}\right]\mathrm{exp}\left(i\overline{k}\xi \right)g_1(\xi ,t)`$
$`f_2(\xi ,t)`$ $``$ $`\mathrm{exp}[i(\mathrm{}\mathrm{\Delta }+ϵ_2{\displaystyle \frac{m_2^{}v_{g,2}^2}{2}}){\displaystyle \frac{t}{\mathrm{}}}]\times `$ (10)
$`\mathrm{exp}\left(i\overline{k}\xi \right)g_2(\xi ,t)`$
with $`\mathrm{}\overline{k}=(m_1^{}v_{g,1}+m_2^{}v_{g,2})/2`$. In the philosophy of the rotating-wave approximation allen87A , we discard oscillating terms after substituting (I-10) into (7), thus retaining only the slowly varying dynamics. Under these approximations, Eq. (7) can be finally cast to the form of Eqs . (LABEL:eq:syssol1 \- LABEL:eq:syssol2) defining adimensional variables $`\eta =\xi /a_0`$ and $`\tau =t/\tau _0`$, with $`\tau _0=\sqrt{|m_1^{}m_2^{}|}a_0^2/\mathrm{}`$, and defining the parameters as $`\alpha =\sqrt{|m_2^{}/m_1^{}|}`$, $`\mathrm{\Delta }k=a_0(m_2^{}v_{g,2}m_1^{}v_{g,1})/\mathrm{}`$, $`\delta =\mathrm{\Delta }\tau _0`$, $`\beta =a_0\tau _0A_0\omega \pi _{1,2}\overline{k}/\mathrm{}`$, and $`u_j=\sqrt{2g_{1D}\mathrm{\Gamma }_{1,2}\tau _0/\mathrm{}}g_j`$ , and $`\gamma _1=\mathrm{\Gamma }_{1,1,1,1}/\mathrm{\Gamma }_{1,1,2,2}`$ and $`\gamma _2=\mathrm{\Gamma }_{2,2,2,2}/\mathrm{\Gamma }_{1,1,2,2}`$.
To check the validity of our approach, we have also performed ab initio integrations of the evolution of a Bose-Einstein condensate in an optical lattice, as given by Eq. (3). In these computations, the transversal trapping is assumed to be constant in time and harmonic with $`\omega _{}=2\pi \times 50Hz`$. Initially a condensate of 400 atoms is prepared in a trap formed by this transversal potential and an axial potential with $`\omega _{||}=2\times 9Hz`$, (point (a) in figure 2). At the starting time, the axial potential is switched off, and the condensate is phase imprinted with a momentum $`\mathrm{}k_0=\mathrm{}0.9\pi /a_0`$ with $`a_0=1.96\mu m`$. In the experiment eierm04A this phase imprinting is replaced by a displacement of the optical lattice with constant velocity. After the phase imprinting, an oscillating optical lattice potential is switched on adiabatically during $`10ms`$ to reach a maximum potential of $`V_0=0.85E_r`$ ($`E_r=\mathrm{}^2\pi ^2/2ma_0^2`$). As the lattice period is $`a_0`$, the wavepacket imprinted with momentum $`\mathrm{}k_0`$ is now located near the edge of the first Brillouin zone (points (b) and (c) in figure 2). The frequency of the lattice oscillation is $`\omega =2\pi \times 64Hz`$ and the oscillation amplitude $`\zeta _0=0.35\mu m`$. With this procedure we obtain a wavepacket with similar characteristics that in the experiment eierm04A but in an oscillating lattice. Figs 3(a) and (b) show the results of this ab initio integration of Eq. (3) in a similar fashion as in plots 1(a) and (b) respectively. In Fig. 3, the vector components $`f_1`$ and $`f_2`$ have been extracted from the time-dependent wavefunction by projecting in the Wannier base of the first and second bands. The mechanism of formation of the solitary wave follows the same steps as in the model eqs. (LABEL:eq:syssol1-LABEL:eq:syssol2): a transient characterized by the emission of radiation, and final state in which the two-component solitary wave is formed. Note that, in contrast with Fig. 1, no oscillations between the vector components are observed. This is not a general case, as our computations show that the oscillating transient behavior can be recovered when the form of the initial wavefunction is chosen to be more similar to the solitary wave shape. We have found, in this case, that the frequency of the oscillations increases with the amplitude $`\zeta _0`$ and as frequency $`\omega `$ decreases (i.e. $`\delta `$ increases). This suggests that these oscillations are of the Rabi type. The absence of them in Fig. 3 can be then explained by the more complex dynamics involved in the convergence of the system as the difference between the shape of the initial wavefunction and the final solitary wave is larger. For the case of scalar gap solitons, it is known that the wavepackets drifts in momentum space as a consequence of the radiation process during the transient plaja04A . Under these circumstances, the parameters of the interband transitions change and, therefore, the Rabi regime is severely distorted.
In conclusion, we have proved the existence of a new kind of vector solitary wave in which the dispersion sign is opposite for each component. Under these circumstances, one of the vector elements is highly dispersive. Selftrapping is obtained through a linear coupling in the vector equations, which allows the highly dispersive component to be stabilized. We have demonstrated that this kind of vector solitary waves can be found in experimentally using Bose-Einstein condensates in oscillating optical lattices.
This work has been supported by the Spanish Ministerio de Ciencia y Tecnología (FEDER funds, grant BFM2002-00033).
|
warning/0506/nucl-ex0506001.html
|
ar5iv
|
text
|
# Near-threshold measurement of the 4He(𝜸,𝒏) reaction
## Abstract
A near-threshold <sup>4</sup>He$`(\gamma ,n)`$ cross-section measurement has been performed at MAX-lab. Tagged photons from 23 $`<`$ $`E_\gamma `$ $`<`$ 42 MeV were directed toward a liquid <sup>4</sup>He target, and neutrons were detected by time-of-flight in two liquid-scintillator arrays. Seven-point angular distributions were measured for eight photon energies. The results are compared to experimental data measured at comparable energies and Recoil-Corrected Continuum Shell Model, Resonating Group Method, and recent Hyperspherical-Harmonic Expansion calculations. The angle-integrated cross-section data are peaked at a photon energy of about 28 MeV, in disagreement with the value recommended by Calarco, Berman, and Donnelly in 1983.
Over the past several decades, many experiments have been performed in an attempt to understand the near-threshold photodisintegration of <sup>4</sup>He. In 1983, a review article by Calarco, Berman, and Donnelly (CBD) assessed all available experimental data and made a recommendation as to the value of the <sup>4</sup>He$`(\gamma ,n)`$ cross section up to a photon energy of 50 MeV. Subsequently, the bulk of the experimental effort has been directed towards measuring either the ratio of the photoproton-to-photoneutron cross sections or simply the photoproton channel. In contrast, only two near-threshold measurements of the photoneutron channel have been published . In this Letter, we report new results obtained for the <sup>4</sup>He$`(\gamma ,n)`$ reaction near threshold, and compare them with the CBD evaluation as well as the post-CBD data. We also demonstrate consistency with previously published higher-energy tagged-photon data . Finally, we compare our data to Recoil-Corrected Continuum Shell Model (RCCSM) calculations , a Resonating Group Method (RGM) calculation , and a recent Hyperspherical-Harmonic (HH) Expansion calculation . A detailed description of the project summarized in this article is given in and will be published in a full article .
The experiment was performed at the MAX-lab tagged-photon facility . A 93 MeV, $``$30 nA, pulse-stretched electron beam with a duty factor of 75% was used to produce quasi-monoenergetic photons via the bremsstrahlung-tagging technique . Post-bremsstrahlung electrons were momentum-analyzed in a magnetic spectrometer equipped with two 32-counter focal-plane scintillator arrays. These arrays tagged a photon-energy interval from 23 $`<`$ $`E_\gamma `$ $`<`$ 42 MeV with a FWHM energy resolution of $``$300 keV. The average instantaneous single-counter rate was 0.5 MHz, and the photon-beam collimation resulted in a tagging efficiency of $``$25%.
A storage-cell cryostat held the liquid <sup>4</sup>He which constituted the target. The cylindrical 75 mm (high) $`\times `$ 90 mm (diameter) cell of 80 $`\mu `$m thick Kapton was mounted with the cylinder axis perpendicular to the photon-beam direction. The cell was surrounded by a heat shield of three layers of 30 $`\mu `$m thick Al foil and multiple layers of the super-insulation NRC-2, all maintained at liquid-N<sub>2</sub> temperature. The assembly sat in a vacuum chamber with 125 $`\mu `$m thick Kapton entrance and exit windows. An identical empty target cell on the movable target ladder enabled measurement of room and non-<sup>4</sup>He background, which turned out to be negligible. Further, a 1 mm thick steel sheet, also mounted on the target ladder, was used to produce relativistic $`e^+e^{}`$ pairs for time-of-flight (TOF) calibration of the neutron detectors (see below). Density fluctuations in the liquid <sup>4</sup>He were negligible , as was the attenuation of the photon flux due to atomic processes within the target materials and the liquid <sup>4</sup>He .
Neutrons were detected in two large solid-angle spectrometers , each consisting of nine 20 cm $`\times `$ 20 cm $`\times `$ 10 cm deep rectangular cells mounted in a 3 $`\times `$ 3 lattice and filled with the liquid scintillator NE213A. Each of these arrays was mounted on a movable platform (45 $`<`$ $`\theta _{\mathrm{neutron}}`$ $`<`$ 135) and encased in Pb, steel, and borated-wax shielding. Plastic scintillators which were 2 cm thick were placed in front of the liquid scintillators and used to identify incident charged particles. The average flight path to the NE213A arrays was 2.6 m, resulting in a 6 msr geometrical solid angle for a single cell and a FWHM TOF neutron-energy resolution of $`<`$2 MeV, which allowed unambiguous identification of two-body <sup>4</sup>He$`(\gamma ,n)`$ events (see the overset in Figure 1). Thus, the neutron energy also provided a cross check on the tagged-photon energy.
Gamma-ray sources were used to calibrate pulse-height output from the NE213A scintillators which was necessary to determine the neutron-detection threshold and thus the neutron-detection efficiency. Pulse-Shape Discrimination (PSD) was employed to distinguish neutrons from photons as the background photon flux on the TOF spectrometers was $``$10<sup>5</sup> times greater than the neutron flux. All events not seen by the veto detector and identified as neutrons by the PSD modules generated a trigger for the data-acquisition system . The data set for each neutron detector consisted of 64 TOF spectra containing real coincidences with the tagger focal plane and a random background (see the overset in Figure 1). The ratio of prompt neutrons to random background (due mainly to photons which survived the PSD rejection and neutrons resulting from untagged bremsstrahlung) was a strong function of photon energy, ranging from 1-to-1 at $`E_\gamma `$ $`=`$ 40.7 MeV to 1-to-10 at $`E_\gamma `$ $`=`$ 24.6 MeV. The 64 TOF spectra were summed in eight groups of eight tagger counters resulting in $``$2.5 MeV wide photon-energy bins, each accumulating $``$10<sup>12</sup> photons over the course of the measurement. The background was fitted by superimposing a periodic ripple (related to the electron beam circuit time within the pulse-stretcher ring ) upon an exponential distribution (due to dead-time effects in the detectors and the single-hit TDCs used to instrument the focal plane).
The background-subtracted neutron yield was corrected for tagger focal-plane dead-time effects . A geant3-based Monte-Carlo simulation was used to determine the neutron-yield attenuation between the reaction vertex and the detector cells as well as the contribution of time-correlated background neutrons scattering into the detectors. The neutron-detection efficiency was determined using the stanton Monte-Carlo code . Cross checks of the predictions made by geant3 and stanton were performed via a dedicated measurement of the neutron-detection efficiency using a <sup>252</sup>Cf fission-fragment source . A summary of the corrections applied to the cross-section data and the corresponding systematic uncertainties is presented in Table 1.
The angular distributions measured at each photon energy were converted from the laboratory to the Center-of-Mass (CM) frame and fitted using
$`{\displaystyle \frac{d\sigma }{d\mathrm{\Omega }_{\mathrm{CM}}}}(\theta _{\mathrm{CM}})=`$ (1)
$`\alpha \left\{\mathrm{sin}^2(\theta _{\mathrm{CM}})\left[1+\beta \mathrm{cos}(\theta _{\mathrm{CM}})+\gamma \mathrm{cos}^2(\theta _{\mathrm{CM}})\right]+\delta +ϵ\mathrm{cos}(\theta _{\mathrm{CM}})\right\}`$
(see Figure 1). This expansion assumes that the photon multipolarities are restricted to $`E1`$, $`E2`$, and $`M1`$, and that the nuclear matrix elements of the $`E`$-multipoles to final states with a channel spin of unity are negligible<sup>1</sup><sup>1</sup>1 Note that Weller et al. claim non-zero interfering $`E1`$ $`S=1`$ strength.. Under these assumptions, $`\alpha `$ arises from the incoherent sum of the $`E1`$, $`E2`$, and $`M1`$ multipoles, $`\beta `$ is due to the interference of the $`E1`$ and $`E2`$ multipoles, $`\gamma `$ results from the $`E2`$ multipole, $`\delta `$ arises from the $`M1`$ multipole, and $`ϵ`$ is zero. Similar to analyses of complementary <sup>4</sup>He$`(\gamma ,p)`$ angular distributions , our angular distributions were constrained to vanish at $`\theta _{\mathrm{CM}}`$ $`=`$ (0,180) $`\mathrm{deg}`$ by forcing the $`\delta `$ and $`ϵ`$ coefficients to be zero.
Figure 2 presents the $`\alpha `$, $`\beta `$, and $`\gamma `$ coefficients (filled circles) together with those extracted from a recent reanalysis of the higher-energy data of Sims et al. (open circles). We stress that these two data sets from MAX-lab are the only tagged-photon data in existence which are differential in angle. Error bars are the statistical uncertainties, while the systematic uncertainties are represented by the bands at the base of each panel. Also shown are RCCSM and RGM calculations. The recent HH calculation does not presently predict angular distributions.
The RCCSM calculations were performed within a continuum shell-model framework in the ($`1p1h`$) approximation, where the transition matrix elements of the $`M1`$ and the spin-independent $`M2`$ multipole operators vanished. Corrections were applied for target recoil. In addition to the Coulomb force, the effective nucleon-nucleon (NN) interaction included central, spin-orbit, and tensor components. Perturbation theory was employed to compute matrix elements for the multipoles and the multipole operators were calculated in the long-wavelength limit. Corrections for spurious CM excitations made these calculations essentially equivalent to the multichannel microscopic RGM calculations. Here, a similar semi-realistic NN force was employed, and the variational principle was used to determine the scattering wave functions. The radiative processes were treated within the Born Approximation, and the electromagnetic transition operators were again taken in the long-wavelength limit. Angular momenta up to $`L=2`$ were allowed in the relative motion of the fragments. Note that the authors of the calculations originally presented their results in the form of Legendre coefficients as a function of CM proton energy.
As can be seen, the data largely reproduce the trends predicted by the calculations. At lower photon energies, the $`E1`$ multipole is completely dominant and the $`\alpha `$ data have a clear resonant structure peaking at about 28 MeV. The RCCSM calculation tends to overestimate these data, but also shows resonant structure peaking at about 25 MeV. The energy dependence of the $`\beta `$ data is reasonably consistent with both the RCCSM and the RGM predictions, given the relatively large systematic uncertainties for $`E_\gamma `$ $`<`$ 26 MeV. Similarly, when accuracy and precision are considered, there is no significant disagreement between the present $`\gamma `$ data and the RCCSM calculation. At higher photon energies, $`E2`$ strength is expected to become more important. Unfortunately, the calculations do not cover the range of the higher-energy data. However, these data do appear to be consistent with the energy-dependent trends of both the lower-energy data and the calculations.
Figure 3 presents the angle-integrated cross-section data (filled circles) together with those extracted from a recent reanalysis of the higher-energy data of Sims et al. (open circles). On average, these angle-integrated data are approximately 7% larger than those which result from simply scaling our $`\theta _{\mathrm{CM}}`$ $`=`$ 90 $`\mathrm{deg}`$ results by $`8\pi /3`$. Also shown is the CBD evaluation , data from a <sup>3</sup>He$`(n,\gamma )`$ measurement , data from a <sup>4</sup>He$`(\gamma ,^3`$He$`)`$ active-target measurement , the recent RCCSM calculation , and the recent HH calculation . Error bars show the statistical uncertainties, while the systematic uncertainties are represented by the bands at the base of the figure.
The recent RCCSM calculation expanded the model space of Ref. to include more reaction channels and all $`p`$-shell nuclei. The HH calculation used a correlated hyperspherical expansion of basis states, with final-state interactions accounted for using the Lorentz Integral Transform Method (which circumvents the calculation of continuum states). For clarity, the small uncertainty in the HH calculation is not shown here. Note that both calculations employ the semirealistic MTI-III potential .
The present <sup>4</sup>He$`(\gamma ,n)`$ excitation function has a clear resonant structure peaking at about 28 MeV. Although data are lacking between 42 and 50 MeV, there is no apparent discontinuity with respect to the re-analyzed MAX-lab data of . Furthermore, the present data extrapolate smoothly to the lower-energy data of . Conversely, the data of exhibit a slow rise which is at odds with all other data, the calculations, and the CBD evaluation. Both the RCCSM and HH calculations are in good agreement with the present data and those of up to the resonant peak at $`E_\gamma `$ $``$ 28 MeV. At higher energies, both calculations tend to overpredict the cross section, although the HH calculation follows the general shape of the excitation function up to 70 MeV reasonably well. Development of the HH formalism continues , and we anticipate new predictions in the near future which use fully realistic NN potential models and which may also include 3N-force effects.
In summary, $`\frac{d\sigma }{d\mathrm{\Omega }}(\theta )`$ for the <sup>4</sup>He$`(\gamma ,n)`$ reaction have been measured with tagged photons and compared to other available measurements and calculations. The energy dependence of the $`\alpha `$, $`\beta `$, and $`\gamma `$ coefficients extracted from the angular distributions agrees reasonably with trends predicted by RCCSM and RGM calculations. The marked resonant behaviour of the present angle-integrated cross section, peaking at about 28 MeV, is in good agreement with recent RCCSM and HH calculations as well as capture data which extend close to the $`(\gamma ,n)`$ threshold. This behaviour disagrees with an evaluation of $`(\gamma ,n)`$ data made in 1983, and recent active-target data .
The authors acknowledge the outstanding support of the MAX-lab staff which made this experiment successful. We also wish to thank Sofia Quaglioni, Winfried Leidemann, and Giuseppina Orlandini (University of Trento, Italy), John Calarco (University of New Hampshire, USA), Gerald Feldman (The George Washington University, USA), Dean Halderson (Western Michigan University, USA), Andreas Reiter (University of Glasgow, Scotland), and Brad Sawatzky (University of Virginia, USA) for valuable discussions. BN wishes to thank Margareta Söderholm and Ralph Hagberg for their unwavering support. The Lund group acknowledges the financial support of the Swedish Research Council, the Knut and Alice Wallenberg Foundation, the Crafoord Foundation, the Swedish Institute, the Wenner-Gren Foundation, and the Royal Swedish Academy of Sciences. The Glasgow group acknowledges the financial support of the UK Engineering and Physical Sciences Research Council.
|
warning/0506/quant-ph0506206.html
|
ar5iv
|
text
|
# Single-shot measurement of quantum optical phase
## Abstract
Although the Canonical phase of light, which is defined as the complement of photon number, has been described theoretically by a variety of distinct approaches, there have been no methods proposed for its measurement. Indeed doubts have been expressed about whether or not it is measurable. Here we show how it is possible, at least in principle, to perform a single-shot measurement of Canonical phase using beam splitters, mirrors, phase shifters and photodetectors.
Quantum-limited phase measurements of the optical field have important applications in precision measurements of small distances in interferometry and in the emerging field of quantum communication, where there is the possibility of encoding information in the phase of light pulses. Much work has been done in attempting to understand the quantum nature of phase. Some approaches have been motivated by the aim of expressing phase as the complement of photon number Leon . Examples of these approaches include the probability operator measure approach Helstrom ; SS , a formalism in which the Hilbert space is doubled Newton , a limiting approach based on a finite Hilbert space PB ; tute and a more general axiomatic approach Leon . Although these approaches are quite distinct, they all lead to the same phase probability distribution for a field in state $`|\psi `$ as a function of the phase angle $`\theta `$ Leon :
$$P(\theta )=\frac{1}{2\pi }|\underset{n=0}{\overset{\mathrm{}}{}}\psi |n\mathrm{exp}(in\theta )|^2$$
(1)
where $`|n`$ is a photon number state. Leonhardt et al. Leon have called this common distribution the “canonical” phase distribution to indicate a quantity that is the canonical conjugate, or complement, of photon number. This distribution is shifted uniformly when a phase shifter is applied to the field and is not changed by a photon number shift Leon . We adopt this definition here and use the term Canonical phase to denote the quantity whose distribution is given by (1).
Much less progress has been made on ways to measure Canonical phase. Homodyne techniques can be used to measure phase-like properties of light but are not measurements of Canonical phase. It is possible in principle to measure the Canonical phase distribution by a series of experiments on a reproducible state of light BPprl but there has been no known way of performing a single-shot measurement. Indeed it is thought that this might be impossible Wiseman . Even leaving aside the practical issues, the concept that a particular fundamental quantum observable may not be measurable, even in principle, has interesting general conceptual ramifications for quantum mechanics. A different approach to the phase problem, which avoids difficulties in finding a way to measure Canonical phase, is to define phase operationally in terms of observables that can be measured Leon . The best known of these operational phase approaches is that of Noh et al. Noh1 ; Noh2 . Although the experiments to measure this operational phase produce excellent results, they were not designed to measure Canonical phase as defined here and, as shown by the the measured phase distribution Noh2 , they do not measure Canonical phase. In this paper we show how, despite these past difficulties, it is indeed possible, at least in principle, to perform a single-shot measurement of Canonical phase in the same sense that the experiments of Noh et al. are single-shot measurements of their operational phase.
A single-shot measurement of a quantum observable must not only yield one of the eigenvalues of the observable, but repeating the measurement many times on systems in identical states should result in a probability distribution appropriate to that state. If the spectrum of eigenvalues is discrete, the probabilities of the results can be easily obtained from the experimental statistics. Where the spectrum is continuous, the probability density is obtainable by dividing the eigenvalue range into a number of small bins and finding the number of results in each bin. As the number of experiments needed to obtain measurable probabilities increases as the reciprocal of the bin size, a practical experiment will require a non-zero bin size and will produce a histogram rather than a smooth curve.
Although the experiments of Noh et al. Noh1 ; Noh2 were not designed to measure Canonical phase, it is helpful to be guided by their approach. In addition to their results being measured and plotted as a histogram, some of the experimental data are discarded, specifically photon count outcomes that lead to an indeterminacy of the type zero divided by zero in their definitions of the cosine and sine of the phase Noh1 . The particular experiment that yields such an outcome is ignored and its results are not included in the statistics.
Concerning the discarding of some results, we note in general that the well-known expression for the probability that a von Neumann measurement on a pure state $`|\psi `$ yields a result $`q`$ is $`\psi |qq|\psi `$, where $`|q`$ is the eigenstate of the measurement operator corresponding to eigenvalue $`q`$. If the state to be measured is a mixed state with a density operator $`\widehat{\rho }`$ the expression becomes Tr($`\widehat{\rho }\widehat{\mathrm{\Pi }}_q)`$ where $`\widehat{\mathrm{\Pi }}_q`$ = $`|qq|`$. The operator $`\widehat{\mathrm{\Pi }}_q`$ is a particular case of an element of a probability operator measure (POM) Helstrom . The sum of all the elements of a POM is the unit operator and the expression Tr($`\widehat{\rho }\widehat{\mathrm{\Pi }}_q)`$ for the probability is based on the premise that all possible outcomes of the measurement are retained for the statistics. If some of the possible outcomes of an experiment are discarded, the probability of a particular result calculated from the final statistics is given by the normalized expression Tr($`\widehat{\rho }\widehat{\mathrm{\Pi }}_q)/_p`$Tr($`\widehat{\rho }\widehat{\mathrm{\Pi }}_p)`$, where the sum is over all the elements of the POM corresponding to outcomes of the measurement that are retained.
We seek now to approximate the continuous distribution (1) by a histogram representing the probability distribution for a discrete observable $`\theta _m`$ such that when the separation $`\delta \theta `$ of consecutive values of $`\theta _m`$ tends to zero the continuous distribution is regained. A way to do this is first to define a state
$$|\theta _m=\frac{1}{(N+1)^{1/2}}\underset{n=0}{\overset{N}{}}\mathrm{exp}(in\theta _m)|n.$$
(2)
There are $`N+1`$ orthogonal states $`|\theta _m`$ corresponding to $`N+1`$ values $`\theta _m=m\delta \theta `$ with $`\delta \theta =2\pi /(N+1)`$ and $`m=0,1,\mathrm{}N`$. This range for $`m`$ ensures that $`\theta _m`$ takes values between $`0`$ and $`2\pi `$. Then, if we can find a measurement technique that yields the result $`\theta _m`$ with a probability of $`|\psi |\theta _m|^2`$, the resulting histogram will approximate a continuous distribution with a probability density of $`|\psi |\theta _m|^2/\delta \theta `$. It follows that, as we let $`N`$ tend to infinity, there will exist a value of $`\theta _m`$ as close as we like to any given value of $`\theta `$ with a probability density approaching $`P(\theta )`$ given by (1). If we keep $`N`$ finite so that we can perform an experiment with a finite number of outcomes, then the value of $`N`$ must to be sufficiently large to give the resolution $`\delta \theta `$ of phase angle required and also for $`|\psi `$ to be well approximated by $`_nn|\psi |n`$ where the sum is from $`n=0`$ to $`N`$. The latter condition ensures that the terms with coefficients $`n|\psi `$ for $`n>N`$ have little effect on the probability $`|\psi |\theta _m|^2`$. As we shall be interested mainly in weak optical fields in the quantum regime with mean photon numbers of the order of unity, the maximum phase resolution $`\delta \theta `$ desired will usually be the determining factor in the choice of $`N`$.
When $`N`$ is finite, the states $`|\theta _m`$ do not span the whole Hilbert space, so the projectors $`|\theta _m\theta _m|`$ will not sum to the unit operator $`\widehat{1}`$. Thus these projectors by themselves do not form the elements of a POM. To complete the POM we need to include an element $`\widehat{1}_m|\theta _m\theta _m|`$. If we discard the outcome associated with this element, that is, treat an experiment with this outcome as an unsuccessful attempt at a measurement in a similar way that Noh et al. Noh1 treated experiments with indeterminate outcomes, then the probability that the outcome of a measurement is the phase angle $`\theta _m`$ is given by
$$Pr(\theta _m)=\frac{\text{Tr}(\widehat{\rho }|\theta _m\theta _m|)}{_p\text{Tr}(\widehat{\rho }|\theta _p\theta _p|)},$$
(3)
where $`p=0,1\mathrm{}N`$. We now require a single-shot measuring device that will reproduce this probability in repeated experiments.
As measurements will be performed ultimately by photodetectors, we seek an optical device that transforms input fields in such a way that photon number measurements at the output ports can be converted to phase measurements. We examine the case of a multi-port device with $`N+1`$ input modes and $`N+1`$ output modes as depicted schematically in Fig. 1. As phase is not an absolute quantity, that is it can only be measured in relation to some reference state, we shall need the field in state $`|\psi _0`$ that is to be measured to be in one input and a reference field in state $`|B_{1\text{ }}`$ to be in another input. We let there be vacuum states $`|0_i`$ with $`i=2,3\mathrm{}N`$ in the other input modes. We let the device be such that the input states are transformed by a unitary transformation $`\widehat{R}`$. We let $`N+1`$ photodetectors that can distinguish between zero photons, one photon and more than one photon be in the output modes.
Consider the case where one photon is detected in each output mode except for mode $`m`$, in which zero photons are detected. The amplitude for this detection event is
$${}_{m}{}^{}0|\left(\underset{jm}{}{}_{j}{}^{}1|)\widehat{R}(\underset{i=2}{\overset{N}{}}|0_i\right)|B_{1}^{}|\psi _0$$
(4)
where the first product is over values of $`j`$ from $`0`$ to $`N`$ excluding the value $`m`$. We require the transformation and the reference state to be such that this amplitude is proportional to $`{}_{0}{}^{}\theta _m|\psi _{0}^{}`$ . We write the photon creation operators acting on the mode $`i`$ as $`\widehat{a}_i^{}`$. Writing $`|1_j`$ as $`\widehat{a}_j^{}|0_j`$ we see that we require $`|\theta _m_0`$ to be proportional to
$${}_{1}{}^{}B|\left(\underset{i=2}{\overset{N}{}}{}_{i}{}^{}0|)\left(\underset{jm}{}\widehat{R}^{}\widehat{a}_j^{}\widehat{R}\right)\widehat{R}^{}(\underset{k=0}{\overset{N}{}}|0_k\right)$$
(5)
We rewrite the unitary transformation in the form
$$\widehat{R}^{}\widehat{a}_j^{}\widehat{R}=\underset{i=0}{\overset{N}{}}U_{ij}\widehat{a}_i^{}$$
(6)
where $`U_{ij}`$ are the elements of a unitary matrix. Reck et al. Reck have shown how it is possible to construct a multi-port device from mirrors, beam splitters and phase shifters that will transform the input modes into the output modes in accord with any $`(N+1)\times (N+1)`$ unitary matrix. We choose such a device for which the associated unitary matrix is
$$U_{ij}=\frac{\omega ^{ij}}{\sqrt{N+1}}$$
(7)
where $`\omega =`$ $`\mathrm{exp}[i2\pi /(N+1)]`$ that is, a $`(N+1)`$th root of unity. Such a device will transform the combined vacuum state to the combined vacuum state so we can delete the $`\widehat{R}^{}`$ on the right of expression (5). Substituting (7) into (6) gives eventually
$`\left({\displaystyle \underset{i=2}{\overset{N}{}}}{}_{i}{}^{}0|)\left({\displaystyle \underset{jm}{}}\widehat{R}^{}\widehat{a}_j^{}\widehat{R}\right)({\displaystyle \underset{k=0}{\overset{N}{}}}|0_k\right)=`$
$`\kappa _1\left[{\displaystyle \underset{jm}{}}(\widehat{a}_0^{}+\omega ^j\widehat{a}_1^{})\right]|0_0|0_1`$ (8)
where $`\kappa _1=(N+1)^{N/2}`$.
To evaluate (8) we divide both sides of the identity
$$X^{N+1}+(1)^N=(X+1)(X+\omega )(X+\omega ^2)\mathrm{}(X+\omega ^N)$$
(9)
by $`X+\omega ^m`$ to give, after some rearrangement and application of the relation $`\omega ^{m(N+1)}=1`$,
$$\underset{jm}{}(X+\omega ^j)=(1)^N\omega ^{mN}\frac{1(X\omega ^m)^{N+1}}{1(X\omega ^m)}\text{ .}$$
(10)
The last factor is the sum of a geometric progression. Expanding this and substituting $`X=x/y`$ gives eventually the identity
$$\underset{jm}{}(x+\omega ^jy)=\underset{n=0}{\overset{N}{}}x^n(\omega ^my)^{Nn}\text{ .}$$
(11)
We now expand $`|B_1`$ in terms of photon number states as
$$|B_1=\underset{n=0}{\overset{N}{}}b_n|n_1$$
(12)
and put $`x=a_0^{}`$ and $`y=a_1^{}`$ in (11). Then from (8) we find that (5) becomes
$`{}_{1}{}^{}B|\left({\displaystyle \underset{i=2}{\overset{N}{}}}{}_{i}{}^{}0|)\left({\displaystyle \underset{jm}{}}\widehat{R}^{}\widehat{a}_j^{}\widehat{R}\right)\widehat{R}^{}({\displaystyle \underset{k=0}{\overset{N}{}}}|0_k\right)=`$
$`\kappa _2{\displaystyle \underset{n=0}{\overset{N}{}}}(1)^{Nn}\left({\displaystyle \genfrac{}{}{0pt}{}{N}{n}}\right)^{1/2}\omega ^{nm}b_{Nn}^{}|n_0`$ (13)
where $`\kappa _2=\kappa _1\omega ^m(N!)^{1/2}`$. We see then that, if we let $`|B_1`$ be the binomial state
$$|B_1=2^{N/2}\underset{n=0}{\overset{N}{}}(1)^n\left(\genfrac{}{}{0pt}{}{N}{n}\right)^{1/2}|n_1,$$
(14)
then expression (13) is proportional to $`_n\omega ^{nm}|n_0`$, that is, to $`|\theta _m_0`$. Thus the amplitude for the event that zero photons are detected in output mode $`m`$ and one photon is detected in all the other output modes will be proportional to $`{}_{0}{}^{}\theta _m|\psi _{0}^{}`$. The probability that the outcome of a measurement is this event, given that only outcomes associated with the $`(N+1)`$ events of this type are recorded in the statistics, will be given by (3), where we note that the proportionality constant $`\kappa _2`$ will cancel from this expression. Thus the measurement event that zero photons are detected in output mode m and one photon is detected in all the other output modes can be taken as the event that the result of the measurement of the phase angle is $`\theta _m`$. Thus the photodetector with zero photocounts, when all other photodetectors have registered one photocount, can be regarded as a digital pointer to the value of the measured phase angle.
We have shown, therefore, that it is indeed possible in principle to conduct a single-shot measurement of Canonical phase to within any given non-zero error, however small. This error is of the order $`2\pi /(N+1)`$ and will determine the value of $`N`$ chosen.
While the aim of this paper is to establish how Canonical phase can be measured in principle, it is worth briefly considering some practical issues. Although we have specified that the photodetectors need only be capable of distinguishing among zero, one and more than one photons, reflecting the realistic case, there are other imperfections such as inefficiency. These will give rise to errors in the phase measurement, just as they will cause errors in a single-shot photon number measurement. In practice, there is no point in choosing the phase resolution $`\delta \theta `$ much smaller than the expected error due to photodetector inefficiencies, thus there is nothing lost in practice in keeping $`N`$ finite. A requirement for the measuring procedure is the availability of a binomial state. Such states have been studied for some time binom but their generation has not yet been achieved. In practice, however, we are usually interested in measuring weak fields in the quantum regime with mean photon numbers around unity Noh1 ; Noh2 and even substantially less Torg . Only the first few coefficients of $`|n_0`$ in (13) will be important for such weak fields. Also, it is not difficult to show that the reference state need not be truncated at $`n=N`$, as indicated in (12), as coefficients $`b_n`$ with $`n>N`$ will not appear in (13). Thus we need only prepare a reference state with a small number of its photon number state coefficients proportional to the appropriate binomial coefficients. Additionally, of course, in a practical experiment we are forced to tolerate some inaccuracy due to photodetector errors, so it will not be necessary for the reference state coefficients to be exactly proportional to the corresponding binomial state coefficients. These factors give some latitude in the preparation of the reference state. The muti-port device depicted in Fig. 1 can be constructed in a variety of ways. Reck et al. Reck provide an algorithm for constructing a triangular array of beam splitters to realize any unitary transformation matrix. Fig. 2 shows, for example, such a device that, with suitable phase shifters in the output modes, will realize the transformation $`U_{ij}`$ in (7) with $`N+1=4`$. Because we are detecting photons, however, these output phase shifters are not actually necessary and are thus not shown. The number of beam splitters needed for a general triangular array increases quadratically with $`N`$. Fortunately, however, our required matrix (7) represents a discrete Fourier transformation and we only require two of the input ports to have input fields that are not in the vacuum state. The device of Törmä and Jex Jex is ideally suited for these specific requirements. This device, which has an even number of input and output ports, is pictured in Ref. Jex . It consists of just $`(N+1)/2`$ ordinary 50:50 beam splitters and two plate beam splitters. The latter are available in the form of glass plates with modulated transmittivity along the direction of the incoming beam propagation Jex .
As a large fraction of the raw data in this procedure is discarded, an interesting question arises as to whether or not there is a relationship between the method of this paper and a limiting case of the operational phase measurements of Torgerson and Mandel Torg2 where it is found that the distribution becomes sharper as more data are discarded. While preliminary analysis indicates that there is not, this will be discussed in more detail elsewhere.
In conclusion we have shown that it is possible in principle to perform a single-shot measurement of the Canonical phase in the same sense that the experiments of Noh et al. are single-shot measurements of operational phase. The technique relies on generating a reference state with some number state coefficients proportional to those of a binomial state.
###### Acknowledgements.
D. T. P. thanks the Australian Research Council for funding.
|
warning/0506/hep-ph0506312.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
In any Grand Unified Theory (GUT) understanding the Higgs sector is not an easy task. Usually these models require the existence of more than one Higgs multiplet. In the minimal $`SU(5)`$ GUT one employs one adjoint 24 -plet ($`\mathrm{\Sigma }`$) and a fundamental 5-plet to break the GUT symmetry down to $`SU(3)_C\times U(1)_{em}`$. The Yukawa couplings of the 5-plet Higgs also generate quark and lepton masses. The Higgs sector becomes somewhat more complicated in larger GUT structures such as $`SO(10)`$ since there is a larger symmetry that needs to be broken. Conventional $`SO(10)`$ models employ at least two different Higgs representations to break the symmetry down to $`SU(3)_C\times SU(2)_L\times U(1)_Y`$ (a 16 or a 126 to change rank, and one of 45, 54 or a 210 to break the symmetry down further ). To achieve electro-weak symmetry breaking and to generate quark and lepton masses an additional 10 dimensional representation is also needed. A minimal $`SO(10)`$ model studied recently, for example, utilizes one 10, one 126 and one 210 Higgs representations to achieve symmetry breaking and to generate masses for the quarks, leptons and the neutrinos.
In this paper we discuss the following question: Is it possible to achieve $`SO(10)`$ symmetry breaking all the way down to the $`SU(3)_C\times U(1)_{em}`$ with a single Higgs representation? We find that this is indeed the case if one employs a 144-plet of Higgs of $`SO(10)`$. The 144-plet is contained in the product $`10\times 16`$, and thus carries one vector and one spinor index. An interesting property of the 144-plet which makes such a symmetry breaking chain possible is that it contains in it an $`SU(5)`$ adjoint with a $`U(1)`$ charge, as well as Standard Model Higgs doublet fields. This can be seen from the following decomposition of 144 under $`SU(5)\times U(1)`$ subgroup of $`SO(10)`$
$`144=\overline{5}(3)+5(7)+10(1)+15(1)+24(5)+40(1)+\overline{45}(3)`$ (1)
It is significant that the $`SU(5)`$ adjoint 24(-5) above also carries a U(1) charge. Once the Standard Model singlet in it acquires a VEV, it would change the rank of the group, leading to a one-step breaking of $`SO(10)`$ down to $`SU(3)_C\times SU(2)_L\times U(1)_Y`$. The sub-multiplets $`\overline{5}(3),5(7)`$ and $`\overline{45}(3)`$ all contain fields with identical quantum numbers as the Standard Model Higgs doublet. If one combination of doublets from these sub-multiplets is made light by fine tuning, it can be used for electro-weak symmetry breaking. Such fine tuning is justified in the context of the multiple vacua of the string landscapes, which has been widely discussed recently. Although consistency of the first stage of symmetry breaking requires the mass-squared of all the physical Higgs particles to be positive (including that of the light Higgs doublet), we show that radiative corrections involving the Higgs self-couplings can turn the mass-squared of the light Higgs doublet negative, facilitating the second stage of symmetry breaking.
Realistic fermion masses can be obtained within this minimal scenario. Recall that the fermions of each family belongs to the 16 dimensional spinor representation of $`SO(10)`$. Under $`SU(5)\times U(1)`$ subgroup the 16 decomposes as follows
$`16=10(1)+\overline{5}(3)+1(5)`$ (2)
Fermion masses will arise from quartic couplings of the $`16_i16_j(144144)`$ and $`16_i16_j`$ $`(144^{}144^{})`$. These couplings would lead to Dirac masses for all the fermions as well as large Majorana masses for the right-handed neutrinos. Since the light Higgs doublet is a linear combination of doublets from 5 and 45 of $`SU(5)`$, the resulting mass pattern is not that of minimal $`SU(5)`$ and is consistent with experimental data, including neutrino oscillations.
The outline of the rest of the paper is as follows: In Sec. 2 we discuss symmetry and mass growth in the $`SU(5)\times U(1)`$ language. In Sec. 3 we discuss the techniques of calculation for the analysis of $`144`$ and $`\overline{144}`$ plet couplings using the method developed in Ref.. Here also we discuss the set of couplings ($`144\times \overline{144}`$), $`(144\times \overline{144})_1(144\times \overline{144})_1`$, $`(144\times \overline{144})_{45}(144\times \overline{144})_{45}`$ and $`(144\times \overline{144})_{210}(144\times \overline{144})_{210}`$. These couplings are needed in the computation of symmetry breaking which is then analysed. In Sec. 4 Higgs phenomenon and mass growth are analysed for the breaking of $`SO(10)`$. Here it is shown that within the landscape scenario with fine tuning one gets a pair of light Higgs doublets exactly as in the minimal supersymmetric standard model (MSSM) while the Higgs triplets and other modes are either absorbed or become super heavy. In Sec. 5 couplings of quarks and leptons are discussed and it is shown that such couplings are quartic in nature. As an illustration the couplings involving $`(16\times 16)_{10}(144\times 144)_{10}`$ and $`(16\times 16)_{10}(\overline{144}\times \overline{144})_{10}`$ are explicitly discussed. It is shown that the resulting masses and mixings are consistent with experimental data. Conclusions are given in Sec. 6.
## 2 Symmetry breaking and mass growth with 144 in the $`\mathrm{𝐒𝐔}(\mathrm{𝟓})\times 𝐔(\mathrm{𝟏})`$ language
Analysis of the symmetry breaking and of fermion mass generation with a 144 of Higgs turns out to be rather complicated. Before we delve into the full detail in the $`SO(10)`$ language, which is presented in the next section, we analyze here these issues in the simpler $`SU(5)\times U(1)`$ subgroup language. We will present our analysis in a non-supersymmetric model. Generalization to supersymmetry require the addition of a $`\overline{144}`$ chiral multiplet, so that the flatness of the D-term potential can be maintained at the GUT scale, leaving supersymmetry intact at that scale. The analysis in this section would also hold for SUSY models with some redefinitions of parameters, provided that the $`144^{}`$ field is identified with the $`\overline{144}`$ of the SUSY $`SO(10)`$ model.
### 2.1 One step GUT symmetry breaking
Since in the $`SU(5)\times U(1)`$ decomposition of $`SO(10)`$ the 144 contains an $`SU(5)`$ adjoint carrying a non-zero $`U(1)`$ charge (see Eq.(1)), it is instructive to analyze symmetry breaking of $`SU(5)\times U(1)`$ with a complex adjoint $`\mathrm{\Sigma }`$. One can construct such a representation from two adjoint representations of $`SU(5)`$: $`\mathrm{\Sigma }=\mathrm{\Sigma }_1+i\mathrm{\Sigma }_2`$. Then $`\mathrm{\Sigma }`$ is not self-adjoint, and we denote the conjugate of $`\mathrm{\Sigma }`$ as $`\mathrm{\Sigma }^{}`$.
The most general $`SU(5)\times U(1)`$ invariant renormalizable potential involving the $`\mathrm{\Sigma }`$ and $`\mathrm{\Sigma }^{}`$ fields is
$`V=M^2tr(\mathrm{\Sigma }\mathrm{\Sigma }^{})+{\displaystyle \frac{\kappa _1}{2}}tr(\mathrm{\Sigma }^2\mathrm{\Sigma }^2)+{\displaystyle \frac{\kappa _2}{2}}(tr(\mathrm{\Sigma }\mathrm{\Sigma }^{}))^2`$
$`+{\displaystyle \frac{\kappa _3}{2}}tr(\mathrm{\Sigma }^2)tr(\mathrm{\Sigma }^2)+{\displaystyle \frac{\kappa _4}{2}}tr(\mathrm{\Sigma }\mathrm{\Sigma }^{}\mathrm{\Sigma }\mathrm{\Sigma }^{}).`$ (3)
Among the possible local minima is the one which preserves the Standard Model gauge symmetry given by the vacuum structure
$`<\mathrm{\Sigma }>=<\mathrm{\Sigma }^{}>=vdiag(2,2,2,3,3)`$ (4)
For some range of the parameters of the potential, this minimum will be the global minimum. Minimization of the potential gives
$`v^2={\displaystyle \frac{M^2}{7(\kappa _1+\kappa _4)+30(\kappa _2+\kappa _3)}}.`$ (5)
Clearly this VEV structure breaks $`SU(5)`$ down to $`SU(3)_C\times SU(2)_L\times U(1)_Y`$. Further, since $`\mathrm{\Sigma }`$ is charged under the $`U(1)`$, its VEV breaks this $`U(1)`$. Thus we see that the $`SU(5)\times U(1)`$ symmetry is broken down to the SM gauge symmetry in one step with one complex adjoint Higgs field. This can also be verified directly by computing the gauge boson masses. The physical Higgs boson masses can all be made positive for some range of parameters of the potential. It is then clear that if an $`SO(10)`$ representation contains sub-multiplets which transform under $`SU(5)\times U(1)`$ symmetry as an adjoint carrying $`U(1)`$ charge, then there is the possibility that this Higgs field can break $`SO(10)`$ all the way down to the SM gauge symmetry in one step. We observe that the 144-plet of $`SO(10)`$ is the simplest representation which has this property<sup>2</sup><sup>2</sup>2The next simplest possibility is to have a 560 of $`SO(10)`$, which contains a 24(-5), 1(-5), as well as $`\overline{5}(3),45(7)`$, and $`\overline{45}(3)`$ under $`SU(5)\times U(1)`$.. Technical details of this assertion in the $`SO(10)`$ language is postponed to the next section.
Identical conclusions can be arrived at for the case of supersymmetric $`SU(5)\times U(1)`$ model. The potential of Eq.(2.1) will become the superpotential (with $`\mathrm{\Sigma }^{}`$ replaced by a chiral superfield $`\overline{\mathrm{\Sigma }}`$), the couplings $`\kappa _i`$ will be nonrenormalizable operators with inverse dimensions of mass, and the mass term $`M^2`$ will be replaced by $`M`$. Thus we conclude that in SUSY $`SO(10)`$ a $`144+\overline{144}`$ pair of chiral superfields can break $`SO(10)`$ in one step down to the supersymmetric Standard Model gauge group.
### 2.2 Electroweak symmetry breaking
Having recognized that a single 144-plet can break non-supersymmetric $`SO(10)`$ down to the SM in one step, we turn our attention to the subsequent electro-weak symmetry breaking. As noted earlier, the 144-plet also contains fields which have the same quantum numbers as the SM Higgs doublet. We explain how these doublet fragments from the 144-plet can be used for the purpose of electro-weak symmetry breaking, thus making the model very economical.
An immediate question that can be raised is how to obtain negative mass-squared for the light Higgs doublet of the SM arising from the 144-plet. Consistency of the GUT symmetry breaking would require positivity of the mass-squared of all the physical Higgs bosons, including that of the light SM doublet. If the surviving symmetry and spectrum below the GUT scale are corresponding to those of the SM, there would be no interactions that turn the Higgs mass-squared negative needed for electroweak symmetry breaking. In analogy to the stop squark quartic couplings turning the Higgs mass-squared negative in the supersymmetric SM, we observe that the quartic couplings between the light Higgs doublet and any fragment of the 144-plet with mass an order of magnitude or so below the GUT scale can turn the Higgs doublet mass-squared negative<sup>3</sup><sup>3</sup>3Yukawa couplings between the Higgs doublet and fermions cannot turn the Higgs mass-squared negative. Cubic self couplings (which are not allowed in the SM Higgs doublet) and/or quartic/cubic scalar couplings involving other fields are necessary for this to happen.. We illustrate this mechanism with a simple $`SU(5)`$ toy model with an adjoint Higgs field below.
Consider a toy model with global $`SU(5)`$ symmetry broken spontaneously by an adjoint Higgs field $`\mathrm{\Sigma }`$. The most general renormalizable potential for this field is given by
$`V=m^2Tr(\mathrm{\Sigma }^2)+{\displaystyle \frac{\kappa _1}{2}}tr(\mathrm{\Sigma }^4)+{\displaystyle \frac{\kappa _2}{2}}(tr(\mathrm{\Sigma }^2))^2+\mu tr(\mathrm{\Sigma }^3).`$ (6)
One possible VEV structure is as shown in Eq.(4). Minimization of the potential of Eq.(6) gives
$`m^2=(7\kappa _1+30\kappa _2)v^2+{\displaystyle \frac{3}{2}}\mu v.`$ (7)
The $`(3,2,5/6)`$ and $`(3^{},2,5/6)`$ components of $`\mathrm{\Sigma }`$ (under the surviving $`SU(3)\times SU(2)\times U(1)`$ symmetry) are Goldstone bosons, while the $`(8,1,0),(1,3,0)`$ and the $`(1,1,0)`$ components have masses given respectively by
$`m_8^2={\displaystyle \frac{15}{2}}\mu v+5\kappa _1v^2,`$
$`m_3^2={\displaystyle \frac{15}{2}}\mu v+20\kappa _1v^2,`$
$`m_1^2={\displaystyle \frac{3}{2}}\mu v+(14\kappa _1+60\kappa _2)v^2.`$ (8)
For a range of parameters, all three squared-masses can be chosen positive, establishing the consistency of symmetry breaking.
It is possible by fine tuning to make the $`SU(2)_L`$ triplet field to be much lighter than the $`SU(5)`$ breaking scale, while keeping the other two components heavy. We wish to ask if such a finely tuned triplet can subsequently break $`SU(2)_L`$ further down to $`U(1)_L`$. This would, however, require that $`m_3^2`$ turn negative at lower scales even though it starts off as being positive at the high scale. Consider the case when $`\kappa _1`$ and $`\mu /v`$ are somewhat smaller than one (say of order 0.01), while $`\kappa _2`$ is of order one. Then $`m_3`$ and $`m_8`$ are generically an order of magnitude below the GUT scale $`v`$, while $`m_1`$ is of order $`v`$. In the momentum range below $`v`$ and above $`m_8`$, the mass parameters $`m_3^2`$ and $`m_8^2`$ will evolve, while the singlet decouples. The RGE for the running of these mass parameters are found to be
$`{\displaystyle \frac{dm_8^2}{dt}}`$ $`=`$ $`{\displaystyle \frac{1}{8\pi ^2}}[{\displaystyle \frac{15}{8}}(\mu +4\kappa _1v)^2+{\displaystyle \frac{5}{2}}(\kappa _1+2\kappa _2)m_8^2+{\displaystyle \frac{3}{2}}\kappa _2m_3^2]`$
$`{\displaystyle \frac{dm_3^2}{dt}}`$ $`=`$ $`{\displaystyle \frac{1}{8\pi ^2}}[4\kappa _2m_8^2+({\displaystyle \frac{1}{2}}\kappa _1+\kappa _2)m_3^2]`$ (9)
where $`t=log(Q)`$. With $`\kappa _2`$ being of order one and $`\kappa _1`$, $`\mu /v<<1`$, we see from these equations that if $`m_3^2`$ at the scale $`v`$ starts off being smaller than $`m_8^2`$, it can turn negative in going down from $`v`$ to the mass scale $`m_8`$. The mass parameter $`m_8^2`$ will remain positive in this case. This example shows that fine tuning of the weak triplet can be done at the scale $`m_8`$ is such a way that its squared-mass turns negative at lower energy scales.
In analogy with this example, if any fragment of the 144-plet of $`SO(10)`$ remains somewhat lighter than the GUT scale, then the quartic couplings involving that fragment and the Higgs doublet would turn the doublet mass-squared negative<sup>4</sup><sup>4</sup>4This statement doas not contradicts Michel-Radicati theorem as we are using radiative correction to turn the doublet mass-squared negative.. It is interesting to note that if such fragments from the 144 that survive below the GUT scale are the color octet(s) and the weak triplet(s), unification of the three SM gauge couplings will occur nicely at a scale of $`(10^{16}10^{17})`$ GeV . The above analysis and conclusions are in the context of a non-SUSY $`SO(10)`$ scenario. In the SUSY $`SO(10)`$ case the top-stop Yukawa couplings will turn the Higgs doublet mass-square negative in the usual way.
### 2.3 Doublet-triplet splitting
We denote the components of the 144 multiplet by $`Q`$’s. As can be seen from Eq.(1) the Higgs fields reside in the multiplets $`Q_i(\overline{5})+Q^i(5)+Q_j^i(24)+Q_{ij}^k(\overline{45})`$. Similarly we denote the components of the $`144^{}`$ multiplet by $`P`$’s and in this case the relevant Higgs multiplets will be $`P_i(\overline{5})+P^i(5)+P_j^i(24)+P_k^{ij}(45)`$. The $`\overline{5},5`$ and $`45`$ representations contain $`SU(2)_L`$ doublets and $`SU(3)_C`$ triplets. Before studying the doublet-triplet splitting in the full $`SO(10)`$ theory, here we analyze this possibility within the simpler $`SU(5)\times U(1)`$ theory, but with couplings motivated by the full $`SO(10)`$ theory. The relevant potential that we consider is given by
$`V_{DT}=m^2tr(Q_iP^i)+m^2tr(Q^iP_i)+m^2tr(Q_{ij}^kP_k^{ij})+\lambda _1tr(Q_iQ^i)tr(Q_j^iQ_i^j)`$
$`+\lambda _2tr(P_iP^i)tr(P_j^iP_i^j)+\lambda _3tr(Q_iP^i)tr(Q_j^iP_i^j)+\lambda _3^{}tr(Q^iP_i)tr(Q_j^iP_i^j)`$
$`+\lambda _4tr(Q_{ij}^kP_k^{ij})tr(Q_j^iP_i^j)+\eta _1tr(Q_iQ^iQ_k^jQ_j^k)+\eta _2tr(P_iP^iP_k^jP_j^k)`$
$`+\eta _3tr(Q_iP^iQ_k^jP_j^k)+\eta _3^{}tr(Q^iP_iQ_k^jP_j^k)+\eta _4tr(Q_{ij}^kP^mQ_k^iP_m^j)+\eta _4^{}tr(P_k^{ij}P_mP_i^kP_j^m)`$
$`+\eta _5tr(Q_{ij}^kQ^mQ_k^iQ_m^j)+\eta _5^{}tr(P^{ij}Q_mP_k^iQ_m^j)+\eta _6tr(P_k^{ij}Q_{lm}^kQ_i^lP_j^m)`$ (10)
where $`\lambda _i`$ and $`\eta _i`$ are dimensionless couplings which represent different types of contractions of indices. It is easily checked that each term in the potential has an overall zero $`U(1)`$ quantum number. There are several Higgs doublets and Higgs triplets and anti-triplets in this model. Thus one set of Higgs doublets and triplets and anti-triplets arise from $`Q_i,Q^i,P_i,P^i`$. Specifically the doublets are $`Q_\alpha ,Q^\alpha ,P_\alpha ,P^\alpha `$, while the triplets (anti-triplets) are $`Q^a,P^a`$($`Q_a,P_a`$). Additional Higgs doublets, triplets and anti-triplets arise from the $`45`$ of $`SU(5)`$ (from the $`\overline{144}`$-plet) and from the $`\overline{45}`$ (from the 144-plet). We discuss the decomposition of these below. The $`45`$ of $`SU(5)`$ embedded in $`\overline{144}`$ has the following $`SU(2)\times SU(3)\times U(1)_Y`$ decomposition
$`45=(2,1)(3)+(1,3)(2)+(3,3)(2)+(1,\overline{3})(8)`$
$`+(2,\overline{3})(7)+(1,\overline{6})(2)+(2,8)(3)`$ (11)
One notices that while there is only one $`SU(2)`$ Higgs doublet ($`\stackrel{~}{P}^\alpha ,\alpha =1,2`$), there is one $`SU(3)_C`$ Higgs triplet $`\stackrel{~}{P}^a`$ ($`a=1,2,3`$) and one anti-triplet $`\stackrel{~}{P}_a`$. Similarly, for the $`\overline{45}`$ embedded in $`144`$ one has one $`SU(2)`$ Higgs doublet ($`\stackrel{~}{Q}_\alpha ,\alpha =1,2`$), one $`SU(3)_C`$ Higgs triplet $`\stackrel{~}{Q}^a`$ (a=1,2,3) and one anti-triplet $`\stackrel{~}{Q}_a`$. The above analysis shows that the Higgs doublet mass matrix will be $`3\times 3`$ while the Higgs triplet mass matrix will be $`4\times 4`$. We focus first on the Higgs doublet mass matrix after $`Q_j^i`$ and $`P_j^i`$ develop VEVs. We display the mass matrix for the Higgs doublets in the basis where the rows are ($`P_\alpha ,Q_\alpha ,h_\alpha `$) and the columns by ($`P^\alpha ,Q^\alpha ,h^\alpha `$)
$`\left[\begin{array}{ccc}30\lambda _2v^2+9\eta _2v^2& m^2+30\lambda _3^{}v^2+9\eta _3^{}v^2& v^2\eta _4^{}c\\ m^2+30\lambda _3v^2+9\eta _3v^2& 30\lambda _1v^2+9\eta _1v^2& v^2\eta _5^{}c\\ v^2\eta _4c& v^2\eta _5c& m^2+30\lambda _4v^2+\frac{21}{4}\eta _6v^2\end{array}\right]`$ (12)
where $`c=15\sqrt{3}/2\sqrt{2}`$. Next, we display the mass matrix for the Higgs triplets in the basis where the rows are labelled ($`P_a,Q_a,\stackrel{~}{Q}_a,\stackrel{~}{P}_a`$) and the columns by ($`P^a,Q^a,\stackrel{~}{P}^a,\stackrel{~}{Q}^a`$). In this basis the Higgs triplet mass matrix is
$`\left[\begin{array}{cccc}30\lambda _2v^2+4\eta _2v^2& m^2+30\lambda _3^{}v^2+4\eta _3^{}v^2& \stackrel{~}{c}\eta _4^{}v^2& 0\\ m^2+30\lambda _3v^2+4\eta _3v^2& 30\lambda _1v^2+4\eta _1v^2& \stackrel{~}{c}\eta _5^{}v^2& 0\\ \stackrel{~}{c}\eta _4v^2& \stackrel{~}{c}\eta _5v^2& m^2+30\lambda _4v^2+\eta _6v^2& 0\\ 0& 0& 0& m^2+30\lambda _4v^2+9\eta _6v^2\end{array}\right]`$ (13)
where $`\stackrel{~}{c}=5\sqrt{2}`$. It is easy to see from the determinants of Eqs.(12) and (13) that one can arrange for a pair of light Higgs doublets while keeping the Higgs triplets heavy. As will be shown in Eqs.(50) and (51) from the full $`SO(10)`$ analysis,we can identify one of the superposition of the doublet fragments as the SM Higgs doublet while keeping the Higgs color triplets fragments all super heavy. It is true that one must fine tune in order to have the Higgs doublet light, but we find it very interesting that with a single 144-plet complete breaking of the $`SO(10)`$ symmetry down to the residual $`SU(3)_C\times U(1)_{em}`$ can be achieved.
### 2.4 Fermion mass growth
For the fermion masses we have the following quartic coupling allowed by gauge invariance, in terms of $`SU(5)\times U(1)`$ decomposition
$`W={\displaystyle \frac{h_1}{M}}10^{ij}10^{kl}\mathrm{\Sigma }_m^jQ^m+{\displaystyle \frac{h_2}{M}}10^{ij}10^{kl}\mathrm{\Sigma }_l^nQ^m+{\displaystyle \frac{h_3}{M}}10^{ij}\overline{5}_i\mathrm{\Sigma }_j^m\overline{P}_m+{\displaystyle \frac{h_4}{M}}10^{ij}\overline{5}_l\mathrm{\Sigma }_i^l\overline{P}_j`$ (14)
From Eq.(14) we see that it is possible to realize Georgi-Jarlskog type relations with appropriate choice of the $`h_i`$ couplings even without the 45 of $`SU(5)`$ acquiring electroweak VEV. In general, there are additional terms involving the 45 of $`SU(5)`$, which would provide additional freedom since the Higgs doublet will now be a linear combination of 5 and 45.
## 3 Calculational Techniques for the full SO(10) analysis
In this section we discuss the breaking of $`SO(10)`$ down to $`SU(3)_C\times U(1)_{em}`$ in a single step by a single pair of $`144+\overline{144}`$. Our analysis will be valid for the supersymmetric $`SO(10)`$ model as well as for the non–supersymmetric model. In the latter case one simply identifies $`\overline{144}`$ with the $`144^{}`$ field. However, for formal reasons we consider a single pair of $`160+\overline{160}`$ where the additional $`16+\overline{16}`$ that reside in $`160+\overline{160}`$ are needed for consistency as we will see below. The analysis involving the $`144+\overline{144}`$ is rather intricate and we use the techniques developed recently in Refs based on the oscillator method of Refs. to compute the desired couplings. There are no cubic couplings of quarks and leptons with the $`144+\overline{144}`$ of Higgs and one needs quartic couplings to grow quark and lepton masses in this scheme. We discuss now the basic ingredients of the model. We begin by displaying the particle content of $`144+\overline{144}`$ in multiplets of $`SU(5)`$. For $`\overline{144}`$ plet one has
$`\overline{144}=\overline{5}(𝒫_i)+5(𝒫^i)+\overline{10}(𝒫_{ij})+\overline{15}(𝒫_{ij}^{(S)})+24(𝒫_j^i)+\overline{40}(𝒫_{ijk}^l)+45(𝒫_k^{ij})`$ (15)
where the subscripts and superscripts $`i,j,k,..`$ are $`SU(5)`$ indices which take on the values $`1,..,5`$. Similarly, for $`144`$ we find
$`144=5(𝒬^i)+\overline{5}(𝒬_i)+10(𝒬^{ij})+15(𝒬_{(S)}^{ij})+24(𝒬_j^i)+40(𝒬_l^{ijk})+\overline{45}(𝒬_{jk}^i)`$ (16)
To make progress we need a field theoretic description of the $`144`$ and $`\overline{144}`$ plet of fields. Possible candidates are the vector-spinors $`|\mathrm{\Psi }_{(\pm )\mu }>`$. However, an unconstrained vector spinor has $`16\times 10=160`$ independent components and is reducible. Thus irreducible vector-spinors with dimensionality 144 must be gotten by removing 16 of the 160 components of the unconstrained vector-spinor. We define the 144-dimensional constrained vector spinors $`|\mathrm{{\rm Y}}_{(\pm )\mu }>`$ by imposition of the constraint
$$\mathrm{\Gamma }_\mu |\mathrm{{\rm Y}}_{(\pm )\mu }>=0$$
(17)
where $`\mathrm{\Gamma }_\mu `$ satisfy a rank-10 Clifford algebra
$$\{\mathrm{\Gamma }_\mu ,\mathrm{\Gamma }_\nu \}=2\delta _{\mu \nu }.$$
(18)
and where $`\mu ,\nu `$ are the $`SO(10)`$ indices and take on the values $`1,..,10`$.
We discuss now further the relationship of $`|\mathrm{\Psi }_{(\pm )\mu }>`$ and $`|\mathrm{{\rm Y}}_{(\pm )\mu }>`$. We begin by writing the 160 and $`\overline{160}`$ component spinor:
$$|\mathrm{\Psi }_{(+)\stackrel{´}{a}\mu }>=|0>𝐏_{\stackrel{´}{a}\mu }+\frac{1}{2}b_i^{}b_j^{}|0>𝐏_{\stackrel{´}{a}\mu }^{ij}+\frac{1}{24}ϵ^{ijklm}b_j^{}b_k^{}b_l^{}b_m^{}|0>𝐏_{\stackrel{´}{a}i\mu }$$
(19)
$$|\mathrm{\Psi }_{()\stackrel{´}{b}\mu }>=b_1^{}b_2^{}b_3^{}b_4^{}b_5^{}|0>𝐐_{\stackrel{´}{b}\mu }+\frac{1}{12}ϵ^{ijklm}b_k^{}b_l^{}b_m^{}|0>𝐐_{\stackrel{´}{b}ij\mu }+b_i^{}|0>𝐐_{\stackrel{´}{b}\mu }^i$$
(20)
where the Latin letters $`i,j,k,l,m,\mathrm{}`$ are $`SU(5)`$ indices and the Greek letters $`\mu ,\nu ,\rho ,\mathrm{}`$ represent $`SO(10)`$ indices. The Latin subscripts $`\stackrel{´}{a},\stackrel{´}{b},\stackrel{´}{c},\stackrel{´}{d}(=1,2,3)`$ are reserved for generation indices. The reducible fields appearing in Eqs.(19) and (20) can be identified in $`SU(5)`$ notation as follows :
$`10=5+\overline{5}:𝐏_\mu =(𝐏_{c_k},𝐏_{\overline{c}_k})(𝐏^k,𝐏_k)`$
$`\overline{10}=5+\overline{5}:𝐐_\mu =(𝐐_{c_k},𝐐_{\overline{c}_k})(𝐐^k,𝐐_k)`$ (21)
$`100=\overline{50}+50:𝐏_\mu ^{ij}=(𝐏_{c_k}^{ij},𝐏_{\overline{c}_k}^{ij})(𝐑^{[ij]k},𝐑_k^{[ij]})`$
$`\overline{100}=\overline{50}+50:𝐐_{\mu ij}=(𝐐_{ijc_k},𝐐_{ij\overline{c}_k})(𝐒_{[ij]}^k,𝐒_{[ij]k})`$
$`50=45+5:𝐑_k^{[ij]}=𝐏_k^{ij}+{\displaystyle \frac{1}{4}}\left(\delta _k^j\widehat{𝐏}^i\delta _k^i\widehat{𝐏}^j\right)`$
$`\overline{50}=\overline{40}+\overline{10}:𝐑^{[ij]k}=ϵ^{ijlmn}𝐏_{lmn}^k+ϵ^{ijklm}\widehat{𝐏}_{lm}`$
$`50=40+10:𝐒_{[ij]k}=ϵ_{ijlmn}𝐐_k^{lmn}+ϵ_{ijklm}\widehat{𝐐}^{lm}`$
$`\overline{50}=\overline{45}+\overline{5}:𝐒_{[jk]}^i=𝐐_{jk}^i+{\displaystyle \frac{1}{4}}\left(\delta _k^i\widehat{𝐐}_j\delta _j^i\widehat{𝐐}_k\right)`$ (22)
$`\overline{50}=25+\overline{25}:𝐏_{i\mu }=(𝐏_{ic_k},𝐏_{i\overline{c}_k})(𝐑_i^k,𝐑_{ik})`$
$`50=25+25:𝐐_\mu ^i=(𝐐_{\overline{c}_k}^i,𝐐_{c_k}^i)(𝐒_k^i,𝐒^{ik})`$
$`25=24+1:𝐑_j^i=𝐏_j^i+{\displaystyle \frac{1}{5}}\delta _j^i\widehat{𝐏},\overline{25}=\overline{10}+\overline{15}:𝐑_{ij}={\displaystyle \frac{1}{2}}\left(𝐏_{ij}+𝐏_{ij}^{(S)}\right)`$
$`25=24+1:𝐒_j^i=𝐐_j^i+{\displaystyle \frac{1}{5}}\delta _j^i\widehat{𝐐},25=10+15:𝐒^{ij}={\displaystyle \frac{1}{2}}\left(𝐐^{ij}+𝐐_{(S)}^{ij}\right)`$ (23)
Further, $`b_i^{}`$ and $`b_i^{}`$ $`(i=1,2,..,5)`$ are the fermionic creation and annihilation operators and obey the anti-commutation rules
$$\{b_i,b_j^{}\}=\delta _i^j;\{b_i,b_j\}=0;\{b_i^{},b_j^{}\}=0$$
(24)
and the $`SU(5)`$ singlet state $`|0>`$ satisfies $`b_i|0>=0`$. $`SO(10)`$ invariance requires the following constraints in general
$$\mathrm{\Gamma }_\mu |\mathrm{\Psi }_{(+)\mu }>=|\mathrm{\Psi }_{()}^{}>$$
(25)
$$\mathrm{\Gamma }_\mu |\mathrm{\Psi }_{()\mu }>=|\mathrm{\Psi }_{(+)}^{}>$$
(26)
where
$`|\mathrm{\Psi }_{()}^{}>=b_1^{}b_2^{}b_3^{}b_4^{}b_5^{}|0>\widehat{𝐏}+{\displaystyle \frac{1}{12}}ϵ^{ijklm}b_k^{}b_l^{}b_m^{}|0>\left(𝐏_{ij}+6\widehat{𝐏}_{ij}\right)`$
$`+b_i^{}|0>\left(𝐏^i+\widehat{𝐏}^i\right)`$ (27)
$`|\mathrm{\Psi }_{(+)}^{}>=|0>\widehat{𝐏}+{\displaystyle \frac{1}{2}}b_i^{}b_j^{}|0>\left(𝐐^{ij}+6\widehat{𝐐}^{ij}\right)`$
$`+{\displaystyle \frac{1}{24}}ϵ^{ijklm}b_j^{}b_k^{}b_l^{}b_m^{}|0>\left(𝐐_i+\widehat{𝐐}_i\right)`$ (28)
We note in passing that to get the $`144`$ and $`\overline{144}`$ spinors, $`|\mathrm{{\rm Y}}_{(\pm )\mu }>`$, we need to impose Eq.(17). This constrain will require setting $`|\mathrm{\Psi }_{(\pm )}^{}>=0`$ and thus require the following constraints
$`\widehat{𝐏}=0,\widehat{𝐏}^i=𝐏^i,\widehat{𝐏}_{ij}={\displaystyle \frac{1}{6}}𝐏_{ij}`$
$`\widehat{𝐐}=0,\widehat{𝐐}_i=𝐐_i,\widehat{𝐐}^{ij}={\displaystyle \frac{1}{6}}𝐐^{ij}`$ (29)
Thus we have
$$|\mathrm{{\rm Y}}_{(\pm )\mu }>=\left(|\mathrm{\Psi }_{(\pm )\mu }>\right)_{constraintofEq.(\text{3})}$$
(30)
However, as we stated already we will be dealing with the full $`160+\overline{160}`$ multiplets.
To normalize the $`SU(5)`$ fields contained in the tensor, $`|\mathrm{\Psi }_{(\pm )\mu }>`$, we carry out a field redefinition
$`\{1\}:\widehat{𝐏}=\sqrt{5}\widehat{𝒫},\{\overline{5}\}:𝐏_i=𝒫_i,\{5\}:𝐏^i=𝒫^i,\widehat{𝐏}^i=2\widehat{𝒫}^i`$
$`\{\overline{10}\}:𝐏_{ij}=\sqrt{2}𝒫_{ij},\widehat{𝐏}_{ij}={\displaystyle \frac{1}{2\sqrt{3}}}\widehat{𝒫}_{ij},\{\overline{15}\}:𝐏_{ij}^{(S)}=\sqrt{2}𝒫_{ij}^{(S)}`$
$`\{24\}:𝐏_j^i=𝒫_j^i,\{\overline{40}\}:𝐏_{ijk}^l={\displaystyle \frac{1}{6}}𝒫_{ijk}^l,\{45\}:𝐏_k^{ij}=𝒫_k^{ij}`$ (31)
$`\{1\}:\widehat{𝐐}=\sqrt{5}\widehat{𝒬},\{5\}:𝐐^i=𝒬^i,\{\overline{5}\}:𝐐_i=𝒬_i,\widehat{𝐐}_i=2\widehat{𝒬}_i`$
$`\{10\}:𝐐^{ij}=\sqrt{2}𝒬_{ij},\widehat{𝐐}^{ij}={\displaystyle \frac{1}{2\sqrt{3}}}\widehat{𝒬}^{ij},\{15\}:𝐐_{(S)}^{ij}=\sqrt{2}𝒬_{(S)}^{ij}`$
$`\{24\}:𝐐_j^i=𝒬_j^i,\{40\}:𝐐_l^{ijk}={\displaystyle \frac{1}{6}}𝒬_l^{ijk},\{\overline{45}\}:𝐐_{ij}^k=𝒬_{ij}^k`$ (32)
In terms of the normalized fields, the kinetic energy of the $`160`$ and $`\overline{160}`$, i.e., $`<_A\mathrm{\Psi }_{(\pm )\mu }|^A\mathrm{\Psi }_{(\pm )\mu }>`$, where $`A`$ is the Lorentz index, takes the form
$`𝖫_{kin}^{\overline{160}}=_A\widehat{𝒫}^{}_A\widehat{𝒫}_A𝒫_i^{}^A𝒫_i_A𝒫^i^A𝒫^i_A\widehat{𝒫}^i^A\widehat{𝒫}^i`$
$`{\displaystyle \frac{1}{2!}}_A𝒫_{ij}^{}^A𝒫_{ij}{\displaystyle \frac{1}{2!}}_A\widehat{𝒫}_{ij}^{}^A\widehat{𝒫}_{ij}{\displaystyle \frac{1}{2!}}_A𝒫_{ij}^{(S)}^A𝒫_{ij}^{(S)}`$
$`_A𝒫_j^i^A𝒫_j^i{\displaystyle \frac{1}{3!}}_A𝒫_{ijk}^l^A𝒫_{ijk}^l{\displaystyle \frac{1}{2!}}_A𝒫_k^{ij}^A𝒫_k^{ij}`$ (33)
$`𝖫_{kin}^{160}=_A\widehat{𝒬}^{}_A\widehat{𝒬}_A𝒬^i^A𝒬^i_A𝒬_i^{}^A𝒬_i_A\widehat{𝒬}_i^{}^A\widehat{𝒬}_i`$
$`{\displaystyle \frac{1}{2!}}_A𝒬^{ij}^A𝒬^{ij}{\displaystyle \frac{1}{2!}}_A\widehat{𝒬}^{ij}^A\widehat{𝒬}^{ij}{\displaystyle \frac{1}{2!}}_A𝒬_{(S)}^{ij}^A𝒬_{(S)}^{ij}`$
$`_A𝒬_j^i^A𝒬_j^i{\displaystyle \frac{1}{3!}}_A𝒬_l^{ijk}^A𝒬_l^{ijk}{\displaystyle \frac{1}{2!}}_A𝒬_{ij}^k^A𝒬_{ij}^k`$ (34)
## 4 Symmetry Breaking
In this section we discuss how $`SO(10)`$ breaks to $`SU(3)_C\times SU(2)_L\times U(1)_Y`$. For this purpose we consider a superpotential of the form
$`𝖶=M(\overline{160}_H\times 160_H)`$
$`+{\displaystyle \frac{\lambda _1}{M^{}}}(\overline{160}_H\times 160_H)_1(\overline{160}_H\times 160_H)_1`$
$`+{\displaystyle \frac{\lambda _{45}}{M^{}}}(\overline{160}_H\times 160_H)_{45}(\overline{160}_H\times 160_H)_{45}`$
$`+{\displaystyle \frac{\lambda _{210}}{M^{}}}(\overline{160}_H\times 160_H)_{210}(\overline{160}_H\times 160_H)_{210}`$ (35)
There are of course many more terms that one can add to Eq.(4) but we consider only the terms displayed in Eq.(4) for simplicity. The relevant terms in the superpotential that accomplish symmetry breaking are
$`𝖶_{_{SB}}=M𝐐_\mu ^i𝐏_{i\mu }+\alpha __1𝐐_\mu ^i𝐏_{i\mu }𝐐_\nu ^j𝐏_{j\nu }+\alpha __2𝐐_\mu ^i𝐏_{j\mu }𝐐_\nu ^j𝐏_{j\nu }`$ (36)
where
$`\alpha __1={\displaystyle \frac{1}{M^{}}}\left(2\lambda _1\lambda _{45}+{\displaystyle \frac{1}{6}}\lambda _{210}\right)`$
$`\alpha __2={\displaystyle \frac{1}{M^{}}}\left(4\lambda _{45}+\lambda _{210}\right)`$ (37)
Expanding into the irreducible components we find
$`W=M𝒬_j^i𝒫_i^j+\alpha __1𝒬_j^i𝒫_i^j𝒬_l^k𝒫_k^l+\alpha __2𝒬_k^i𝒫_j^k𝒬_l^j𝒫_i^l+M\widehat{𝒬}\widehat{𝒫}+2\left(\alpha __1+{\displaystyle \frac{2}{5}}\alpha __2\right)𝒬_l^k𝒫_k^l\widehat{𝒬}\widehat{𝒫}`$
$`+{\displaystyle \frac{2}{\sqrt{5}}}\alpha __2𝒬_k^i𝒫_j^k𝒬_i^j\widehat{𝒫}+{\displaystyle \frac{2}{\sqrt{5}}}\alpha __1𝒬_k^i𝒫_j^k𝒫_i^j\widehat{𝒬}+{\displaystyle \frac{1}{5}}\alpha __2𝒬_l^k𝒬_k^l\widehat{𝒫}\widehat{𝒫}+{\displaystyle \frac{1}{5}}\alpha __1𝒫_l^k𝒫_k^l\widehat{𝒬}\widehat{𝒬}`$
$`+\left(\alpha __1+{\displaystyle \frac{1}{5}}\alpha __2\right)\widehat{𝒬}\widehat{𝒫}\widehat{𝒬}\widehat{𝒫}`$ (38)
In the minimization we look for solutions of the type
$`<𝒬_j^i>=qdiag(2,2,2,3,3),<𝒫_j^i>=pdiag(2,2,2,3,3)`$ (39)
One finds the following results from the minimization of the potential
$`Mp+2p^2q\left(30\alpha __1+7\alpha __2\right)+2\left(\alpha __1+{\displaystyle \frac{2}{5}}\alpha __2\right)pQ__0P__0`$
$`+{\displaystyle \frac{1}{75}}\alpha __2qP__0^2++{\displaystyle \frac{26}{5\sqrt{5}}}\alpha __2(pQ__0+2qP__0)p=0`$ (40)
$`Mq+2q^2p\left(30\alpha __1+7\alpha __2\right)+2\left(\alpha __1+{\displaystyle \frac{2}{5}}\alpha __2\right)qQ__0P__0`$
$`+{\displaystyle \frac{1}{75}}\alpha __2pQ__0^2++{\displaystyle \frac{26}{5\sqrt{5}}}\alpha __2(qP__0+2pQ__0)q=0`$ (41)
$`60\left[\left(\alpha __1+{\displaystyle \frac{2}{5}}\alpha __2\right)qp+M\right]P__0+{\displaystyle \frac{2}{5}}\alpha __2p^2Q__0+{\displaystyle \frac{156}{\sqrt{5}}}\alpha __2p^2q`$
$`+2\left(\alpha __1+{\displaystyle \frac{1}{5}}\alpha __2\right)Q__0P__0^2=0`$ (42)
and
$`60\left[\left(\alpha __1+{\displaystyle \frac{2}{5}}\alpha __2\right)qp+M\right]Q__0+{\displaystyle \frac{2}{5}}\alpha __2q^2P__0+{\displaystyle \frac{156}{\sqrt{5}}}\alpha __2pq^2`$
$`+2\left(\alpha __1+{\displaystyle \frac{1}{5}}\alpha __2\right)Q__0^2P__0=0`$ (43)
where $`Q__0=<\widehat{𝒬}>`$ and $`P__0=<\widehat{𝒫}>`$. The D-flatness condition $`<144>=<\overline{144}>`$ gives $`q=p`$. With the above vacuum expectation value (VEV), spontaneous breaking occurs so that $`SO(10)SU(3)_C\times SU(2)_L\times U(1)_Y`$. We note that the VEVs $`Q_0`$ and $`P_0`$ do not play a role in the above breakdown as this breakdown will occur even when $`Q_0=0=P_0`$.
## 5 Higgs Phenomenon and Mass Growth
We outline here the Higgs phenomenon and the mass growth associated with the spontaneous breaking given by Eqs.(4-4). The fields that participate in the Higgs phenomenon include the 45 vector super multiplet that belongs to the adjoint representation of $`SO(10)`$ and the $`144+\overline{144}`$ chiral superfields. For the analysis of the Higgs phenomenon it is useful to decompose the 45-plet of $`SO(10)`$ in multiplets of $`SU(5)`$ so that
$`45=1+10+\overline{10}+24`$ (44)
After spontaneous symmetry breaking the $`10_{45}`$ massless vector super multiplet absorbs the $`10_{144}`$ chiral multiplet to become a $`10_{45}`$ massive vector super multiplet with spins $`(1,\frac{1}{2},0)`$. Similarly, the $`\overline{10}_{45}`$ vector super multiplet absorbs the $`\overline{10}_{\overline{144}}`$ chiral multiplet to become the $`\overline{10}_{45}`$ massive vector super multiplet. Now the 24 plet of $`SU(5)`$ decomposes under $`SU(3)_C\times SU(2)_L`$ as follows
$`24=(8,1)+(\overline{3},2)+(3,2)+(1,3)+(1,1)`$ (45)
After spontaneous breaking the super vector multiplets with the quantum numbers $`(\overline{3},2)+(3,2)`$ absorb one linear combination of the chiral multiplets $`((\overline{3},2)+(3,2))_{144}`$ and $`((\overline{3},2)+(3,2))_{\overline{144}}`$ becoming a massive $`(\overline{3},2)+(3,2)`$ vector super multiplets while the orthogonal linear combination of $`((\overline{3},2)+(3,2))_{144}`$ and $`((\overline{3},2)+(3,2))_{\overline{144}}`$ which is not absorbed becomes massive. The vector super multiplets corresponding to $`(8,1)+(1,3)+(1,1)`$ remain massless. The chiral super multiplets corresponding to $`(8,1)+(1,3)`$ become massive. (The $`(1,1)`$ components of $`24_{144}`$ and $`24_{\overline{144}}`$ require special treatment and we return to it below). Thus we have accounted for the mass growth of the $`10+\overline{10}`$ vector super multiplet and the mass growth of the 12 components $`(\overline{3},2)+(3,2)`$ of the 24 plet vector super multiplet. This leaves us to discuss mass growth of the singlet vector super multiplet in Eq.(44). This mass growth comes about by absorption of the chiral superfield combination $`(\frac{2}{5}\mathrm{\Sigma }_a^a\frac{3}{5}\mathrm{\Sigma }_\alpha ^\alpha )`$ where the repeated indices are summed ($`a`$ is the color index which takes on values 1,2,3 and $`\alpha `$ is the $`SU(2)`$ index and takes on values 4,5), and $`\mathrm{\Sigma }_j^i`$ is a linear combination of $`𝒬_j^i`$ and $`𝒫_j^i`$. Since $`\mathrm{\Sigma }`$ is traceless the above equals $`\mathrm{\Sigma }_a^a=\mathrm{\Sigma }_\alpha ^\alpha `$. Thus the singlet vector super multiplet absorbs a linear combination of $`𝒫_a^a`$ and $`𝒬_a^a`$ becoming a singlet massive vector super multiplet while the orthogonal combination of the chiral superfields $`𝒫_a^a`$ and $`𝒬_a^a`$ becomes massive. Thus after the symmetry breaking and Higgs phenomenon only the $`(1,8)+(1,3)+(1,1)`$ vector supermultiplet remains massless and the remaining components of $`45`$ of the vector super multiplet become massive. Similarly, all the unabsorbed components of the $`144+\overline{144}`$ become massive. At this stage the gauge group $`SO(10)`$ has broken down to $`SU(3)_C\times SU(2)_L\times U(1)_Y`$. To accomplish the breaking of the electro-weak symmetry we need a pair of Higgs doublets. Such a possibility arises for $`𝒬_\alpha `$, $`𝒫^\alpha `$.
To exhibit this we need to compute masses for $`𝒬_i`$, $`𝒫^i`$. It is also instructive to compute masses for $`𝒬^i`$, $`𝒫_i`$, $`\widehat{𝒬}_i`$, $`\widehat{𝒫}^i`$, $`𝒬_{jk}^i`$ and $`𝒫_k^{ij}`$ The relevant terms in the superpotential are
$`𝖶_{_{mass}}=\left[M+{\displaystyle \frac{1}{M^{}}}\left(4\lambda _1+6\lambda _{45}{\displaystyle \frac{1}{3}}\lambda _{210}\right)<𝐒_n^m𝐑_m^n>\right]\left(𝒬_i𝒫^i+𝒬^i𝒫_i\right)`$
$`\left[{\displaystyle \frac{8}{3}}{\displaystyle \frac{\lambda _{210}}{M^{}}}<𝐒_m^i𝐑_j^m>\right]𝒬_i𝒫^j+\left[{\displaystyle \frac{1}{M^{}}}\left(8\lambda _{45}{\displaystyle \frac{2}{3}}\lambda _{210}\right)<𝐒_m^i𝐑_j^m>\right]𝐒_{[in]}^l𝐑_l^{[nj]}`$
$`+\left[{\displaystyle \frac{1}{2}}M+{\displaystyle \frac{1}{M^{}}}\left(2\lambda _1+\lambda _{45}{\displaystyle \frac{1}{6}}\lambda _{210}\right)<𝐒_n^m𝐑_m^n>\right]𝐒_{[ij]}^k𝐑_k^{[ij]}`$ (46)
Eq.(5) makes it apparent why for technical reasons we need to keep the $`160+\overline{160}`$ multiplet. To see this let us set all the couplings $`\lambda `$ to zero in Eq.(5) so that the only terms surviving are proportional to $`M`$. Next suppose we impose on $`𝐒_{[ij]}^k`$ and $`𝐑_k^{[ij]}`$ the constraint of Eq.(3) so that we are strictly considering only the $`144+\overline{144}`$ multiplet. Then we see that the last line of Eq.(5) contributes an additional mass term $`\frac{M}{4}𝒬_i𝒫^i`$ while there is no such term for $`𝒬^i𝒫_i`$. This additional term is clearly not desired and the reason for its appearance is that we are identifying the $`5`$ plet in $`𝐑_k^{[ij]}`$ with $`𝒫^i`$ and $`\overline{5}`$ plet in $`𝐒_{[ij]}^k`$ with $`𝒬_i`$ because of Eq.(3) which feeds in the undesired additional term. Thus for book keeping we must not impose the constraint of Eq.(3) in the beginning. Returning to Eq.(5), the corresponding mass terms in the Lagrangian are given by
$`𝖫=\left|{\displaystyle \frac{𝖶_{_{mass}}}{𝒬_i}}\right|^2+\left|{\displaystyle \frac{𝖶_{_{mass}}}{𝒫^i}}\right|^2+\left|{\displaystyle \frac{𝖶_{_{mass}}}{𝒬^i}}\right|^2+\left|{\displaystyle \frac{𝖶_{_{mass}}}{𝒫_i}}\right|^2+\left|{\displaystyle \frac{𝖶_{_{mass}}}{𝐒_{[ij]}^k}}\right|^2+\left|{\displaystyle \frac{𝖶_{_{mass}}}{𝐑_k^{[ij]}}}\right|^2`$ (47)
Explicitly we have
$`𝖫=\left|\sigma \right|^2\left(𝒫_i𝒫_i^{}+𝒬^i𝒬^i\right)`$
$`+\left|\sigma +\omega __1\beta \right|^2\left(𝒫^\alpha 𝒫^\alpha +𝒬_\alpha 𝒬_\alpha ^{}\right)`$
$`+\left|\sigma +\omega __2\beta \right|^2\left(𝒫^a𝒫^a+𝒬_a𝒬_a^{}\right)`$
$`+{\displaystyle \frac{1}{2}}\left[\left|\rho \omega __1\gamma \right|^2+3\left|\rho {\displaystyle \frac{1}{2}}\left(\omega __1+\omega __2\right)\gamma \right|^2\right]\left(\widehat{𝒫}^\alpha \widehat{𝒫}^\alpha +\widehat{𝒬}_\alpha \widehat{𝒬}_\alpha ^{}\right)`$
$`+\left[\left|\rho \omega __1\gamma \right|^2+\left|\rho {\displaystyle \frac{1}{2}}\left(\omega __1+\omega __2\right)\gamma \right|^2\right]\left(\widehat{𝒫}^a\widehat{𝒫}^a+\widehat{𝒬}_a\widehat{𝒬}_a^{}\right)`$
$`+\left|\rho \omega __1\gamma \right|^2\left(𝒫_k^{\alpha \beta }𝒫_k^{\alpha \beta }+𝒬_{\alpha \beta }^k𝒬_{\alpha \beta }^k\right)`$
$`+\left|\rho \omega __2\gamma \right|^2\left(𝒫_k^{ab}𝒫_k^{ab}+𝒬_{ab}^k𝒬_{ab}^k\right)`$
$`+2\left|\rho {\displaystyle \frac{1}{2}}\left(\omega __1+\omega __2\right)\gamma \right|^2\left(𝒫_k^{a\alpha }𝒫_k^{a\alpha }+𝒬_{a\alpha }^k𝒬_{a\alpha }^k\right)`$
$`+\left[\right|\rho {\displaystyle \frac{1}{2}}(\omega __1+\omega __2)\gamma |^2|\rho \omega __1\gamma |^2](𝒫_\alpha ^{a\alpha }\widehat{𝒫}^a+𝒬_{a\alpha }^\alpha \widehat{𝒬}_a^{}+H.C.)`$
$`+\left[\right|\rho {\displaystyle \frac{1}{2}}(\omega __1+\omega __2)\gamma |^2|\rho \omega __2\gamma |^2](𝒫_a^{\alpha a}\widehat{𝒫}^\alpha +𝒬_{\alpha a}^a\widehat{𝒬}_\alpha ^{}+H.C.)`$ (48)
where
$`\sigma =M+{\displaystyle \frac{qp}{M^{}}}\left(4\lambda _1+6\lambda _{45}{\displaystyle \frac{1}{3}}\lambda _{210}\right)\left(30+{\displaystyle \frac{P__0Q__0}{pq}}\right)`$
$`\rho ={\displaystyle \frac{1}{2}}M+{\displaystyle \frac{qp}{M^{}}}\left(2\lambda _1+\lambda _{45}{\displaystyle \frac{1}{6}}\lambda _{210}\right)\left(30+{\displaystyle \frac{P__0Q__0}{pq}}\right)`$
$`\omega __1=qp\left[9{\displaystyle \frac{3}{\sqrt{5}}}\left({\displaystyle \frac{P__0}{p}}+{\displaystyle \frac{Q__0}{q}}\right)+{\displaystyle \frac{1}{5}}{\displaystyle \frac{P__0Q__0}{pq}}\right]`$
$`\omega __2=qp\left[4+{\displaystyle \frac{2}{\sqrt{5}}}\left({\displaystyle \frac{P__0}{p}}+{\displaystyle \frac{Q__0}{q}}\right)+{\displaystyle \frac{1}{5}}{\displaystyle \frac{P__0Q__0}{pq}}\right]`$
$`\gamma ={\displaystyle \frac{1}{M^{}}}\left(8\lambda _{45}{\displaystyle \frac{2}{3}}\lambda _{210}\right)`$
$`\beta ={\displaystyle \frac{8}{3}}{\displaystyle \frac{\lambda _{210}}{M^{}}}`$ (49)
The masses for the fields $`𝒫^i,𝒬_i`$, $`𝒬^i,𝒫_i`$, $`𝒬_{ij}^k`$, and $`𝒫_k^{ij}`$ can be computed from Eqs.(4-4) and (5). It is easily checked that the masses vanish unless one has a non-vanishing $`\lambda _{45}`$ or $`\lambda _{210}`$. A scrutiny of the mass growth above shows that the masses of the Higgs doublets $`𝒬_\alpha ,𝒫^\alpha `$ are split from the Higgs triplets $`𝒬_a,𝒫^a`$. Indeed this allows one to fine tune the masses of the Higgs doublets to zero by the condition
$`{\displaystyle \frac{MM^{}}{qp}}+\left(120\lambda _1+180\lambda _{45}\right)\left(1+{\displaystyle \frac{1}{30}}{\displaystyle \frac{P__0Q__0}{pq}}\right)`$
$`+\lambda _{210}\left[34+{\displaystyle \frac{8}{\sqrt{5}}}\left({\displaystyle \frac{P__0}{p}}+{\displaystyle \frac{Q__0}{q}}\right){\displaystyle \frac{13}{15}}{\displaystyle \frac{P__0Q__0}{pq}}\right]=0`$ (50)
With this constraint one finds that the Higgs doublets $`𝒬_\alpha ,𝒫^\alpha `$ are massless while the Higgs triplets $`𝒬_a,𝒫^a`$ are massive. Thus the Higgs triplet mass $`M_{H_3}`$ of the fields $`𝒬_a,𝒫^a`$, under the constraint that the Higgs doublet $`𝒬_\alpha ,𝒫^\alpha `$ be massless, is given by
$$M_{H_3}=\frac{40}{3}\frac{qp}{M^{}}\left[1\frac{1}{\sqrt{5}}\left(\frac{P__0}{p}+\frac{Q__0}{q}\right)\right]\lambda _{210}$$
(51)
We note that it is not possible to achieve a doublet-triplet splitting for the multiplets $`𝒬^i`$ and $`𝒫_i`$. Thus the doublet-triplet splitting we are considering is unique. Further, we find that $`𝒬^i`$, $`𝒫_i`$ develop a mass
$`M_{𝒬^i,𝒫_i}=24{\displaystyle \frac{qp}{M^{}}}\left[1{\displaystyle \frac{1}{3\sqrt{5}}}\left({\displaystyle \frac{P__0}{p}}+{\displaystyle \frac{Q__0}{q}}\right)+{\displaystyle \frac{1}{45}}{\displaystyle \frac{P__0Q__0}{pq}}\right]\lambda _{210}`$ (52)
In the above $`Q_0`$ and $`P_0`$ are crucial in getting a pair of light Higgs doublets. Thus suppose we have $`Q_0=0=P_0`$ in Eqs.(4-4). In this case one finds an additional constraint. Thus from Eqs.(4) and (4) one finds that $`Q_0=0=P_0`$ imply that $`\alpha _2=0`$. Under this constraint Eq.(50) is not consistent with Eqs.(4) and (4). i.e., one cannot find a pair of light Higgs doublets. We note that this result is a consequence of considering only a limited number of couplings in Eq.(4). Inclusion of a larger set of couplings should allow one to get consistent solutions without inclusion of the singlet VEVs.
To summarize the results thus far we have here a complete breaking of $`SO(10)`$ to $`SU(3)_C\times SU(2)_L\times U(1)_Y`$. To get a pair of light Higgs we need to invoke a string landscape scenario which has been extensively discussed recently. Effectively in this framework we use a fine tuning to keep one pair of Higgs doublets light while keeping the Higgs triplets heavy. With the usual radiative electroweak symmetry breaking mechanism, the light doublets of Higgs can develop VEVs breaking the $`SU(2)_L\times U(1)_Y`$ symmetry down to $`U(1)_{em}`$. Thus with the above mechanism one can break $`SO(10)`$ down to $`SU(3)_C\times U(1)_{em}`$ with just one pair of $`144+\overline{144}`$ of Higgs. A similar analysis shows that one gets masses for the remaining parts of the $`144+\overline{144}`$ chiral multiplets not absorbed by the vector bosons which become heavy. The heavy spectrum of this model differs significantly from the standard $`SU(5)`$ and the standard $`SO(10)`$ models. It consists of several parts: (i) the super heavy lepto-quarks associated with the breaking of the $`SO(10)SU(3)_C\times SU(2)_L\times U(1)_Y`$, (ii) the super heavy Higgs triplet field, (iii) the components of 24 plet fields $`𝒫_j^i,𝒬_j^i`$ which are unabsorbed by the Higgs phenomenon and become super heavy, (iv) the remaining components of $`144+\overline{144}`$, aside from a 24 plets of $`SU(5)`$ each in $`144`$ and $`\overline{144}`$ discussed in (iii) and excluding the Higgs doublets $`𝒬_\alpha ,𝒫^\alpha `$, which become super heavy, and (v) $`16+\overline{16}`$ plet of super heavy fields. Gauge coupling unification will require a careful analysis of contribution of each of the above. We do not address this question further here.
We discuss briefly the issue of split vs non-split supersymmetry breaking. Eq.(50) is a tree level relation and its imposition to achive light Higgs doublets presumes that the loop corrections to the Higgs masses are small. Specifically it requires that the scale of supersymmetry breaking is not high, e.g., the masses of of squarks are at the electroweak scale and not at the GUT scale, An alternative possibility is that one may impose Eq.(50) but with loop correction included. In this case we can allow for split supersymmetry scenario where the masses of the squarks are superheavy while gaugino masses may lie at the electroweak scale. Thus the model we are considering can accommodate a split or a non-split supersymmetry breaking scenario.
## 6 Couplings of Quarks and Leptons with 144 and $`\overline{\mathrm{𝟏𝟒𝟒}}`$ of Higgs
The $`144`$ and $`\overline{144}`$ plets of Higgs have no $`SO(10)`$ invariant trilinear coupling with the 16 plet of matter. However one can write quartic couplings of the type $`(1/M_P)(16\times 16)(144\times 144)`$ and $`(1/M_P)(16\times 16)(\overline{144}\times \overline{144})`$, where $`M_P`$ is a super heavy mass. These interactions will generate effective cubic couplings with the light Higgs doublets $`𝒬_\alpha ,𝒫^\alpha `$ after spontaneous breaking of $`SO(10)`$ discussed in Secs.3 and 4. The size of these couplings is $`O(M_G/M_P)`$. The most general set of quartic couplings involving quarks and leptons are<sup>5</sup><sup>5</sup>5We have not included the $`(16\times 16)_{120}(144\times 144)_{120}`$ and $`(16\times 16)_{120}(\overline{144}\times \overline{144})_{120}`$ couplings here since these couplings are anti-symmetric in the generation indices and vanish with only a single $`144+\overline{144}`$.
$`(16\times 16)_{10}(144\times 144)_{10},(16\times 16)_{10}(\overline{144}\times \overline{144})_{10},`$
$`(16\times 16)_{\overline{126}}(144\times 144)_{126},(16\times 16)_{\overline{126}}(\overline{144}\times \overline{144})_{126}`$ (53)
The couplings $`\left(16\times 16\right)_{10}\left(144\times 144\right)_{10}`$ and $`\left(16\times 16\right)_{10}\left(\overline{144}\times \overline{144}\right)_{10}`$ arise from the following structures
$`𝖶^{(10)}={\displaystyle \frac{1}{2}}\mathrm{\Phi }_{\nu 𝒰}_{𝒰𝒰^{}}^{^{(10)}}\mathrm{\Phi }_{\nu 𝒰^{}}+h_{\stackrel{´}{a}\stackrel{´}{b}}^{^{(10)}}<\mathrm{{\rm Y}}_{(+)\stackrel{´}{a}\mu }^{}|B\mathrm{\Gamma }_\nu |\mathrm{{\rm Y}}_{(+)\stackrel{´}{b}\mu }>k__𝒰^{^{(10)}}\mathrm{\Phi }_{\nu 𝒰}`$
$`+f_{\stackrel{´}{a}\stackrel{´}{b}}^{^{(10)}}<\mathrm{\Psi }_{(+)\stackrel{´}{a}}^{}|B\mathrm{\Gamma }_\nu |\mathrm{\Psi }_{(+)\stackrel{´}{b}}>l__𝒰^{^{(10)}}\mathrm{\Phi }_{\nu 𝒰}++\overline{h}_{\stackrel{´}{a}\stackrel{´}{b}}^{^{(10)}}<\mathrm{{\rm Y}}_{()\stackrel{´}{a}\mu }^{}|B\mathrm{\Gamma }_\nu |\mathrm{{\rm Y}}_{()\stackrel{´}{b}\mu }>\overline{k}__𝒰^{^{(10)}}\mathrm{\Phi }_{\nu 𝒰}`$ (54)
where the indices $`𝒰,𝒰^{}`$ run over several Higgs representations of the same kind, $`^{^{(10)}}`$ represents the mass matrix and $`f^{^{(10)}}`$, $`k^{^{(10)}}`$, $`\overline{k}^{^{(10)}}`$, and $`l^{^{(10)}}`$ are constants. $`B`$ is the usual $`SO(10)`$ charge conjugation operator defined by $`B=_{\mu =odd}\mathrm{\Gamma }_\mu `$. The semi-spinors $`\mathrm{\Psi }_{(\pm )}`$ transforms as a 16($`\overline{16}`$)-dimensional irreducible representation of $`SO(10)`$ and contain $`1+\overline{5}+10`$($`1+5+\overline{10}`$) in its $`SU(5)`$ decomposition. $`|\mathrm{\Psi }_{(+)\stackrel{´}{a}}>`$ is given by
$$|\mathrm{\Psi }_{(+)\stackrel{´}{a}}>=|0>𝐌_{\stackrel{´}{a}}+\frac{1}{2}b_i^{}b_j^{}|0>𝐌_{\stackrel{´}{a}}^{ij}+\frac{1}{24}ϵ^{ijklm}b_j^{}b_k^{}b_l^{}b_m^{}|0>𝐌_{\stackrel{´}{a}i}$$
(55)
Elimination of the super heavy fields $`\mathrm{\Phi }_{\nu 𝒰}`$ using the F-flatness condition gives
$`𝖶_{dim5}^{(10)}=𝖶^{^{^{(16\times 16)_{10}(144\times 144)_{10}}}}+𝖶^{^{^{(16\times 16)_{10}(\overline{144}\times \overline{144})_{10}}}}`$ (56)
where
$`𝖶^{^{^{(16\times 16)_{10}(\overline{144}\times \overline{144})_{10}}}}=2\xi _{\stackrel{´}{a}\stackrel{´}{b},\stackrel{´}{c}\stackrel{´}{d}}^{^{(10)}}<\mathrm{\Psi }_{(+)\stackrel{´}{a}}^{}|B\mathrm{\Gamma }_\rho |\mathrm{\Psi }_{(+)\stackrel{´}{b}}><\mathrm{{\rm Y}}_{(+)\stackrel{´}{c}\nu }^{}|\mathrm{\Gamma }_\rho |\mathrm{{\rm Y}}_{(+)\stackrel{´}{d}\nu }>`$
$`=4\xi _{\stackrel{´}{a}\stackrel{´}{b},\stackrel{´}{c}\stackrel{´}{d}}^{^{(10)}}[<\mathrm{\Psi }_{(+)\stackrel{´}{a}}^{}|Bb_i|\mathrm{\Psi }_{(+)\stackrel{´}{b}}><\mathrm{{\rm Y}}_{(+)\stackrel{´}{c}\nu }^{}|Bb_i^{}|\mathrm{{\rm Y}}_{(+)\stackrel{´}{d}\nu }>`$
$`+<\mathrm{\Psi }_{(+)\stackrel{´}{a}}^{}|Bb_i^{}|\mathrm{\Psi }_{(+)\stackrel{´}{b}}><\mathrm{{\rm Y}}_{(+)\stackrel{´}{c}\nu }^{}|Bb_i|\mathrm{{\rm Y}}_{(+)\stackrel{´}{d}\nu }>]`$
$`=2\xi _{\stackrel{´}{a}\stackrel{´}{b},\stackrel{´}{c}\stackrel{´}{d}}^{^{(10)(+)}}[ϵ_{jklmn}𝐌_{\stackrel{´}{a}i}^𝐓𝐌_{\stackrel{´}{b}}^{ij}𝐏_{\stackrel{´}{c}\mu }^{kl𝐓}𝐏_{\stackrel{´}{d}\mu }^{mn}8𝐌_{\stackrel{´}{a}i}^𝐓𝐌_{\stackrel{´}{b}}^{ij}𝐏_{\stackrel{´}{c}j\mu }^𝐓𝐏_{\stackrel{´}{d}\mu }`$
$`+ϵ_{jklmn}𝐌_{\stackrel{´}{a}}^{kl𝐓}𝐌_{\stackrel{´}{b}}^{mn}𝐏_{\stackrel{´}{c}i\mu }^𝐓𝐏_{\stackrel{´}{d}\mu }^{ij}8𝐌_{\stackrel{´}{a}j}^𝐓𝐌_{\stackrel{´}{b}}𝐏_{\stackrel{´}{c}i\mu }^𝐓𝐏_{\stackrel{´}{d}\mu }^{ij}]`$ (57)
and
$`𝖶^{^{^{(16\times 16)_{10}(144\times 144)_{10}}}}=2\zeta _{\stackrel{´}{a}\stackrel{´}{b},\stackrel{´}{c}\stackrel{´}{d}}^{^{(10)}}<\mathrm{\Psi }_{(+)\stackrel{´}{a}}^{}|B\mathrm{\Gamma }_\rho |\mathrm{\Psi }_{(+)\stackrel{´}{b}}><\mathrm{{\rm Y}}_{()\stackrel{´}{c}\nu }^{}|B\mathrm{\Gamma }_\rho |\mathrm{{\rm Y}}_{()\stackrel{´}{d}\nu }>`$
$`=4\zeta _{\stackrel{´}{a}\stackrel{´}{b},\stackrel{´}{c}\stackrel{´}{d}}^{^{(10)}}[<\mathrm{\Psi }_{(+)\stackrel{´}{a}}^{}|Bb_i|\mathrm{\Psi }_{(+)\stackrel{´}{b}}><\mathrm{{\rm Y}}_{()\stackrel{´}{c}\nu }^{}|Bb_i^{}|\mathrm{{\rm Y}}_{()\stackrel{´}{d}\nu }>`$
$`+<\mathrm{\Psi }_{(+)\stackrel{´}{a}}^{}|Bb_i^{}|\mathrm{\Psi }_{(+)\stackrel{´}{b}}><\mathrm{{\rm Y}}_{()\stackrel{´}{c}\nu }^{}|Bb_i|\mathrm{{\rm Y}}_{()\stackrel{´}{d}\nu }>]`$
$`=2\zeta _{\stackrel{´}{a}\stackrel{´}{b},\stackrel{´}{c}\stackrel{´}{d}}^{^{(10)(+)}}[8𝐌_{\stackrel{´}{a}i}^𝐓𝐌_{\stackrel{´}{b}}^{ij}𝐐_{\stackrel{´}{c}\mu }^{k𝐓}𝐐_{\stackrel{´}{d}kj\mu }8𝐌_{\stackrel{´}{a}i}^𝐓𝐌_{\stackrel{´}{b}}𝐐_{\stackrel{´}{c}\mu }^{i𝐓}𝐐_{\stackrel{´}{d}\mu }`$
$`𝐌_{\stackrel{´}{a}}^{ij𝐓}𝐌_{\stackrel{´}{b}}^{kl}𝐐_{\stackrel{´}{c}kl\mu }^𝐓𝐐_{\stackrel{´}{d}ij\mu }+𝐌_{\stackrel{´}{a}}^{ij𝐓}𝐌_{\stackrel{´}{b}}^{kl}𝐐_{\stackrel{´}{c}ik\mu }^𝐓𝐐_{\stackrel{´}{d}jl\mu }`$
$`𝐌_{\stackrel{´}{a}}^{ij𝐓}𝐌_{\stackrel{´}{b}}^{kl}𝐐_{\stackrel{´}{c}il\mu }^𝐓𝐐_{\stackrel{´}{d}jk\mu }+ϵ^{ijklm}𝐌_{\stackrel{´}{a}i}^𝐓𝐌_{\stackrel{´}{b}}𝐐_{\stackrel{´}{c}jk\mu }^𝐓𝐐_{\stackrel{´}{d}lm\mu }`$
$`+ϵ_{ijklm}𝐌_{\stackrel{´}{a}}^{ij𝐓}𝐌_{\stackrel{´}{b}}^{kl}𝐐_{\stackrel{´}{c}\mu }^{m𝐓}𝐐_{\stackrel{´}{d}\mu }]`$ (58)
and where
$`\xi _{\stackrel{´}{a}\stackrel{´}{b},\stackrel{´}{c}\stackrel{´}{d}}^{^{(10)}}=f_{\stackrel{´}{a}\stackrel{´}{b}}^{^{(10)}}h_{\stackrel{´}{c}\stackrel{´}{d}}^{^{(10)}}l__𝒰^{^{(10)}}\stackrel{~}{}_{𝒰𝒰^{}}^{^{(10)}}k__𝒰^{}^{^{(10)}}`$
$`\zeta _{\stackrel{´}{a}\stackrel{´}{b},\stackrel{´}{c}\stackrel{´}{d}}^{^{(10)}}=f_{\stackrel{´}{a}\stackrel{´}{b}}^{^{(10)}}\overline{h}_{\stackrel{´}{c}\stackrel{´}{d}}^{^{(10)}}l__𝒰^{^{(10)}}\stackrel{~}{}_{𝒰𝒰^{}}^{^{(10)}}\overline{k}__𝒰^{}^{^{(10)}}`$
$`\stackrel{~}{}^{^{(10)}}=\left[^{^{(10)}}+\left(^{^{(10)}}\right)^𝐓\right]^1`$ (59)
We note that $`\xi _{\stackrel{´}{a}\stackrel{´}{b},\stackrel{´}{c}\stackrel{´}{d}}^{^{(10)(+)}}`$, and $`\zeta _{\stackrel{´}{a}\stackrel{´}{b},\stackrel{´}{c}\stackrel{´}{d}}^{^{(10)(+)}}`$, are the same as defined by Eq.(6), provided we replace $`h`$’s, $`\overline{h}`$’s, $`f`$’s by $`h^{^{(+)}}`$’s, $`\overline{h}^{^{(+)}}`$’s, $`f^{^{(+)}}`$’s respectively where (+) indicates that the couplings are symmetric, i.e., $`h_{\stackrel{´}{a}\stackrel{´}{b}}^{^{(10)(\pm )}}=\frac{1}{2}\left(h_{\stackrel{´}{a}\stackrel{´}{b}}^{^{(10)}}\pm h_{\stackrel{´}{b}\stackrel{´}{a}}^{^{(10)}}\right)`$. The above couplings produce quark and lepton masses after GUT symmetry breaking followed by spontaneous breaking of the electroweak symmetry. A preliminary analysis shows that the relation on Yukawa couplings such as $`h_b=h_t`$ does not hold. It would be interesting to study the quark-lepton textures in this framework. Further, the preliminary analysis shows that baryon and lepton number violating dimension five operators in this theory are rather different than what one has in the usual $`SU(5)`$ and $`SO(10)`$ models. Thus, for example, in $`SU(5)`$ unified models the baryon and lepton number violating dimension five operators arise from the Higgs triplet exchange from pairs of $`5+\overline{5}`$. In the present model there are several sources of baryon and lepton number violations, including Higgs triplets from $`5`$ and $`\overline{5}`$ and from the exchange of $`45+\overline{45}`$ present in $`144+\overline{144}`$ of Higgs. A full analysis of this issue involves additional couplings where the mediation occurs via 120, 126 etc and is beyond the scope of this paper. An analysis of this will be given elsewhere. We note here that if in Eq. (53) only couplings involving $`\overline{144}`$ are kept, the resulting fermion mass matrices will have the same structure as in with a single 10 and one $`\overline{126}`$ coupling to fermions. Such matrices are fully consistent with experimental data.
## 7 Conclusion
In conclusion we have investigated a new class of $`SO(10)`$ models where the breaking of $`SO(10)`$ down to $`SU(3)_C\times SU(2)_L\times U(1)_Y`$ can occur in a single step. Further, it is possible to achieve with fine tuning, a pair of light Higgs doublets which is justifiable within a string based landscaped scenario. The light Higgs doublets allow one to break the electroweak symmetry $`SU(2)_L\times U(1)_Y`$ down to $`U(1)_{em}`$ and thus $`SO(10)`$ can break to $`SU(3)_C\times U(1)_{em}`$. The cubic interactions of the light Higgs doublets with quarks and leptons arise from quartic interactions of the type $`(16.16)(144.144)`$ and $`(16.16)(\overline{144}.\overline{144})`$ after spontaneous breaking of $`SO(10)`$. In this scenario the baryon and lepton number violating dimension five operators receive contributions not just from the conventional Higgs triplet fields but also from the exchange of $`45+\overline{45}`$ components of $`144+\overline{144}`$. The above feature distinguishes the above scenario from the conventional models and would lead to different estimates on the proton lifetime and in conventional GUTs. A more detailed analysis of the quark -lepton masses as well as of the proton life time is outside the scope of this paper and will be dealt with elsewhere. Finally we note that above the GUT scale one has a large number of degrees of freedom and the renormalization group evolution is very rapid and thus the theory becomes nonperturbative. We view this theory as descending directly from a theory of quantum gravity where the scale of quantum gravity (e.g., string scale) may lie close to the GUT scale.
ACKNOWLEDGEMENTS
The work of KB is supported in part by DOE grant DE-FG03-98ER-41076. The work of IG is supported in part by NSF grant PHY-0098791. The work of PN and RS is supported in part by NSF grant PHY-0139967. Part of this work was done while PN was visting the Max Planck Institute, Munich. He acknowledges support from the Alexander von Humboldt Foundation and thanks the MPI for the hospitality extended him.
|
warning/0506/math-ph0506031.html
|
ar5iv
|
text
|
# Reciprocal relativity of noninertial frames and the quaplectic group
## 1. Introduction
This year is the one hundredth anniversary of Einstein’s special relativity theory that changed our concept of space and time by combining the separate concepts of space and time into a unified four dimensional space-time. Newtonian mechanics has the concept of an absolute inertial rest frame that all observers agree upon, and furthermore all observers agree upon a single universal concept of time. Special relativity eliminates the absolute rest frame and the universal concept of time, but an absolute inertial frame that all observers agree upon continues to exist.
General relativity eliminates the need for an absolute inertial frame for gravitational forces by requiring particles under the influence of gravity to be in a curved space-time manifold. In this curved space-time, gravitating particles follow geodesics and therefore are locally inertial. In this sense, gravity is no longer a force that causes particles to transition to noninertial frames. This sidesteps the issue of a global inertial frame by eliminating the concept of noninertial frames as all frames for particles that are only under the influence of gravity are locally inertial.
However, if other forces are considered, the question of the absolute inertial frame reemerges. Return again to the case where the underlying space-time manifold is flat. Special relativity eliminates the problem of the absolute rest frame, simply by showing that there is no absolute rest frame. Velocities are meaningful only between observers associated with particle states and the relative velocity is bounded by $`c`$. Can we follow this approach of special relativity and simply eliminate the concept of a global inertial frame? The answer to this, remarkably, is yes. Forces can be made relativistic also so that rates of change of momentum are meaningful only between particle states (and not relative to some absolute inertial frame) and furthermore are bounded by a universal constant $`b`$. However, with the elimination of the absolute inertial frame, space-time takes on even more unusual properties than the four dimensional continuum introduced by special relativity.
The first step is to formulate the transformations of noninertial frames as the action of a more general group. The key to this is the observation that nonrelativistic Hamilton’s mechanics is a continuous unfolding of a canonical transformation. As this represents all possible motions of a classical point particle, these frames are generally noninertial. Hamilton’s equations may be formulated as a group of transformations of frames on time-position-momentum-energy space. These equations leave invariant the symplectic metric, and are therefore canonical, as well as the invariant nonrelativistic line element, $`ds^2=dt^2`$. These transformations describe general noninertial frames in the nonrelativistic context where there is an absolute inertial rest frame.
Next, consider frames on time-position-momentum-energy space under special relativistic transformations. Special relativity eliminates the concept of the absolute rest frame by requiring the invariance of the line element $`\mathrm{𝑑𝑠}^2=\mathrm{𝑑𝑡}^2+\frac{1}{c^2}\mathrm{𝑑𝑞}^2`$ on position-time space and the line element $`d\mu ^2=\frac{1}{c^2}\mathrm{𝑑𝑒}^2+\mathrm{𝑑𝑝}^2`$ on energy-momentum space. However, requiring these line elements to be independently invariant requires the canonical frame to be an inertial frame. To regain the general noninertial transformation equations of the nonrelativistic case, we must combine these two line elements into a single line element defined by the Born-Green metric $`\mathrm{𝑑𝑠}^2=\mathrm{𝑑𝑡}^2+\frac{1}{c^2}\mathrm{𝑑𝑞}^2+\frac{1}{b^2}(\mathrm{𝑑𝑝}^2\frac{1}{c^2}\mathrm{𝑑𝑒}^2)`$. Now we have two metrics that must be invariant under the transformations, the symplectic metric and the Born-Green orthogonal metric. The required group in $`n`$ dimensions is $`𝒰(1,n)`$ with $`n=3`$ corresponding to the usual physical case.
The introduction of the Born-Green metric enables the group of transformations to again include noninertial frames on time-position-momentum-energy space. The full group, $`𝒰(1,n),`$describes the transformations between frames that are canonical and noninertial. The remarkable fact is that, for these transformations, there is no longer an absolute inertial frame nor an absolute rest frame. Forces have become relative and are bounded by $`b`$. Velocities continue to be relative and are bounded by $`c`$. We call this reciprocal relativity. There is also no longer an absolute concept of position-time space that all observers agree on. The determination of the position-time subspace of the time-position-momentum energy space has become observer frame dependent. Under extreme noninertial conditions, that is, very strongly interacting particles, all of the time, position, momentum and energy degrees of freedom are mixed under this group of transformations that takes us from one noninertial observer to another. Simply put, momentum and energy can be transformed into position and time.
This generalization is very analogous to time becoming relative in the special relativity theory. In Newtonian mechanics, the Lorentz group contracts in the limit $`c\mathrm{}`$ to the Euclidean group. The Euclidean group acting on the frame leaves time invariant. Consequently, time is absolute in the sense that the determination of the time subspace of the position-time manifold is common to all observers. Under the action of the Lorentz group in special relativity, time is observer frame dependant and so there is no longer an absolute concept of time.
These transformations reduce to the canonical transformations between inertial frames of special relativity if the rate of change of momentum between the observers is zero. In this case, the transformation equations reduce to the usual special relativity Lorentz group transformations that act separately on position-time and energy-momentum space. The concept of a position-time space and energy-momentum space, on which all inertial observers agree, reemerges.
Up until this point we have been discussing the homogeneous group that generalizes the concept of the Lorentz group to noninertial frames. The basic wave or field equations of special relativistic quantum mechanics arise from the study of the unitary irreducible representations of the Poincaré group . The Poincaré group is the semidirect product of the Lorentz group with the abelian translation group. The Hermitian representations of the eigenvalue equations of the Casimir invariant operators define the free particle (or inertial frame) basic wave equations, the Klein-Gordon, Dirac, Maxwell equations and so forth.
Simply considering the semidirect product of the $`𝒰(1,n)`$ group of reciprocal relativity with the abelian translation group does not work. First, a basic principle of quantum mechanics is that the position and momentum and energy and time degrees of freedom do not commute. In fact, in nonrelativistic quantum mechanics, position and momentum are realized as the Hermitian representation of the algebra associated with the unitary irreducible representations of the Heisenberg group. The Heisenberg group itself is a real matrix Lie group that is the semidirect product of two translation groups, $`(n)=𝒯(n)_s𝒯(n+1)`$. As a real matrix group, the Heisenberg Lie algebra is
$$[P,Q]=I,[E,T]=I$$
The quaplectic group is then defined as the semidirect product
$$𝒬(1,n)=𝒰(1,n)_s\left(n+1\right).$$
It is important to emphasize that the non-commutative property of the position and momentum and the energy and time degrees of freedom therefore appear already in the classical (i.e. non quantum) formulation. That is, this may be viewed simply as a noncommutative geometry on the $`2n+4`$ dimensional space $`𝒬(1,n)/𝒮𝒰(1,n)`$ analogous to Minkowski space $`𝒫(1,n)/𝒮𝒪(1,n)`$. For $`n=3`$, this space has 10 dimensions.
The quantum theory of the particles in the noninertial frames then follows directly by considering the unitary irreducible representations of the quaplectic group. Note that the expected Heisenberg relations follow directly from the Hermitian representations of the algebra corresponding to unitary irreducible representations of the Heisenberg group. The wave equations that result are derived in . By way of example, the equivalent in this theory of the scalar equation is the relativistic oscillator. It is derived directly from the Hermitian representations of the eigenvalue equation for the Casimir invariants in the scalar representation. The Casimir invariant is by definition an invariant of the quaplectic group, and in particular, is invariant under Heisenberg nonabelian translations. This is exactly the method used to compute the free particle wave equations from the Poincaré group.
Returning again to the classical (ie non-quantum), but nonabelian theory, there is another fundamental motivation for why the inhomogeneous group cannot be the abelian translations. Clearly, if a general relativity type theory is to be considered in this framework, it must be possible to admit manifolds that are not flat. Schuller has proven a no-go theorem that states that a manifold that is a tangent bundle with a symplectic and (Born-Green) orthogonal metric must be flat. One way to invalidate the no-go theorem is to make the manifold noncommutative. For the physical case $`n=3`$, this results in a 10 dimensional noncommutative curved manifold. How to formulate this differential geometry is an open question and the topic of a subsequent investigation.
A final observation about the choice of the Heisenberg group as the normal subgroup of the semidirect product. Unlike the translation group where we can construct the inhomogeneous general linear group, the group "$`𝒢(n)_s(n)`$" does not exist. This is because the automorphisms of the quaplectic group require the homogeneous group in this semidirect product to be the symplectic group, or one of its subgroups. Thus the requirement for the normal subgroup to be the Heisenberg group automatically requires the symplectic metric that has been fundamental in our discussion.
This paper seeks to motivate the above description in the simplest possible manner. It looks at a system that has only one position dimension and traces the above arguments through to show how the quaplectic group arises. The companion paper then determines the unitary representations of the $`n`$ dimensional quaplectic group, the associated Hilbert spaces and the field equations that arise from the Hermitian representation of the eigenvalue equations of the Casimir invariant operators.
## 2. Homogeneous group: Relativity
### 2.1. Basic special relativity
The special relativity transformation equations between inertial observers. As we are considering the one dimensional case $`x=\{t,q\}𝕄^2`$. The global transforms are
$$\stackrel{~}{t}=\left(1\left(\frac{v}{c}\right)^2\right)^{1/2}\left(t+\frac{v}{c^2}q\right),\stackrel{~}{q}=\left(1\left(\frac{v}{c}\right)^2\right)^{1/2}\left(q+vt\right)$$
These expressions are made local by lifting to the cotangent space acting on a frame $`\{dt,dq\}`$.
$$d\stackrel{~}{t}=\left(1\left(\frac{v}{c}\right)^2\right)^{1/2}\left(dt+\frac{v}{c^2}dq\right),d\stackrel{~}{q}=\left(1\left(\frac{v}{c}\right)^2\right)^{1/2}\left(dq+vdt\right)$$
(1)
This may be written in matrix notation as $`d\stackrel{~}{x}=\mathrm{\Lambda }(v)dx`$ where $`dx=\{dt,dq\}`$, $`dxT_{}^{}{}_{x}{}^{}𝕄`$ and $`\mathrm{\Lambda }(v)`$ is a $`2\times 2`$ matrix
$$\mathrm{\Lambda }(v)=\left(1\left(\frac{v}{c}\right)^2\right)^{1/2}\left(\begin{array}{cc}1\hfill & \frac{v}{c^2}\hfill \\ v\hfill & 1\hfill \end{array}\right)$$
(2)
These transformations leave invariant the line elements $`\mathrm{𝑑𝑠}^2=\mathrm{𝑑𝑡}^2+\frac{1}{c^2}\mathrm{𝑑𝑞}^2`$. The usual addition law for special relativity results directly from the group composition law
$$\mathrm{\Lambda }(\stackrel{~}{\stackrel{~}{v}})=\mathrm{\Lambda }(\stackrel{~}{v})\mathrm{\Lambda }(v),\mathrm{\Lambda }^1(v)=\mathrm{\Lambda }(v)$$
(3)
Multiplying out the matrices gives the expected result
$$\stackrel{~}{\stackrel{~}{v}}=\frac{\stackrel{~}{v}+v}{1+\stackrel{~}{v}v/c^2}$$
(4)
Physically, if $`v`$ is the velocity between inertial frame of observers 1 and 2, $`\stackrel{~}{v}`$ the relative velocity between inertial frames of observers 2 and 3, then $`\stackrel{~}{\stackrel{~}{v}}`$ given by this expression is the relative velocity observed between inertial frames of observers 1 and 3.
Alternatively, from (1),
$$\frac{d\stackrel{~}{q}}{d\stackrel{~}{t}}=\left(dq+vdt\right)/\left(dt+\frac{v}{c^2}dq\right)=\left(\frac{dq}{dt}+v\right)/\left(1+\frac{v}{c^2}\frac{dq}{dt}\right)$$
(5)
Comparing this equation with (4) leads to the identification of $`v`$ with the rate of change of position, $`dq/dt`$.
Now the nonrelativistic case is given simply by requiring the limit
$$\mathrm{\Phi }(v)=\underset{c\mathrm{}}{lim}\mathrm{\Lambda }(v)=\left(\begin{array}{cc}1\hfill & 0\hfill \\ v\hfill & 1\hfill \end{array}\right)$$
(6)
Multiplying out the matrices
$$\mathrm{\Phi }(\stackrel{~}{\stackrel{~}{v}})=\mathrm{\Phi }(\stackrel{~}{v})\mathrm{\Phi }(v),\mathrm{\Phi }^1(v)=\mathrm{\Phi }(v)$$
(7)
gives the expected result
$$\stackrel{~}{\stackrel{~}{v}}=\stackrel{~}{v}+v$$
(8)
and so the nonrelativistic transformations between inertial frames are $`d\stackrel{~}{x}=\mathrm{\Phi }(v)dx`$ is just the expected
$$\begin{array}{c}d\stackrel{~}{t}=dt\hfill \\ d\stackrel{~}{q}=dq+vdt\hfill \end{array}$$
(9)
These equations leave invariant the classical line element
$$ds^2=\underset{c\mathrm{}}{lim}\left(\mathrm{𝑑𝑡}^2+\frac{1}{c^2}\mathrm{𝑑𝑞}^2\right)=dt^2$$
(10)
For this reason, we say that Newtonian mechanics has a concept of absolute time. Then, as in the special relativistic case (5), divide by $`dt`$ to obtain the usual velocity addition (8)
$$\frac{d\stackrel{~}{q}}{d\stackrel{~}{t}}=\left(\frac{dq}{dt}+v\right)$$
(11)
The key difference between special relativity and Newtonian mechanics is that the latter has a universal ‘inertial rest frame’ whereas in the former, velocity is relative and so there is no absolute rest frame. However, in special relativity, there continues to be a global inertial frame.
### 2.2. Inertial canonical frame transformations of special relativity
The above basic arguments may be repeated on time-position-momentum-energy space. In the following argument, this combination is somewhat formal as these transformations never mix. However, it motivates the section that follows.
The special relativity transformation equations between inertial observers and consider also the momentum, energy equations. $`z=\{t,q,p,e\}^4`$.
$$\stackrel{~}{t}=\left(1\left(\frac{v}{c}\right)^2\right)^{1/2}\left(t+\frac{v}{c^2}q\right),\stackrel{~}{q}=\left(1\left(\frac{v}{c}\right)^2\right)^{1/2}\left(q+vt\right)$$
$$\stackrel{~}{p}=\left(1\left(\frac{v}{c}\right)^2\right)^{1/2}\left(p+\frac{v}{c^2}e\right),\stackrel{~}{e}=\left(1\left(\frac{v}{c}\right)^2\right)^{1/2}\left(e+vp\right)$$
In these expressions, $`t`$ is time, $`q`$ is position, $`p`$ is momentum and, with a little unconventional notation, $`e`$ is energy. Again, we make the expressions local, applying the same argument also to the momentum and energy degrees of freedom, by lifting to the cotangent space with a frame $`\{dt,dq,dp,de\}`$.
$$d\stackrel{~}{t}=\left(1\left(\frac{v}{c}\right)^2\right)^{1/2}\left(dt+\frac{v}{c^2}dq\right),d\stackrel{~}{q}=\left(1\left(\frac{v}{c}\right)^2\right)^{1/2}\left(dq+vdt\right)$$
$$d\stackrel{~}{p}=\left(1\left(\frac{v}{c}\right)^2\right)^{1/2}\left(dp+\frac{v}{c^2}de\right),d\stackrel{~}{e}=\left(1\left(\frac{v}{c}\right)^2\right)^{1/2}\left(de+vdp\right)$$
This may be written in matrix notation as $`d\stackrel{~}{z}=\mathrm{\Gamma }(v)dz`$ where $`dz=\{dt,dq,dp,de\}`$, $`dzT_{}^{}{}_{z}{}^{}`$ and $`\mathrm{\Gamma }(v)`$ is a $`4\times 4`$ matrix
$$\mathrm{\Gamma }(v)=\left(\begin{array}{cc}\mathrm{\Lambda }(v)\hfill & 0\hfill \\ 0\hfill & \mathrm{\Lambda }(v)\hfill \end{array}\right),\mathrm{\Lambda }(z)=\left(1\left(\frac{v}{c}\right)^2\right)^{1/2}\left(\begin{array}{cc}1\hfill & \frac{v}{c^2}\hfill \\ v\hfill & 1\hfill \end{array}\right)$$
Expanding this out gives the $`4\times 4`$ matrix
$$\mathrm{\Gamma }(v)=\left(1\left(\frac{v}{c}\right)^2\right)^{1/2}\left(\begin{array}{cccc}1\hfill & \frac{v}{c^2}\hfill & 0\hfill & 0\hfill \\ v\hfill & 1\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 1\hfill & \frac{v}{c^2}\hfill \\ 0\hfill & 0\hfill & v\hfill & 1\hfill \end{array}\right)$$
(12)
As the matrix is simply the direct sum, the subspaces spanned by $`\{dt,dq\}`$ and $`\{dp,de\}`$ do not mix. These transformations leave invariant the line elements $`\mathrm{𝑑𝑠}^2=\mathrm{𝑑𝑡}^2+\frac{1}{c^2}\mathrm{𝑑𝑞}^2`$ and independently $`d\mu ^2=\frac{1}{c^2}\mathrm{𝑑𝑒}^2+\mathrm{𝑑𝑝}^2`$. The symplectic metric continues to be invariant also
$$\begin{array}{cc}\hfill {}_{}{}^{t}dz\zeta dz& =dedt+dpdq\hfill \end{array}$$
(13)
Thus, in addition to being inertial, the frames are canonical in the sense that the symplectic metric has the simple form given. The appearance of the symplectic metric is not arbitrary but directly required if a semidirect product group with the Heisenberg group as the normal subgroup is to be constructed for the inhomogeneous case. This is described in Section 3.3.
The usual addition law for special relativity results directly from the group composition law, $`\mathrm{\Lambda }(\stackrel{~}{\stackrel{~}{v}})=\mathrm{\Lambda }(\stackrel{~}{v})\mathrm{\Lambda }(v)`$. Multiplying out the matrices gives the expected result (4). Physically, if $`v`$ is the velocity between inertial frame of observers 1 and 2, $`\stackrel{~}{v}`$ the relative velocity between inertial frames of observers 2 and 3, then $`\stackrel{~}{\stackrel{~}{v}}`$ given by this expression is the relative velocity observed between inertial frames of observers 1 and 3. The identification with velocity is verified by
$$\frac{d\stackrel{~}{q}}{d\stackrel{~}{t}}=(\frac{dq}{dt}+v)/(1+\frac{v}{c^2}\frac{dq}{dt})$$
(14)
$$\frac{\stackrel{~}{e}}{\stackrel{~}{p}}=\left(\frac{e}{p}+v\right)/\left(1+\frac{v}{c^2}\frac{e}{p}\right)$$
(15)
Now the nonrelativistic case is given simply by taking the limit
$$\mathrm{\Phi }(v)=\underset{c\mathrm{}}{lim}\mathrm{\Gamma }(v)=\left(\begin{array}{cccc}1\hfill & 0\hfill & 0\hfill & 0\hfill \\ v\hfill & 1\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 1\hfill & 0\hfill \\ 0\hfill & 0\hfill & v\hfill & 1\hfill \end{array}\right)$$
(16)
and the nonrelativistic transformations between inertial frames, $`d\stackrel{~}{z}=\mathrm{\Phi }(v)dz`$ are the expected
$$\begin{array}{c}d\stackrel{~}{t}=dt\hfill \\ d\stackrel{~}{q}=dq+vdt\hfill \\ d\stackrel{~}{p}=dp\hfill \\ d\stackrel{~}{e}=de+vdp\hfill \end{array}$$
(17)
These equations leave invariant the line element $`ds^2=dt^2`$ (10) and also the symplectic metric (13). $`\mathrm{\Phi }(v)`$ is just a realization of the one dimensional translation group and so again, the velocity addition law is $`\mathrm{\Phi }(\stackrel{~}{\stackrel{~}{v}})=\mathrm{\Phi }(\stackrel{~}{v})\mathrm{\Phi }(v)`$ and it follows that $`\stackrel{~}{\stackrel{~}{v}}=\stackrel{~}{v}+v`$ .
Again, these transformation equations are for frames that are both inertial and canonical. For heuristic purposes, note that (15) leads to the identification $`v=\frac{dq}{dt}=\frac{e}{p}`$. We will return to this more carefully shortly.
### 2.3. Noninertial frame transformations of nonrelativistic mechanics
Nonrelativistic Hamilton’s mechanics does not have the requirement for frames to be inertial. Given the Hamiltonian function, particles with complex noninertial motion can be described. All point particle states in classical mechanics, and the associated noninertial frames, must comply with Hamilton’s equations. The basic idea of Hamilton’s mechanics is that the particle motion is the continuous unfolding of a canonical transformation. Hamilton’s equations are
$$v=\frac{dq(t)}{dt}=\frac{H(p,q,t)}{p},f=\frac{dp(t)}{dt}=\frac{H(p,q,t)}{q},r=\frac{H(p,q,t)}{t}$$
(18)
Consider the co-ordinate transformation $`\stackrel{~}{z}=\phi (z)`$. Then, with simple matrix notation
$$d\stackrel{~}{z}=\frac{\phi (z)}{z}\mathrm{𝑑𝑧}=\mathrm{\Phi }(v,f,r)dz$$
(19)
and so in component form
$$\frac{\phi ^\alpha }{z^\beta }=\mathrm{\Phi }(v,f,r)_\beta ^\alpha ,\alpha ,\beta =1,2,3,4.$$
(20)
The solution of Hamilton’ s equations defines the canonical transformations $`\phi ^\alpha (t,q,p,e)`$ that have the form
$$\begin{array}{cc}\stackrel{~}{t}=\phi ^1(t,q,p,e)=t\hfill & \\ \stackrel{~}{q}=\phi ^2(t,q,p,e)=q+q(t)\hfill & q(0)=0\hfill \\ \stackrel{~}{p}=\phi ^3(t,q,p,e)=p+p(t)\hfill & p(0)=0\hfill \\ \stackrel{~}{e}=\phi ^4(t,q,p,e)=e+H(p,q,t)\hfill & H(0,0,0)=0\hfill \end{array}$$
(21)
The notation is is being abused slightly using $`q,p`$ for the initial points and $`q(t),p(t)`$ for the functions giving the time evolution. Substituting (21) into (19) and using Hamilton’s equations (18) gives the result
$$\mathrm{\Phi }(v,f,r)\left(\begin{array}{cccc}1\hfill & 0\hfill & 0\hfill & 0\hfill \\ v\hfill & 1\hfill & 0\hfill & 0\hfill \\ f\hfill & 0\hfill & 1\hfill & 0\hfill \\ r\hfill & f\hfill & v\hfill & 1\hfill \end{array}\right)$$
(22)
Writing out the transformations using $`d\stackrel{~}{z}=\mathrm{\Phi }(v,f,r)dz`$ yields
$$\begin{array}{c}d\stackrel{~}{t}=dt\hfill \\ d\stackrel{~}{q}=dq+vdt\hfill \\ d\stackrel{~}{p}=dp+fdt\hfill \\ d\stackrel{~}{e}=de+vdpfdq+rdt\hfill \end{array}$$
(23)
Conversely, starting with (23) in (19), one can derive (21) and Hamilton’s equations (18). These equations define the transformations between noninertial frames. Only noninertial frames that satisfy these transformations, and hence Hamilton’s equations, are physical.
These transformations leave invariant the line element $`ds^2=dt^2={}_{}{}^{t}dz\eta _cdz`$ and also the symplectic metric
$$\begin{array}{cc}\hfill {}_{}{}^{t}d\stackrel{~}{z}\zeta d\stackrel{~}{z}& =d\stackrel{~}{e}d\stackrel{~}{t}+d\stackrel{~}{p}d\stackrel{~}{q}\hfill \\ & =\left(de+vdpfdq+rdt\right)dt+\left(dp+fdt\right)\left(dq+vdt\right)\hfill \\ & =dedt+dpdq={}_{}{}^{t}dz\zeta dz\hfill \end{array}$$
In fact, $`\mathrm{\Phi }`$ may be derived from the requirement that $`{}_{}{}^{t}\mathrm{\Phi }\zeta \mathrm{\Phi }=\zeta `$ and $`{}_{}{}^{t}\mathrm{\Phi }\eta _c\mathrm{\Phi }=\eta _c`$ where
$$\zeta =\left(\begin{array}{cccc}0\hfill & 0\hfill & 0\hfill & 1\hfill \\ 0\hfill & 0\hfill & 1\hfill & 0\hfill \\ 0\hfill & 1\hfill & 0\hfill & 0\hfill \\ 1\hfill & 0\hfill & 0\hfill & 0\hfill \end{array}\right),\eta _c=\left(\begin{array}{cccc}1\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill \end{array}\right)$$
(24)
Certainly the momentum transformation for these frames, that are canonical but not necessarily inertial, is the form expected. The energy transformation has a kinetic term $`vdp`$, as in the inertial equations above, a new work term $`fdq`$ and the explicit power term $`rdt`$.
Direct matrix multiplication defines the group composition laws and verifies that this is a matrix Lie group
$$\begin{array}{c}\mathrm{\Phi }(\stackrel{~}{\stackrel{~}{v}},\stackrel{~}{\stackrel{~}{f}},\stackrel{~}{\stackrel{~}{r}})=\mathrm{\Phi }(\stackrel{~}{v},\stackrel{~}{f},\stackrel{~}{r})\mathrm{\Phi }(v,f,r)=\mathrm{\Phi }(\stackrel{~}{v}+v,\stackrel{~}{f}+f,\stackrel{~}{r}+r+\stackrel{~}{v}f\stackrel{~}{f}v),\hfill \\ \mathrm{\Phi }^1(v,f,r)=\mathrm{\Phi }(v,f,r)\hfill \end{array}$$
(25)
We call this the Hamilton group in one dimension, $`\mathrm{\Phi }(v,f,r)a(1)`$. As in the inertial case, this defines the generalized addition laws for velocity $`v`$, force, $`f`$ and power, $`r`$ for three different observers. That is if we have relative $`(v,f,r`$) between noninertial observer frames $`1`$ and $`2`$ and relative $`(\stackrel{~}{v},\stackrel{~}{f},\stackrel{~}{r})`$ between observer frames 2 and 3, then observer frames 1 and 3 are related by $`(\stackrel{~}{\stackrel{~}{v}},\stackrel{~}{\stackrel{~}{f}},\stackrel{~}{\stackrel{~}{r}})`$ where
$$\begin{array}{c}\stackrel{~}{\stackrel{~}{v}}=\stackrel{~}{v}+v\hfill \\ \stackrel{~}{\stackrel{~}{f}}=\stackrel{~}{f}+f\hfill \\ \stackrel{~}{\stackrel{~}{r}}=\stackrel{~}{r}+v\stackrel{~}{f}f\stackrel{~}{v}+r\hfill \end{array}$$
(26)
As time is invariant under the transformations, we can simply divide (23) by $`dt`$ to obtain.
$$\frac{d\stackrel{~}{q}}{d\stackrel{~}{t}}=\frac{dq}{dt}+v,\frac{d\stackrel{~}{p}}{d\stackrel{~}{t}}=\frac{dp}{dt}+f,\frac{d\stackrel{~}{e}}{d\stackrel{~}{t}}=\frac{de}{dt}+v\frac{dp}{dt}f\frac{dq}{dt}+r$$
(27)
Comparing with (25), this lead to the identification of $`v`$ with rate of change of position, $`f`$ with rate of change of momentum and $`r`$ with rate of change of energy with respect to the invariant time element $`dt`$ as asserted.
The matrix Lie algebra of $`a(1)`$ follows directly from the Lie algebra valued one forms $`d\mathrm{\Phi }(v,f,r)|_0=\mathrm{𝑑𝑣}G+d\mathrm{𝑓𝐹}+\mathrm{dr}R`$ where the matrix realization of the generators is given by
$$\begin{array}{cc}G=\left(\begin{array}{cccc}0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 1\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 1\hfill & 0\hfill \end{array}\right),\hfill & F=\left(\begin{array}{cccc}0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 1\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 1\hfill & 0\hfill & 0\hfill \end{array}\right),\hfill \\ R=\left(\begin{array}{cccc}0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 1\hfill & 0\hfill & 0\hfill & 0\hfill \end{array}\right)\hfill & \end{array}$$
(28)
with
$$G=\frac{\mathrm{\Phi }}{v},F=\frac{\mathrm{\Phi }}{f},R=\frac{\mathrm{\Phi }}{r}$$
The Lie bracket of a matrix Lie group is given simply by $`[A,B]=ABBA`$ and these may be directly computed as
$$[G,F]=2R,[R,F]=0,[R,G]=0$$
(29)
These may also be viewed as an abstract Lie algebra satisfying the above commutation relations. The manner in which this formulation carries over to Lagrangian mechanics is summarized in Section 5.1.
### 2.4. Reciprocal relativity
This is where we depart from a simple review of standard physics. We now introduce the Born-Green conjecture of the combined line element and derive the transformation equations that encompass reciprocal relativity.
Two conditions must be satisfied by the transformation equations. As, always, the symplectic metric $`\mathrm{𝑑𝑒}\mathrm{𝑑𝑡}+\mathrm{𝑑𝑝}\mathrm{𝑑𝑞}`$ must be invariant. Furthermore, in the classical nonrelativistic theory, we have the invariant $`\mathrm{𝑑𝑡}^2`$ while in the special relativistic case we have the invariants $`\mathrm{𝑑𝑡}^2+\frac{1}{c^2}\mathrm{𝑑𝑞}^2`$ and $`\frac{1}{c^2}\mathrm{𝑑𝑒}^2+\mathrm{𝑑𝑝}^2`$. Following Born-Green, this suggests we investigate that case of further combining these invariants to define a non-degenerate (pseudo) orthogonal metric on the 4 dimensional space
$$ds^2=dt^2+\frac{1}{c^2}dq^2+\frac{1}{b^2}\left(\frac{1}{c^2}de^2+dp^2\right)={}_{}{}^{t}dz\eta dz$$
(30)
where $`\eta =\mathrm{diag}\{1,1,1,1\}`$.
This is the only new physical assumption that has been introduced in this paper. $`b`$ is a universal constant with the dimensions of force. Usually $`(c,\mathrm{},G)`$ are taken to be the independent dimensional constants. In terms of this dimensional basis,
$$b=\alpha _G\frac{c^4}{G}.$$
(31)
If $`\alpha _G=1`$, this is merely a notational change. However, $`\alpha _G`$ is a dimensionless parameter that must be determined by theory or experiment. We take the independent dimensional constants to be $`(c,\mathrm{},b)`$ in which case $`G=\alpha _G\frac{c^4}{b}`$.
#### 2.4.1. Transformation Equations
The transformations leaving the symplectic form invariant are $`𝒮p(4)`$ and the transformations leaving the orthogonal metric invariant are $`𝒪(2,2)`$. The group of transformations leaving both invariant is
$$𝒰(1,1)𝒰(1)𝒮𝒰(1,1)𝒮p(4)𝒪(2,2)$$
Consider first the $`𝒮𝒰(1,1)`$ transformation equations
$$\mathrm{\Xi }(v,f,r)=\left(1w^2\right)^{1/2}\left(\begin{array}{cccc}1\hfill & \frac{v}{c^2}\hfill & \frac{f}{b^2}\hfill & \frac{r}{b^2c^2}\hfill \\ v\hfill & 1\hfill & \frac{r}{b^2}\hfill & \frac{f}{b^2}\hfill \\ f\hfill & \frac{r}{c^2}\hfill & 1\hfill & \frac{v}{c^2}\hfill \\ r\hfill & f\hfill & v\hfill & 1\hfill \end{array}\right).$$
(32)
The transformation equations are then given $`d\stackrel{~}{z}=\mathrm{\Xi }dz`$
$$d\stackrel{~}{t}=\left(1w^2\right)^{1/2}\left(\mathrm{𝑑𝑡}+\frac{v}{c^2}\mathrm{𝑑𝑞}+\frac{f}{b^2}\mathrm{𝑑𝑝}\frac{r}{b^2c^2}\mathrm{𝑑𝑒}\right),$$
$$d\stackrel{~}{q}=\left(1w^2\right)^{1/2}\left(\mathrm{𝑑𝑞}+v\mathrm{𝑑𝑡}+\frac{r}{b^2}\mathrm{𝑑𝑝}\frac{f}{b^2}\mathrm{𝑑𝑒}\right),$$
$$d\stackrel{~}{p}=\left(1w^2\right)^{1/2}\left(\mathrm{𝑑𝑝}+f\mathrm{𝑑𝑡}\frac{r}{c^2}\mathrm{𝑑𝑞}+\frac{v}{c^2}\mathrm{𝑑𝑒}\right)$$
$$d\stackrel{~}{e}=(1w^2)^{1/2}(\mathrm{𝑑𝑒}f\mathrm{𝑑𝑞}+v\mathrm{𝑑𝑝}+r\mathrm{𝑑𝑡})$$
where $`w^2=v^2/c^2+f^2/b^2r^2/b^2c^2`$. It may be directly verified that these equations leave invariant the symplectic metric (23), $`{}_{}{}^{t}\mathrm{\Xi }\zeta \mathrm{\Xi }=\zeta `$ and Born-Green orthogonal metric (30), $`{}_{}{}^{t}\mathrm{\Xi }\eta \mathrm{\Xi }=\eta `$.
The usual special relativity equations take a convenient form when parameterized by angles and hyperbolic angles. This is true also in this formulation as described in the Appendix 5.2.
Again, these frames are associated with particles that are simply curves or trajectories in the space of time, position, momentum and energy $`z=(t,q,p,e)^4`$. Each particle has associated with it an observer with a frame that is a basis of the cotangent space $`\mathrm{𝑑𝑧}=(\mathrm{𝑑𝑡},\mathrm{𝑑𝑞},\mathrm{𝑑𝑝},\mathrm{𝑑𝑒})T_{}^{}{}_{z}{}^{}`$. Consider two observers with trajectories $`c::sz=c(s)`$ and $`\stackrel{~}{c}::s\stackrel{~}{z}=\stackrel{~}{c}(s)`$ that pass in the neighborhood of each other at some point in $``$. In this case, we assume that their frames are related by a rate of change of position, $`v`$, a rate of change of momentum, $`f`$ and a rate of change of energy $`r`$ with time $`t`$. These generalize the usual transformations of special relativity where it is assumed that the relative rate of change of momentum and energy are zero. The transformations no longer have position-time and momentum-energy invariant subspaces but mix all the degrees of freedom. Thus the dilation and contraction concepts of special relativity are generalized to the full space.
These transformations no longer have position-time as an invariant subspace. The position-time subspace of time-position-momentum-energy space is observer frame dependent. Extreme noninertial frames of very strongly interacting particles relative to the scale $`b`$ have energy and momentum degrees of freedom of particle states are transforming into position and time degrees of freedom of particle states.
To provide heuristic plausibility that these degrees of freedom could mix in the manner described, consider this in the context of the very early universe where the energy and momentum degrees of freedom where huge and position and time degrees of freedom were very small. We are now in a universe where the energy and momentum degrees of freedom of particle states are generally relatively small but the position and time degrees of freedom are very large relative to these scales.
The tangent vectors are
$$\frac{d\stackrel{~}{q}}{d\stackrel{~}{t}}=\left(\frac{\mathrm{𝑑𝑞}}{\mathrm{𝑑𝑡}}+v+\frac{r}{b^2}\frac{\mathrm{𝑑𝑝}}{\mathrm{𝑑𝑡}}\frac{f}{b^2}\frac{\mathrm{𝑑𝑒}}{\mathrm{𝑑𝑡}}\right)/\left(1+\frac{v}{c^2}\frac{\mathrm{𝑑𝑞}}{\mathrm{𝑑𝑡}}+\frac{f}{b^2}\frac{\mathrm{𝑑𝑝}}{\mathrm{𝑑𝑡}}\frac{r}{b^2c^2}\frac{\mathrm{𝑑𝑒}}{\mathrm{𝑑𝑡}}\right),$$
(33)
$$\frac{d\stackrel{~}{p}}{d\stackrel{~}{t}}=\left(\frac{\mathrm{𝑑𝑝}}{\mathrm{𝑑𝑡}}+f\frac{r}{c^2}\frac{\mathrm{𝑑𝑞}}{\mathrm{𝑑𝑡}}+\frac{v}{c^2}\frac{\mathrm{𝑑𝑒}}{\mathrm{𝑑𝑡}}\right)/\left(1+\frac{v}{c^2}\frac{\mathrm{𝑑𝑞}}{\mathrm{𝑑𝑡}}+\frac{f}{b^2}\frac{\mathrm{𝑑𝑝}}{\mathrm{𝑑𝑡}}\frac{r}{b^2c^2}\frac{\mathrm{𝑑𝑒}}{\mathrm{𝑑𝑡}}\right),$$
(34)
$$\frac{d\stackrel{~}{e}}{d\stackrel{~}{t}}=\left(\frac{\mathrm{𝑑𝑒}}{\mathrm{𝑑𝑡}}f\frac{\mathrm{𝑑𝑞}}{\mathrm{𝑑𝑡}}+v\frac{\mathrm{𝑑𝑝}}{\mathrm{𝑑𝑡}}+r\right)/\left(1+\frac{v}{c^2}\frac{\mathrm{𝑑𝑞}}{\mathrm{𝑑𝑡}}+\frac{f}{b^2}\frac{\mathrm{𝑑𝑝}}{\mathrm{𝑑𝑡}}\frac{r}{b^2c^2}\frac{\mathrm{𝑑𝑒}}{\mathrm{𝑑𝑡}}\right)$$
(35)
With the identification $`v\frac{\mathrm{𝑑𝑞}}{\mathrm{𝑑𝑡}}`$, $`\frac{\mathrm{𝑑𝑓}}{\mathrm{𝑑𝑡}}f`$ and $`\frac{\mathrm{𝑑𝑒}}{\mathrm{𝑑𝑡}}r`$ this leads to the rate of change of position, momentum and energy relativity transformation laws
$$\stackrel{~}{\stackrel{~}{v}}=\left(\stackrel{~}{v}+v+\frac{r\stackrel{~}{f}}{b^2}\frac{f\stackrel{~}{r}}{b^2}\right)/\left(1+\frac{v\stackrel{~}{v}}{c^2}+\frac{f\stackrel{~}{f}}{b^2}\frac{r\stackrel{~}{r}}{b^2c^2}\right),$$
$$\stackrel{~}{\stackrel{~}{f}}=\left(\stackrel{~}{f}+f\frac{r\stackrel{~}{v}}{c^2}+\frac{v\stackrel{~}{r}}{c^2}\right)/\left(1+\frac{v\stackrel{~}{v}}{c^2}+\frac{f\stackrel{~}{f}}{b^2}\frac{r\stackrel{~}{r}}{b^2c^2}\right),$$
$$\stackrel{~}{\stackrel{~}{r}}=\left(\stackrel{~}{r}f\stackrel{~}{v}+v\stackrel{~}{f}+r\right)/\left(1+\frac{v\stackrel{~}{v}}{c^2}+\frac{f\stackrel{~}{f}}{b^2}\frac{r\stackrel{~}{r}}{b^2c^2}\right)$$
Again as in the usual special relativity case, these are bounded. If $`v=\stackrel{~}{v}=c`$, $`f=\stackrel{~}{f}=b`$ and $`r=\stackrel{~}{r}=bc`$, then
$$\stackrel{~}{\stackrel{~}{v}}=\left(c+c+\frac{bcb}{b^2}\frac{bbc}{b^2}\right)/\left(1+\frac{cc}{c^2}+\frac{bb}{b^2}\frac{bcbc}{b^2c^2}\right)=c$$
$$\stackrel{~}{\stackrel{~}{f}}=\left(b+b\frac{bcc}{c^2}+\frac{cbc}{c^2}\right)/\left(1+\frac{cc}{c^2}+\frac{bb}{b^2}\frac{bcbc}{b^2c^2}\right)=b$$
$$\stackrel{~}{\stackrel{~}{r}}=\left(bcbc+cb+bc\right)/\left(1+\frac{cc}{c^2}+\frac{bb}{b^2}\frac{bcbc}{b^2c^2}\right)=bc$$
In this formulation, rates of change of position, momentum and energy with respect to time are all relative. Forces are only meaningful between particle states. There is no absolute inertial frame or absolute rest frame. Position-time, or as we usually say, space-time itself is relative.
That this identification is correct may be verified by computing the matrix multiplication for the four dimensional matrix realization of the group and verifying that it is precisely the above transformation rules in (35) that are required for the matrix identity
$$\mathrm{\Xi }(\stackrel{~}{\stackrel{~}{v}},\stackrel{~}{\stackrel{~}{f}},\stackrel{~}{\stackrel{~}{r}})=\mathrm{\Xi }(\stackrel{~}{v},\stackrel{~}{f},\stackrel{~}{r})\mathrm{\Xi }(v,f,r)$$
(36)
to be satisfied. This is exactly the same reasoning as in the usual special velocity relativity case given in (3)-(5).
#### 2.4.2. The $`𝒰(1)`$ transformations
The full group $`𝒰(1,1)`$ of transformations leaving both the symplectic and orthogonal metrics invariant includes the $`𝒰(1)`$ subgroup. This group commutes with the $`𝒮𝒰(1,1)`$ transformations which enables us to consider it separately. The transformation equations for this group are
$$\begin{array}{c}d\stackrel{~}{t}=\mathrm{cos}\theta \mathrm{𝑑𝑡}\frac{1}{bc}\mathrm{sin}\theta \mathrm{𝑑𝑒},\hfill \\ d\stackrel{~}{q}=\mathrm{cos}\theta \mathrm{𝑑𝑞}\frac{c}{b}\mathrm{sin}\theta \mathrm{𝑑𝑝},\hfill \\ d\stackrel{~}{p}=\mathrm{cos}\theta \mathrm{𝑑𝑝}+\frac{b}{c}\mathrm{sin}\theta \mathrm{𝑑𝑞},\hfill \\ d\stackrel{~}{e}=\mathrm{cos}\theta \mathrm{𝑑𝑒}+bc\mathrm{sin}\theta dt.\hfill \end{array}$$
(37)
Setting $`\mathrm{tan}\theta =\frac{a}{bc}`$, the matrix realization for this is
$$\mathrm{\Xi }^{}(a)\left(1+\left(\frac{a}{bc}\right)^2\right)^{1/2}\left(\begin{array}{cccc}1\hfill & 0\hfill & 0\hfill & \frac{a}{b^2c^2}\hfill \\ 0\hfill & 1\hfill & \frac{a}{b^2}\hfill & 0\hfill \\ 0\hfill & \frac{a}{c^2}\hfill & 1\hfill & 0\hfill \\ a\hfill & 0\hfill & 0\hfill & 1\hfill \end{array}\right)$$
(38)
and the full $`𝒰(1,1)`$ matrix realization are
$$\mathrm{\Xi }(v,f,r,a)=\mathrm{\Xi }^{}(a)\mathrm{\Xi }(v,f,r)$$
(39)
The full $`𝒰(1,1)`$ transformation equations are
$$d\stackrel{~}{z}=\mathrm{\Xi }(v,f,r,a)dz$$
(40)
The $`𝒰(1)`$ term does not appear in the classical theory and the transformation equations are therefore restricted to the $`𝒮𝒰(1,1)`$ case. However, what is quite remarkable, is that they appear in an essential way in the nonabelian theory that we shall consider shortly.
#### 2.4.3. Limiting forms
As we have noted above $`\mathrm{\Xi }(v,0,0)=\mathrm{\Lambda }(v)`$ and so in this special case, the reciprocal relativistic transformation equations reduce to the usual special relativity equations.
$$\mathrm{\Xi }(v,0,0)=\mathrm{\Lambda }(v)=\left(1\left(\frac{v}{c}\right)^2\right)^{1/2}\left(\begin{array}{cccc}1\hfill & \frac{v}{c^2}\hfill & 0\hfill & 0\hfill \\ v\hfill & 1\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 1\hfill & \frac{v}{c^2}\hfill \\ 0\hfill & 0\hfill & v\hfill & 1\hfill \end{array}\right)$$
(41)
This is a realization of $`𝒮𝒪(1,1)`$ subgroup on this 4 dimensional space. Note also that
$$\mathrm{\Xi }(0,f,0)=\mathrm{\Lambda }^{}(f)=\left(1\left(\frac{f}{b}\right)^2\right)^{1/2}\left(\begin{array}{cccc}1\hfill & 0\hfill & \frac{f}{b^2}\hfill & 0\hfill \\ 0\hfill & 1\hfill & 0\hfill & \frac{f}{b^2}\hfill \\ f\hfill & 0\hfill & 1\hfill & 0\hfill \\ 0\hfill & f\hfill & 0\hfill & 1\hfill \end{array}\right)$$
(42)
This is also a realization of a different $`𝒮𝒪(1,1)`$ subgroup on this 4 dimensional space. It acts on the $`(t,p)`$ and $`(e,q)`$ subspaces.
Clearly, arranging for $`v=r=0`$ with $`f`$ non-zero will only happen at isolated points on the trajectory. Think of an oscillator describing a circle in $`(p,q)`$ space and a corkscrew when you add in time. There are points on this trajectory where $`f=r=0`$ with $`v`$ non-zero and also points with for $`v=r=0`$ with $`f`$ non-zero. The special relativity case (41) applies for the first case and the reciprocally conjugate case (42) for the second. Note that this latter subgroup leaves invariant the line elements
$$\mathrm{𝑑𝑡}^2+\frac{1}{b^2}\mathrm{𝑑𝑝}^2,\mathrm{𝑑𝑞}^2\frac{1}{b^2}\mathrm{𝑑𝑒}^2$$
Note that these equations reduce as required to the nonrelativistic equations described in Section 2.1. In particular, (32) reduces in the limit to
$$\begin{array}{cc}\hfill \underset{b,c\mathrm{}}{lim}\mathrm{\Xi }(v,f,r)& =\underset{b,c\mathrm{}}{lim}(1w^2)^{1/2}\left(\begin{array}{cccc}1\hfill & \frac{v}{c^2}\hfill & \frac{f}{b^2}\hfill & \frac{r}{b^2c^2}\hfill \\ v\hfill & 1\hfill & \frac{r}{b^2}\hfill & \frac{f}{b^2}\hfill \\ f\hfill & \frac{r}{c^2}\hfill & 1\hfill & \frac{v}{c^2}\hfill \\ r\hfill & f\hfill & v\hfill & 1\hfill \end{array}\right)\hfill \\ & =\left(\begin{array}{cccc}1\hfill & 0\hfill & 0\hfill & 0\hfill \\ v\hfill & 1\hfill & 0\hfill & 0\hfill \\ f\hfill & 0\hfill & 1\hfill & 0\hfill \\ r\hfill & f\hfill & v\hfill & 1\hfill \end{array}\right)=\mathrm{\Phi }(v,f,r)\hfill \end{array}$$
(43)
This is just the expression in (22). The rate of change of position, momentum and energy in (35) in the limit $`b,c\mathrm{}`$ are simply
$$\begin{array}{c}\stackrel{~}{\stackrel{~}{v}}=\stackrel{~}{v}+v\hfill \\ \stackrel{~}{\stackrel{~}{f}}=\stackrel{~}{f}+f\hfill \\ \stackrel{~}{\stackrel{~}{r}}=\stackrel{~}{r}+v\stackrel{~}{f}f\stackrel{~}{v}+r\hfill \end{array}$$
These are precisely the equations given in (26).
We can also look at the case where rates of change of momentum is small with respect to $`b`$ and the rate of change of energy is small with respect to $`bc`$. This is the limit $`b\mathrm{}`$. In this case, we have
$$\begin{array}{cc}\hfill \mathrm{{\rm Y}}(v,f,r)& =\underset{b\mathrm{}}{lim}\mathrm{\Xi }(v,f,r)\hfill \\ & =\left(1\left(\frac{v}{c}\right)^2\right)^{1/2}\left(\begin{array}{cccc}1\hfill & \frac{v}{c^2}\hfill & 0\hfill & 0\hfill \\ v\hfill & 1\hfill & 0\hfill & 0\hfill \\ f\hfill & \frac{r}{c^2}\hfill & 1\hfill & \frac{v}{c^2}\hfill \\ r\hfill & f\hfill & v\hfill & 1\hfill \end{array}\right)\hfill \end{array}$$
(44)
The rate of change of position, momentum and energy are given by
$$\begin{array}{c}\stackrel{~}{\stackrel{~}{v}}=\left(\stackrel{~}{v}+v\right)/\left(1+\frac{v\stackrel{~}{v}}{c^2}\right),\hfill \\ \stackrel{~}{\stackrel{~}{f}}=\left(\stackrel{~}{f}+f+\frac{1}{c^2}\left(r\stackrel{~}{v}v\stackrel{~}{r}\right)\right)/\left(1+\frac{v\stackrel{~}{v}}{c^2}\right),\hfill \\ \stackrel{~}{\stackrel{~}{r}}=\left(\stackrel{~}{r}+rf\stackrel{~}{v}+v\stackrel{~}{f}\right)/\left(1+\frac{v\stackrel{~}{v}}{c^2}\right)\hfill \end{array}$$
(45)
It can also be verified that this defines a matrix group. That is $`\mathrm{{\rm Y}}(\stackrel{~}{v},\stackrel{~}{f},\stackrel{~}{r})\mathrm{{\rm Y}}(v,f,r)=\mathrm{{\rm Y}}(\stackrel{~}{\stackrel{~}{v}},\stackrel{~}{\stackrel{~}{f}},\stackrel{~}{\stackrel{~}{r}})`$ with the addition law defined in (44). Clearly $`\mathrm{\Lambda }(v)=\mathrm{{\rm Y}}(v,0,0)`$ is a subgroup.
The action of $`\mathrm{\Xi }`$ on $`T_{}^{}{}_{z}{}^{}`$, the cotangent vector space spanned by the basis $`\mathrm{𝑑𝑧}=\{\mathrm{𝑑𝑡},\mathrm{𝑑𝑞},\mathrm{𝑑𝑝},\mathrm{𝑑𝑒}\}`$ has no invariant subspaces. The transformations mix all four of the degrees of freedom. Thus it is no longer meaningful to talk about a position-time or space-time manifold. In the limit $`b\mathrm{}`$, corresponding to the physical case of arbitrary velocities but relatively small rates of change of momentum relative to $`b`$ and rates of change of energy relative to $`bc`$, the subspace spanned by $`\{\mathrm{𝑑𝑡},\mathrm{𝑑𝑞}\}`$ is invariant. In this case, all observers have a common view of position-time space, and the usual concept of space-time is meaningful. In the limit $`b,c\mathrm{}`$, corresponding to relatively small rates of change of position, momentum and energy, there are three invariant subspaces, $`\{\mathrm{𝑑𝑡}\}`$, $`\{\mathrm{𝑑𝑡},\mathrm{𝑑𝑞}\}`$ and $`\{\mathrm{𝑑𝑡},\mathrm{𝑑𝑝}\}`$. This means that in the classical limit there is an absolute notion of time on which all observers agree upon and also all observers agree on the position-time subspace and the momentum-time subspace.
#### 2.4.4. Lie Algebra
The Lie algebra may be computed directly as in the nonrelativistic case using $`d\mathrm{\Xi }(v,f,r,a)|_0=dvK+dfN+drM+daU`$
$$\begin{array}{cc}K=\left(\begin{array}{cccc}0\hfill & \frac{1}{c^2}\hfill & 0\hfill & 0\hfill \\ 1\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & \frac{1}{c^2}\hfill \\ 0\hfill & 0\hfill & 1\hfill & 0\hfill \end{array}\right),\hfill & N=\left(\begin{array}{cccc}0\hfill & 0\hfill & \frac{1}{b^2}\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & \frac{1}{b^2}\hfill \\ 1\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 1\hfill & 0\hfill & 0\hfill \end{array}\right)\hfill \\ M=\left(\begin{array}{cccc}0\hfill & 0\hfill & 0\hfill & \frac{1}{c^2b^2}\hfill \\ 0\hfill & 0\hfill & \frac{1}{b^2}\hfill & 0\hfill \\ 0\hfill & \frac{1}{c^2}\hfill & 0\hfill & 0\hfill \\ 1\hfill & 0\hfill & 0\hfill & 0\hfill \end{array}\right),\hfill & U=\left(\begin{array}{cccc}0\hfill & 0\hfill & 0\hfill & \frac{1}{b^2c^2}\hfill \\ 0\hfill & 0\hfill & \frac{1}{b^2}\hfill & 0\hfill \\ 0\hfill & \frac{1}{c^2}\hfill & 0\hfill & 0\hfill \\ 1\hfill & 0\hfill & 0\hfill & 0\hfill \end{array}\right)\hfill \end{array}$$
(46)
with
$$K=\frac{\mathrm{\Xi }}{v},N=\frac{\mathrm{\Xi }}{f},M=\frac{\mathrm{\Xi }}{r},U=\frac{\mathrm{\Xi }}{a}$$
These may also be viewed as an abstract Lie algebra satisfying the commutation relations
$$\begin{array}{c}[K,N]=2M,[M,N]=2K,[M,K]=2N\hfill \\ [U,N]=0,[U,K]=0,[U,M]=0,\hfill \end{array}$$
(47)
Again, the limiting forms
$$F=\underset{b\mathrm{}}{lim}N,\widehat{M}=\underset{b\mathrm{}}{lim}M$$
satisfy the algebra
$$[K,F]=2\widehat{M},[\widehat{M},F]=0,[\widehat{M},K]=2N$$
(48)
and the nonrelativistic case is
$$G=\underset{c\mathrm{}}{lim}K,R=\underset{b,c\mathrm{}}{lim}M=\underset{b,c\mathrm{}}{lim}U$$
with the corresponding algebra given in (29).
$$[G,F]=2R,[R,F]=0,[R,G]=0$$
(49)
#### 2.4.5. Dimensions
The theory presented in the preceding section describes the behavior as the rate of change of momentum approaches a constant $`b`$ and the rate of change of energy approaches $`cb`$ in addition to the usual special relativity behavior as the rate of change of position approaches $`c`$. The constant $`b`$ is a new universal physical constant. It may be defined in terms of the existing constants through a dimensionless parameter $`\alpha _G=\frac{bG}{c^4}`$ as defined in (31). Just as we can define Planck scales of time, position, momentum and energy in terms of $`\{c,\mathrm{},G\}`$, we can define them in terms of $`\{c,\mathrm{},b\}`$as
$$\lambda _t=\sqrt{\frac{\mathrm{}}{bc}},\lambda _q=\sqrt{\frac{\mathrm{}c}{b}},\lambda _p=\sqrt{\frac{\mathrm{}b}{c}},\lambda _e=\sqrt{\mathrm{}bc}$$
(50)
The relationship between the dimensional scales may be conveniently represented in a quad.
$$\begin{array}{ccccc}\lambda _t\hfill & \hfill & c=\lambda _q/\lambda _t\hfill & \hfill & \lambda _q\hfill \\ \hfill & \hfill & & \hfill & \hfill \\ b=\lambda _p/\lambda _t\hfill & & \mathrm{}=\lambda _t\lambda _e=\lambda _q\lambda _p\hfill & & b=\lambda _e/\lambda _q\hfill \\ \hfill & \hfill & & \hfill & \hfill \\ \lambda _p\hfill & \hfill & c=\lambda _e/\lambda _p\hfill & \hfill & \lambda _e\hfill \end{array}$$
(51)
If $`\alpha _G=1`$, then this is simply a rewriting of the usual Planck scales defined in terms of $`\{c,\mathrm{},G\}`$. The basis of the cotangent space $`\{\mathrm{𝑑𝑡},\mathrm{𝑑𝑞},\mathrm{𝑑𝑝},\mathrm{𝑑𝑒}\}`$ and the parameters $`\{v,f,r\}`$ may be made dimensionless simply by dividing by these scales
$$\{d\stackrel{ˇ}{t},d\stackrel{ˇ}{q},d\stackrel{ˇ}{p},d\stackrel{ˇ}{e}\}=\{\frac{1}{\lambda _t}\mathrm{𝑑𝑡},\frac{1}{\lambda _q}\mathrm{𝑑𝑞},\frac{1}{\lambda _p}\mathrm{𝑑𝑝},\frac{1}{\lambda _e}\mathrm{𝑑𝑒}\}$$
(52)
and
$$\{\stackrel{ˇ}{v},\stackrel{ˇ}{f},\stackrel{ˇ}{r}\}=\{\frac{1}{c}v,\frac{1}{b}f,\frac{1}{bc}r\}$$
(53)
Clearly $`\stackrel{ˇ}{v}=1`$ when $`v=c`$ and likewise $`\stackrel{ˇ}{f}=1`$ when $`f=b`$ and $`\stackrel{ˇ}{r}=1`$ when $`r=cb`$. In terms of these dimensionless basis, all of the $`c,b`$ constants that appear in the preceding section may be eliminated by being simply set to 1. For example, (32) becomes
$$\begin{array}{c}d\stackrel{ˇ}{\stackrel{~}{t}}=\left(1\stackrel{ˇ}{w}^2\right)^{1/2}\left(d\stackrel{ˇ}{t}+\stackrel{ˇ}{v}d\stackrel{ˇ}{q}+\stackrel{ˇ}{f}d\stackrel{ˇ}{p}\stackrel{ˇ}{r}d\stackrel{ˇ}{e}\right),\hfill \\ d\stackrel{ˇ}{\stackrel{~}{q}}=\left(1\stackrel{ˇ}{w}^2\right)^{1/2}\left(d\stackrel{ˇ}{q}+\stackrel{ˇ}{v}d\stackrel{ˇ}{t}+\stackrel{ˇ}{r}d\stackrel{ˇ}{p}\stackrel{ˇ}{f}d\stackrel{ˇ}{e}\right),\hfill \\ d\stackrel{ˇ}{\stackrel{~}{p}}=\left(1\stackrel{ˇ}{w}^2\right)^{1/2}\left(d\stackrel{ˇ}{p}+\stackrel{ˇ}{f}d\stackrel{ˇ}{t}+\stackrel{ˇ}{r}d\stackrel{ˇ}{q}+\stackrel{ˇ}{v}d\stackrel{ˇ}{e}\right)\hfill \\ d\stackrel{ˇ}{\stackrel{~}{e}}=(1\stackrel{ˇ}{w}^2)^{1/2}(d\stackrel{ˇ}{e}\stackrel{ˇ}{f}d\stackrel{ˇ}{q}+\stackrel{ˇ}{v}d\stackrel{ˇ}{p}+\stackrel{ˇ}{r}d\stackrel{ˇ}{t})\hfill \end{array}$$
(54)
with $`\stackrel{ˇ}{w}=\stackrel{ˇ}{v}^2+\stackrel{ˇ}{f}^2\stackrel{ˇ}{r}^2`$. This is true for all the equations with respect to the natural scales. In this sense, the constants $`c,b`$, and as we will see, $`\mathrm{}`$ are no different than the constants we would have to introduce if we defined the scales in the $`{}_{}{}^{}x_{}^{}`$ position direction to be feet and the $`{}_{}{}^{}y_{}^{}`$ position direction to be meters. Using dimensionless quantities defines in terms of these constants eliminates their appearance in all the equations. In particular, the symplectic and orthogonal metrics are simply
$$\begin{array}{c}\mathrm{𝑑𝑠}^2={}_{}{}^{t}dz\eta dz=d\stackrel{ˇ}{t}^2+d\stackrel{ˇ}{q}^2+d\stackrel{ˇ}{p}^2d\stackrel{ˇ}{e}^2\hfill \\ {}_{}{}^{t}dz\zeta dz=d\stackrel{ˇ}{e}d\stackrel{ˇ}{t}+d\stackrel{ˇ}{p}d\stackrel{ˇ}{q}\hfill \end{array}$$
## 3. Inhomogeneous group with Heisenberg nonabelian ‘translations’
### 3.1. Translation group and inhomogeneous groups
The discussion up until this point has been concerned with the homogeneous group acting as a transformation group on the co-tangent space. We can cast these transformations in purely group theoretic terms by introducing the notions of a semidirect product group.
Consider a Lie group $`𝒢`$, a normal closed subgroup $`𝒩𝒢`$ has the property that, for all $`g𝒢`$ and $`n𝒩`$, that $`g^1ng𝒩`$. If in addition there is another closed subgroup $`𝒦𝒢`$ with $`𝒦𝒩=𝒢`$ and $`𝒩𝒦=e`$ where $`e`$ is the trivial group containing only the identity, then $`𝒢`$ is the semidirect product group $`𝒢=𝒦_s𝒩`$. Note that as the automorphism group $`𝒜ut(𝒩)`$ of $`𝒩`$ is the group of elements with the property that for $`n𝒩`$, $`g^1ng𝒩`$. Clearly $`𝒢𝒜ut(𝒩)`$.
The group $`𝒩`$ has an algebra that may be identified with the Lie algebra valued one forms $`dn𝒂(𝒩)T_e𝒩`$. The group $`k𝒦`$ acts this algebra through the adjoint action $`k^1dnk𝒂(𝒩)`$. If $`k=e^{dk}`$, then the infinitesimal transformations are given by $`d\stackrel{~}{n}=dn+[dk,dn]`$.
The four dimensional translation group $`𝒯(4)`$ may be realized as a $`5\times 5`$ matrix group.
$$\mathrm{T}(\stackrel{ˇ}{t},\stackrel{ˇ}{q},\stackrel{ˇ}{p},\stackrel{ˇ}{e})=\left(\begin{array}{ccccc}1\hfill & 0\hfill & 0\hfill & 0\hfill & \stackrel{ˇ}{t}\hfill \\ 0\hfill & 1\hfill & 0\hfill & 0\hfill & \stackrel{ˇ}{q}\hfill \\ 0\hfill & 0\hfill & 1\hfill & 0\hfill & \stackrel{ˇ}{p}\hfill \\ 0\hfill & 0\hfill & 0\hfill & 1\hfill & \stackrel{ˇ}{e}\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 1\hfill \end{array}\right)$$
(55)
The algebra is given by
$$\begin{array}{cc}\hfill d\mathrm{T}(\stackrel{ˇ}{t},\stackrel{ˇ}{q},\stackrel{ˇ}{p},\stackrel{ˇ}{e})& =d\stackrel{ˇ}{t}T+d\stackrel{ˇ}{q}Q+d\stackrel{ˇ}{p}P+d\stackrel{ˇ}{e}E\hfill \\ & =\frac{1}{\lambda _t}dtT+\frac{1}{\lambda _q}dqQ+\frac{1}{\lambda _p}dpP+\frac{1}{\lambda _e}deE\hfill \end{array}$$
(56)
where the basis $`\{T,Q,P,E\}`$ of the abelian algebra $`𝒂(𝒯(4))`$ have the matrix realizations
$$\begin{array}{cc}T=\left(\begin{array}{ccccc}0\hfill & 0\hfill & 0\hfill & 0\hfill & 1\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \end{array}\right)\hfill & Q=\left(\begin{array}{ccccc}0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 1\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \end{array}\right)\hfill \\ P=\left(\begin{array}{ccccc}0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 1\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \end{array}\right)\hfill & E=\left(\begin{array}{ccccc}0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 1\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \end{array}\right)\hfill \end{array}$$
(57)
More compactly, with $`\{Z_\alpha \}=\{T,Q,P,E\}`$ and $`\{z^\alpha \}=\{\stackrel{ˇ}{t},\stackrel{ˇ}{q},\stackrel{ˇ}{p},\stackrel{ˇ}{e}\}`$,
$$d\mathrm{T}(z)=dz^\alpha Z_\alpha =\left(\begin{array}{cc}0\hfill & dz\hfill \\ 0\hfill & 0\hfill \end{array}\right)$$
Consider the semidirect product group $`𝒢=𝒦_s𝒯(4)`$. The automorphism group of $`𝒯(n)`$ is $`𝒢(n)_s𝒯(n)`$ and therefore $`𝒦𝒢(4)`$. Therefore elements $`k𝒦`$ may be realized by nonsingular $`4\times 4`$ matrices $`K`$ and elements $`g𝒢`$ may be realized by the $`5\times 5`$ matrices $`\mathrm{\Gamma }`$
$$\mathrm{\Gamma }(K,z)=\left(\begin{array}{cc}K\hfill & Kz\hfill \\ 0\hfill & 1\hfill \end{array}\right)$$
(58)
with group product $`\mathrm{\Gamma }(\stackrel{~}{K},\stackrel{~}{z})\mathrm{\Gamma }(K,z)=\mathrm{\Gamma }(\stackrel{~}{K}K,\stackrel{~}{K}z+\stackrel{~}{z})`$ and inverse $`\mathrm{\Gamma }^1(K,z)=\mathrm{\Gamma }(K^1,z)`$.
#### 3.1.1. Lie algebra
Now, we can consider, in particular, the group $`𝒦𝒰(1,1)`$ and set $`\mathrm{K}=\mathrm{\Xi }(v,f,r,a)`$ from (39). The action of $`\mathrm{\Xi }`$ on the element of the algebra $`d\mathrm{T}(z)`$ is
$$\begin{array}{cc}\hfill \mathrm{\Xi }d\mathrm{T}(z)\mathrm{\Xi }^1& =\left(\begin{array}{cc}\mathrm{\Xi }\hfill & 0\hfill \\ 0\hfill & 1\hfill \end{array}\right)\left(\begin{array}{cc}0\hfill & dz\hfill \\ 0\hfill & 0\hfill \end{array}\right)\left(\begin{array}{cc}\mathrm{\Xi }^1\hfill & 0\hfill \\ 0\hfill & 1\hfill \end{array}\right)\hfill \\ & =\left(\begin{array}{cc}0\hfill & \mathrm{\Xi }dz\hfill \\ 0\hfill & 0\hfill \end{array}\right)=d\mathrm{T}(\stackrel{~}{z})\hfill \end{array}$$
(59)
This is precisely the transformation equations given in (32), $`d\stackrel{~}{z}=\mathrm{\Xi }\mathrm{𝑑𝑧}`$.
The generators $`\{Z_\alpha \}=\{T,Q,P,E\}`$ are given in (57) and the generators $`\{K,N,M,U\}`$ of the Lie algebra are given in (46) with the embedding of the $`4\times 4`$ matrices in the $`5\times 5`$ matrices. The nonzero generators of the algebra of $`𝒰(1,1)_s𝒯(4)`$ are $`d\mathrm{\Xi }(v,f,r,a)|_0=dvK+dfN+drM+daU`$
$$\begin{array}{cccc}[K,N]=2M,\hfill & [M,N]=2K,\hfill & [M,K]=2N\hfill & \\ [K,T]=Q,\hfill & [K,Q]=T,\hfill & [K,P]=E,\hfill & [K,E]=P,\hfill \\ [N,T]=P,\hfill & [N,Q]=E,\hfill & [N,P]=T,\hfill & [U,E]=Q,\hfill \\ [M,T]=E,\hfill & [M,Q]=P,\hfill & [M,P]=Q,\hfill & [M,E]=T,\hfill \\ [U,T]=E,\hfill & [U,Q]=P,\hfill & [U,P]=Q,\hfill & [U,E]=T,\hfill \end{array}$$
(60)
The infinitesimal transformation equations of (32) are then given by the action of the algebra of the $`𝒮𝒰(1,1)`$ subgroup
$$\begin{array}{cc}\hfill d\mathrm{T}(\stackrel{~}{z})& =d\mathrm{T}(z)+[d\mathrm{\Xi }(v,f,r)|_0,d\mathrm{T}(z)]\hfill \\ & =dz^\alpha Z_\alpha +dz^\alpha [\mathrm{𝑑𝑣}K+d\mathrm{𝑓𝑁}+drM,Z_\alpha ]\hfill \\ & =dz^\alpha (Z_\alpha +dv[K,Z_\alpha ]+df[N,Z_\alpha ]+dr[M,Z_\alpha ])\hfill \end{array}$$
(61)
This gives the expected result for the infinitesimal transformations
$$\begin{array}{c}d\stackrel{~}{t}=\mathrm{𝑑𝑡}+\frac{1}{c^2}dv\mathrm{𝑑𝑞}+\frac{1}{b^2}df\mathrm{𝑑𝑝}\frac{1}{b^2c^2}dr\mathrm{𝑑𝑒},\hfill \\ d\stackrel{~}{q}=\mathrm{𝑑𝑞}+dv\mathrm{𝑑𝑡}+\frac{1}{b^2}dr\mathrm{𝑑𝑝}\frac{1}{b^2}df\mathrm{𝑑𝑒},\hfill \\ d\stackrel{~}{p}=\mathrm{𝑑𝑝}+df\mathrm{𝑑𝑡}\frac{1}{c^2}dr\mathrm{𝑑𝑞}+\frac{1}{c^2}dv\mathrm{𝑑𝑒},\hfill \\ d\stackrel{~}{e}=\mathrm{𝑑𝑒}df\mathrm{𝑑𝑞}+dv\mathrm{𝑑𝑝}+drdt.\hfill \end{array}$$
(62)
The lowest order Casimir invariant for this algebra is $`T^2+Q^2+P^2E^2`$ which is precisely the form of the orthogonal metric. This enables all the essential properties to be formulated in purely group theoretic terms.
### 3.2. Heisenberg group
The theory considered so far does not take into account the nonabelian nature of phase space. Physical observations tell us that the position and momentum degrees of freedom and the time and energy degrees of freedom cannot be measured simultaneously. Mathematically, this means that the abelian group $`𝒯(4)`$ in the previous section must be replaced by a Heisenberg group $`(2)=𝒯(2)_s𝒯(3)`$. This is a 5 dimensional group with an algebra that has the generators $`\{T,Q,P,E,I\}`$ satisfying the Lie algebra
$$[P,Q]=I,[T,E]=I$$
(63)
The group $`(2)`$ is just a matrix group like all the groups we have encountered so far. It can be realized by the $`6\times 6`$ matrices
$$\mathrm{H}(\stackrel{ˇ}{t},\stackrel{ˇ}{q},\stackrel{ˇ}{p},\stackrel{ˇ}{e},\iota )=\left(\begin{array}{cccccc}1\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & \stackrel{ˇ}{t}\hfill \\ 0\hfill & 1\hfill & 0\hfill & 0\hfill & 0\hfill & \stackrel{ˇ}{q}\hfill \\ 0\hfill & 0\hfill & 1\hfill & 0\hfill & 0\hfill & \stackrel{ˇ}{p}\hfill \\ 0\hfill & 0\hfill & 0\hfill & 1\hfill & 0\hfill & \stackrel{ˇ}{e}\hfill \\ \stackrel{ˇ}{e}\hfill & \stackrel{ˇ}{p}\hfill & \stackrel{ˇ}{q}\hfill & \stackrel{ˇ}{t}\hfill & 1\hfill & 2\iota \hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 1\hfill \end{array}\right)$$
(64)
This may be written more compactly as
$$\mathrm{H}(z,\iota )=\left(\begin{array}{ccc}I\hfill & 0\hfill & z\hfill \\ {}_{}{}^{t}z\zeta \hfill & 1\hfill & 2\iota \hfill \\ 0\hfill & 0\hfill & 1\hfill \end{array}\right)$$
(65)
It follows that the group product and inverse are
$$\mathrm{H}(\stackrel{~}{z},\stackrel{~}{\iota })\mathrm{H}(z,\iota )=\mathrm{H}(\stackrel{~}{z}+z,\stackrel{~}{\iota }+\iota +\frac{1}{2}{}_{}{}^{t}\stackrel{~}{z}\zeta z),\text{ }\mathrm{H}(z,\iota )^1=\mathrm{H}(z,\iota )$$
(66)
and the algebra is
$$d\mathrm{H}(z,\iota )=\left(\begin{array}{ccc}0\hfill & 0\hfill & dz\hfill \\ {}_{}{}^{t}(\zeta dz)\hfill & 0\hfill & 2d\iota \hfill \\ 0\hfill & 0\hfill & 0\hfill \end{array}\right)=dz^\alpha Z_\alpha +d\iota I$$
(67)
Expanding out again, this means explicitly that the generators are given by the $`6\times 6`$ matrices
$$\begin{array}{cc}T=\left(\begin{array}{cccccc}0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 1\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 1\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \end{array}\right)\hfill & Q=\left(\begin{array}{cccccc}0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 1\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 1\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \end{array}\right)\hfill \\ P=\left(\begin{array}{cccccc}0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 1\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 1\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \end{array}\right)\hfill & E=\left(\begin{array}{cccccc}0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 1\hfill \\ 1\hfill & 0\hfill & 0\hfill & 1\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \end{array}\right)\hfill \\ I=\left(\begin{array}{cccccc}0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 2\hfill \\ 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill & 0\hfill \end{array}\right)\hfill & \end{array}$$
(68)
These matrices satisfy the algebra of the algebra of the Heisenberg matrix Lie group. In quantum mechanics where the Heisenberg algebra normally appears, we are using the unitary representations of the group on a Hilbert space. In general an element of a Lie group $`g𝒢`$ is given in the neighborhood of the identity by $`g=e^X`$ where $`X𝒂(𝒢)`$ is an element of the algebra. A unitary representation $`\varrho `$ of the group as unitary operators on a Hilbert space $`𝑯^\varrho `$. The unitary irreducible representations determine the Hilbert space and so it is labelled by the representation. The unitary operators $`\varrho (g)^{}=\varrho (g)^1`$induce anti-Hermitian representations $`\varrho ^{}(X)^{}=\varrho ^{}(X)`$ of the algebra. Inserting an $`i`$ maps these onto Hermitian operators $`\varrho (g)=e^{i\varrho (g)}`$ normally used in quantum mechanics. Then if the Lie algebra is $`[X,Y]=Z`$, then the Lie algebra of the Hermitian representation is $`[\varrho ^{}(X),\varrho ^{}(Y)]=i\varrho ^{}(Z)`$. This is where the $`i`$ comes from in the Heisenberg algebra as we generally deal with the unitary representations in a quantum mechanics context where the Hilbert space is $`𝑯^\varrho =L^2(,)`$ and the representations $`\widehat{Q}=\varrho ^{}(Q),`$$`\widehat{P}=\varrho ^{}(P)`$ and $`\widehat{I}=\varrho ^{}(I)`$ of the algebra satisfy $`[\widehat{P},\widehat{Q}]=i\mathrm{}\widehat{I}`$ and are realized in the position diagonal basis by the Hermitian operators $`\widehat{Q}=q`$ and $`\widehat{P}=i\mathrm{}\frac{}{q}`$ or the momentum diagonal basis by $`\widehat{Q}=i\mathrm{}\frac{}{p}`$ and $`\widehat{P}=p`$ .
The Heisenberg group itself, before the unitary representations are considered, is just a straightforward real matrix group as are the rotation, symplectic and translation groups with which we are familiar.
### 3.3. Automorphisms of the Heisenberg group
We can now proceed as in the case of the translation group and construct a semidirect product group that contains the Heisenberg group as the normal subgroup. This group must be a subgroup of the automorphism group of the Heisenberg group. That is for $`a𝒜ut((2))`$, $`a^1ha(2)`$ for all $`h(2)`$. Constructing a general $`6\times 6`$ matrix and computing the product shows that the automorphism group must be of the form
$$𝒜ut((2))𝒟_s\left(𝒜(1)_s\left(𝒮p(4)_s(2)\right)\right)$$
(69)
$`𝒟`$ is the discrete group of automorphisms that is considered further in Section 4.5 below. $`𝒜`$ is the abelian group of dilations that are represented by the $`6\times 6`$ matrices $`A(ϵ)𝒜`$
$$A(ϵ)=\left(\begin{array}{ccc}I\hfill & 0\hfill & 0\hfill \\ 0\hfill & e^ϵ\hfill & 0\hfill \\ 0\hfill & 0\hfill & e^ϵ\hfill \end{array}\right)$$
(70)
The product is $`A(\stackrel{~}{ϵ})A(ϵ)=A(\stackrel{~}{ϵ}+ϵ)`$ and inverse $`A(ϵ)^1=A(ϵ)`$. The action as an automorphism of the Heisenberg group is
$$A(ϵ)\mathrm{H}(z,\iota )A(ϵ)^1=\left(\begin{array}{ccc}I\hfill & 0\hfill & e^ϵz\hfill \\ {}_{}{}^{t}(\zeta e^ϵz)\hfill & 1\hfill & 2e^{2ϵ}\iota \hfill \\ 0\hfill & 0\hfill & 1\hfill \end{array}\right)=\mathrm{H}(e^ϵz,e^{2ϵ}\iota )$$
(71)
The remaining automorphisms are the $`6\times 6`$ matrices $`\mathrm{{\rm Y}}𝒮p(4)(2)`$
$$\mathrm{{\rm Y}}(K,z,\iota )=\left(\begin{array}{ccc}K\hfill & 0\hfill & Kz\hfill \\ {}_{}{}^{t}z\zeta \hfill & 1\hfill & 2\iota \hfill \\ 0\hfill & 0\hfill & 1\hfill \end{array}\right)$$
(72)
where $`K𝒮p(4)`$ are the $`4\times 4`$ matrices with the property that $`{}_{}{}^{t}K\zeta K=\zeta `$. This property may be used to determine the group multiplication and inverse. As this is a matrix group, the group product and inverse is computed directly from the matrix product and inverse is calculated us
$$\mathrm{{\rm Y}}(\stackrel{~}{K},\stackrel{~}{z},\stackrel{~}{\iota })\mathrm{{\rm Y}}(K,z,\iota )=\mathrm{{\rm Y}}(\stackrel{~}{K}K,\stackrel{~}{z}+\stackrel{~}{K}z,\iota +\stackrel{~}{\iota }+\frac{1}{2}{}_{}{}^{t}\stackrel{~}{z}\zeta z)$$
$$\mathrm{{\rm Y}}(K,z,\iota )^1=\mathrm{{\rm Y}}(K^1,z,\iota )$$
Finally the automorphisms are
$$\begin{array}{cc}\hfill \mathrm{{\rm Y}}(\stackrel{~}{K},\stackrel{~}{z},\stackrel{~}{\iota })\mathrm{H}(z,\iota )\mathrm{{\rm Y}}(\stackrel{~}{K},\stackrel{~}{z},\stackrel{~}{\iota })^1& =\left(\begin{array}{ccc}I\hfill & 0\hfill & \stackrel{~}{K}z\hfill \\ {}_{}{}^{t}z\zeta \stackrel{~}{K}^1\hfill & 1\hfill & 2\left(\iota +{}_{}{}^{t}\stackrel{~}{z}\zeta z\right)\hfill \\ 0\hfill & 0\hfill & 1\hfill \end{array}\right)\hfill \end{array}$$
(73)
This is an element of $`(2)`$ only if $`{}_{}{}^{t}z\zeta \stackrel{~}{K}^1={}_{}{}^{t}(\stackrel{~}{K}z)\zeta ={}_{}{}^{t}z{}_{}{}^{t}\stackrel{~}{K}\zeta `$. That is, $`\zeta \stackrel{~}{K}^1={}_{}{}^{t}\stackrel{~}{K}\zeta `$ or equivalently $`\zeta ={}_{}{}^{t}\stackrel{~}{K}\zeta \stackrel{~}{K}`$. This is the condition for $`\stackrel{~}{K}(2)`$ and the automorphisms are then
$$\begin{array}{cc}\hfill \mathrm{{\rm Y}}(\stackrel{~}{K},\stackrel{~}{z},\stackrel{~}{\iota })\mathrm{H}(z,\iota )\mathrm{{\rm Y}}(\stackrel{~}{K},\stackrel{~}{z},\stackrel{~}{\iota })^1& =\mathrm{H}(\stackrel{~}{K}z,\iota +{}_{}{}^{t}\stackrel{~}{z}\zeta z)\hfill \end{array}$$
(74)
### 3.4. Quaplectic group
Then, the group $`𝒬(1,1)=𝒰(1,1)_s(2)`$ is a subgroup of the group of continuous automorphisms with elements realized by the matrices
$$\mathrm{\Theta }=\left(\begin{array}{ccc}\mathrm{\Xi }\hfill & 0\hfill & \mathrm{\Xi }z\hfill \\ {}_{}{}^{t}z\zeta \hfill & 1\hfill & 2\iota \hfill \\ 0\hfill & 0\hfill & 1\hfill \end{array}\right)$$
(75)
This is an element of the quaplectic group in a matrix realization.
The full group including the discrete transformations and the scaling transformations is the extended quaplectic group $`\widehat{𝒬}(1,1)=(𝒟𝒜b𝒰(1,1))_s(2)`$ with elements realized by the matrices
$$\widehat{\mathrm{\Theta }}=\varsigma \left(\begin{array}{ccc}\mathrm{\Xi }\hfill & 0\hfill & \mathrm{\Xi }z\hfill \\ e^ϵ{}_{}{}^{t}z\zeta \hfill & e^ϵ\hfill & 2e^ϵ\iota \hfill \\ 0\hfill & 0\hfill & e^ϵ\hfill \end{array}\right)=\varsigma A\mathrm{\Xi }\mathrm{H}$$
(76)
where $`\varsigma 𝒟`$ is an element of the finite abelian discrete group that is defined in the following section.
The quaplectic group gives the transformation equations
$$\begin{array}{cc}\hfill \mathrm{\Xi }d\mathrm{H}(z)\mathrm{\Xi }^1& =\left(\begin{array}{ccc}\mathrm{\Xi }\hfill & 0\hfill & 0\hfill \\ 0\hfill & 1\hfill & 0\hfill \\ 0\hfill & 0\hfill & 1\hfill \end{array}\right)\left(\begin{array}{ccc}0\hfill & 0\hfill & dz\hfill \\ {}_{}{}^{t}dz\zeta \hfill & 0\hfill & 2d\iota \hfill \\ 0\hfill & 0\hfill & 0\hfill \end{array}\right)\left(\begin{array}{ccc}\mathrm{\Xi }^1\hfill & 0\hfill & 0\hfill \\ 0\hfill & 1\hfill & 0\hfill \\ 0\hfill & 0\hfill & 1\hfill \end{array}\right)\hfill \\ & =\left(\begin{array}{ccc}0\hfill & 0\hfill & \mathrm{\Xi }\mathrm{𝑑𝑧}\hfill \\ {}_{}{}^{t}(\mathrm{\Xi }\mathrm{𝑑𝑧})\zeta \hfill & 0\hfill & 2d\iota \hfill \\ 0\hfill & 0\hfill & 0\hfill \end{array}\right)=d\mathrm{H}(\stackrel{~}{z})\hfill \end{array}$$
(77)
where we have used $`{}_{}{}^{t}dz\zeta \mathrm{\Xi }^1={}_{}{}^{t}dz{}_{}{}^{t}\mathrm{\Xi }{}_{}{}^{t}\mathrm{\Xi }_{}^{1}\zeta \mathrm{\Xi }^1={}_{}{}^{t}(\mathrm{\Xi }\mathrm{𝑑𝑧})\zeta `$ as $`{}_{}{}^{t}\mathrm{\Xi }\zeta \mathrm{\Xi }=\zeta `$ as $`\mathrm{\Xi }𝒮p(4)`$ and therefore $`{}_{}{}^{t}\mathrm{\Xi }_{}^{1}\zeta \mathrm{\Xi }^1=\zeta `$ with $`\zeta ^1=\zeta `$. These are precisely the transformation equations $`d\stackrel{~}{z}=\mathrm{\Xi }\mathrm{𝑑𝑧}`$ in (40).
#### 3.4.1. Lie Algebra
The Lie algebra of the quaplectic group is
$$\begin{array}{c}d\mathrm{\Theta }=d\mathrm{\Xi }+dz^\alpha Z_\alpha +d\iota I\hfill \\ d\mathrm{\Xi }=dvK+dfN+drM+daU\hfill \end{array}$$
(78)
The generators $`\{Z_\alpha ,I\}=\{T,Q,P,E,I\}`$ are given in (68) and the generators $`\{K,N,M,M^{}\}`$ are given in (46) with the obvious embedding of the $`4\times 4`$ matrices in the $`6\times 6`$ matrices. The nonzero generators of the full quaplectic algebra are
$$\begin{array}{cccc}[K,N]=2M,\hfill & [M,N]=2K,\hfill & [M,K]=2N\hfill & \\ [K,T]=Q,\hfill & [K,Q]=T,\hfill & [K,P]=E,\hfill & [K,E]=P,\hfill \\ [N,T]=P,\hfill & [N,Q]=E,\hfill & [N,P]=T,\hfill & [U,E]=Q,\hfill \\ [M,T]=E,\hfill & [M,Q]=P,\hfill & [M,P]=Q,\hfill & [M,E]=T,\hfill \\ [U,T]=E,\hfill & [U,Q]=P,\hfill & [U,P]=Q,\hfill & [U,E]=T,\hfill \\ [P,Q]=I,\hfill & [E,T]=I\hfill & & \end{array}$$
(79)
The infinitesimal transformation equations of (32) are then given by
$$\begin{array}{cc}\hfill d\mathrm{H}(\stackrel{~}{z})& =d\mathrm{H}(z)+[d\mathrm{\Xi }(v,f,r,a)|_0,d\mathrm{H}(z)]\hfill \\ & =dz^\alpha Z_\alpha +\mathrm{𝑑𝑧}^\alpha [\mathrm{𝑑𝑣}K+d\mathrm{𝑓𝑁}+\mathrm{dr}M+\mathrm{𝑑𝑎𝑈},Z_\alpha ]\hfill \\ & =dz^\alpha (Z_\alpha +dv[K,Z_\alpha ]+df[N,Z_\alpha ]+dr[M,Z_\alpha ]+da[U,Z_\alpha ])\hfill \end{array}$$
(80)
This gives the expected result on the non abelian manifold
$$\begin{array}{c}d\stackrel{~}{t}=\mathrm{𝑑𝑡}+\frac{1}{c^2}dv\mathrm{𝑑𝑞}+\frac{1}{b^2}df\mathrm{𝑑𝑝}\frac{1}{b^2c^2}\left(dr+da\right)\mathrm{𝑑𝑒},\hfill \\ d\stackrel{~}{q}=\mathrm{𝑑𝑞}+dv\mathrm{𝑑𝑡}+\frac{1}{b^2}\left(drda\right)\mathrm{𝑑𝑝}\frac{1}{b^2}df\mathrm{𝑑𝑒},\hfill \\ d\stackrel{~}{p}=\mathrm{𝑑𝑝}+df\mathrm{𝑑𝑡}\frac{1}{c^2}\left(drda\right)\mathrm{𝑑𝑞}+\frac{1}{c^2}dv\mathrm{𝑑𝑒},\hfill \\ d\stackrel{~}{e}=\mathrm{𝑑𝑒}df\mathrm{𝑑𝑞}+dv\mathrm{𝑑𝑝}+\left(dr+da\right)dt.\hfill \end{array}$$
(81)
In this case however, the lowest order Casimir invariant is simply $`C_1=I`$ as it commutes with all the generators and the second order Casimir invariant is
$$C_2=\frac{1}{2}\left(T^2Q^2+P^2E^2\right)IU$$
(82)
This can be viewed as a metric on a 6 dimensional space (or with the $`n=3`$ case, a $`10`$ dimensional space.) The additional term is required as the $`Q`$ and $`P`$ and the $`T`$ and $`E`$ do not commute. However, this additional generator associated with the $`𝒰(1)`$ subgroup provides precisely the term to cancel out the resulting term when considering Lie brackets of the form $`[Z_\alpha ,C_2]`$. That is
$$[T,C_2]=2\frac{1}{2}E[T,E]I[T,U]=EI+IE=0$$
(83)
and so forth.
This is a very essential change in the structure of the theory. One of the most profound of these is that the abelian theory with a hermitian metric does not admit a non-trivial manifold structure to enable the mathematical development of a general theory on this space corresponding to the general relativity generalization of special relativity. This is the no go theorem of Schuller . However, this nonabelian theory does not appear to be constrained by this no go theorem and it is possible to investigate generalizations to general curved noncommutative manifolds. The construction of a nonabelian geometry where locally the nonabelian Heisenberg group are the local translations that generalize to a nonabelian connection is a very interesting follow on problem.
### 3.5. Discrete transformations
The parity, time reversal and charge conjugation (PCT) discrete transformations play an important role in the standard special relativistic theory. These transformations
$$\begin{array}{cc}\varsigma _P\{\mathrm{𝑑𝑡},\mathrm{𝑑𝑞}\}=\{\mathrm{𝑑𝑡},\mathrm{𝑑𝑞}\},\hfill & \varsigma _T\{\mathrm{𝑑𝑡},\mathrm{𝑑𝑞}\}=\{\mathrm{𝑑𝑡},\mathrm{𝑑𝑞}\},\hfill \\ \varsigma _C\{\mathrm{𝑑𝑡},\mathrm{𝑑𝑞}\}=\{dt,dq\}\hfill & \end{array}$$
(84)
are a discrete abelian group satisfying $`\varsigma _{P}^{}{}_{}{}^{2}=\varsigma _{T}^{}{}_{}{}^{2}=\varsigma _{C}^{}{}_{}{}^{2}=\varsigma _0`$, $`\varsigma _P\varsigma _T=\varsigma _C`$, $`\varsigma _C\varsigma _P=\varsigma _T`$, $`\varsigma _T\varsigma _C=\varsigma _P`$. These transformations leave the Lorentz metric invariant and are automorphisms of the group
$$\varsigma _P\mathrm{\Lambda }(v)\varsigma _{P}^{}{}_{}{}^{1}=\mathrm{\Lambda }(v),\varsigma _T\mathrm{\Lambda }(v)\varsigma _{T}^{}{}_{}{}^{1}=\mathrm{\Lambda }(v),\varsigma _C\mathrm{\Lambda }(v)\varsigma _{C}^{}{}_{}{}^{1}=\mathrm{\Lambda }(v)$$
(85)
These transformations carry over directly to the $`𝒰(1,3)`$ group of discrete automorphisms $`𝒟^{}`$. Again, define $`dz=\{\mathrm{𝑑𝑡},\mathrm{𝑑𝑞},\mathrm{𝑑𝑝},\mathrm{𝑑𝑒}\}`$,
$$\begin{array}{cc}\varsigma _P\left(\mathrm{𝑑𝑧}\right)=\{\mathrm{𝑑𝑡},\mathrm{dq},\mathrm{𝑑𝑝},\mathrm{𝑑𝑒}\},\hfill & \varsigma _T\left(\mathrm{𝑑𝑧}\right)=\{\mathrm{𝑑𝑡},\mathrm{𝑑𝑞},\mathrm{𝑑𝑝},\mathrm{𝑑𝑒}\},\hfill \\ \varsigma _C\left(\mathrm{𝑑𝑧}\right)=\{\mathrm{𝑑𝑡},\mathrm{𝑑𝑞},\mathrm{𝑑𝑝},\mathrm{𝑑𝑒}\}\hfill & \end{array}$$
(86)
These satisfy the above group product relations given above in (84). There are now in addition 3 new discrete transformations, the Born reciprocity transformations (using units $`b=c=\mathrm{}=1)`$
$$\begin{array}{cc}\varsigma _E\left(\mathrm{𝑑𝑧}\right)=\{\mathrm{𝑑𝑒},\mathrm{𝑑𝑞},\mathrm{𝑑𝑝},\mathrm{𝑑𝑡}\}\hfill & \varsigma _Q\left(\mathrm{𝑑𝑧}\right)=\{\mathrm{𝑑𝑡},\mathrm{𝑑𝑝},\mathrm{𝑑𝑞},\mathrm{𝑑𝑒}\},\hfill \\ \varsigma _R\left(\mathrm{𝑑𝑧}\right)=\{\mathrm{𝑑𝑒},\mathrm{𝑑𝑝},\mathrm{𝑑𝑞},\mathrm{𝑑𝑡}\}\hfill & \end{array}$$
(87)
The group products for these transformations are $`\varsigma _{R}^{}{}_{}{}^{2}=\varsigma _C`$, $`\varsigma _{Q}^{}{}_{}{}^{2}=\varsigma _P`$, $`\varsigma _{E}^{}{}_{}{}^{2}=\varsigma _T`$, $`\varsigma _Q\varsigma _E=\varsigma _R`$, $`\varsigma _R\varsigma _Q=\varsigma _E`$, $`\varsigma _E\varsigma _R=\varsigma _Q`$.
These then multiplied with the PCT transformations to define 6 additional elements $`\varsigma _{\alpha \beta }=\varsigma _\alpha \varsigma _\beta `$ with $`\alpha \{P,T,C\}`$ and $`\beta \{Q,E,R\}`$ labels of the abelian group. The transformations equations for these follow immediately from (86) and (87). With these 13 elements, the abelian discrete group closes. The multiplication table is simply worked out using these multiplication rules
$$\varsigma _{\alpha \beta }\varsigma _{\gamma \delta }=\varsigma _\alpha \varsigma _\beta \varsigma _\gamma \varsigma _\delta =\left(\varsigma _\alpha \varsigma _\gamma \right)\left(\varsigma _\beta \varsigma _\delta \right),$$
(88)
where $`\alpha ,\gamma \{P,T,C\}`$ and $`\beta ,\delta \{Q,E,R\}`$.
The group elements $`\varsigma _P,\varsigma _T,\varsigma _Q,\varsigma _E`$ may be realized by the $`4\times 4`$ matrices.
$$\begin{array}{cc}\varsigma _P\left(\begin{array}{cccc}1\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 1\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 1\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 1\hfill \end{array}\right),\hfill & \varsigma _T\left(\begin{array}{cccc}1\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 1\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 1\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 1\hfill \end{array}\right),\hfill \\ \varsigma _Q\left(\begin{array}{cccc}1\hfill & 0\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 1\hfill & 0\hfill \\ 0\hfill & 1\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 0\hfill & 1\hfill \end{array}\right),\hfill & \varsigma _E\left(\begin{array}{cccc}0\hfill & 0\hfill & 0\hfill & 1\hfill \\ 0\hfill & 1\hfill & 0\hfill & 0\hfill \\ 0\hfill & 0\hfill & 1\hfill & 0\hfill \\ 1\hfill & 0\hfill & 0\hfill & 0\hfill \end{array}\right)\hfill \end{array}$$
(89)
with $`\varsigma _0=I,\varsigma _C=I`$ and $`\varsigma _R=\zeta `$. The remaining matrices may be computed from these four elements using the group multiplication rules given above as these four elements generate the group. These 4 elements generate the group and so we need consider only these elements further. Elements of the discrete group leave invariant the symplectic and Born-Green orthogonal metrics and are automorphisms of $`𝒰(1,3)`$group.
$$\begin{array}{c}\varsigma _P\mathrm{\Xi }(v,f,r,a)\varsigma _{P}^{}{}_{}{}^{1}=\mathrm{\Lambda }(v,f,r,a),\hfill \\ \varsigma _T\mathrm{\Lambda }(v,f,r,a)\varsigma _{T}^{}{}_{}{}^{1}=\mathrm{\Lambda }(v,f,r,a),\hfill \\ \varsigma _C\mathrm{\Lambda }(v,f,r,a)\varsigma _{C}^{}{}_{}{}^{1}=\mathrm{\Lambda }(v,f,r,a),\hfill \\ \varsigma _Q\mathrm{\Lambda }(v,f,r,a)\varsigma _{Q}^{}{}_{}{}^{1}=\mathrm{\Lambda }(f,v,r,a),\hfill \\ \varsigma _E\mathrm{\Lambda }(v,f,r,a)\varsigma _{E}^{}{}_{}{}^{1}=\mathrm{\Lambda }(f,v,r,a),\hfill \\ \varsigma _R\mathrm{\Lambda }(v,f,r,a)\varsigma _{R}^{}{}_{}{}^{1}=\mathrm{\Lambda }(v,f,r,a)\hfill \end{array}$$
(90)
Finally, consider the finite discrete abelian group for extended quaplectic case, we consider transformations of the basis $`\{dz,d\iota \}=\{\mathrm{𝑑𝑡},\mathrm{𝑑𝑞},\mathrm{𝑑𝑝},\mathrm{𝑑𝑒}\}`$ as embedded in the $`6\times 6`$ matrices. The above generators may be embedded in the $`6\times 6`$ matrices and the additional generators incorporating the additional discrete symmetry in (76) may be defined as
$$\begin{array}{cc}\varsigma _\alpha ^+=\left(\begin{array}{ccc}\varsigma _\alpha \hfill & 0\hfill & 0\hfill \\ 0\hfill & 1\hfill & 0\hfill \\ 0\hfill & 0\hfill & 1\hfill \end{array}\right)\hfill & \varsigma _\alpha ^{}=\left(\begin{array}{ccc}\varsigma _\alpha \hfill & 0\hfill & 0\hfill \\ 0\hfill & 1\hfill & 0\hfill \\ 0\hfill & 0\hfill & 1\hfill \end{array}\right)\hfill \end{array}$$
(91)
The full finite abelian discrete group $`𝒟`$ for the extended quaplectic group is the 26 element abelian group with elements $`\varsigma _\alpha ^\pm 𝒟`$, with $`\alpha `$ taking values in the set of labels of the 13 elements of $`𝒟^{}`$. These transformations leave invariant the metric for the nonabelian space defined by the Casimir invariant (82) and are automorphisms of the quaplectic group.
## 4. Discussion
### 4.1. $`n`$ Dimensional Case and Limits
For simplicity and clarity, we have studied the case with 1 position dimension. The theory clearly generalizes to $`n`$ dimensions with
$$𝒞(1,n)=𝒰(1,n)_s(n+1)=\left(𝒰(1)𝒮𝒰(1,n)\right)_s(n+1)$$
(92)
where
$$𝒰(1,n)=𝒪(2,2n))𝒮p(2n+2)$$
(93)
In the physical case where $`n=3`$, $`𝒞(1,3)`$ is $`25`$ dimensional. In the limiting case $`b,c\mathrm{}`$,
$$\underset{b,c\mathrm{}}{lim}𝒰(1,n)=𝒮𝒪(n)_s(n)$$
Note that for $`n=1`$,
$$\underset{b,c\mathrm{}}{lim}𝒮𝒰(1,1)=(1)$$
(94)
which is why (25) is the Heisenberg group composition law for $`(1)`$. This is the counterpart of the usual relation for the orthogonal group
$$\underset{c\mathrm{}}{lim}𝒮𝒪(1,n)=(n)=𝒮𝒪(n)_s𝒯(n)$$
(95)
### 4.2. Comments on Quantum Mechanics
A quantum theory may be constructed from the unitary representations of a dynamical group. The wave equations and Hilbert space of basic special relativistic quantum mechanics follow from the unitary irreducible representations of the Poincaré group . The unitary representations of the Poincaré group determine infinite dimensional Hilbert spaces that define the states of free particles. The eigenvalue equations for the Hermitian representation of the Casimir invariants are the basic field or wave equations of physics: Maxwell, Dirac, Klein-Gordon and so forth. The eigenvalues labeling the irreducible representations are the fundamental concepts of mass and spin.
One can likewise construct a quantum theory of the quaplectic group by considering the unitary irreducible representations of the quaplectic group. Again, the Hilbert space of particle states is determined from these irreducible unitary representations. However, in this case, these states include free and interacting particles with correspondingly nonintertial frames. Again, the eigenvalue equations for the Hermitian representation of the Casimir invariants are the basic field or wave equations of physics of this theory and the eigenvalues must define basic physical properties.
This theory is presented in a companion paper . The field equations are determined and shown to satisfy certain basic criteria for being reasonable. The Schrödinger-Robinson inqualities that generalize the Heisenberg uncertainty relations may be shown to be invariant under the quaplectic group and this may be used to study the semiclassical limit in terms of coherent states The scalar case in this theory is the relativistic oscillator. While the quaplectic invariant wave equations have been determined in , these equations have not yet been explored to understand their physical meaning. This is work that remains to be undertaken.
### 4.3. Summary
There are many higher dimensional theories in the literature. The theory presented is just a higher dimensional theory. Before relativity, one would say that we live in a three dimensional space with a universal time parameter. Relativity changed that to a four dimensional space-time, or as we use the word space more generally here, position-time manifold. Special relativity eliminates the concept of an absolute rest frame but continues to have the concept of an absolute inertial frame. Eliminating the absolute rest frame requires the four dimensional space-time continuum that no longer has an absolute sense of time. We simply generalized this approach to noninertial frames. We consider the local transformations on the time-position-momentum-energy space and show that the expected transformations result under the nonrelativistic and special relativistic assumptions. The general case of transformation between noninertial frames results from requiring an invariant symplectic metric and Born-Green orthogonal metric. An investigation of the equations shows that the need for absolute inertial frame is eliminated, forces and rates of change of energy are relative, and bounded by $`b`$ and $`bc`$. For the physical case $`n=3`$, this group is $`𝒰(1,3)`$.
The basic wave equations of special relativistic quantum theory arises from considering the unitary representations of the Poincaré group (or more accurately, its universal cover). This leads us to consider the inhomogeneous group on the time-position-momentum-energy space. However, position and momentum and time and energy do not commute. This leads us to consider the nonabelian Heisenberg group that is the semidirect product of two translation groups. The remarkable fact emerges that constructing a semidirect product with the Heisenberg group as the normal group requires an invariant symplectic metric. One need only hypothesize the Born-Green orthogonal metric. The group $`𝒰(1,n)_s(1+n)`$ is the quaplectic group.
The Born-Green orthogonal metric requires the introduction of a new physical constant $`b`$ that can be taken to have the dimensions of force. There are only three dimensionally independent physical constants. These may be taken to be $`c,b`$ and $`\mathrm{}`$. The remarkable fact is that the quaplectic group introduces a relativity principal that is reciprocal, in the sense of Born, to the usual special relativity. Now, in addition to rates of change of position with time being bounded by $`c`$, rates of change of momentum are bounded by $`b`$ and rates of change of momentum are bounded by $`bc`$. The position-time subspace is now observer frame dependent and these effects become manifest for noninertial interacting particles with rates of change of momentum approaching $`b`$.
A quantum theory is constructed by considering the unitary representations. The unitary irreducible representations determines the Hilbert space of particle states. The eigenvalue equations of the Hermitian representations of the Casimir invariant operators determines the field or wave equations of the theory. The eigenvalues characterize basic particle properties. This is discussed in a companion paper .
The nonabelian generalization of the time-position-momentum-energy space that is characterized in this paper for the one dimensional case has a rich and subtle structure. There is a very simple heuristic model that may be helpful to visualize it. Special relativity was greatly simplified with the introduction of the four vector notation. One might think that the natural extension here is, for the $`n=3`$ case, an eight or ten vector notation. However, the nonabelian structure of the space is best represented heuristically by a quad. Only the degrees of freedom on each of the four faces of the quad commute. Compare also with the quad of dimensions previously given in (51).
$$\begin{array}{cc}T\hfill & Q\hfill \\ P\hfill & E\hfill \end{array}$$
The parity, time reversal, and the Born reciprocity transformations that generate the discrete automorphism group have the action on the quad given by
$$\varsigma _P:\begin{array}{cc}T\hfill & Q\hfill \\ P\hfill & E\hfill \end{array}\begin{array}{cc}T\hfill & Q\hfill \\ P\hfill & E\hfill \end{array},\varsigma _T:\begin{array}{cc}T\hfill & Q\hfill \\ P\hfill & E\hfill \end{array}\begin{array}{cc}T\hfill & Q\hfill \\ P\hfill & E\hfill \end{array},$$
$$\varsigma _Q:\begin{array}{cc}T\hfill & Q\hfill \\ P\hfill & E\hfill \end{array}\begin{array}{cc}T\hfill & P\hfill \\ Q\hfill & E\hfill \end{array},\varsigma _E:\begin{array}{cc}T\hfill & Q\hfill \\ P\hfill & E\hfill \end{array}\begin{array}{cc}E\hfill & Q\hfill \\ P\hfill & T\hfill \end{array}$$
Now, continuing in this heuristic manner, we know that the universal constant $`c`$ is associated with a relativity on the $`(T,Q)`$ and $`(P,E)`$ subspaces that is locally governed by the Lorentz group. The remarkable fact is that the constant $`b`$ introduced above is associated with a relativity on the $`(T,P)`$ and $`(Q,E)`$ subspaces that is also locally governed by a Lorentz group. For this reason, we call the relativity of velocity and forces reciprocal relativity. Both of these are subgroups of the more general pseudo-unitary group.
The etymology of quaplectic is the following. Qua is a seldom used English word that means in the character of or simply as as in Sin qua non. Plectic has origins in Greek which means to pleat or to fold diagonally. So quaplectic means in the character of folding diagonally. It also invokes Quad, Quantum and Symplectic, all of which play a role.
The author thanks P. Jarvis for stimulating discussion and comments on this paper and B. Hall for encouraging a simple physical description.
## 5. Appendix
### 5.1. Comment on Lagrangian mechanics
Lagrangian mechanics introduces the bias to a position-time manifold formalism. In this section, we show that the bias is not intrinsic in the basic mathematical formulation as one can simply Legendre transform also to an equivalent momentum-time Lagrangian formulation.
We have been considering a formulation on $`z=\{t,q,p,e\}^4`$ with frames $`dz=\{dt,dq,dp,de\}T_{}^{}{}_{z}{}^{}`$.
The symplectic 2-form $`d\stackrel{~}{e}d\stackrel{~}{t}+d\stackrel{~}{p}d\stackrel{~}{q}`$ in (23) may be integrated to define the 1-form
$$ed\stackrel{~}{t}+pd\stackrel{~}{q}=\left(H(p,q,t)+p(t)\frac{dq(t)}{dt}\right)dt=L(q(t),\frac{dq(t)}{dt},t)$$
(96)
where Hamilton’s equations $`\frac{\mathrm{𝑑𝑞}(t)}{\mathrm{𝑑𝑡}}=\frac{H(q,p,t)}{p}`$ are used to solve for $`p=p(q,\frac{\mathrm{𝑑𝑞}}{\mathrm{𝑑𝑡}})`$ provided that the Hessian is nonsingular (we use the general expressions here which carry over to $`n`$ dimensions even though in our current context we are in one dimension and the determinant is trivial.)
$$\mathrm{Det}\frac{H(p,q,t)}{pp}0$$
The Lagrangian satisfies the variational principle $`\delta `$$`L(q,\frac{\mathrm{𝑑𝑞}}{\mathrm{𝑑𝑡}},t)`$=0 to yield the Euler-Lagrange equations
$$\frac{L(q,v,t)}{q}\frac{d}{\mathrm{𝑑𝑡}}\frac{L(q,v,t)}{v}=0$$
(97)
Geometrically, a number of things have happened here. Let $`𝕄^2`$ with $`(t,q)𝕄`$ and then $`e\mathrm{𝑑𝑡}+p\mathrm{𝑑𝑞}`$ is an element of $`T^{}𝕄`$. The Lagrangian on the other hand, is a function on the tangent space $`T𝕄`$. This is possible as $`\{\mathrm{𝑑𝑡}\}`$ and $`\{\mathrm{𝑑𝑡},\mathrm{𝑑𝑞}\}`$ are invariant subspaces under the action of $`\mathrm{\Phi }`$. That is, there is an absolute notion of time that all observers agree on and furthermore all observers agree on the position-time subspace of the full space $``$.
Note also that $`\{\mathrm{𝑑𝑡},\mathrm{𝑑𝑝}\}`$ is an invariant subspace and therefore all observers agree on the momentum-time subspace. The corresponding integration of the symplectic two form is $`e\mathrm{𝑑𝑡}q\mathrm{𝑑𝑝}`$ is an element of $`T^{}\stackrel{ˇ}{𝕄}`$ with $`\stackrel{ˇ}{𝕄}^2`$ with $`(t,p)\stackrel{ˇ}{𝕄}`$. Applying the transformations gives
$$ed\stackrel{~}{t}qd\stackrel{~}{p}=\left(H(p,q,t)+q(t)\frac{dp(t)}{dt}\right)dt=L(p(t),\frac{dp(t)}{dt},t)$$
(98)
where Hamilton’s equations $`\frac{\mathrm{𝑑𝑝}(t)}{\mathrm{𝑑𝑡}}=\frac{H(q,p,t)}{q}`$ are used to solve for $`q=q(p,\frac{\mathrm{𝑑𝑝}}{\mathrm{𝑑𝑡}})`$ provided that the Hessian is nonsingular
$$\mathrm{Det}\frac{H(p,q,t)}{qq}0$$
The Lagrangian satisfies the variational principle $`\delta `$$`L(p,\frac{\mathrm{𝑑𝑝}}{\mathrm{𝑑𝑡}},t)`$=0 to yield the Euler-Lagrange equations
$$\frac{L(p,f,t)}{p}\frac{d}{\mathrm{𝑑𝑡}}\frac{L(p,f,t)}{f}=0$$
(99)
Clearly, for a free particle, this latter Hessian is singular. However, in the presence of long range $`1/r^2`$ type forces, there are no truly free particles. Consequently, for classical systems with these types of forces present, a free particle is an abstraction and in principle one could formulate all of basic mechanics on this momentum-time space. We are heavily biased by our Newtonian heritage to consider space, that is position space,as the fundamental arena of physics. This is re-enforced by special relativity that brings time onto the same footing as position to define space-time (that is position-time space). However, the basic Hamilton’s equations put momentum and position on completely equal footing and this remains true in Dirac’s nonrelativistic transformation theory of quantum mechanics . In fact, as just shown, there is a reciprocal conjugate Lagrangian formulation on momentum-time space that is equally valid in the nonrelativistic case as the position-time Lagrangian formulation.
In this heuristic sense, the simplest classical Hamiltonian formulation satisfies Born reciprocity in position and momentum, there is no distinction in the mathematical formulation that biases the formulation to position-time over momentum-time.
### 5.2. Comment on the $`𝒰`$(1,1) transformation equations
The transformation equations $`𝒰(1,1)=𝒰(1)𝒮𝒰(1,1)`$ may also be written in terms of hyperbolic trig functions.
The $`𝒮𝒰(1,1)`$ transformation equations that leave this invariant are
$$\begin{array}{c}d\stackrel{~}{t}=\mathrm{cosh}\omega \mathrm{𝑑𝑡}+\frac{\mathrm{sinh}\omega }{\omega }\left(\frac{\beta }{c}\mathrm{𝑑𝑞}+\frac{\gamma }{b}\mathrm{𝑑𝑝}\frac{\vartheta }{bc}\mathrm{𝑑𝑒}\right),\hfill \\ d\stackrel{~}{q}=\mathrm{cosh}\omega \mathrm{𝑑𝑞}+\frac{\mathrm{sinh}\omega }{\omega }\left(c\beta \mathrm{𝑑𝑡}\frac{\gamma }{b}\mathrm{𝑑𝑒}+\frac{c\vartheta }{b}\mathrm{𝑑𝑝}\right),\hfill \\ d\stackrel{~}{p}=\mathrm{cosh}\omega \mathrm{𝑑𝑝}+\frac{\mathrm{sinh}\omega }{\omega }\left(b\gamma \mathrm{𝑑𝑡}+\frac{\beta }{c}\mathrm{𝑑𝑒}\frac{b\vartheta }{c}\mathrm{𝑑𝑞}\right)\hfill \\ d\stackrel{~}{e}=\mathrm{cosh}\omega \mathrm{𝑑𝑒}+\frac{\mathrm{sinh}\omega }{\omega }(b\gamma \mathrm{𝑑𝑞}+c\beta \mathrm{𝑑𝑝}+\mathrm{𝑏𝑐}\vartheta \mathrm{𝑑𝑡}),\hfill \end{array}$$
(100)
where $`\omega ^2=\beta ^2+\gamma ^2\vartheta ^2`$. Note immediately that if $`\gamma =\vartheta =0`$, these reduce to the usual special relativity equations. It may be directly verified that these equations leave invariant the symplectic metric
$$\begin{array}{cc}\hfill d\stackrel{~}{e}d\stackrel{~}{t}+d\stackrel{~}{p}d\stackrel{~}{q}& =\mathrm{𝑑𝑒}\mathrm{𝑑𝑡}+\mathrm{𝑑𝑝}\mathrm{𝑑𝑞}\hfill \end{array}$$
(101)
and the Born-Green metric line element in (30). The group is a matrix group with elements $`d\stackrel{~}{z}=\mathrm{\Xi }(\beta ,\gamma ,\vartheta )\mathrm{𝑑𝑧}`$ realized by the matrix
$$\mathrm{\Xi }(\beta ,\gamma ,\vartheta )=\left(\begin{array}{cccc}\mathrm{cosh}\omega \hfill & \frac{\beta \mathrm{sinh}\omega }{\omega c}\hfill & \frac{\gamma \mathrm{sinh}\omega }{\omega b}\hfill & \frac{\vartheta \mathrm{sinh}\omega }{\omega bc}\hfill \\ \frac{c\beta \mathrm{sinh}\omega }{\omega }\hfill & \mathrm{cosh}\omega \hfill & \frac{c\vartheta \mathrm{sinh}\omega }{b\omega }\hfill & \frac{\gamma \mathrm{sinh}\omega }{\omega b}\hfill \\ \frac{b\gamma \mathrm{sinh}\omega }{\omega }\hfill & \frac{b\vartheta \mathrm{sinh}\omega }{c\omega }\hfill & \mathrm{cosh}\omega \hfill & \frac{\beta \mathrm{sinh}\omega }{c\omega }\hfill \\ \frac{bc\vartheta \mathrm{sinh}\omega }{\omega }\hfill & \frac{b\gamma \mathrm{sinh}\omega }{\omega }\hfill & \frac{c\beta \mathrm{sinh}\omega }{\omega }\hfill & \mathrm{cosh}\omega \hfill \end{array}\right)$$
(102)
In order that the transformations are hyperbolic and not the usual sine or cosine functions that would result in oscillations, we must have $`\omega ^2=\beta ^2+\gamma ^2\vartheta ^20`$. This formalism is equivalent to the discussion in the main body of text by defining the parameterization
$$v=\frac{c\beta }{\omega }\mathrm{tanh}\omega ,f=\frac{b\gamma }{\omega }\mathrm{tanh}\omega ,r=\frac{bc\vartheta }{\omega }\mathrm{tanh}\omega $$
(103)
Then, define
$$w^2=\left(\frac{v}{c}\right)^2+\left(\frac{f}{c}\right)^2\left(\frac{r}{bc}\right)^2=\left(\mathrm{tanh}\omega \right)^2$$
(104)
Note that $`\omega 0`$ implies that $`ds^20`$. It follows immediately that $`\mathrm{cosh}\omega =(1w^2)^{1/2}`$ and $`\mathrm{sinh}\omega =w(1w^2)^{1/2}`$ . Also note that
$$\frac{\beta }{\omega }=\frac{v}{cw},\frac{\gamma }{\omega }=\frac{f}{bw},\frac{\vartheta }{\omega }=\frac{r}{bcw}$$
(105)
Equation (32) follows directly.
|
warning/0506/astro-ph0506343.html
|
ar5iv
|
text
|
# A Spectroscopic Search for the non-nuclear Wolf-Rayet Population of the metal-rich spiral galaxy M 83 Based on observations made with ESO Telescopes at the Paranal Observatory under programme ID 69.B-0125
## 1 Introduction
Massive stars play a major role in the ecology of galaxies via radiative, mechanical and chemical feedback (Smith, 2005). Wolf-Rayet (WR) stars in particular, albeit rare and short-lived, make a significant contribution to their environment via the mechanical return of nuclear processed material to the interstellar medium (ISM) through their exceptionally powerful stellar winds.
Metallicity, $`Z`$, is a key factor in determining the number and subtype distribution of a WR population. Although the metallicity dependence of WR wind properties still remains unclear, mass-loss prior to this phase has been established to depend on metallicity, with the latest models predicting Ṁ $`Z^{0.8}`$ for O stars (Vink et al., 2001). Evolutionary models for single stars predict that the minimum mass cut-off required for WR formation should decrease as metal content increases. It is anticipated that the minimum mass required for progression to the WR phase decreases from $`32\text{M}_{}`$ for a SMC-like metallicity to $`21\text{M}_{}`$ for super-Solar metallicity (Meynet & Maeder, 2004). Single star predictions are broadly consistent with the initial masses of WR stars in the Milky Way, LMC and SMC inferred from cluster membership (Massey et al., 2000, 2001). The fractional distribution of carbon-rich (WC) to nitrogen-rich (WN) stars is also known to increase with metallicity, such that empirically one would expect to observe a large WC population in a metal-rich environment (Massey & Johnson, 1998). The formation of WR stars at low metallicity is anticipated primarily via close binary Roche Lobe Overflow (RLOF), yet the observed WR binary fraction in the SMC does not differ from that of the Milky Way (Foellmi et al., 2003).
With the aim of substantiating such predictions, surveys of WR stars in Local Group galaxies (typically 0.2–1Z) have been carried over the last two decades. At sub-Solar metallicities, the LMC and SMC have been well sampled (Breysacher et al., 1999; Massey et al., 2003), as has M 33 (Massey & Johnson, 1998; Abbott et al., 2004). In contrast, M 31 is the only Local Group member with super-Solar metallicity, but its unfavourable inclination and large spatial extent makes surveying the complete WR population very challenging. In order to increase the variety of galaxies sampled, our group has begun to look beyond the Local Group (e.g. NGC300, Schild et al., 2003).
Galaxies hosting substantive WR populations are known as ‘WR galaxies’ (Kunth & Sargent, 1981; Schaerer et al., 1999b), where the number of WR stars ranges from $``$35 in NGC1569-A (Gonzalez Delgado et al., 1997) to 2$`\times 10^4`$ in Mrk 309 (Schaerer et al., 2000). Within specifically metal-rich environments, previous studies of WR populations have generally been restricted to integrated spectra from bright star forming knots (e.g. Schaerer et al., 1999a) or H ii regions (e.g. Pindao et al., 2002). Here we present the results of a deep imaging and spectroscopic survey of the disk WR population within the metal-rich galaxy M 83, in which WR signatures have previously been identified by Rosa & Richter (1988) and Bresolin & Kennicutt (2002).
M 83 (NGC 5236) is a massive, grand-design southern spiral (SBc(s) II) with on-going star formation in its spiral arms plus an active nuclear starburst (Elmegreen et al., 1998; Harris et al., 2001). M 83 is the principal member of a small galaxy group ($`11`$ members) within the Centaurus A complex (Karachentsev et al., 2002). Located at a distance of 4.5$`\pm `$0.3Mpc (Thim et al., 2003), its favourable inclination and apparently high metal abundance of log(O/H)+12=9.2 (Bresolin & Kennicutt, 2002) makes M 83 an ideal candidate for studies of massive stellar populations at high metallicity.
More recently, oxygen abundances in metal-rich galaxies have been revised downward (Pilyugin et al., 2004; Bresolin et al., 2004), such that M 83 may have a metal abundance closer to log(O/H)+12=9.0 (Bresolin, 2004, priv. comm.), i.e. approximately twice the Solar oxygen abundance of log(O/H)+12=8.66 recently derived by Asplund et al. (2004).
We present the results of an imaging and spectroscopic survey of the WR content of M 83 using the ESO Very Large Telescope (VLT). The present paper complements the initial findings of this study reported in Crowther et al. (2004, hereafter Paper I). In Sect. 2 we briefly describe the observations and data reduction techniques employed. Section 3 discusses the method followed to obtain a global WR population of M 83. Sect. 4 discusses the properties of metal rich WR stars with those of Local Group galaxies and evolutionary models. Finally, conclusions are drawn in Sect. 5.
## 2 Observations and data reduction
We have observed M 83 with the ESO Very Large Telescope UT4 (Yepun) and Focal Reduced/Low Dispersion Spectrograph #2 (FORS2). The detector consists of a mosaic of two $`2048\times 1024`$ MIT/LL CCDs which in conjunction with the standard collimator provides a field-of-view $`6.8^{}\times 6.8^{}`$ and an image scale of 0.126<sup>′′</sup>/pixel. Photometric observations of M 83 were made between May–June 2002 with follow-up spectroscopic data being acquired during April–June 2003.
### 2.1 Imaging
M 83 subtends 12.9 by 11.5 on the sky, preventing it being imaged by a single FORS2 frame. In order to obtain complete coverage, the galaxy was divided into four overlapping regions, covering the NE (Field A), NW (B), SE (C) and SW (D) as indicated in Fig. 1. Occulting bars were positioned in Field C to prevent detector saturation by bright foreground stars. The central 15<sup>′′</sup> appears saturated on all images obtained, and as a result the WR population of the nucleus can not be discussed further.
FORS2 was used on 2 June 2002 to obtain narrow-band images with central wavelengths 4684Å, 4781Å and band widths of 66Å and 68Å respectively. These were obtained consecutively for each Field in seeing conditions between 0.6 – 0.8<sup>′′</sup><sup>1</sup><sup>1</sup>10.8<sup>′′</sup> corresponds to a linear scale of 18pc at the distance of M 83 (Thim et al., 2003) with individual exposures of 1800s. The $`\lambda `$4684 filter is coincident with the strong WR emission features which incorporates the N iii ($`\lambda 4640`$Å), C iii ($`\lambda 4650`$Å) and Heii ($`\lambda 4686`$Å) emission lines, whereas the latter samples a wavelength region relatively free from emission, providing a measure of the continuum level. In addition to these, 2 exposures (60s and 600s) were taken using narrow-band on- and off-$`\text{H}\alpha `$ filters $`(\lambda 6563,\mathrm{\hspace{0.17em}6665}\text{Å},\text{FWHM}=61,\mathrm{\hspace{0.17em}65}\text{Å}`$) on 16 May 2002. Finally, in order to supplement the primary dataset, 2 exposures (60s and 120s) were also acquired using a Bessell B filter on 21 May 2002.
### 2.2 Photometry
Images were prepared following standard procedures i.e. debiased, flat field corrected and cosmic ray cleaned. Photometry of individual sources within M 83 was performed using the package daophot, a point-spread function (PSF) fitting routine within iraf. Absolute photometry in the broad-band B filter was achieved with the aid of photometric standard fields Ru 152 and PG 1528+062 (containing a total of 10 photometric standards, 11.9 $``$ B $``$16.3). For the narrow-band images such standards are not available and photometric zero-points have been obtained by observing spectrophotometric standards LTT7987 (B = 12.2) and G138-31 (B = 16.5).
The majority of our sources appear point-like on the ground-based images. However, a number of bright sources are surrounded by a faint, extended halo, which was not accounted for in the PSF photometry and as a result only a lower limit to the magnitude is given, based on PSF photometry. A further subset of the bright sources are spatially extended, indicating that PSF photometry is inappropriate, as indicated in Table A1 in the appendix.
Typical formal photometric errors range between 0.02 mag ($``$18 mag), 0.05 mag ($``$20 mag) and 0.08 mag ($``$22 mag). Significantly higher errors, of up to 0.15 mag, are obtained for regions of the galaxy where the background levels are high, or they are located in spatially crowded regions.
As a consistency check we have compared results obtained for the two Bessell B exposures (for which the PSF model was based on different template stars) and also derived magnitudes for objects which appear in multiple fields. Excellent agreement was observed in both cases, with results agreeing to within the formal errors. In a minority of cases this was not achieved due to severe crowding.
### 2.3 Candidate Selection
WR candidates were identified by searching for He ii / C iii excess emission (at $`\lambda `$4684) relative to the continuum ($`\lambda `$4781), i.e. a negative value of $`\mathrm{\Delta }m`$ = $`m_{\lambda 4684}`$\- $`m_{\lambda 4781}`$. The optimal method of identifying suitable candidates was found to be via ‘blinking’ individual $`\lambda `$4781 and $`\lambda `$4684 frames together with the difference image obtained by subtracting the $`\lambda `$4781 image from the $`\lambda `$4684 frame. In total, 283 candidate $`\lambda `$4684 emission sources were identified.
For 75% of our candidates we have obtained a magnitude in at least the $`\lambda `$4684 filter. For cases where we did not obtain photometry, the object was either too faint or was located in a spatially crowded region. In addition, for a significant fraction of the fainter sources it was not possible to measure a $`\lambda `$4871 magnitude.
Candidates were grouped according to continuum brightness, $`\mathrm{\Delta }m`$, and association with underlying H ii regions. To ensure we spectroscopically observed a representative sample, a selection from each group was chosen for spectroscopic follow up. In Fig. 2 we show $`\mathrm{\Delta }m`$ as a function of continuum magnitude for the sources in which WR signatures were either spectroscopically confirmed, rejected or no spectroscopy was obtained, i.e. the remaining candidates. The majority of confirmed sources have a $`\lambda `$4684 excess between –1.5 $`\mathrm{\Delta }m`$ –0.4 mag, although a few do exhibit rather smaller values of $`\mathrm{\Delta }m`$. In contrast, all rejected regions have $`\mathrm{\Delta }m`$ –0.2 mag, suggesting that remaining candidates which display a moderate $`\lambda `$4684 excess should represent regions that genuinely host WR stars, together with a subset of those for with $`\mathrm{\Delta }m`$ 0.0 mag.
### 2.4 Spectroscopy
Spectroscopic data was obtained using FORS2 with the Multi Object Spectroscopy (MOS) mode. MOS datasets of individual WR candidates were obtained during seeing conditions of $``$0.5–1.0<sup>′′</sup>, using a slit width of $`0.8^{\prime \prime }`$. The CCD was binned by a factor of 2 in the dispersion direction, resulting in a dispersion of 3.3Å pixel<sup>-1</sup> with the 300V grism and a spectral resolution of $`7`$Å, as measured from comparison arc lines. The wavelength range of individual targets depended on their position within the MOS mask but typical wavelength coverage was $``$3700Å to $``$7500Å .
MOS allows the spectra of up to 19 candidates to be recorded simultaneously. However, due to positional limitations this was generally restricted to $``$15, supplemented where possible by H ii regions. In total, 198 candidates have been spectroscopically observed using 17 different MOS masks. To maximise continuum S/N, sources were grouped according to brightness, with total on-source integration times ranging from 720s for the brightest objects to 4800s for the faintest. Details of the spectroscopic log can be found in Table 1, and includes DIMM seeing measurements. The MOS masks were labelled according to the region of M 83 in which they were observed, i.e. Field A was observed using 5 different masks labelled A1 to A5.
Datasets were prepared and processed using standard iraf and figaro packages i.e. the data were bias subtracted, flat field corrected, spectra were traced, extracted, and subsequently wavelength and flux calibrated. For very faint sources, where no continuum was evident, identification was made solely on the basis of strong emission lines, a neighbouring continuum source was used as the trace.
Spectroscopic magnitudes (hereafter m<sub>spec</sub>) were obtained by convolving the individual spectra with the transmission curves of the imaging filters. A comparison between the spectroscopic and the PSF photometry at $`\lambda 4684`$ permitted absolute flux calibration. For 160 sources brighter than 24 mag the average slit correction factor was found to be 3.1 ($`\sigma =1.9`$). This large factor is due in part to the often crowded nature of sources, such that the full profile extent was not extracted. For 20 spectroscopically observed regions, where photometry was unavailable, we corrected the spectroscopy by a factor of 3$`\pm `$1.5.
The blue ($``$4500Å) continuum S/N ranged from $``$80 for the brightest sources, to $``$1 for the faintest sources. Nevertheless, lines were typically detected even in the faintest sources at the 5–10$`\sigma `$ level, and a source was only confirmed if WR emission lines were detected at a $`3\sigma `$ level.
## 3 The Wolf-Rayet Population of M 83
The WR content of the disk of M 83 has been determined by visually inspecting the extracted spectra for the characteristic WR emission signatures, i.e. N iii$`\lambda `$4634–41, N v$`\lambda `$4603–20, C iii$`\lambda 4650`$ – Heii$`\lambda 4686`$ blend (hereafter blue WR features) and/or C iii $`\lambda `$5696, C iv$`\lambda 5812`$ features (hereafter yellow WR features).
Of the 198 sources spectroscopically observed, 132 contain WR emission features at a 3$`\sigma `$ level. These are presented in a catalogue in Table A1, in the Appendix, which includes coordinates, PSF (or spectroscopic) photometry, interstellar reddening, line measurements and WR populations. Deprojected distances are included using M 83 properties presented in Table 1 of Lundgren et al. (2004). We present finding charts for all confirmed sources in the (electronic only) Appendix, Figures B1–B17.
Of the remaining 66 sources, 40 displayed an early-type spectrum with no WR emission present, 16 resembled that of a late-type carbon star, while 10 sources revealed WR features below the 3$`\sigma `$ level, or lacked the blue region necessary for WN identification. The latter two groups, along with the 79 regions which were not spectroscopically observed, are given in our candidate list (Table A2 in the Appendix).
### 3.1 Interstellar reddening
Estimates of the interstellar reddening for our confirmed WR sources have generally been derived using measurements of the nebular H$`\alpha `$ (accounting for nearby \[N ii\] emission) and H$`\beta `$ features present in the extracted spectra.
Assuming Case B recombination theory for typical electron densities of $`10^2\text{cm}^3`$ and a temperature of $`10^4\text{K}`$ (Hummer & Storey, 1987), we obtain 0.2$``$E(B-V)=*c*(H$`\beta `$)/1.46 $``$ 0.8 mag for the majority of the sources, with a few outliers, and typical formal uncertainty of $`\pm `$0.02 mag. Where Balmer emission was observed, typical H$`\beta `$ equivalent widths lay in the range $`\text{and}20\text{to}150`$Å. Consequently, the underlying stellar absorption components ($``$1Å at H$`\beta `$) are neglected.
In 41 sources no nebular lines were observed. For those with a well defined continuum, E(B–V) was estimated by assuming an intrinsic optical flux distribution equivalent to a late O-type star, with typical uncertainty of $`\pm `$0.05–0.1 mag. In 15 cases, the continuum S/N was insufficient for this comparison and an average reddening of E(B–V)= 0.5$`\pm `$0.3 was adopted. Correction for reddening adopt a standard Seaton (1979) extinction law with R=3.1=$`A_\mathrm{V}`$/E(B-V).
### 3.2 Spectral classification
In order to classify and quantify the WR population within each region, we have fit Gaussian line profiles to the blue and yellow WR features, revealing line fluxes, equivalent widths and FWHM. An example of the fits to the blue and yellow WR features is presented in Fig. 3, where a source (#74) hosting a mixed WN and WC population is presented.
In general, it was straightforward to distinguish between WN (strong Heii$`\lambda 4686`$) and WC subtypes (strong C iii$`\lambda `$4650 and C iii$`\lambda `$5696 and/or C iv$`\lambda `$ 5801-12). The following classification scheme was applied for further subdivision. In a minority of cases it was not possible to separate the $`\lambda `$4650 – $`\lambda `$4686 features into individual components, and as a result an overall blend was measured. Since WC subtypes were assigned on the basis of $`\lambda 5696`$ and $`\lambda 5812`$ features, this did not prevent accurate classification.
Late and early WN subtypes were assigned if Heii$`\lambda 4686`$ emission was accompanied by N iii$`\lambda 463441`$ or N v$`\lambda 460320`$ emission, respectively. If nitrogen lines were undetected, we assigned a WNE subtype if FWHM (Heii$`\lambda 4686`$) $`>`$ 20Å, and WNL otherwise. For WC stars, WC4 – 6 was assigned if C iv$`\lambda 580112`$ was present along with either weak or absent C iii$`\lambda 5696`$. For $`0.25F_\lambda `$ (C iii$`\lambda `$5696/C iv$`\lambda `$5801-12) $`0.8`$ sources were classified WC7, and WC8–9 if C iii$`\lambda 5696`$ was present, with C iv$`\lambda 580112`$ weak or absent.
To ensure consistency with previous studies (e.g. Schaerer et al., 1999a; Bresolin & Kennicutt, 2002; Chandar et al., 2004) we have derived WR populations based on individual line fluxes adapted from Schaerer & Vacca (1998). As discussed in Paper I, we adopt He ii $`\lambda `$4686 lines fluxes of 5.2$`\times 10^{35}`$ erg s<sup>-1</sup> and 1.6$`\times 10^{36}`$ erg s<sup>-1</sup> for WN2-5 and WN6-10 stars, respectively. For WC stars, we adopt C iv $`\lambda `$5801 line fluxes of 1.6$`\times 10^{36}`$ erg s<sup>-1</sup> and 1.4$`\times 10^{36}`$ erg s<sup>-1</sup> for WC4–6 and WC7 stars, respectively, and a C iii $`\lambda `$5696 line flux of 7.1$`\times 10^{35}`$ erg s<sup>-1</sup> for WC8–9 stars. WR contents of individual sources then follow, with populations rounded to the nearest integer ($``$1). In one source (#117), we were unable to reliably extract the spectrum since it was located at the very edge of the slit, and so a measure of the reddening/line flux was not possible. Nevertheless, broad He ii $`\lambda `$4686 is clearly present, with no WC signature, such that we indicate a population of $`1`$ early-type WN star.
In Fig. 4 we compare sources containing representative late, mid and early WC stars from M 83 with extinction corrected Milky Way counterparts, scaled to the distance of M 83. Large line widths amongst M 83 members hosting late WC stars are apparent, particularly for #32 versus HD 192103 (WC8) and #81 versus HD 164270 (WC9). In contrast, sources containing WC4–7 stars indicate similar line widths to individual Galactic counterparts. Comparisons between sources containing WNL and WNE stars in M 83 and two Galactic counterparts are shown in Fig. 5, revealing similar spectral morphologies. Other examples of sources hosting WN and WC populations are presented in Fig.1 of Paper I.
### 3.3 The M 83 WR population – individual stars, binaries, complexes or clusters?
What is the nature of the 132 sources in M 83 that are known to host WR stars? In Fig. 6(a) we compare the spectroscopic continuum magnitude to the spectroscopic excess, $`\mathrm{\Delta }m_{\mathrm{spec}}=`$$`m_{\lambda 4684}`$$`m_{\lambda 4781}`$, for all sources. This is more complete than Fig. 2, since it was generally possible to estimate the spectral $`m_{\lambda 4781}`$ magnitude for the fainter sources, where PSF-photometry was not available.
The brightest confirmed WR sources in our sample ($`m_{\lambda 4781}`$$``$20 mag) exhibit –0.3 $`\mathrm{\Delta }m_{\mathrm{spec}}`$0.0 mag. Such values are consistent with luminous complexes, greatly diluting the WR emission signature. In contrast, the faintest confirmed sources ($`m_{\lambda 4781}`$$``$25 mag) possess large spectroscopic excesses of –2 $`\mathrm{\Delta }m_{\mathrm{spec}}`$ –0.5 mag, consistent with isolated, single or binary WR systems. Intermediate brightness sources span the full range in excess, corresponding to less luminous regions hosting a few WR stars to those containing large WR populations.
Fig. 6(b) compares the spectroscopic $`\lambda 4686`$ excess to the C iii$`\lambda `$4650/Heii$`\lambda `$4686 equivalent width, confirming the expected tight correlation between line strength and $`\mathrm{\Delta }m_{\mathrm{spec}}`$, where the scatter indicates the observational accuracy. Typical excesses of –0.2 mag equate to small line equivalent widths of $``$10Å, whilst an excess of –1.0 mag corresponds to $``$100Å, and the largest excesses equate to $``$500Å. For comparison, single Galactic and LMC WR stars possess C iii$`\lambda `$4650/Heii$`\lambda `$4686 equivalent widths of 10–500Å (WN subtypes) or 150–2000Å (WC subtypes).
### 3.4 The global disk WR population of M 83
We identify 1035$`\pm `$300 WR stars, comprising 564$`\pm `$170 WC and 471$`\pm `$130 WN stars, within our 132 spectroscopically observed regions, where errors quoted here were obtained from simply adding individual uncertainties for all regions.
The most important discovery of our spectroscopic survey is the dominant late-type WC population of M 83. Over half of the spectroscopically identified WR stars in M 83 fall into the WC8–9 subtype, with few WC4–7 stars identified. For comparison, no WC8–9 stars are observed in the SMC, LMC or M 33 and the total number of such stars in the Milky Way and M 31 is less than 50 (van der Hucht, 2001; Moffat & Shara, 1987). The distribution among late- and early-type WN stars is more even, with WNL/WNE $`1`$. This value is much greater than that observed in the SMC ($``$0) and LMC ($``$0.25), but comparable to that of $`1.3`$ determined for the Milky Way (van der Hucht, 2001).
How robust is this derived WR population for M 83? For each source, we have propagated uncertainties in the distance, reddening, photometry and line flux measurements. Together, these translate to a typical uncertainty of $`2030`$%, or somewhat higher for regions in which an interstellar reddening or a slit loss correction factor have been adopted.
One of the main limitations in estimating the content of an unresolved WR population is the conversion from WR line flux to WR content. Given the large late WC population identified in M83, we have reconsidered the line flux of individual WC8–9 stars determined by Schaerer & Vacca (1998). From unpublished data for 5 Galactic, and 2 M 31 WC8–9 stars, each with well derived distances, we find a mean $`\lambda 5696`$ flux of $`5.1\times 10^{35}\text{ergs}^1`$ and $`4.7\times 10^{35}\text{ergs}^1`$ respectively. This is $``$30% lower than Schaerer & Vacca, and suggests that, if anything, we may be underestimating the true WC population of M 83.
We have also estimated the WC population using the alternative C iii$`\lambda 4650`$ line. Based on individual WR $`\lambda 4650`$ line fluxes of $`3.4\times 10^{36},\mathrm{\hspace{0.17em}4.5}\times 10^{36}\text{and}\mathrm{\hspace{0.17em}1.0}\times 10^{35}\text{ergs}^1`$ for individual WC4–6, WC7 and WC8–9 subtypes, respectively (Schaerer & Vacca, 1998), populations of individual sources were found to agree to within a factor of 2, relative to the yellow features. The total WC population was calculated to be 594 using C iii$`\lambda `$4650, in excellent agreement with that of 564 obtained from C iii$`\lambda 5696`$ and C iv $`\lambda `$5808.
Turning to the candidates for which spectroscopy was not obtained, all regions in Fig. 2 with $`\mathrm{\Delta }\mathrm{m}`$ –0.3 mag correspond to spectroscopically confirmed WR complexes. Therefore, we would expect that at least 25 out of the 49 candidates, for which $`\lambda `$4684 and $`\lambda `$4781 photometry is available, also possess WR stars. Adopting the same fraction for regions where PSF photometry is not available, we expect $``$50 of the remaining 89 candidate regions to contain WR stars. Indeed, #159 has already been observed by Bresolin & Kennicutt (2002). Designated M83-5 in their study, WR emission is spectroscopically confirmed and a population of 2 WCL and 6 WNL stars (scaled to a distance of 4.5Mpc) is inferred from its line luminosity. On average, our confirmed sources host $``$5 WR stars, such that we expect $``$250 WR stars await identification in M 83, bringing the total disk population to $``$1300.
The inferred WR population of M 83 is greater than that known in the entire Local Group, to date (Massey & Johnson, 1998). As anticipated from Figs. 2 and 6, some sources host a single WR star, whilst others contain larger WR populations ($`10`$). Regions which contain an exceptionally large WR population will be discussed in more detail in the next section.
### 3.5 Complexes hosting large WR populations
In the Milky Way, the most massive open clusters (e.g. Arches, Westerlund 1) host at most 10–20 WR stars (Blum et al., 2001; Negueruela & Clark, 2005). Similar numbers are observed in the largest H ii regions of M 33, and 30 Doradus in the LMC. We identify 10 regions in M 83 with large ($``$20), or mixed, WR populations. Mixed WN and WC populations are observed in a total of 5 complexes, #66 (8$`\pm `$2 WNL, 4$`\pm `$1 WC7), #38 (7$`\pm `$2 WNL, 21$`\pm `$6 WCL), #41 (14$`\pm `$4 WNL, 13$`\pm `$6 WC7), #86 (9$`\pm `$4 WNL, 24$`\pm `$10 WCL) and #74 which will be discussed separately.
Are the sources that host WR stars in M 83 compact clusters (e.g. Arches) or extended, giant H ii regions (30 Doradus)? Massive compact clusters are generally rare in normal disk galaxies, although M 83 is known to host many examples, from HST imaging (Larsen, 2004). Of the 60 bright H ii regions in M 83 identified by de Vaucouleurs et al., between 28–38 host WR populations. Indeed, the 3 complexes hosting the largest WR populations are all associated with H ii regions identified by de Vaucouleurs et al.. Optical spectroscopy of these were presented in Paper I, together with an estimate of their O star population.
#### 3.5.1 Source #74
From our sample #74 is exceptional, with 230$`\pm `$50 late-type WN and WC stars inferred from the de-reddened line fluxes (recall Fig.3). This source has the highest interstellar reddening of our entire sample with E(B-V)=1.0$`\pm `$0.03, although it is closest to the nucleus. However, the H$`\alpha `$/H$`\beta `$ nebular value is supported from fitting its stellar continuum to a young ($``$4 Myr) instantaneous burst model at Z=0.04 from Starburst99 (Leitherer et al., 1999). In Paper I, we estimated a Lyman continuum flux of 8$`\times 10^{51}`$ s<sup>-1</sup> from the de-reddened H$`\alpha `$ flux, such that #74 has an ionizing flux equivalent to the giant H ii region 30 Doradus. However, it possesses a WR content which is a factor of ten times larger, i.e. N(WR)/N(O)$``$0.25 versus 0.02 in 30 Doradus.
We have inspected archival HST/Advanced Camera for Surveys (ACS) Wide Field Camera (WFC) F475W datasets of M 83 (Proposal 9299, P.I. H. Ford). This revealed that #74 is very compact, with a FWHM of $``$0.2 arcsec or $``$4.5 pc (for a distance of 4.5Mpc). For H ii regions with solar or super-solar metallicities, WR signatures are expected to be present in bursts of age 3–6Myr. We have compared the absolute F475W magnitude of #74 with evolutionary synthesis models for an instantaneous burst of age 3–5Myr (Leitherer et al., 1999), from which we estimate a mass of 1.4–2$`\times 10^5M_{}`$. Therefore, its mass and size indicate that it is a young massive compact cluster, or Super Star Cluster (Whitmore, 2003).
In Fig.7 we present $`5\times 5`$ arcsec ($`110\times 110`$ pc) images of #74 obtained with FORS2 and ACS. It is apparent that the peak H$`\alpha `$ source, i.e. H ii region #35 from de Vaucouleurs et al. (1983), lies $``$2 arcsec to the S-W of the brightest continuum source (the WR cluster). The spectrum presented in Fig.2 of Paper I is that of the WR cluster, whilst the H$`\alpha `$ flux, and corresponding O7V star content of $``$810 represents the integrated total from both regions. The WR cluster provides approximately 1/3 of the total H ii luminosity, such that the WR/O ratio of this region approaches unity, comparable to the WR cluster NGC 3125-1 (Chandar et al., 2004).
#### 3.5.2 Other Clusters in M 83
Larsen (2004) has identified $``$80 young massive clusters in M 83 based on HST/WFPC2 images. Three such regions are in common with our catalogue of sources containing WR stars, namely n5236-607 (#61), -617 (#73) and -277 (#79), although none host more than a few WR stars. Larsen (priv. comm.) has compared the UBVI colours of these clusters with Solar metallicity Bruzual & Charlot (1993) models, suggesting age estimates of $`\mathrm{log}(\tau )=6.20\pm `$0.51, 6.90$`\pm `$0.54 and 9.89$`\pm `$1.87, respectively. The first two are fully consistent with a young cluster which contains WR stars, while the third suggests a dominant old population.
Five additional clusters from Larsen (2004) are also in common with our remaining candidates, namely n5236-169 (#193), -805 (#179), -818 (#163), -1011 (#157) and -1027 (#173). Of course, such candidates have the potential to also host a large WR population - indeed three of these clusters appear young ($``$1.5–6Myr) from UBVI photometry (Larsen, priv. comm.), i.e. #193, #179 and #157. Note that #179 is one of two clusters for which dynamical mass estimates has been made by Larsen & Richtler (2004). Follow-up spectroscopic observations would be required for the identification of additional WR rich clusters.
### 3.6 Comparisons with previous studies
To date, there have only been two previous studies relating to WR stars within M 83. Rosa & Richter (1988) and Bresolin & Kennicutt (2002) have both studied stellar populations within M 83 and identify six H ii regions which exhibit WR characteristics. Four of these have been re-examined in this study. Rosa & Richter obtain optical spectra with very poor signal-to-noise preventing a quantitative discussion, consequently we shall restrict any comparisons solely to results obtained by Bresolin & Kennicutt (2002).
Both studies followed a similar methodology in estimating the WR population, except that Bresolin & Kennicutt adopted a distance of 3.2Mpc to M 83 (versus 4.5Mpc adopted here). This introduces a factor of 2 between intrinsic line luminosities observed in this study and that by Bresolin & Kennicutt.
* *#40* (M83-2) – The derived WR population for this region is estimated to be 6$`\pm `$2 WC8–9, contrasting that of 1–2 WNL obtained by Bresolin & Kennicutt. We achieve a 3$`\sigma `$ detection for the $`\lambda 5696`$ and $`\lambda 5812`$ carbon features, suggesting that poor signal-to-noise in the original investigation prevented positive WC identification.
* *#41* (M83-3) – We confirm the detection of 14$`\pm `$4 WNL stars identified in region M83-3. In addition we estimate the presence of 13$`\pm `$6 WC7 stars. Bresolin & Kennicutt state that C iii may be present, but not at a significant level (versus $`5\sigma `$ here).
* *#74* (M83-8) – Bresolin & Kennicutt failed to detect any WR emission in this H ii region. However, we find the largest individual WR population of any source, namely 230 stars. As stated in Sect.3.5 the WR emission is offset by several arc-secs to the N-E of the peak $`\text{H}\alpha `$emission. Since the Bresolin & Kennicutt concentrated on bright H ii regions, their slit was probably centred on the peak $`\text{H}\alpha `$emission, such that the WR signature was missed.
* *#103* (M83-9) – Both investigations infer a late WN population. The present study obtains a population of 29$`\pm `$9 WNL stars, in agreement with that estimated by Bresolin & Kennicutt after allowing for differences in the assumed distance.
## 4 Discussion
We have identified up to $``$200 regions in the disk of M 83 that host WR stars. We now compare the properties of WR stars at the high metallicity of M 83 with those of Local Group galaxies, attempt to explain the dominant late subtypes amongst WC stars, and make comparisons with current evolutionary models.
### 4.1 Properties of WR stars at high metallicity
How do the line strengths and widths of sources containing WR stars in M 83 compare with those of other galaxies? In Fig. 8, we show the classification ratio $`\text{W}_\lambda `$ (C iv 5808) / $`\text{W}_\lambda `$ (C iii 5696) versus FWHM (C iii 5696). Data for WC7–9 stars in four Local Group galaxies are included, along with subtype boundaries as derived by Crowther et al. (1998). This figure highlights the dominance of WC8 and WC9 subtypes, which comprise 95% of the total WC content of M 83, by number.
The Galactic WC9 population is very homogeneous, centred on a FWHM of $``$25Å and $`\text{W}_\lambda `$ (C iv 5808) / $`\text{W}_\lambda `$ (C iii 5696) $``$0.3. In contrast, the WC9 population of M 83 is very heterogeneous, spanning a wide range of both FWHM and $`\text{W}_\lambda `$ (C iv 5808) / $`\text{W}_\lambda `$ (C iii 5696), with the maximum FWHM reaching $`60`$Å, 2.5 times greater than the typical Galactic WC9 star. Line widths of sources hosting WC8 stars are much greater than typical Galactic WC8 stars, whilst the few sources containing WC7 subtypes are more indicative of Milky Way counterparts.
In Fig. 9 we present C iii$`\lambda `$5696 line width versus line strength for WCL stars observed in M 83, along with data for single or binary WCL stars in Local Group galaxies. The majority of WR complexes observed in M 83 display evidence for line dilution from underlying stellar continua, since line strengths fall well below those observed in Local Group counterparts. Some WR complexes in M 83 do display similar line strengths to those in the Milky Way or M31/M33, suggesting little evidence of line dilution in those cases.
In Fig. 10 we compare $`\text{W}_\lambda `$ (C iv$`\lambda 5808`$) to FWHM (C iv$`\lambda 5808`$) for early-type WC stars in M 83, with (mostly single) Galactic and LMC counterparts. Again, the observed line strengths of sources containing early WC stars in M 83 fall below those of Galactic and LMC stars. This is again attributed to line dilution by the underlying continua from early-type stars.
From Fig. 10, the observed WCE line widths of M 83 members are generally comparable to, or lower than, those of other Local Group members, in contrast with WC8, and especially, WC9 subtypes. Indeed, there are no cases for which FWHM (C iv $`\lambda `$5808) $``$ 100Å, corresponding to WO subtypes in the Milky Way/LMC (Kingsburgh et al., 1995; Drew et al., 2004), except possibly #6 for which no evidence of O vi $`\lambda `$3811-34 is observed.
Finally, in Fig.11 we compare the equivalent width and FWHM of Heii$`\lambda `$4686 for all WN sources identified in M 83. Again, we have included data for single/binary Galactic and LMC WN stars. Aside from the effect of line dilution, some late-type WN stars in M 83 possess unusually large line widths. In some M83 complexes hosting multiple WN stars, line widths are a factor of two greater than Galaxy or LMC counterparts. These are reminiscent of observations of broad, strong N iii$`\lambda `$4640 in unresolved WR galaxies (Schmutz & Vacca, 1999).
### 4.2 Origin of late WC stars at high metallicity?
In Fig. 12 we compare the fractional distribution of WC8–9 to WC4–7 stars in galaxies with a wide range of metallicity. This clearly illustrates the extreme WCL population hosted by M 83, indicating that WCL stars are uniquely associated with metal-rich environments. In M 83 the relative number of late to early WC stars is found to be $`9`$, much greater than 0.9 and $`0.2`$ observed for the Milky Way and M 31 respectively. C iii $`\lambda `$5696 has been observed in a small number of metal-rich WR galaxies (Phillips & Conti, 1992; Pindao et al., 2002), but as these represent integrated populations, the ‘average’ WC subtype is generally WC7–8.
It has long been recognised that Milky Way WC9 stars are universally observed towards the Galactic Centre. Smith & Maeder (1991) argued that the apparent trend towards later subtypes was due to heavy mass-loss, revealing WC subtypes at an earlier evolutionary phase, assuming the surface (C+O)/He ratio decreases from early to late WC subtypes. However, subsequent spectral analysis failed to confirm any systematic trend in C/He with subtype (Kosterke & Hamann, 1995), arguing against late WC stars being exposed earlier due to prior mass-loss.
Instead, Crowther et al. (2002) claimed that WC subtypes resulted from primarily metallicity-dependent wind strengths. They suggested that the strength of C iii $`\lambda `$5696 scales very sensitively with wind density. If wind strengths increase with increasing (heavy element) metallicity, as already established for OB stars, stars which are otherwise identical will only reveal strong C iii $`\lambda `$5696 emission at high metallicities, with a corresponding late subtype.
Indeed, late-type WC stars are observed across the disk of M 83 at a lower average galactocentric distance of 2.9 $`\pm `$ 0.9 arcmin ($`\rho `$ = 0.4 $`\pm `$ 0.1 $`\rho _0`$) versus 3.6 $`\pm `$ 0.9 arcmin ($`\rho `$ = 0.5 $`\pm `$ 0.1 $`\rho _0`$) for early-type WC stars. This can be explained by the weak metallicity gradient observed in M 83 (Pilyugin et al., 2004) since stars at smaller galactocentric distances will be more metal-rich than those at larger galactocentric distances.
As discussed in the introduction, the lower mass limit for stars that ultimately become WR stars decreases with increasing metal content, i.e. the lifetimes of low (initial) mass WR stars are greatly enhanced relative to those at lower metallicity. Could late WC stars be the descendants of such (low initial mass) stars, such that they greatly outnumber the (initially more massive) early WC stars?
In this scenario, one would expect late WC stars to be observed in Milky Way clusters with low mass turn-off’s. Schild & Maeder (1986) identified the Galactic WC stars WR77 (WC8) and WR95 (WC9) in open clusters with progenitor masses as low as 35$`M_{}`$, whilst early WC stars appear to originate from more massive progenitors ($``$ 60$`M_{}`$). More recently, Massey et al. (2001) identified the late WC WR93 (WC7) with a very high progenitor mass of $`120M_{}`$, comparable to or higher than early WC subtypes. Unfortunately, WC8–9 stars were not included in their study. In addition, the late WC component of WR11 ($`\gamma `$ Vel, WC8+O7.5) originates from a mass somewhat in excess of 30$`M_{}`$, the current mass of the O star companion (De Marco & Schmutz, 1999). Overall, there is limited evidence for a distinction between the progenitor masses of early and late WC stars. Since nebular H$`\alpha `$ emission scales inversely with age, the complexes hosting early- and late-type WC stars appear to be located in both young and old clusters, such that they do indeed originate from the same parent population.
Consequently, the observed WC population in M 83 can most naturally be explained if WC winds are metallicity dependent. Increased mass-loss would elevate the strength of the classification line C iii $`\lambda `$5696 resulting in predominantly late WC subtypes. Increased mass-loss rates would, of course, have implications for the life-times of WC stars. For an initial mass of 40 M and $`Z=0.04`$, a WR star would spend 50% less time in the WC phase with a metal dependent stellar wind (See Meynet & Maeder, 2004). At the current stage, however, it is not possible to firmly exclude different progenitor masses for early and late WC populations.
Finally, a similar metallicity effect was earlier proposed by Crowther (1998), i.e. WO subtypes would be restricted to environments with weak winds (i.e. low metallicities) due to the inverse sensitivity of O vi $`\lambda `$3811-34 with increased mass-loss. The absence of WO stars in M 83 is also consistent with the inverse sensitivity of the classification line O vi $`\lambda `$3811-34 to increasing wind strength due to higher metallicity.
### 4.3 Comparison with evolutionary predictions
Surveys for WR stars in Local Group galaxies over the past three decades have revealed a strong correlation between the relative number of WC to WN stars and oxygen content of the host galaxy (Massey, 1996). Extrapolating from previous observations, one would expect N(WC)/N(WN) $`1`$ for a galaxy forming stars continuously with $``$twice the Solar oxygen content (Massey & Johnson, 1998).
A summary of results for the disk population of M 83 are presented in Table 2 together with Local Group members, such that the observed N(WC)/N(WN) ratio is presented in Fig.13. Indeed, M 83 continues the observed trend rather well with N(WC)/N(WN)$``$1.2. Undoubtedly, completeness should obviously be kept in mind given that WC stars are more readily identified in external galaxies than WN stars due to their intrinsically stronger lines. Nevertheless, our approach is optimised for net emission at $`\lambda `$4686, such that we achieve cases of 4$`\sigma `$ spectroscopic WNL detections with $`W_\lambda `$(He ii $`\lambda `$4686) $``$ 1Å.
Recent bursts of star formation may cause strong deviations from the general trend via a strong enhancement of the WC population at an age of $``$5Myr (Pindao et al., 2002). Consequently, galaxies in which there is a significant recent starburst episode may strongly deviate from this correlation. IC 10 strongly deviates from the overall trend in Fig. 13 due to an apparent ‘galaxy-wide’ starburst (Crowther et al., 2003), although the WN population may be significantly incomplete (Massey & Holmes, 2002).
In the case of M 83, since we have no information on the nuclear starburst, the above statistics should be reasonable for the quiescent star forming regions, except that the presence of a recent starburst in #74 represents a non-negligible fraction of the total disk WR population. If we were to exclude #74 from our statistics, we would obtain N(WC)/N(WN)=1 (equivalent to the Milky Way) and N(WC8–9)/N(WC4–7)=6 (vs 0.9 in the Milky Way) for the quiescent disk. The subtype ratios remain far from current evolutionary predictions at high metallicity (Meynet & Maeder, 2004).
Recently, Meynet & Maeder (2004) have constructed a set of evolutionary models for rotating (initially $`v_{\mathrm{sin}i}=300`$ km s<sup>-1</sup>) massive stars from $`Z`$=0.004 (SMC) to 0.04 ($``$M 83). For low metallicity Local Group galaxies, predictions from rotating models are in good agreement with observed WC to WN ratios (see Fig. 13). However, at higher metallicities, even allowing for metallicity dependent WR winds, evolutionary models fail and dramatically underestimate the number of WC stars, i.e. WC/WN=0.36 at $`Z`$=0.04. In fact, non-rotating models provide a better match to our observations of M 83, although such models are unsuccessful in predicting the correct fraction of Type Ib/c to Type II Supernovae at high metallicity. Consequently, there remains a significant discrepancy between the observed and predicted WR populations above Solar metallicities.
Evolutionary models distinguish between late and early WN subtypes via the presence or absence of hydrogen, whilst spectroscopic definitions relate to the observed ionization of nitrogen lines. Determinations of hydrogen content are possible for Local Group WN stars, but the strong nebulosity and potential multiplicity for sources at the distance of M 83 prevent such measurements. For a twice Solar metallicity ($`Z`$=0.04), Meynet & Maeder (2004) predicts WNL/WNE $``$ 4 by number, allowing for metallicity dependent WR wind strengths. This should provide a reasonable analogue to the observational statistics for M 83, on the basis of a good correspondence between late WN stars (with hydrogen) and early WN stars (without hydrogen) in the Milky Way. Consequently, as with the WC to WN ratio, the predicted distribution amongst WN subtypes ($``$4) differs from observations ($``$1) by a significant factor.
Nitrogen in WN stars is of course partially processed from carbon and oxygen, such that the CNO equilibrium abundance linearly scales with metallicity. As a consequence, the abundance of nitrogen in M 83 WN stars will exceed that in the Milky Way and other Local Group galaxies. Crowther (2000) demonstrated that for otherwise identical parameters, N iii $`\lambda `$4634–41 reacts more sensitively than N iv $`\lambda `$4058 to increased nitrogen content, i.e. a later WN subtype results. If WN winds are also metallicity dependent, the effect will be magnified such that one will expect a predominantly late WN population at high metallicities.
Indeed, given the weak metallicity gradient of M 83 (Pilyugin et al., 2004), late-type WN stars are located within the inner regions of M 83, at an average galactocentric distance of 2.2$`\pm `$0.9 arcmin ($`\rho `$ = 0.3$`\pm `$0.1 $`\rho _0`$) versus 3.5$`\pm `$1.3 arcmin ($`\rho `$ = 0.5$`\pm `$0.2 $`\rho _0`$) for early-type WN stars.
## 5 Summary
Our analysis of the VLT/FORS2 imaging and spectroscopic data indicates that the disk of M 83 hosts a large WR population. Using narrow-band optical images we have identified 283 candidate WR regions within M 83, of which 198 have been spectroscopically observed. Of these we find that 132 regions contain WR stars. Absolute WR populations have been derived using line flux conversions adapted from Schaerer & Vacca (1998). We estimate a total WR population of 1035$`\pm `$300, consisting of 564$`\pm `$170 WC and 471$`\pm `$130 WN stars from this population, i.e. a quiescent N(WC)/N(WN) ratio of $``$1.2, or $``$1.0 excluding the starburst cluster #74. This differs greatly from current evolutionary predictions at high metallicity, which suggest N(WC)/N(WN)$``$0.36, even allowing for metallicity dependent WR mass-loss rates (Meynet & Maeder, 2004).
The observed statistics exclude both the potentially large WR population in the central starburst, plus the population of perhaps $``$250 WR stars resulting from remaining candidates. Pellerin (2004) has carried out spectral synthesis of $`FUSE`$ observations of the nucleus of M 83(30$`{}_{}{}^{\prime \prime }\times 30^{\prime \prime }`$) suggesting a mass of 1.5$`\times 10^6M_{}`$ and age 3.5 Myr, with an inferred WR population of 1700. This appears plausible given that the nuclear starburst has a star formation rate approaching that of the disk (Harris et al., 2001; Bell & Kennicutt, 2001). Consequently, the total WR population of M 83 may exceed 3000.
Using the WR population derived in this study, the global surface density of M 83 is found to be $``$3 WR/kpc<sup>2</sup>, typical of that observed in the Solar Neighbourhood and M33 (Massey & Johnson, 1998). If the nucleus and remaining candidates are accounted for this likely to increase this by a factor of $``$3.
The WC population of M 83 is dominated by late-type stars. The relative number of WC8–9 to WC4–7 stars is found to be $`9`$ (or $``$6 excluding cluster #74), outnumbering that of any other Local Group galaxy tenfold, as illustrated in Fig. 12. WO stars are not observed in M 83 suggesting that there is a genuine trend to later subtype at higher metallicities. Observed line widths in early WC subtypes appear to be comparable to those observed in the Milky Way and LMC. This population is most readily explained by a metallicity dependent wind strength amongst WC subtypes.
M 83 has a substantial WN population, evenly split between early and late subtypes. The high WNL population likely results from the sensitivity of nitrogen diagnostics to the high global metallicity, whilst evidence in favour of metallicity dependent winds amongst WN subtypes is less clear. Evolutionary models, in contrast, predict a far higher late WN population at high metallicities, with WNL/WNE$``$4–5 (Meynet & Maeder, 2004).
From the present study we infer 9 complexes in M 83 which contain a large WR population ($`>20`$), with #74 hosting over 200 late-type WN and WC stars, outnumbering 30 Doradus tenfold. HST/ACS images indicate that #74 is a compact cluster, whilst several other massive compact clusters from Larsen (2004) host more modest WR populations.
Within galaxies located at up to $``$10 Mpc, how unusual are the sources in M 83 with regard to WR content? Schaerer et al. (1999a) identified $``$40 WN and WC stars in clusters A and B of NGC 5253, similar to #41 in M 83, whilst He 2–10 dwarfs even #74, with 1100 WN and $`>`$250 WC stars. Recently, Chandar et al. (2004) use HST/STIS UV spectroscopy to claim NGC 3125-1 hosts 5000 WNL stars, a factor of ten times greater than optically derived by Schaerer et al. (1999a), and comparable to the global WR population in M 83. Consequently, the WR population of M 83, and source #74 in particular, is extreme only with respect to Local Group galaxies. Where our results for M 83 stand out from previous studies is the ability to resolve the disk WR population into $``$200 regions.
Given such a large WR population in the optically visible disk of M 83, it is likely that one such massive star will undergo a core-collapse each century, given typical WR lifetimes of $``$10<sup>5</sup> yr. Indeed, SN1983N (Porter & Filippenko, 1987), one of the six optically visible SN that have been reported in M83 since 1923, was a type Ib SN, for which WR stars are considered as the most probable precursors.
In order to directly witness a WR explode as a supernova on a shorter timeframe of a decade or so, a survey of the WR population in 10–20 nearby ($`<`$10Mpc) massive star forming galaxies would be required. The present observational program will continue towards this goal, complementing existing broad-band pre-Supernova surveys (Smartt et al., 2003; Van Dyk et al., 2003).
###### Acknowledgements.
LH acknowledges financial support from PPARC and a Royal Society summer studentship award, PAC acknowledges financial support from the Royal Society. We thank Soeren Larsen for providing ages estimates of clusters from broad-band photometry. Some of the data presented in this paper were obtained from the Multimission Archive at the Space Telescope Science Institute (MAST). STScI is operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS5-26555. Support for MAST for non-HST data is provided by the NASA Office of Space Science via grant NAG5-7584 and by other grants and contracts.
## Appendix A Catalogue
Tables A1 and A2 list properties of confirmed and candidate regions containing WR stars in M 83.
## Appendix B Finding charts (electronic only)
Fig. B1 provides a master key to individual finding charts B2–B17, obtained from FORS2 $`\lambda `$4684 images.
|
warning/0506/math0506334.html
|
ar5iv
|
text
|
# On the X-rays of Permutations
## 1 Introduction
Let $`𝒮_n`$ be the set of all permutations of $`[n]=\{1,2,\mathrm{},n\}`$ and let $`P_\pi `$ be the permutation matrix corresponding to $`\pi 𝒮_n`$. For $`k=2,\mathrm{},2n`$, the $`(k1)`$-th *antidiagonal sum* of $`P_\pi `$ is $`x_{k1}(\pi )=_{i+j=k}[P_\pi ]_{i,j}`$. The sequence of nonnegative integers $`x(\pi )=x_1(\pi )x_2(\pi )\mathrm{}x_{2n1}(\pi )`$ is called the (*antidiagonal*) *X-ray* of $`\pi `$. The *diagonal X-ray* of $`\pi `$, denoted by $`x_d(\pi )`$, is similarly defined. Note that $`x(\pi )=x(\pi ^1)`$, for every $`\pi 𝒮_n`$. The sequence $`x(\pi )`$ may be also seen as a word over the alphabet $`[n]`$. As an example, the following table contains the X-rays of all permutations in $`𝒮_3`$:
| $`\pi `$ | $`x\left(\pi \right)`$ | $`\pi `$ | $`x\left(\pi \right)`$ | $`\pi `$ | $`x\left(\pi \right)`$ | $`\pi `$ | $`x\left(\pi \right)`$ | $`\pi `$ | $`x\left(\pi \right)`$ |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| $`123`$ | $`10101`$ | $`231,312`$ | $`01110`$ | $`132`$ | $`10020`$ | $`213`$ | $`02001`$ | $`321`$ | $`00300`$ |
Although X-rays of permutations are interesting object on their own, among the reasons why they are of general interest in Discrete Tomography is that many types of integral matrices can be written as linear combinations of permutation matrices (for example, binary matrices with equal row-sums and column-sums, like the adjacency matrices of Cayley graphs). Deciding whether for a given word $`w=w_1\mathrm{}w_{2n1}`$ there exists $`\pi 𝒮_n`$ such that $`w=x(\pi )`$ is an NP-complete problem (see also ). The complexity is polynomial if the permutation matrix is promised to be wrapped around a cylinder . It is necessary to keep into account that permutations are not generally specified by their X-rays: just consider the permutation $`\pi =73142865`$ and $`\sigma =72413865`$; we have $`x(\pi )=x(\sigma )=000110200002100`$, $`x_d(\pi )=x_d(\sigma )=00021111100010`$ and $`\pi \sigma ^1`$. This hints that an issue concerning X-rays is to quantify how much information about $`\pi `$ is contained in $`x(\pi )`$. In this paper we present some connections between X-rays of permutations and a variety of combinatorial objects. From a practical perspective, this may be useful in isolating and approaching special cases of the above problem.
The remainder of the paper is organized as follows. In Section 2 we consider the problem of counting X-rays. We prove a bijection between X-rays and nondecreasing differences of permutations. We define the *degeneracy* of an X-ray $`x(\pi )`$ as the number of permutations $`\sigma `$ such that $`x(\pi )=x(\sigma )`$, and we characterize the X-rays with the maximum degeneracy. We prove a bijection between X-ray of length $`4k+1`$ having maximum degeneracy and zero-sum arrays. In Section 3 we consider the notion of simple permutations. This notion seems to provide a good framework to study the degeneracy of X-rays, but the relation between simple permutations and X-rays with small degeneracy remains unclear. Section 4 is devoted to binary X-rays, that is X-rays whose entries are only zeros and ones. We characterize the X-rays of circulant permutation matrices of odd order. Moreover, we present a relation between binary X-rays, the $`n`$-queens problem (see, *e.g.*, ), the score sequences of tournaments on $`n`$ vertices (see \[10, Sequence A000571\]), and extremal Skolem sequences, see \[7, Conjecture 2.2\].
A number of conjectures and open problems will be explicitly formulated or will simply stand out from the context. We use the standard notation for integers sequences from the OEIS .
## 2 Counting X-rays
We begin by addressing the following natural question: what is the number of different X-rays of permutations in $`𝒮_n`$? Although we are unable to find a generating function for the sequence, we show a bijection between X-rays and nondecreasing differences of permutations. The *difference* of permutations $`\pi ,\sigma 𝒮_n`$ is the integers sequence $`\pi \sigma =(w_1,w_2,\mathrm{},w_n)`$, where $`w_1=\pi _1\sigma _1,w_2=\pi _2\sigma _2,\mathrm{},w_n=\pi _n\sigma _n`$. For example, if $`\pi =1234`$ and $`\sigma =2413`$, we have $`e2413=(1,2,2,1)`$. Let $`x_n`$ be the numbers of different X-rays of permutations in $`𝒮_n`$. Let $`d_n`$ be the number of nondecreasing differences of permutations in $`𝒮_n`$. The number $`d_n`$ equals the number of different differences $`e\sigma `$ with entries rearranged in the nondecreasing order. In other words, $`d_n`$ equals the number of different multisets of the form $`M(\sigma )=\{1\sigma _1,2\sigma _2,\mathrm{},n\sigma _n\}`$, with entries rearranged in the nondecreasing order. The entries of $`x(\pi )`$ are then the entries of the vector $`e_{1\sigma _1}+e_{2\sigma _2}+\mathrm{}+e_{n\sigma _n}`$, where $`e_i`$ is the $`i`$-th coordinate vector of length $`2n1`$. For example, for $`\pi =3124`$ we have $`x(3124)=0101200`$ and $`e_{13}+e_{21}+e_{32}+e_{44}=(0,1,0,0,0,0,0)+(0,0,0,0,1,0,0)+(0,0,0,0,1,0,0)+(0,0,0,1,0,0,0)=(0,1,0,1,2,0,0)`$. On the basis of this reasoning we can state the following result.
###### Proposition 1
The number $`x_n`$ of different X-rays of permutations in $`𝒮_n`$ is equal to the number $`d_n`$ (see \[10, Sequence A019589\]) of nondecreasing differences of permutations in $`𝒮_n`$.
Let us define and denote the *degeneracy* of an X-ray $`x(\pi )`$ by
$$\delta (x(\pi ))=|\{\sigma :x(\sigma )=x(\pi )\}|.$$
If $`x(\pi )`$ is such that $`\delta (x(\pi ))\delta (x(\sigma ))`$ for all $`\sigma S_n`$, we write $`x_{\mathrm{max}}^n=x(\pi )`$ and we say that $`x(\pi )`$ has *maximum degeneracy*. The following table contains $`x_n`$, $`x_{\mathrm{max}}^n`$ and $`\delta (x_{\mathrm{max}}^n)`$ for $`n=1,\mathrm{},8`$.
| $`n`$ | $`x_n`$ | $`x_{\mathrm{max}}^n`$ | $`\delta (x_{\mathrm{max}}^n)`$ | $`n`$ | $`x_n`$ | $`x_{\mathrm{max}}^n`$ | $`\delta (x_{\mathrm{max}}^n)`$ |
| --- | --- | --- | --- | --- | --- | --- | --- |
| $`1`$ | $`1`$ | $`1`$ | $`1`$ | $`5`$ | $`59`$ | $`001111100`$ | $`6`$ |
| $`2`$ | $`2`$ | $`020,101`$ | $`1`$ | $`6`$ | $`246`$ | $`00011211000`$ | $`12`$ |
| $`3`$ | $`5`$ | $`01110`$ | $`2`$ | $`7`$ | $`1105`$ | $`0001111111000`$ | $`28`$ |
| $`4`$ | $`16`$ | $`0012100`$ | $`3`$ | $`8`$ | $`5270`$ | $`000011121110000`$ | $`76`$ |
It is not difficult to characterize the X-rays with maximum degeneracy. One can verify by induction that for $`n`$ even,
$$x_{\mathrm{max}}^n=00\mathrm{}011\mathrm{}121\mathrm{}110\mathrm{}00,$$
with $`n/2`$ left-zeros and right-zeros, and $`n/21`$ ones; for $`n`$ odd,
$$x_{\mathrm{max}}^n=\mathrm{00..011}\mathrm{}\mathrm{110..00},$$
with $`(n1)/2`$ left-zeros and right-zeros, and $`n`$ ones. Notice that if $`x(\pi )=x_{\mathrm{max}}^n`$ (for $`n`$ odd) then $`P_\pi `$ can be seen as an hexagonal lattice with all sides of length $`\left(n+1\right)/2`$. In each cell of the lattice there is $`0`$ or $`1`$, and $`1`$ is in exactly $`n`$ cells; the column-sums are $`1`$ and the diagonal and anti-diagonal sums are $`0`$. This observation describes a bijection between permutations of odd order whose X-ray is $`x_{\mathrm{max}}^n`$ and zero-sum arrays. An $`(m,2n+1)`$-*zero-sum* array is an $`m\times (2n+1)`$ matrix whose $`m`$ rows are permutations of the $`2n+1`$ integers $`n,n+1,\mathrm{},n`$ and in which the sum of each column is zero . The matrix
$$\left[\begin{array}{ccc}\hfill 1& \hfill 0& \hfill 1\\ \hfill 0& \hfill 1& \hfill 1\\ \hfill 1& \hfill 1& \hfill 0\end{array}\right]$$
is an example of $`(3,3)`$-zero-sum array. Thus we have the next result.
###### Proposition 2
The number $`\delta (x_{\mathrm{max}}^n)`$ for $`n`$ odd is equal to the number of $`(3,2n+1)`$-zero-sum arrays (see \[10, Sequence A002047\]).
Before concluding the section, it may be interesting to notice that if we sum entry-wise the X-rays of all permutations in $`𝒮_n`$ we obtain the following sequence of $`2n1`$ terms:
$$(n1)!,2(n1)!,\mathrm{},(n1)(n1)!,n!,(n1)(n1)!,\mathrm{},2(n1)!,(n1)!.$$
The meaning of the terms of this sequence is clear.
## 3 Simple permutations and X-rays
In the previous section we have considered the X-rays with maximum degeneracy. What can we say about X-rays with degeneracy $`1`$? If $`\delta (x(\pi ))=1`$ then $`\pi `$ is an involution (in such a case $`P_\pi =P_\pi ^1`$) but the converse if not necessarily true. In fact consider the involution $`\pi =1267534`$. One can verify that $`x(\pi )=x(\sigma )=x(\rho )=1010000212000`$, for $`\rho =1275634`$ and $`\sigma =1267453`$. In a first approach to the problem, it seems useful to study what kind of operations can be done “inside” a permutation matrix $`P_\pi `$ in order to obtain another permutation, say $`P_\sigma `$, such that $`x(\pi )=x(\sigma )`$ and $`P_\pi P_\sigma ^1`$. A intuitively good framework for this task is provided by the notion of *block permutation*. A *segment* and a *range* of a permutation are a set of consecutive positions and a set of consecutive values. For example, in the permutation $`34512`$, the segment formed by the positions $`2,3,4`$ is occupied by the values $`4,5,1`$; the elements $`1,2,3`$ form a range. A *block* is a segment whose values form a range. Every permutation has singleton blocks together with the block $`12\mathrm{}n`$. A permutation is called *simple* if these are the only blocks . A permutation is said to be a *block permutation* if it is not simple. Note that if $`\pi `$ is simple then it is $`\pi ^1`$. Let $`S=(\pi _1𝒮_{n_1},\mathrm{},\pi _k𝒮_{n_k})`$ be an ordered set and let $`\pi 𝒮_k`$. We assume that in $`S`$ there exists $`1ik`$ such that $`n_i>1`$. We denote by $`P(\pi ,S)`$ the $`(n_1+\mathrm{}+n_k)`$-dimensional permutation matrix which is partitioned in $`k^2`$ blocks, $`B_{1,1},\mathrm{},B_{k,k}`$, such that $`B_{i,j}=P_{\pi _i}`$ if $`\pi (i)=j`$ and $`B_{i,j}=0`$, otherwise. We denote by $`\pi [\pi _1,\mathrm{},\pi _k]`$ (or equivalently by $`(\pi )[S]`$) the permutation corresponding to $`P(\pi ,S)`$. For example, let $`S=(231,21,312)`$ and $`\pi =231`$. Then
$$P(231,S)=\left[\begin{array}{ccc}0& P_{231}& 0\\ 0& 0& P_{21}\\ P_{312}& 0& 0\end{array}\right]$$
and $`231[S]=231[231,21,312]=56487312`$. The matrix $`P(231,S)`$ can be modified leaving the X-ray of $`(\pi )[S]`$ invariant:
$$P(231,(312,21,312))=\left[\begin{array}{ccc}0& P_{312}=P_{231}^T& 0\\ 0& 0& P_{21}\\ P_{312}& 0& 0\end{array}\right].$$
It is clear that $`56487312`$ is a block permutation. Let $`\pi `$ be a simple permutation then possibly $`\delta (x(\pi ))>1`$. In fact, the permutation $`\pi =531642`$ is simple, but $`\delta (x(\pi ))=6`$, since $`x(\pi )=00111011100=x(526134)=x(461253)`$, plus the respective inverses. The permutations $`526134`$ and $`461253`$ are decomposable. This means that there possibly exists a decomposable permutation $`\sigma `$ such that $`x(\sigma )=x(\pi )`$, even if $`\pi `$ is simple. There relation between simple permutations and X-rays of small degeneracy is not clear. Intuitively, a simple permutation allows less “freedom of movement” than a block permutation. It is also intuitive that we have low degeneracy when the nonzero entries of the X-ray are “distributed widely” among the $`2n1`$ coordinates. The following result is easily proved.
###### Proposition 3
Let $`\sigma =\pi [S]=\pi [\pi _1,\mathrm{},\pi _k]`$ be a block permutation. Then $`\delta (x(\pi ))>1`$ if one of the following two conditions is satisfied:
(1) If $`\pi 12\mathrm{}n`$ then there is at least one $`\pi _iS`$ which is not an involution;
(2) If $`\pi =12\mathrm{}n`$ then there are at least two $`\pi _i,\pi _iS`$ which are not involution.
Proof. (1) Let $`\pi 12\mathrm{}n`$ be any permutation. Take $`\pi _i^1`$ for some $`\pi _iS`$. Let $`\rho =\pi [\pi _1,\mathrm{},\pi _i^1,\mathrm{},\pi _k]`$. Since $`\sigma `$ is a block permutation, $`x(\sigma )=x(\rho )`$. However, if $`\pi _i\pi _i^1`$ then $`\sigma \rho `$ and $`\sigma ^1\sigma `$. It follows that $`x(\sigma )`$ does not specify $`\sigma `$. (2) Let $`\pi =12\mathrm{}n`$. Let all elements of $`S`$ be involutions except $`\pi _i`$. Take $`\pi _i^1`$. Let $`\rho =\pi [\pi _1,\mathrm{},\pi _i^1,\mathrm{},\pi _k]`$. Again, $`x(\sigma )=x(\rho )`$, but this time $`\rho =\sigma ^1`$. Then $`x(\sigma )`$ possibly specifies $`\sigma `$. If, for distinct $`i,j`$, there are $`\pi _i,\pi _jS`$ such that $`\pi _i\pi _i^1`$ and $`\pi _j\pi _j^1`$ then
$$x(\sigma ^{})=x(\pi [\pi _1,\mathrm{},\pi _i^1,\mathrm{},\pi _j^1,\mathrm{},\pi _k])=x(\sigma ),$$
but $`x(\sigma )`$ does not specify $`\sigma `$, given that $`\rho \sigma ^1`$.
This is however not a sufficient condition for having $`\delta (x(\pi ))>1`$. Permutations with equal X-rays are said to be in the same *degeneracy class*. The table below contain the number of permutations in $`𝒮_n`$ which are in each degeneracy class, and the number of different degeneracy classes with the same cardinality, for $`n=2,\mathrm{},8`$. These numbers provide a partition on $`n!`$. We denote by $`C(n)`$ the total number of degeneracy classes. We write $`a(b)`$, where $`a`$ is the number of permutations in the degeneracy class and $`b`$ the number of degeneracy classes of the same cardinality:
| $`C(2)=1`$: 1(2) |
| --- |
| $`C(3)=2`$: 1(4),2(1) |
| $`C(4)=3`$: 1(9),2(6),3(1) |
| $`C(5)=5`$: 1(20),2(26),3(6),4(6),6(1) |
| $`C(6)=10`$: 1(49),2(100),3(19),4(43),5(1),6(19),7(2),8(11),9(1),2(1) |
| $`C(7)=20`$: 1(114),2(345),3(60),4(229),5(18),6(118),7(11),8(98),10(29) |
| 11(2),12(33),14(13),16(14),18(6),20(4),21(1),22(2),26(1),28(1). |
.
We conjecture that if $`\delta (x(\pi ))=1`$ then $`x(\pi )`$ does not have more than $`2`$ adjacent nonzero coordinates. However the converse is not true if $`\pi 𝒮_n`$ for $`n8`$: for $`\pi =17543628`$ and $`\sigma =16547328`$, we have $`x(\pi )=x(\sigma )=100000320010001`$, but there are no more than $`2`$ adjacent coordinates.
## 4 Binary X-rays
In general, it does not seem to be an easy task to characterize X-rays. A special case is given by X-rays associated with circulant permutation matrices, for which is available an exact characterization. An X-ray $`x(\pi )`$ is said to be *binary* if $`x_i(\pi )\{0,1\}`$ for every $`1i2n1`$. The set all permutations in $`𝒮_n`$ with binary X-ray is denoted by $`_n`$. Counting binary X-rays means solving a modified version of the $`n`$-queens problem (see, *e.g.*, ) in which two queens do not attack each other if they are in the same NorthWest-SouthEst diagonal. The permutations with binary X-rays associated to circulant matrices are characterized in a straightforward way. Let $`C_n`$ be the permutation matrix associated with the permutation $`c_n=23\mathrm{}n1`$, that is the *basic circulant permutation matrix*. The matrices in the set $`𝒞_n=\{C_n^0,C_n,C_n^2,\mathrm{},C_n^{n1}\}`$ ($`C_n^0`$ is the identity matrix) are called the *circulant permutation matrices*. The matrix $`C_n^k`$ is associated to $`c_n^k`$. Observe that $`x(\pi )`$ can be seen as a binary number, since $`x_i(\pi )\{0,1\}`$ for every $`i`$. Let
$$d_j(\pi )=2^{2n1j}x_j(\pi ),j=1,2,\mathrm{},2n1,$$
and $`d(\pi )=_{i=1}^{2n1}d_i(\pi )`$, that is the decimal expansion of $`x(\pi )`$. The table below lists the X-rays of $`𝒞_3,𝒞_5`$ and $`𝒞_7`$, and their decimal expansions:
$$\begin{array}{cccccc}\pi \hfill & x(\pi )\hfill & d(\pi )\hfill & \pi \hfill & x(\pi )\hfill & d(\pi )\hfill \\ 123\hfill & 10101\hfill & 21\hfill & 12345\hfill & 101010101\hfill & 341\hfill \\ 231\hfill & 01110\hfill & 14\hfill & 23451\hfill & 010111010\hfill & 186\hfill \\ & & & 34512\hfill & 001111100\hfill & 124\hfill \end{array}.$$
For $`\pi =c_n^k`$, one can verify that
$$\begin{array}{cc}d(\pi )\hfill & =\frac{1}{6}2^{\frac{3}{2}n+\frac{1}{2}+k}\frac{1}{6}2^{\frac{1}{2}n+\frac{1}{2}+k}+\frac{1}{3}2^{\frac{3}{2}n+\frac{1}{2}k}\frac{1}{3}2^{\frac{1}{2}n+\frac{1}{2}k}\hfill \\ & =a(k)\left(2^n1\right)\left(2^n1\right)2^{\frac{1}{2}nk+\frac{1}{2}},\hfill \end{array}$$
where $`a(k)=(2^{2k1}+1)/3`$ (A007583).
In the attempt to count binary X-rays, we are able to establish a bijection between these objects and score sequences of tournaments. A *tournament* is a loopless digraph such that for every two distinct vertices $`i`$ and $`j`$ either $`(i,j)`$ or $`(j,i)`$ is an arc . The *score sequence* of an tournament on $`n`$ vertices is the vector of length $`n`$ whose entries are the out-degrees of the vertices of the tournament rearranged in nondecreasing order.
###### Proposition 4
Let $`b_n`$ be the number of binary X-rays of permutations in $`S_n`$ and let $`s_n`$ be the number of different score sequences of tournaments on $`n`$ vertices (see \[10, Sequence A000571\]). Then $`b_ns_n`$.
Proof. The number $`s_n`$ equals the number of integers lattice points $`(p_0,\mathrm{},p_n)`$ in the polytope $`P_n`$ given by the inequalities $`p_0=p_n=0`$, $`2p_ip_{i+1}p_{i1}1`$ and $`p_i0`$, for $`i=1,\mathrm{},n1`$, see . Let $`x_1,\mathrm{},x_n`$ be the coordinates related to $`p_1,\mathrm{},p_n`$ by $`p_i=x_1+\mathrm{}+x_ii^2`$, for $`i=1,\mathrm{},n`$. We can rewrite the inequalities defining the polytope $`P_n`$ in these coordinates as follows: $`x_1+\mathrm{}+x_ii^2`$, $`x_{i+1}x_i+1`$ and $`x_1+\mathrm{}+x_n=n^2`$. For a permutation $`w𝒮_n`$ with a binary X-ray, let $`l_i=l_i(w)`$ be the position of the $`i`$-th ‘$`1`$’ in its X-ray. In other words, the sequence $`(l_1,\mathrm{},l_n)`$ is the increasing rearrangement of the sequence $`(w_1,w_2+1,w_3+2,\mathrm{},w_n+n1)`$. Then the numbers $`l_1,\mathrm{},l_n`$ satisfy the inequalities defining the polytope $`P_n`$ (in the $`x`$-coordinates). Indeed, $`l_1+\mathrm{}+l_n=w_1+(w_2+1)+\mathrm{}+(w_n+n1)=n^2`$; $`l_{i+1}l_i+1`$; and the minimal possible value of $`l_1+\mathrm{}+l_i`$ is $`(1+0)+(2+1)+\mathrm{}+(i+(i1))=i^2`$. This finishes the proof. In order, to prove that $`b_n=s_n`$ it is enough to show that, for any integer point $`(x_1,\mathrm{},x_n)`$ satisfying the above inequalities, we can find a permutation $`w𝒮_n`$ with $`x_i=l_i(w)`$.
###### Conjecture 5
All binary X-rays of permutations in $`𝒮_n`$ are in a bijective correspondence with integer lattice points $`(x_1,\mathrm{},x_n)`$ of the polytope given by the inequalities
$$\begin{array}{c}x_1+\mathrm{}+x_ii^2,i=1,\mathrm{},n;\hfill \\ x_1+\mathrm{}+x_n=n^2,\hfill \\ x_{i+1}x_i1,i=1,\mathrm{},n1.\hfill \end{array}$$
For a permutation $`wS_n`$, the corresponding sequence $`(x_1,\mathrm{},x_n)`$ is defined as the increasing rearrangement of the sequence $`(w_1,w_2+1,w_3+2,\mathrm{},w_n+n1)`$.
Again, it is clear that X-rays injectively map into the integer points of the above polytope. One needs to show that there will be no gaps in the image. Also, it can be shown that the above conjecture is equivalent to Conjecture 2.2 from concerning extremal Skolem sequences. The conjecture turns out to be false, when not restricted to binary X-rays.
We conjecture also that the number of different X-rays of permutations in *$`𝒮_n`$* whose possible entries are $`0`$ and $`2`$ is equal to the number of score sequences in tournament with $`n`$ players, when 3 points are awarded in each game (see \[10, Sequence A047729\]).
## 5 Palindromic X-rays
What can we say about X-rays with special symmetries? The *reverse* of $`x(\pi )`$, denoted by $`\overline{x}(\pi )`$, is the mirror image of $`x(\pi )`$. If $`x(\pi )=\overline{x}(\pi )`$ then $`\pi `$ is said to be *palindromic*. The *reverse* of $`\pi `$, denoted by $`\overline{\pi }`$, is mirror image of $`\pi `$. For example, if $`\pi =25143`$ then $`\overline{\pi }=34152`$. The permutation matrix $`P_{\overline{\pi }}`$ is obtained by writing the rows of $`P_\pi `$ in reverse order. In general $`\overline{x}(\pi )x(\overline{\pi })`$. In fact, for $`\pi =25143`$, we have $`x(\pi )=0011001200`$, $`\overline{x}(\pi )=0021001100`$ and $`x(\overline{\pi })=0020011010`$. We denote by $`|M`$ and $`\underset{¯}{M}`$ the matrices obtained by writing the columns and the rows of a matrix $`M`$ in reverse order, respectively.
###### Proposition 6
Let $`l_n`$ be the number of permutations in $`𝒮_n`$ with palindromic X-rays and let $`i_n`$ be the number of involutions in $`𝒮_n`$ (see \[10, Sequence A000085\]). Then, in general, $`l_n>i_n`$.
Proof. Recall that a permutation $`\pi `$ is an *involution* if $`\pi =\pi ^1`$. Since $`P_\pi =P_\pi ^T`$, it is clear that the diagonal X-ray of an involution $`\pi `$ is palindromic. The X-ray of $`\sigma `$ such that $`P_\sigma =|P_\pi `$ is then also palindromic. This shows that $`l_ni_n`$. Now, consider a permutation matrix of the form
$$P_\sigma =\left[\begin{array}{cc}P_\rho & 0\\ 0& P_\rho ^T\end{array}\right],$$
for some permutation $`\rho `$ which is not an involution. Then $`P_\rho P_\rho ^T`$, $`P_\sigma P_\sigma ^T`$ and $`\sigma `$ is not an involution, but the diagonal X-ray of $`\sigma `$ is palindromic. The X-ray of $`\pi `$ such that $`P_\pi =|P_\sigma `$ is then also palindromic. This proves the proposition.
The next contains the values of $`l_n`$ for small $`n`$:
n
ln
n
ln
n
ln
n
ln2241261288211034532743698814.fragments
n
ln
n
ln
n
ln
n
ln2241261288211034532743698814\begin{tabular}[]{||l|l||l|l||l|l||l|l||}\hline\cr$n$&$l_{n}$&$n$&$l_{n}$&$n$&$l_{n}$&$n$&$l_{n}$\\
\hline\cr$2$&$2$&$4$&$12$&$6$&$128$&$8$&$2110$\\
\hline\cr$3$&$4$&$5$&$32$&$7$&$436$&$9$&$8814$\\
\hline\cr\end{tabular}\ .
###### Proposition 7
Let $`l_{n,A=D}`$ be the number of permutations in $`𝒮_n`$ with:
(1) equal diagonal and antidiagonal X-rays;
(2) palindromic X-rays.
Let $`r_n`$ be the number of permutations in $`𝒮_n`$ invariant under the operation of first reversing and then taking the inverse (see \[10, Sequnce A097296\]). Then, in general, $`l_{n,A=D}>r_n`$.
Proof. We first construct the permutations which are invariant under the operation of first reversing and then taking the inverse. Let $`\pi 𝒮_n`$ where $`n=2k`$. We look at $`P_\pi `$ as partitioned in $`4`$ blocks:
$$P_\pi =\left[\begin{array}{cc}A& B\\ C& D\end{array}\right].$$
If
$$P_\pi =(\underset{¯}{P_\pi })^T=\left[\begin{array}{cc}A& B\\ C& D\end{array}\right]^T=\left[\begin{array}{cc}\underset{¯}{B}& \underset{¯}{A}\\ \underset{¯}{D}& \underset{¯}{C}\end{array}\right]^T=\left[\begin{array}{cc}(\underset{¯}{B})^T& (\underset{¯}{D})^T\\ (\underset{¯}{A})^T& (\underset{¯}{C})^T\end{array}\right]$$
then $`A=(\underset{¯}{B})^T,B=(\underset{¯}{D})^T,C=(\underset{¯}{A})^T`$ and $`D=(\underset{¯}{C})^T`$. This implies the X-ray of $`P_\pi `$ being palindromic and, moreover, the diagonal and antidiagonal X-rays being equal. Note that we can construct $`P_\pi `$ only if $`n0(\mathrm{m}od4)`$, and in this case $`r_n0`$. However, fixed $`n0(\mathrm{m}od4)`$, we have $`r_n=r_{n+1}`$, since the permutation matrix
$$P_\sigma =\left[\begin{array}{ccc}A& & \mathrm{𝟎}\\ & 1& \\ \mathrm{𝟎}& & D\end{array}\right]+\left[\begin{array}{ccc}\mathrm{𝟎}& & B\\ & 1& \\ C& & \mathrm{𝟎}\end{array}\right]$$
can be always constructed from $`P_\pi `$. (Permutation matrices like $`P_\pi `$ and $`P_\sigma `$ provide the solutions of the “rotationally invariant” $`n`$-rooks problem. This points out that A097296 and A037224 are indeed the same sequence.) Now, the proposition is easily proved by observing that, for $`\rho =369274185`$, $`P_\rho `$ is not of the form of $`P_\sigma `$. A direct calculation shows that $`r_9=12`$ and $`l_{9,A=D}=20`$.
|
warning/0506/astro-ph0506283.html
|
ar5iv
|
text
|
# IMF biases and how to correct them
## 1. Introduction
In order to obtain the mass function of a stellar population from photometric data one starts by placing the stars in a color-magnitude or in a theoretical (temperature-luminosity) HR diagram along with the evolutionary tracks and the corresponding isochrones. For a simple population one can then find the appropriate isochrone and obtain the masses for each star. For a complex star formation history, one has to obtain the full transformation from temperature and luminosity (or color and magnitude) to mass and age. The number of stars as a function of mass (corrected for age effects if necessary) can then be used to obtain the initial mass function (or IMF).
A number of problems and biases can arise along the way to obtaining the true IMF due to the oversimplification of the assumptions (e.g. using isochrone fitting for a population with an age spread), the existing intrinsic degeneracies (e.g. the complicated topology of evolutionary tracks, metallicity effects, and rotation can make two stars of different masses and ages have the same temperature and luminosity), as well as for other reasons. In this work we will analyze three sources of systematic effects: the numerical bias introduced by the use of constant-size bins for the fitting of the IMF, the “mass diffusion” from low to high masses due to photometric uncertainties, and the existence of unresolved multiples. Those three effects go in the same direction of making the measured IMF flatter than the real one and can affect different samples to different degrees, thus introducing the possibility of yielding a dispersion of measured IMF values where only a single one exists in reality.
## 2. Binning biases
The first type of bias we will discuss is purely numerical and affects not only IMF determinations but the fitting of any function to binned data that follows Poisson or multinomial statistics (Wheaton et al. 1995; Lucy 2000) and was analyzed by Maíz Apellániz & Úbeda (2005). Suppose we have measured the masses ($`m`$) for a set of stars to which we want to fit a power law of the form:
$$\frac{dn}{dm}=Am^\gamma ,$$
(1)
where $`dn`$ is the number of stars with mass in the interval $`m`$ to $`m+dm`$. Integrating both sides of the equation and taking logarithms we arrive at the expression:
$$\mathrm{log}_{10}N_i=\mathrm{log}_{10}\left(\frac{A}{\gamma +1}\left[\left(x_i+\frac{\mathrm{\Delta }m_i}{2}\right)^{\gamma +1}\left(x_i\frac{\mathrm{\Delta }m_i}{2}\right)^{\gamma +1}\right]\right),$$
(2)
where $`x_i`$ is the mass at the center of an interval of width $`\mathrm{\Delta }m_i`$ that contains $`N_i`$ stars. Equation 2 is the expression that can be used to derive the IMF slope, $`\gamma `$, by measuring the number of stars in each bin and fitting the data, which can be done by minimizing a $`\chi ^2`$ statistic and deriving the associated uncertainties. Since $`N_i`$ follows a binomial distribution (with $`N`$ being the total number of stars summed over all bins), the associated weight for bin $`i`$ for the $`\chi ^2`$ fit is given by:
$$w_i=\frac{N_iN}{(NN_i)(\mathrm{log}_{10}e)^2}.$$
(3)
Two warnings should be given here. The first one is that for two bins $`i`$ and $`j`$, $`N_i`$ and $`N_j`$ are correlated because they are both part of a joint multinomial distribution (see e.g. Lucy 2000). The second and most important one is that the value of $`N_i`$ that should be strictly used in Eq. 3 is the one calculated from the fit, not the one measured from the data (Wheaton et al. 1995). Otherwise, one runs into the possibility of introducing biases in the measurement of $`\gamma `$ unless some precautions are taken because of the differences between the real and the assumed weights assigned to bins with a small number of stars in them.
Using the fit $`N_i`$ instead of the data $`N_i`$ for the weights requires iterating and, therefore, complicates the IMF calculation. An alternative strategy was analyzed by Maíz Apellániz & Úbeda (2005): rather than trying to find the right weights for each bin, one can select the bins in such a way that they all have similar weights and, therefore, the obtained value of $`\gamma `$ is nearly independent of the weights themselves. In this manner, the weights derived from the data should give very similar results to the weights derived from the fit, hence eliminating the need for an iterative procedure. This strategy can be implemented by using bins with object equipartition, i.e. selecting the bins in such a way that they all have the same number of stars in them. A graphical example of the difference between constant bin-size (the standard approach to fitting binned data) and variable bin-size with object equipartition can be seen in Fig. 1.
In order to test whether variable bins provide a significant improvement over constant bins, Maíz Apellániz & Úbeda (2005) designed a series of Montecarlo numerical experiments in which they also analyzed the importance of the choice of the lower and upper mass limits. The reader is referred to that article for details; here we provide a summary in Tables 1 and 2 and in Figs. 2 and 3 of two of the experiments. In the first experiment, they generated 1000 realizations from a distribution with Salpeter ($`\gamma =2.35`$) slope using either $`N`$ = 30, 100, 300, or 1000 stars. Each realization was then binned into 3, 5, 10, 30, and 50 constant-size bins and a power law was fitted using $`\chi ^2`$ minimization. Finally, for the 1000 realizations of each $`N`$ \+ number of bins combinations the normalized bias $`b`$ was computed using the definition:
$$b=\frac{1}{1000}\underset{k=1}{\overset{1000}{}}\frac{\gamma _k+2.35}{\sigma _k},$$
(4)
where $`\sigma _k`$ is the uncertainty in $`\gamma _k`$ derived from $`\chi ^2`$ minimization and the $`k`$ index is used to denote the realization number. $`b`$ is a sensible choice to judge the existence of biases. If $`|b|1`$, then the fitting method will be unbiased because it will yield values that will be larger than the real one on $``$50% of the occasions and smaller on another $`50`$%. If, on the other hand, $`|b|1`$ or larger, a significant bias will exist.
As seen in Table 1, significant biases exist for most $`N`$ \+ number of bins combinations when using constant bins. In the right panel of Fig. 2 we see that $`b`$ is a strong function of $`\overline{N}_{i,\mathrm{min}}`$, the mean $`N_i`$ in the bin with the lowest number of counts (which for constant-size bins and a Salpeter power law will be the rightmost one). Note that at least 10 stars in the rightmost bin are required for this method to yield small biases, thus making its use impractical for most applications. Also note that the bias always points in the direction of making the measured IMF flatter than what it really is and that, for a fixed number of bins, the effect is larger when there are fewer stars.
The second experiment (third in the numbering of Maíz Apellániz & Úbeda 2005) was a repetition of the first one using bin equipartition instead of constant bins. As a comparison of Tables 1 and 2 shows, the biases for the second experiment are much smaller (even for the extreme case of dividing 30 stars into 30 one-star bins, $`b`$ is only 0.264). The effect is also perceptible in the right panel of Fig. 3, where the vertical scale is a factor of 10 smaller than in the equivalent panel of Fig. 2. Furthermore, the left panel of Fig 2 shows that the distribution of $`(\gamma _k+2.35)/\sigma _k`$ is well approximated by a Gaussian with a dispersion of 1.0, indicating that the random uncertainties derived from $`\chi ^2`$ minimization can also be trusted.
We conclude that the bin equipartition strategy proposed by Maíz Apellániz & Úbeda (2005) for the fitting of power laws with Salpeter slopes yields results that (a) are nearly bias-free and (b) produce correct uncertainty estimates. On the other hand, the standard uniform-size binning introduces biases that are dependent on the number of stars per bin. The power of the equipartition technique extends to small samples, since it is possible to obtain accurate values with reasonable precisions for the IMF slope even when as few as 30 stars are available for analysis.
We would also like to point out that, given the purely numerical nature of the analysis, these results could be extended to other similar problems. For example, the mass function for young stellar clusters can be rather well approximated by a power law with a slope of $`2.0`$ (see e.g. Fall & Zhang 2001), which is quite close to $`2.35`$, so the same type of biases should be present there as well. In general, we recommend that biases be evaluated for any function fitted to binned data through $`\chi ^2`$ minimization by means of specific numerical experiments analogous to the ones in this article.
## 3. Mass diffusion
The second type of bias we will discuss is caused by photometric uncertainties and detection limits. When one tries to use photometric data to derive statistical properties of a stellar population, one finds that corrections for the undetected stars must be included in the calculation because it is easier to detect bright stars than dim ones. Such an incompleteness correction is usually handled by crowded-field photometry packages such as DAOphot (Stetson 1987) or HSTphot (Dolphin 2000) by doing experiments in which the code is run with artificial stars added and the percentage of recovered objects as a function of magnitude and color is calculated. The correction is then applied to the observed stellar statistics.
Obviously, ignoring an incompleteness correction can result in a large bias in the measurement of the IMF and this well-known fact is taken into consideration in modern works on the subject. However, a related bias which is more subtle is not always taken into account. Suppose that we observe several times a star that has a real magnitude $`m`$. Due to Poisson, detector, and background noise, in some of our observations we will measure a magnitude $`m^{}>m`$ and in others we will measure $`m^{}<m`$. If our detector is well calibrated, the first circumstance will happen 50% of the time and the second one the remaining 50%. From our analysis of the detector we should be capable of estimating from a single measurement of the star an uncertainty $`\sigma _m`$ in such a way that $`m^{}\pm \sigma _m`$ behaves in an approximately Gaussian way, e.g. the single-measurement values $`m^{}`$ will be within $`m\sigma _m`$ and $`m+\sigma _m`$ for approximately 2/3 of the sample and outside ($`m2\sigma _m,m+2\sigma _m`$) for approximately 5% of the sample. Now, for most of the stellar mass range the IMF has a negative slope, meaning that there are more low-mass (dim) stars than high-mass (bright) ones. Therefore, if we measure a star to have magnitude $`m^{}`$, there should be a higher probability that its real magnitude $`m`$ is dimmer than $`m^{}`$ than that it is brighter (i.e. there are more dim stars disguised as bright ones at a given measured magnitude than bright stars disguised as dim ones) because the underlying real luminosity distribution provides more dim stars to start with. We will call this effect mass diffusion because it acts in a manner analogous to a diffusion process, smoothing a gradient by shifting objects from where they are more abundant to where they are more scarce.
Computing the correction required to eliminate mass-diffusion effects from an IMF is not straightforward because two intermediate steps are required. First, one has to translate uncertainties in the measured magnitudes and colors into uncertainties in temperature and luminosity (or bolometric magnitudes). Second, the uncertainties in the theoretical HR diagram have to be converted into uncertainties in mass. The first step involves applying extinction and bolometric corrections and taking into consideration the correlations between them and the measured magnitudes. An example of such effects is shown in Fig. 4, where we have plotted the temperatures and bolometric magnitudes of two stars in NGC 4214 derived from multiband HST/WFPC2 stellar photometry (Úbeda et al. 2005). A strong correlation is observed between temperatures and bolometric magnitudes: most of this correlation is caused by the strong temperature dependence of the bolometric correction. The data plotted in Fig. 4 was calculated using CHORIZOS (Maíz Apellániz 2004), a code specifically designed for the task of transforming from measured magnitudes to physical properties such as temperature, age, or extinction. Note that the lower uncertainty ellipse is larger than the upper one: most of this effect is caused by the higher uncertainties in the measured magnitudes of dim stars compared to bright ones.
The second step (calculating the uncertainties for the masses) can be achieved by producing a point-to-point coordinate conversion between temperature + luminosity and mass + age (with the caveats about one-to-one correspondence previously mentioned), generating a distribution of temperatures and luminosities for each star according to its uncertainty ellipse, transforming those values into masses and ages using the conversion above, and calculating the mean and standard deviation of the derived mass distribution for each star. Those values can then be used as the mass ($`M_i`$) and its uncertainty ($`\sigma _{M_i}`$) for each star.
Once the masses and corresponding uncertainties have been computed for each star, one can calculate the correction due to mass diffusion in the following way. First, a function $`\sigma _M(M)`$ is computed from the values for the individual stars ($`M_i`$, $`\sigma _{M_i}`$) by fitting a simple function such as a parabola. Then, one generates a series of realizations of the IMF with different values of the slope (which we will call $`\gamma _{\mathrm{real}}`$) which are then smoothed using a Gaussian kernel with $`\sigma _M(M)`$. The output is fitted using $`\chi ^2`$ minimization and the corresponding fitted value of the slope (which we will call $`\gamma _{\mathrm{fit}}`$) is obtained. Finally, as shown in Fig. 5, a polynomial is fit to $`\gamma _{\mathrm{fit}}\gamma _{\mathrm{real}}`$ as a function of $`\gamma _{\mathrm{fit}}`$ which is the correction that needs to be applied. Note that since $`\sigma _M(M)`$ is data-dependent, an individual correction has to be applied to each specific observation.
We show in Fig. 5 the correction for the specific case of the NGC 4214 data previously mentioned. The magnitude of the correction can be taken to be typical for HST photometry beyond the Magellanic Clouds obtained with a few orbits of exposure time. The bias produced by not applying the correction is in the same direction as the one caused by constant-size bins: the measured IMF appears to be flatter than the real one.
We recommend that the correction described here be applied to the calculation of IMFs in general. However, we should point out that, ideally, one would like the correction to be as small as possible. One (obvious) way to achieve this is to obtain photometry with better S/N ratio. Another one is to use as many filters as possible to adequately characterize the temperature (and possibly gravity and metallicity) of each star and to adequately correct for extinction and then to process the data using a code like CHORIZOS, instead on relying on conversions from single-color + magnitude diagrams to temperature + luminosity equivalents.
## 4. Multiplicity
The third type of bias that will be discussed here is that caused by multiplicity. It has been known for a long time that a large fraction of stars are located in multiple systems. For example, Kouwenhoven et al. (2005) measured that at least 61% of the stars in the Scorpius-Centaurus OB association are in a multiple system (the number could actually be higher due to incompleteness and selection effects). Interestingly, the single-star fraction decreases for early spectral types: those authors measured that 41% of the systems in which the primary is a B4-B9 are single but the fraction decreases to 12% if the primary is a B1-B3. The large multiplicity at the high-mass end of the stellar mass spectrum is not a new result: Mason et al. (1998) measured that at least 75% of the stellar systems with O stars in clusters or associations are multiple. Furthermore, those authors recognized that with the current instrumentation capabilities there is still enough discovery space between visual and spectroscopic binaries to allow for basically all O stars in clusters and associations to be part of multiple systems.
The existence of unresolved binaries artificially flattens the IMF due to a combination of two effects. First, an unresolved binary of any mass is shifted from a lower mass to a higher mass in a mass histogram. Second, the flattening should be further enhanced by the apparent increase of multiplicity with mass in the range between a few and several tens of solar masses, which is the range for which Salpeter slopes are reported by most authors (Chabrier 2003 and references therein).
The flattening due to unresolved binaries has been estimated by Kroupa (2001) to produce a change in $`\gamma `$ between 0.0 and 1.3 for low-mass stars and brown dwarfs. Given the large multiplicity fractions observed for OB stars, the effect must also be significant for intermediate- and high-mass stars. At the highest end of the stellar mass spectrum, the problem of the calculation of the IMF slope is coupled with another one: is there an upper limit for stellar masses or is the highest mass in a cluster simply determined by statistical sampling in a quasi-Salpeter power law that extends to infinite masses? A few years ago there was a large discrepancy between the highest mass measured from orbital motion (the most reliable method of measuring masses), which yielded values around 60 M, and the masses measured from photometry and spectral classification in R136 by Massey & Hunter (1998). Those authors measured stellar masses in the range 120-150 M and claimed that the data were compatible with a Salpeter IMF that extended beyond there. That gap has been recently narrowed in both directions. On the one hand, the 60 M barrier for spectroscopic-binary masses has been broken and the current heavyweight champion, WR 20a, lies around 80 M (Rauw et al. 2004; Bonanos et al. 2004). On the other hand, new statistical analyses indicate that there appears to be an upper mass limit somewhere in the 120-200 M range (Weidner & Kroupa 2004; Oey & Clarke 2005; Figer 2005).
In order to study the effect of unresolved binaries on the slope of the IMF at its upper-mass end and to obtain a better constraint on the stellar upper mass limit, we are currently engaged in an HST GO program (10602) which is a continuation of a previous one (10205). We are obtaining multi-filter imaging of all known Galactic O2/O3/O3.5 stars with the High Resolution Channel (HRC) on the Advanced Camera for Surveys (ACS). The HRC has a pixel size of 0$`.^{\prime \prime }`$027 and its PSF is well sampled and very stable across the detector (Anderson & King 2004). Furthermore, its geometric distortion is very well characterized, allowing for a relative astrometric precision of 0.005 pixels for very bright stars.
We present here our first results on the core of Trumpler 14 (Fig. 6), a young cluster in the Carina Nebula Association that contains at least three very-early O-type stars, including HD 93129 A, of spectral type O2 If\*, the closest known O2 star (Walborn et al. 2002; Maíz Apellániz et al. 2004), and HD 93129 B, which is an O3.5 V((f+)). Those two stars are separated by 2$`.^{\prime \prime }`$7 and appeared to be single in ground-based speckle interferometry (Mason et al. 1998). Recently, however, Nelan et al. (2004) were able to split HD 93129 A into two components using the Fine Guidance Sensor (FGS) on HST. They obtained a separation of 55 $`\pm `$ 3 mas at a position angle of 356 $`\pm `$ 4 degrees (measured from N towards E) and a magnitude difference of 0.90 $`\pm `$ 0.05 in the visible. In the same data HD 93129 B is unresolved.
We applied a PSF-fitting IDL photometry code especially written for this purpose to the HRC data for HD 93129 A and B. The code was applied to two dithered exposures in each of the F220W and F435W filters, thus yielding four independent measurements for each star. If a single component is used for the fit, the residuals for HD 93129 B are very small but those of HD 93129 A are very large, as expected for a binary system (Figs. 7and 8). On the other hand, a two-component fit yields very small residuals for HD 93129 A, hence confirming the binary character detected with FGS.
We present the results of the two-component PSF fitting to HD 93129 A in Table 3. The final row shows the proposed values for the separation, position angle, and $`\mathrm{\Delta }m`$ derived from the four independent HRC exposures. Our position angle is 18 $`\pm `$ 4 degrees to the E of that measured by Nelan et al. (2004), suggesting that we are detecting the relative motion of HD 93129 Ab with respect to Aa. From the 2.4 year difference between the epochs of the FGS and HRC observations we derive a very preliminary orbital period of $``$50 years for the system but, obviously, observations at other epochs will be needed to confirm and measure an orbit. We point out that this is the first O2/O3/O3.5 star ever measured to be an astrometric binary.
The magnitude difference between HD 93129Aa and Ab measured from the HRC data is similar to but slightly larger than the one measured with FGS (which uses light with longer wavelengths). The two components have very similar F220W-F435W colors, with Ab being redder only by 0.0407 $`\pm `$ 0.0072 magnitudes. Given their proximity, it appears unlikely that the relative color is caused by differences in the amount or type of extinction. On the other hand, such a difference in color is equivalent to that between 50 000 K and 44 000 K for TLUSTY models (Lanz & Hubeny 2003) with $`\mathrm{log}g=4.75`$ and solar metallicity, indicating that both components are likely to be early-O stars. We are currently working on a more detailed analysis of the photometry using CHORIZOS (Maíz Apellániz 2004).
It is important to note that HD 93129 A has not been identified as a spectroscopic binary. With a separation between its two components of about 150 AU, one would expect relative velocities of the order of 30 km s<sup>-1</sup> if the inclination is large. Therefore, its non-identification is not surprising, since one would require a separation an order of magnitude smaller to allow for a clear detection of radial velocity variations<sup>1</sup><sup>1</sup>1Of course, close binaries are easier to detect not only because the radial velocity variations are larger but because they occur on shorter time scales. (see e.g. Bonanos & Stanek 2005). This also means that we cannot even discard the possibility that either HD 93129 Aa or Ab are binaries themselves.
What does this mean for the biases in the top end of the IMF induced by multiplicity? Trumpler 14 is at an approximate distance of 2.7 kpc. If it were located at the same distance as the Galactic Center or NGC 3603, HD 93129 A will likely appear unresolved with HRC or FGS. At the distance of the Magellanic Clouds, Aa and Ab would be unresolved and A and B could be resolved but only with HST or adaptive optics. Moving to M31 or M33, HD 93128 (another O3 star in the cluster outside the field in Fig. 6) and HD 93129 would have an angular separation similar to that of Aa and Ab at its actual distance, with the rest of the stars (likely of late-O and B type) in Fig. 6 in between. This yields a total of (at least) four early-type stars blended together in a cluster that is quite massive, but far less than R136. The reader can easily deduce from these simple calculations how much he/she can trust IMF derivations of extragalactic young clusters that do not take into account multiplicity corrections.
## 5. Conclusions
We have discussed three sources of biases in the determination of the initial mass function (IMF): the use of constant-size bins, the uncertainty in the determinations of masses, and the existence of unresolved multiple systems. Those three effects tend to produce IMFs that are flatter than the real one, all of them large enough to potentially introduce significant systematic errors in the derived power-law slope. In the first case we present a technique that can get rid of the bias almost completely, even after the data have been obtained. In the second case we present a method that can also be used a posteriori to estimate the biases and we give some advice as to how to reduce them by using multifilter data. The third case, as demonstrated by the example of HD 93129 A, is harder to correct, given that the current capabilities do not allow us to detect all multiple systems in the stellar clusters and associations of interest, even those in our own Galaxy. Nevertheless, that should not preclude us from making an effort to close the current gap between visual and spectroscopic binaries or from trying to estimate the contribution of of unresolved multiple systems to the IMF slope.
### Acknowledgments.
Support for this work was provided by NASA through grants GO-09419.01, AR-09553.02, and GO-10205.01 from the Space Telescope Science Institute, Inc., under NASA contract NAS5-26555.
|
warning/0506/gr-qc0506106.html
|
ar5iv
|
text
|
# The universe seen at different scales
## 1 Different scale descriptions: coarse–graining the gravitational field
Any mathematical description of a physical system depends on an averaging scale characterizing the nature of the envisaged model. This averaging scale is usually hidden from view: it is taken to be understood. Thus, when a fluid is described as a continuum, this assumes one is using an averaging scale large enough that the size of individual molecules is negligible. If the averaging scale is close to molecular scale, small changes in the position or size of the averaging volume lead to large changes in the measured density and velocity of the matter, as individual molecules are included or excluded from the reference volume. Then the fluid approximation is no longer applicable; rather one is using a detailed description of the fluid where individual molecules are represented. Usual work referring to the fluid density and velocity assumes a medium–size averaging scale: not so small that molecular effects matter, but not so large that spatial gradients in the properties of the fluid are significant (, p.5). The actual averaging scale, or rather the acceptable range of averaging scales, is not explicitly stated but is in fact a key–feature underlying the description used, and hence the effective macroscopic dynamical laws investigated. Indeed, different types of physics (particle physics, atomic physics, molecular physics, macroscopic physics, astrophysics) correspond to different assumed averaging scales. Thus, instead of referring to a density function $`\varrho `$, one should really refer to a function $`\varrho _L`$: the density averaged over volumes characterized by scale length $`L`$. The key–point about the fluid approximation is that, provided this length scale is in the appropriate domain, then its actual value does not matter; i.e. when it is in this range, then changing $`L`$ by a factor of $`10`$, $`100`$, or even much more makes no difference: the measured density and average velocity will not change. But if you change $`L`$ by a very large amount until outside this range, this is no longer true. Hence, there is a range of validity $`L_1<L<L_2`$ where the fluid approximation holds and explicit mention of the associated averaging scale may be omitted.
In electromagnetic theory, polarization effects result from a large–scale field being applied to a medium with many microscopic charges. The macroscopic field $`E`$ differs from the point–to–point microscopic field, which acts on the individual charges because of a fluctuating internal field $`E_i`$, the total internal field at each point being $`D=E+E_i`$ (, p.116). Spatially averaging, one regains the average field because the internal field cancels out: $`E=D`$, indeed this is how the macroscopic field is defined (implying invariance of the background field under averaging: $`E=E)`$. On a microscopic scale, however, the detailed field $`D`$ is the effective physical quantity, and so is the field “measured” by electrons and protons at that scale. Thus, the way different test objects respond to the field crucially depends on their scale (a macroscopic device will measure the averaged field).
Now, exactly the same issue arises with regard to the gravitational field. Applications such as the solar system tests of general relativity theory, and in particular Einstein’s triumphant prediction of light bending by the Sun, are at solar system scales. We apply gravitational theory, however, at many other scales: to star clusters, galaxies, clusters of galaxies, and large–scale structures (walls and voids), as well as to black holes (occurring at solar system and star cluster scales, and possibly at much smaller scales).
Cosmology utilizes the largest scale averaging envisaged in astrophysics: a representative scale is assumed that is a significant fraction of the Hubble scale, and the cosmological velocity and density functions are defined by averaging on such scales (, p.111). Einstein first introduced the fluid approximation in his 1917 static universe model, as well as a highly idealized macroscopic model of the large–scale (smoothed) geometry of the universe. This geometrical idealization was then canonized via Milne’s cosmological principle (, p.408), or a somewhat more general Copernican principle (, pp.134, 350); the resulting locally isotropic, constant–curvature Robertson–Walker geometries (, Sect. 5.3) are nowadays taken to be a good description of the known region of the universe. The best justification of this assumption is the measured high degree of isotropy of the cosmic blackbody background radiation, taken together with a Copernican assumption (see , pp.351–3, for the argument in the case of exact isotropy, and or , Sect. 8.5 for the case of almost–isotropy).
However, a range of scales of description are relevant to cosmology. There are levels of approximation in modelling the universe, each with a hidden averaging scale. One can have a description in which every star is represented, or every galaxy (the stars averaged over), or only the largest scale cosmological structures (even galaxies averaged over, as in the fluid approximation). A typical cosmological simulation of “dark matter” gravitational clustering uses Newtonian theory and resolves fluid elements that still contain $`10^{60}`$ dark matter particles. This implicit coarse–graining can be made explicit within a Newtonian kinetic description: introducing filtering scales for a distribution of N self–gravitating particles in phase space reveals that the washed out small–scale degrees of freedom are represented by additional force terms that account for the dynamical coupling to these degrees of freedom ; they can also be modelled by phenomenological noise and/or stochastic forces , , and can lead to drastic qualitative changes of the system. However, this kind of calculation would be much more difficult in a General Relativity context.
The General Relativistic cosmological perturbation solutions used to study structure formation embody two interacting levels: the background (zero–order) model, almost always a Robertson–Walker metric, and the perturbed (first–order) model representing the growth of inhomogeneities, represented by a perturbed Robertson–Walker metric. The question then is how do models on two or more different scales relate to each other in Einstein’s gravitational theory . This is a difficult issue both because of the non–linearity of Einstein’s equations, and because of the lack of a fixed background spacetime – one of the core features of Einstein’s theory. This causes major problems in defining suitable averaging processes as needed in studying these processes. While there have been many analyses of this problem, there are still issues to be resolved in relation both to observations and dynamics, and in how this relates to gravitational entropy and the arrow of time.
## 2 Non–commutativity of averaging and observations
The usual analysis of cosmological observations is based on the Mattig equations relating apparent magnitude and redshift , , derived from analyzing the behaviour of null geodesics in Robertson–Walker spacetimes. In terms of the Sachs optical scalar equations, for hypersurface–orthogonal null geodesics in a general spacetime, the basic equations are:
$$\frac{d\theta }{dv}=R_{ab}K^aK^b2\sigma ^2\frac{1}{2}\theta ^2;\frac{d\sigma _{mn}}{dv}=E_{mn},$$
(1)
where $`\theta `$ is the rate of expansion of the null geodesics with tangent vector $`K^a=dx^a/dv`$ and affine parameter $`v`$, $`\sigma _{mn}`$ is their shear, $`R_{ab}`$ the Ricci tensor, and $`E_{mn}`$ a matrix of Weyl tensor components (, p.88; , pp.108–9). In the idealized Robertson–Walker case, the Weyl tensor $`C_{abcd}`$ vanishes, but the Ricci tensor is non–zero, being given via the Einstein field equations from the matter present. Thus, $`C_{abcd}=\mathrm{\hspace{0.17em}0}E_{mn}=0`$, and the relevant solutions are shear–free:
$$\sigma ^2=\mathrm{\hspace{0.17em}0}\frac{d\theta }{dv}=R_{ab}K^aK^b\frac{1}{2}\theta ^2.$$
(2)
Integration gives the Mattig relations applicable to Friedmann universe models (, pp.134–7), also elegantly obtainable from the geodesic deviation equations with vanishing Weyl Tensor .
However, in the real universe, observations take place via null geodesics lying in the empty spacetime between galaxies (you can’t see the further galaxy, if there is one in the foreground). Thus, the real situation in a universe with no intergalactic medium (all the matter is concentrated in galaxies) is the opposite of that above: in the region of spacetime traversed by the geodesics, the Ricci tensor vanishes, so
$$\frac{d\theta }{dv}=2\sigma ^2\frac{1}{2}\theta ^2,$$
(3)
but the non–zero Weyl tensor (the tidal field caused by nearby matter) generates shear that then causes focussing. Thus, the microscopic description of the focussing ($`\sigma 0,`$ $`R_{ab}=\mathrm{\hspace{0.17em}0},`$ $`E_{mn}0`$) is radically different from the macroscopic one ($`\sigma =\mathrm{\hspace{0.17em}0},`$ $`R_{ab}0,`$ $`E_{mn}=0`$), and the area distance–redshift relation may be expected to be different on microscopic scales (i.e. the small solid angle bundles of null geodesics actually used in observations of individual objects), as compared with macroscopic scales (averaging over large solid angles).
Various proposals have been made to deal with this. The most popular is the Dyer–Roeder distance , obtained by assuming only a fraction $`f`$ of the total mass density is encountered by the light–rays but ignoring the shear. Thus, in (2) one replaces $`R_{ab}K^aK^b`$ by $`fR_{ab}K^aK^b`$ and works out the corresponding area distance (, pp.138–143; ). This may be a good approximation if galaxies are embedded in a fairly uniform intergalactic medium of dark matter, but clearly does not take shear effects properly into account. How good it is will depend on the nature of clustering in the universe and how the averaged distribution impacts along the line of sight .
One can approach the topic in other ways: for example by using stochastic methods , or detailed examination of geodesics in Swiss–Cheese universe models . It has been suggested that energy conservation will imply that the effect averages out over the entire sky , but this calculation assumed that areas of a bundle of null geodesics were the same in the perturbed and background models, which will not be the case when one takes the effect of caustics into account . Indeed, areas increase slower than in a Robertson–Walker model in the empty spaces between matter, where the Ricci term is zero, and faster in the high–density regions where matter is concentrated, so one might think these effects cancel out. However, the strongly lensed rays soon go through a caustic and emerge highly divergent, so that areas are rapidly increasing again. It is plausible that on average the overall effect is always an increase in area, that is a lesser area distance than in the smooth background model.
The potential importance of this effect is in relation to the interpretation of the Supernova data , which is usually taken to imply the existence of a cosmological constant or quintessence causing acceleration of the universe at recent times . Kantowski has obtained analytic expressions for distance–redshift relations that have been corrected for the effects of inhomogeneities in the density. The values of the density parameter and cosmological constant inferred from a given set of observations depends on the fractional amount of matter in inhomogeneities and can significantly differ from those obtained by using the Mattig relations. As an example, a determination of $`\mathrm{\Omega }_0`$ made by applying the homogeneous distance–redshift relation to SN 1997ap at z = 0.83 could be as much as 50% lower than its true value. It could be that the apparent acceleration term detected is at least partly due to this optical effect: focussing of null geodesics is different in a lumpy universe than in a smooth one. Clearly, this effect needs careful investigation.
## 3 Non–commutativity of averaging and dynamics
The key–point in considering dynamical effects is that the two processes involved in relating the field equations at different scales do not commute . These processes are:
* calculating the Einstein tensor $`G_{1ab}:=R_{1ab}\frac{1}{2}R_1g_{1ab}`$ from a metric tensor $`g_{1ab}`$, and, hence, determining the quantity $`E_{1ab}:=G_{1ab}\kappa T_{1ab}`$ for $`g_{1ab},`$ where $`T_{1ab}`$ is the matter tensor appropriate to the scale represented by $`g_{1ab}`$;
* averaging the metric tensor $`g_{1ab}`$ to produce a smoothed metric tensor $`g_{2ab}:`$ $`g_{2ab}=g_{1ab}`$ and the matter tensor $`T_{1ab}`$ to produce a corresponding smoothed matter tensor $`T_{2ab}:`$ $`T_{2ab}=T_{1ab}`$.
Now in general the averaging process does not commute with taking derivatives: for a function $`g`$, usually $`_ig_ig`$ (see equations (8), (9) below for specific examples). Furthermore the inverse metric $`g_2^{ab}`$ (non–linearly dependent on the metric tensor components $`g_{1ab})`$ is not the smoothed version of $`g_1^{ab}.`$ The resulting Christoffel terms $`\mathrm{\Gamma }_{2bc}^a`$ are therefore not the smoothed version of $`\mathrm{\Gamma }_{1bc}^a,`$ hence the Ricci tensor components $`R_{2ab},`$ non–linearly dependent on $`\mathrm{\Gamma }_{2bc}^a`$, are not the smoothed versions of $`R_{1ab}`$. Extra non–linearities occur in calculating the Einstein tensor $`G_{2ab}=R_{2ab}\frac{1}{2}R_2g_{2ab}`$ from the Ricci tensor $`R_{2ab}`$. Thus, if you smooth first and then calculate the field equations, you get a different answer than if you calculate the field equations first and then smooth; symbolically $`𝐀(𝐄(g_{1ab}))𝐄(𝐀(g_{1ab}))`$ .
Suppose the field equations are true at the first scale: $`E_{1ab}=\mathrm{\hspace{0.33em}0}`$, then they will not be true at the second scale: $`E_{2ab}:=G_{2ab}\kappa T_{2ab}0`$. Thus, there will be an extra term in the equations at the smoother scale. We can either regard it as an extra term on the left–hand–side,
$$G_{2ab}E_{2ab}=\kappa T_{2ab},$$
(4)
representing a modified curvature term, or as an extra term on the right–hand–side,
$$G_{2ab}=\kappa T_{2ab}+E_{2ab},$$
(5)
where it is regarded as an extra contribution to the matter tensor. Which is the more appropriate interpretation depends on the context.
Szekeres developed a polarization formulation for a gravitational field acting in a medium, in analogy to electromagnetic polarization. He showed that the linearized Bianchi identities for an almost flat spacetime may be expressed in a form that is suggestive of Maxwell’s equations with magnetic monopoles. Assuming the medium to be molecular in structure, it is shown how, on performing an averaging process on the field quantities, the Bianchi identities must be modified by the inclusion of polarization terms resulting from the induction of quadrupole moments on the individual “molecules”. A model of a medium whose molecules are harmonic oscillators is discussed and constitutive equations are derived. This results in the form:
$$E^{2ab}=Q_{;cd}^{abcd},$$
(6)
that is $`E_{2ab}`$ is expressed as the double divergence of an effective quadrupole gravitational polarization tensor $`Q^{abcd}`$ with suitable symmetries:
$$Q^{abcd}=Q^{[ab][cd]}=Q^{cdab}.$$
(7)
Gravitational waves are demonstrated to slow down in such a medium.
The problem with such averaging procedures is that they are not covariant. They can be defined in terms of the background unperturbed space, usually either flat spacetime or a Robertson–Walker geometry, and so will be adequate for linearized calculations where the perturbed quantities can be averaged in the background spacetime (although even here the gauge problem arises, see below). But the procedure is inadequate for non–linear cases, where the integral needs to be done over a generic lumpy (non–linearly perturbed) spacetime that are not “perturbations” of a high–symmetry background. However, it is precisely in these cases that the most interesting effects will occur.
The only tensor integrals that are well–defined over a generic spacelike surface or spacetime region (and one interesting issue is which of these one should use) are for scalars , unless one uses the bitensors associated with Synge’s world function , based on parallel propagation along geodesics, to compare tensors at different points in a normal neigbourhood. The problem is that they cannot be used for averaging the metric tensor, for it is the metric tensor itself that defines the parallel propagation used in this process, and so is left invariant by it (since $`g_{ab;c}=0`$). So, one has to devise a procedure in which either the field equations are represented only in terms of scalars, possible for example if one takes components relative to a covariantly uniquely defined tetrad, or else bitensors are used to define averages of quantities other than the metric.
Zalaletdinov has taken this issue seriously, and provided the only sustained such attempt based on bitensors . He proposes a macroscopic description of gravitation based on a covariant spacetime averaging procedure. The geometry of the macroscopic spacetime follows from averaging Cartan’s structure equations, leading to a definition of correlation tensors. Macroscopic field equations (averaged Einstein equations) can be derived in this framework. It is claimed that use of Einstein’s equations with a hydrodynamic stress–energy tensor means neglecting all gravitational field correlations, and a system of macroscopic gravity equations is given when the correlations are taken into consideration. This approach has not won many adherents, but is nevertheless a systematic and coherent attempt to set up the problem generically.
## 4 Gravitational radiation
So far, there are two main applications of dynamical averaging. The first is to the issue of gravitational radiation. When electromagnetic radiation is present, one can characterize it as the high–frequency part of the electromagnetic field , and assign to it an energy density and momentum. This leads to an energy–momentum tensor that then serves as a source of curvature in the Einstein field equations. An obvious question is if one can do the same for gravitational radiation: can one identify it locally, and then assign to it an energy density and momentum? If so, there should be a form of the gravitational equations where this high–frequency part of the gravitational field acts as an effective source of spacetime curvature. But this is a version of the problem described above: it is just the definition of a contribution $`E_{2ab}`$ to the macroscopic gravitational field due to the fine–scale structure of the high–frequency radiation.
The problem is that gravitational radiation is only easily determined in linearly perturbed spacetimes; in more general spacetimes it is not easy to define the gravitational radiation part of the curvature, except near infinity in asymptotically flat spacetimes. Isaacson considered the case of linear perturbations about flat spacetime, determining the backreaction due to the gravitational radiation in this case. He obtained a close analogy with the electromagnetic situation: the ‘shortwave approximation’ shows how the stress–energy in the waves creates background curvature (, Sect. 35.13). A similar process can be applied to gravitational radiation in cosmological backgrounds, and backreaction by low–frequency gravitational radiation has been discussed by Dautcourt . To understand non–linear phenomena in gravitational radiation, the possibility of solitonic solutions and caustics should also be of concern, since these phenomena are presumably easier to detect.
## 5 Cosmology: Backreaction
The second application is to understand the nature of the backreaction of perturbations in cosmology . Unlike the gravitational radiation case, where one averages over tensor perturbations, here one first thinks of averaging over scalar quantitites like the density or the rate of expansion, in order to get control on cosmological parameters in an inhomogeneous universe model. As long as one works with exact equations for the evolution of those fields in a given foliation of spacetime, such an averaging procedure is covariant, e.g. for idealized cases like dust or a perfect fluid we can work in the ‘covariant fluid gauge’ . For these cases generalized forms of Friedmann’s equations can be employed to study backreaction . As soon as we invoke explicit model assumptions, e.g. perturbation theory, one runs directly into the gauge problem for cosmological perturbations, as well as the covariance question mentioned above.
The gauge problem is the following: when you perturb a smooth background cosmological metric $`\overline{g}_{1ab}`$ to obtain a perturbed metric $`g_{ab}=\overline{g}_{1ab}+h_{1ab}`$, the inverse relation is not unique: there is no agreed averaging or fitting process that will give back a unique background metric $`\overline{g}_{ab}`$ back from the “lumpy” metric $`g_{ab}`$ . Some other smooth metric $`\overline{g}_{2ab}`$ could have been chosen as the background metric instead, leading to a different definition of the perturbations: $`h_{2ab}:=g_{ab}\overline{g}_{2ab}`$, instead of $`h_{1ab}:=g_{ab}\overline{g}_{1ab}`$. The choice of background metric $`\overline{g}_{ab}`$ for a specific “lumpy” metric $`g_{ab}`$ is called a ‘gauge choice’. The backreaction problem will look very different if described in terms of different gauges.
The best way to look at this is to think of a gauge choice as a mapping of a smooth background metric $`\overline{g}_{1ab}`$ into the lumpy universe with metric $`g_{ab}`$ . At each point in the real spacetime the density perturbation $`\delta \varrho `$ is then defined by $`\delta \varrho :=\varrho \overline{\varrho }`$, where $`\varrho `$ is the actual density at that point, and $`\overline{\varrho }`$ the background density at the same point. The key–issue is the choice of surfaces of constant time in the perturbed spacetime, conventionally taken to represent the image of surfaces of constant density of the background spacetime. It then becomes clear that one can for example set the density perturbation to zero by choosing the mapping so that the surfaces of constant background density $`\overline{\varrho }`$ are the same as the surfaces of constant real density: for then at each point $`\varrho =\overline{\varrho }`$ $`\delta \varrho =0`$. However, with this choice, the fluid flow lines will not be orthogonal to the surfaces of constant density, so there will still be a non–zero density variation measured by comoving observers. Gauge issues arising in treating multi–component fluids raise extra issues because of the multiple possible choices of reference velocity field .
The remedy to this disconcerting behaviour is to choose gauge invariant variables, for example a set of covariantly defined variables that vanish in the background spacetime . While many studies have been carried out for quantifying backreaction effects in cosmology, where the smoothed–out effect of the small–scale perturbations causes extra terms in the Friedmann equations for the background metric, none have been done that both fully and clearly take the gauge issue into account and go beyond linear order. This is a key–issue waiting to be resolved. One certainly wants to go at least to second order in understanding the effects of non–linear perturbations, and while linear perturbations are well–understood, there are still many competing second order methods without a proper consensus on their implications emerging yet. Many of the crucial results at linear order no longer hold, for example scalar, vector and tensor perturbations are no longer independent of each other at second order , and then the backreaction in turn affects the perturbations themselves .
Isaacson’s method mentioned above has been used in the cosmological context , as has Zalaletdinov’s . In Zalaletdinov’s approach to the averaging problem in cosmology, the Einstein field equations on cosmological scales are modified by appropriate gravitational correlation terms. For a spatially homogeneous and isotropic macroscopic spacetime, the correlation tensor is of the form of a spatial curvature term. However, it is not clear how this approach relates to the gauge problem.
There is no doubt that interesting effects occur. How can we design a strategy that allows both making contact with the well–developed inventory of Friedmannian cosmology and quantifying backreaction effects? Cosmological parameters like the rate of expansion or the mass density are to be considered as volume–averaged quantities, and only these can be compared with cosmological observations. For this reason we expect that the relevant parameters are intrinsically scale–dependent unlike the situation in a Friedmannian cosmology. Averaging scalar characteristics on a Riemannian spatial domain delivers the effective dynamical sources that an observer would measure, but although he measures within the lumpy spacetime, he – due to a lack of better standards – is going to interpret his observations within a Friedmannian fitting model. This suggests a logical division of the averaging problem into 1) calculating averages in the real manifold, and 2) determining the mapping between averages in the real manifold and averages in the Friedmannian model. The first averaging is straightforward for scalars, as we mentioned above, and it encounters what we may call non–commutativity of averaging and time–evolution: this is a purely kinematical property that can be expressed, for a scalar field $`\psi `$, through the rule
$$_t\psi _t\psi =\theta \psi \theta \psi .$$
(8)
The fluctuation part on the right–hand–side of this rule produces the kinematical backreaction, which is now studied in the context of the dark energy problem (see below). The second “averaging” is more adequately thought of as a rescaling of the tensorial geometry. A (Lagrangian) smoothing as opposed to (Eulerian) rescaling of the metric on regional spatial domains has been proposed by Buchert and Carfora , using a global Ricci deformation flow for the metric. The smoothing of geometry implies a renormalization of averaged spatial variables, determining the effective cosmological parameters as they appear in the Friedmannian fitting model. Two effects that quantify the difference between background and real parameters were identified: curvature backreaction and volume effect . Both are the result of an inherent non–commutativity of averaging and spatial rescaling. In this way we look at the averaging problem in two directions in function space: time–evolution (as a deformation in direction of the extrinsic curvature of the space sections encoding the kinematical variables), and scale–“evolution” (as a deformation in direction of the intrinsic 3–Ricci curvature). With regard to a proper relation of those averages to observations, however, the possibility of averaging on the lightcone has to be seriously considered .
Employing such a logical split, it is reasonable to ask whether the universe described by a kinematically averaged model accelerates, independently of the question of whether we think that the universe accelerates because we may be using the wrong fitting model. With regard to the dark energy problem, the question of whether a cosmological constant is needed in the standard fitting model is then related to all of the effects mentioned above, while the question addressed to the realistic model only depends on the quantitative importance of the kinematical backreaction compared with the other averaged sources . It should be emphasized that in Newtonian cosmology global kinematical backreaction is absent, since (for Euclidean space sections) the fluctuating source term is a total divergence and thus vanishes for the periodic–boundary architecture of Newtonian models .
The key–issue is whether these backreaction effects are significant in cosmology. On the one hand, they might play a significant role in the inflationary era . On the other, it can possibly help explain the apparent dynamical existence either of dark energy and/or of dark matter as effective terms in the macroscopic dynamics at recent times. Various papers suggest the effect may indeed be significant, for example the observed acceleration of the universe could possibly be the result of the backreaction of cosmological perturbations rather than the effect of a negative–pressure dark energy fluid , . However, other studies obtain different results , , , , . Gauge effects are problematic , and many astrophysicists doubt the effect is significant. The issue still has to be resolved. In any case, a detailed investigation of backreaction effects helps to improve fitting models on regional scales for a better interpretation of observational data.
## 6 Entropy and coarse–graining
Related to all this is the puzzling question of gravitational entropy. The spontaneous structure growth in the expanding universe due to gravitational attraction appears to be contrary to all the statements about entropy in standard textbooks . This must somehow be related to the nature of the entropy of the gravitational field itself, not just the entropy of matter in a gravitational field.
Now the key–feature regarding the entropy of matter, as clearly explained by Penrose , is that it is associated with the loss of information that occurs with any coarse–grained description of matter. The most likely macroscopic states will be those that correspond to the largest numbers of microscopic states; that is to the largest volumes of phase space. This is made clear in Boltzmann’s definition of entropy: $`S=k\mathrm{ln}V_\mathrm{\Gamma }`$ where $`k`$ is Boltzmann’s constant and $`V_\mathrm{\Gamma }`$ the volume of phase space with points indistinguishable from each other by means of macroscopic observations of some macro (coarse–grained) variable to some accuracy $`\epsilon `$. The dynamics of the system is accompanied by an increase of this entropy as the representative point in phase space moves from less probable to more probable states.
One might therefore expect that a proper definition of gravitational entropy would similarly be related to some kind of coarse–graining of the gravitational field. However, most attempts at definitions of gravitational entropy in the cosmological context (e.g. , ) build on Penrose’s proposal that it be related to the magnitude of the Weyl tensor, with no introduction of coarse–graining. This is quite puzzling, given the persuasiveness of Penrose’ arguments that in the case of matter descriptions, entropy is always related to such coarse–graining. In our view this is one of the most fundamental missing aspects of gravitational theory: a satisfactory relation of gravitational entropy for a general gravitational field in terms of a coarse–grained description of that field, therefore relating to all the issues mentioned in the preceding sections.
A promising start has been made by Hosoya et al. : if we are only concerned with averaging the matter inhomogeneities on an inhomogeneous geometry, one can deduce an entropy measure for the distinguishability of the density distribution from its average value directly from the non–commutativity rule:
$$_t\varrho _t\varrho =\frac{1}{V}_tS\{\varrho ||\varrho \};S\{\varrho ||\varrho \}:=\varrho \mathrm{ln}\frac{\varrho }{\varrho },$$
(9)
where the functional $`S\{\varrho ||\varrho \}`$ is known in information theory as the Kullback–Leibler relative entropy, spatial averaging and integration is performed over a domain with volume $`V`$.
The conjecture has been made that this functional is, after a sufficient period of time, always globally increasing. This (in view of canonical considerations, e.g. in isolated Markovian systems) counter–intuitive statement is justified in a self–gravitating system because gravity is long–range, the averaging domain is not isolated, and gravity invokes a negative feedback: structural inhomogeneities are amplified due to gravitational instability. We may expect that the information content in the matter inhomogeneities is always increasing.
Given such a definition, the problem is to determine whether increasing total entropy (in the gravitational field and in the matter distribution) occurs always, or whether this is true only for special initial conditions. As discussed by Penrose , it seems plausible that the latter is the case, with the arrow of time in physics arising from boundary conditions at the start and end of the universe: specifically, the Weyl tensor taking a special form at the start of the expansion of the universe but a generic form at the end. The specific details of this proposal have never been clarified, and it is possible that the relation is not due to the Weyl tensor per se but rather due to a spatial integral of the divergence of the Electric part of the Weyl tensor (Ellis and Tavakol, unpublished). A further problem is then relating the arrow of time for structure growth in the universe to that for electromagnetic and gravitational radiation . Here again coarse–graining is crucial, for this relates to the kind of multi–scale description of the gravitational field envisaged by Isaacson, as discussed above.
Entropy and the associated arrow of time are fundamental to macroscopic physics. Their foundations in relation to microphysics remain mysterious in the case of general gravitational fields. The entropy of black holes is of course well understood, but this is an extreme case that does not by itself help us understand the relation of entropy to spontaneous structure formation in the expanding universe. Until this is solved, we cannot claim to properly understand the nature of entropy in the cosmological context .
GE acknowledges support by the NRF, and TB by the Sonderforschungsbereich SFB 375 ‘Astroparticle physics’ of the German Science Foundation DFG.
|
warning/0506/math0506292.html
|
ar5iv
|
text
|
# On the global stable manifold
## 1 Notations, Definitions and Basic Facts
### Linear Operators and splittings.
Let $`(E,||_E)`$ and $`(F,||_F)`$ be Banach spaces. We denote by $`(E,F)`$ the Banach space of all linear bounded operators from $`E`$ to $`F`$, endowed with the operator norm $`T:=sup_{|x|_E1}|Tx|_F`$. If $`E=F`$ we simply write $`(E)`$ for $`(E,E)`$. A linear subspace $`X`$ (necessarily closed) of $`E`$ splits if and only if there exists a subspace $`Y`$ such that $`E=XY`$. If $`L(E,F)`$ and $`R(F,E)`$ are such that $`LR=1_F`$, $`L`$ is called a left inverse of $`R`$ or a linear retraction and $`R`$ is called a right inverse of $`L`$ or a linear section. Then $`L`$ is surjective, $`R`$ is injective and $`E`$ decomposes as $`E=\mathrm{ker}L\mathrm{ran}R`$, with projections $`P_{\mathrm{ran}R}=RL`$ and $`P_{\mathrm{ker}L}=1_ERL`$. Conversely, if $`R(F,E)`$ is injective and $`E=X\mathrm{ran}R`$ for some subspace $`X`$ of $`E`$, then $`L:=R^1P_{\mathrm{ran}R}(E,F)`$ is a right inverse of $`R`$, with $`\mathrm{ker}L=X`$. Similarly, if $`L(E,F)`$ is surjective and $`E=\mathrm{ker}LY`$ for some subspace $`Y`$ of $`E`$, then $`R:=(L_{|Y})^1`$ is a right inverse of $`L`$ with $`\mathrm{ran}R=Y`$. The set of linear sections $`L(E,F)`$ and the set of linear retractions $`R(E,F)`$ are open in $`(E,F)`$.
### Immersions and submersions.
Let $`M`$, $`N`$ be differentiable manifolds of class $`C^k`$, $`1k\omega `$, modeled on the Banach space $`E`$, respectively $`F`$. A map $`f:MN`$ is a local immersion (resp. a local submersion) at $`p`$, if $`f`$ is a linear section (resp. a linear retraction) in local charts at $`p`$, meaning that there exist a local chart at $`p`$, $`\phi :U\phi (U)E`$, a local chart at $`q:=f(p)`$, $`\psi :V\psi (V)F`$, and a linear operator $`A(E,F)`$ which is a linear section (resp. a linear retraction) and such that $`\psi f\phi ^1=A_{|\phi (U)}`$. Then $`Y:=f(U)`$ is submanifold of $`N`$ and its tangent space at $`q`$ is $`T_qY=\mathrm{ran}Df(p)`$ (resp. $`X:=f^1(q)U`$ is a submanifold of $`M`$ and its tangent space at $`p`$ is $`T_pX=\mathrm{ker}Df(p)`$ ). The map $`f`$ is said to be simply an immersion (resp. a submersion), if it is a local immersion (resp. a local submersion) at any $`pM`$. In the first case, if $`f`$ is also injective, $`f(M)`$ is said to be an immersed submanifold of $`N`$. If $`f`$ is a local submersion at every $`pf^1(q)`$, then $`f^1(q)`$ is an embedded submanifold of $`M`$.
The implicit mapping theorem implies the usual criterion for local immersions and submersions, stating that $`f`$ is a local immersion (resp. a local submersion) at $`p`$ if and only if $`Df(p)(T_pM,T_qN)`$ is a linear section (resp. a linear retraction). A standard reference is \[Lan99\], section II, §2.
The criterion for local submersions has the following immediate consequence:
###### Proposition 1.1
Let $`f,g:MN`$ be $`C^k`$ maps between $`C^k`$ Banach manifolds, $`1k\omega `$, and set
$$W=\{pMf(p)=g(p)\}.$$
If for every $`pW`$, the operator $`Df(p)Dg(p)(T_pM,T_{f(p)}N)`$ is a linear retraction, then $`W`$ is a $`C^k`$ submanifold of $`M`$, with $`T_pW=\mathrm{ker}(Df(p)Dg(p))`$.
Indeed, the matter being local, we may assume that $`N`$ is an open subset of the Banach space $`F`$, so that $`W`$ is the zero set of the map $`fg`$, which is by hypothesis a local submersion at every $`pW`$.
### Discrete convolutions on $`\mathrm{}_𝐩`$ classes.
If $`(E,||)`$ is a Banach space, the $`\mathrm{}_p`$-norm of $`u:\mathrm{}E`$ is $`u_p:=\left(_n\mathrm{}|u(n)|^p\right)^{\frac{1}{p}}`$, for $`1p<\mathrm{}`$, or $`u_{\mathrm{}}:=sup_n\mathrm{}|u(n)|`$. Then $`\mathrm{}_p(\mathrm{},E)`$ denotes the Banach space of all $`u:\mathrm{}E`$ such that $`u_p<\mathrm{}`$. The set
$$c_0(\mathrm{},E):=\{u:\mathrm{}E\underset{|n|\mathrm{}}{lim}u(n)=0\}$$
is a closed subspace of $`\mathrm{}_{\mathrm{}}(\mathrm{},E)`$ and for all $`1pq<\mathrm{}`$, $`\mathrm{}_p(\mathrm{},E)\mathrm{}_q(\mathrm{},E)c_0(\mathrm{},E)\mathrm{}_{\mathrm{}}(\mathrm{},E)`$. The analogous class $`\{u:\mathrm{}Elim_n\mathrm{}u(n)=0\}`$ is denoted simply by $`c_0(E)`$; it can be viewed as a closed splitting subspace of $`c_0(\mathrm{},E)`$. Indeed, the identity mapping on $`c_0(E)`$ factors as $`c_0(E)\stackrel{j}{}c_0(\mathrm{},E)\stackrel{\rho }{}c_0(E)`$, the inclusion $`j`$ being given by zero-extension, the map $`\rho `$ being given by restriction to $`\mathrm{}\mathrm{}`$.
If $`g\mathrm{}_1(\mathrm{},(E))`$ and $`u\mathrm{}_{\mathrm{}}(\mathrm{},E)`$, their convolution product $`gu`$ is defined by
$$(gu)(n):=\underset{h\mathrm{}}{}g(nh)u(h).$$
Young’s inequality $`gu_pg_1u_p`$ implies that for any $`p[1,+\mathrm{}]`$ the convolution product is continuous as a bilinear map $`\mathrm{}_1(\mathrm{},(E))\times \mathrm{}_p(\mathrm{},E)\mathrm{}_p(\mathrm{},E)`$. Furthermore, $`guc_0(\mathrm{},E)`$ whenever $`g\mathrm{}_1(\mathrm{},(E))`$ and $`uc_0(\mathrm{},E)`$. <sup>3</sup><sup>3</sup>3This follows immediately by approximating $`g`$ with the sequence $`g_n:=\mathbb{𝟙}_{[n,+n]}g`$, for $`g_ng`$ in $`\mathrm{}_1`$, $`g_nuc_0(\mathrm{},E)`$ and by Young’s inequality $`gug_nu_{\mathrm{}}=(gg_n)u_{\mathrm{}}g_ng_1u_{\mathrm{}}0`$, so $`gu=lim_n\mathrm{}g_nuc_0(\mathrm{},E)`$.
Notice that the convolution with $`g\mathrm{}_1(\mathrm{},(E))`$ defines a bounded linear operator $`R_g`$ on $`c_0(E)`$ by $`ugu`$ (more precisely, $`R_gu=\rho (gj(u))`$).
### Manifolds of sequences.
Let $`M`$ be a $`C^k`$ manifold modeled on the Banach space $`E`$, let $`xM`$, and let $`c_x(M)`$ be the set of sequences $`u:\mathrm{}M`$ which converge to $`x`$. Equivalently, denoting by $`\overline{\mathrm{}}=\mathrm{}\{\mathrm{}\}`$ the one-point compactification of the set of natural numbers, $`c_x(M)`$ is the set
$$c_x(M)=\{uC^0(\overline{\mathrm{}},M)u(\mathrm{})=x\},$$
so it can be endowed with the restriction of the compact-open topology of $`C^0(\overline{\mathrm{}},M)`$. The space $`c_x(M)`$ has the structure of a $`C^k`$ manifold modeled on the Banach space $`c_0(E)`$. Indeed, given $`C^k`$ local charts $`\phi _n:U_n\phi _n(U_n)E`$, $`n=0,\mathrm{},m`$, where $`xU_m`$ and $`\phi _m(x)=0`$, consider the open subset of $`c_x(M)`$
$$𝒰=𝒰(U_0,\mathrm{},U_m)=\{uc_x(M)u(n)U_nn=0,\mathrm{},m1,u(n)U_mnm\},$$
and the homeomorphism $`\mathrm{\Phi }=\mathrm{\Phi }(\phi _0,\mathrm{},\phi _m):𝒰\mathrm{\Phi }(𝒰)c_0(E)`$ defined by
$$\mathrm{\Phi }(u)(n)=\phi _n(u(n))\text{ if }0nm1,\mathrm{\Phi }(u)(n)=\phi _m(u(n))\text{ if }nm.$$
It is easy to check that the collection of homeomorphisms $`\mathrm{\Phi }(\phi _0,\mathrm{},\phi _m)`$ constitute a $`C^k`$ atlas of $`c_x(M)`$. The tangent bundle of $`c_x(M)`$ is
$$Tc_x(M)=c_{0_x}(TM),$$
where $`0_x`$ is the zero element of $`T_xMTM`$, and its fibers are
$$T_uc_x(M)=\{v:\mathrm{}TMv(n)T_{u(n)}Mn\mathrm{},\underset{n\mathrm{}}{lim}v(n)=0_x\}.$$
In particular, the tangent space of $`c_x(M)`$ at the constant sequence $`x`$ is $`T_xc_x(M)=c_0(T_xM)`$.
The (left) shift operator $`𝒮:c_x(M)c_x(M)`$, $`𝒮(u)(n)=u(n+1)`$, is of class $`C^k`$, and its differential at $`x`$ is the (left) shift linear operator $`S`$ on $`c_0(T_xM)`$.
Also the evaluation at zero, $`\mathrm{ev}_0:c_x(M)M`$, $`uu(0)`$, is a map of class $`C^k`$, and its differential at $`u`$ is the linear evaluation at zero $`D\mathrm{ev}_0(u)[v]=v(0)`$.
Finally, every continuous map $`f:MN`$ with $`f(x)=y`$ induces by composition a continuous map $`f_{}:c_x(M)c_y(N)`$, $`f_{}(u):=fu`$. If $`f`$ is of class $`C^k`$, $`1k\omega `$, so is $`f_{}`$. Indeed, by local charts we are reduced to the case $`M=E`$, $`N=F`$, $`x=y=0`$, where the $`h`$-th differential of $`f_{}`$ at $`uc_0(E)`$ is given by the formula
$$\left(D^hf_{}(u)[v]^h\right)(n)=D^hf(u(n))[v(n)]^h.$$
In particular, the differential of $`f_{}`$ at the constant sequence $`x`$ is the multiplication operator by $`Df(x)`$,
$$Df_{}(x):T_xc_x(M)=c_0(T_xM)T_yc_y(N)=c_0(T_yN),Df_{}(x)[u](n)=Df(x)[u(n)].$$
### Hyperbolic fixed points.
An invertible operator $`T(E)`$ is said to be hyperbolic if its spectrum does not meet the unit circle: $`\sigma (T)\{|z|=1\}=\mathrm{}`$. Then $`\sigma (T)`$ consists of the two disjoint closed subsets $`\sigma (T)\{|z|<1\}`$ and $`\sigma (T)\{|z|>1\}`$, so $`E`$ has the $`T`$-invariant spectral decomposition $`E=E^sE^u`$, where $`\sigma (T|_{E^s})=\sigma (T)\{|z|<1\}`$ and $`\sigma (T|_{E^u})=\sigma (T)\{|z|>1\}`$.
A fixed point $`x`$ of a diffeomorphism $`f:Mf(M)M`$ is said to be hyperbolic if the differential of $`f`$ at $`x`$, $`Df(x)(T_xM)`$, is a hyperbolic operator. The corresponding spectral decomposition of the tangent space at $`x`$ is denoted by $`T_xM=E^sE^u`$.
## 2 Proof of the stable manifold theorem
Let us prove Theorem A. By definition, $`𝒲`$ is the set
$$𝒲=\{uc_x(M)𝒮(u)=f_{}(u)\}.$$
We start by studying the linear map $`D𝒮(x)Df_{}(x)(T_xc_x(M))`$. By the discussion of section 1, this is the linear operator
$$SDf(x)_{}:c_0(T_xM)c_0(T_xM).$$
Let us simplify the notation by setting $`E=T_xM`$, $`T=Df(x)(E)`$. Denote by $`P^s`$ and $`P^u`$ the spectral projections associated to the decomposition $`E=E^sE^u`$. The following lemma uses the fact that $`T`$ is a hyperbolic operator.
###### Lemma 2.1
Set, for $`n\mathrm{}`$,
$$g(n):=T^{n1}\left(\mathbb{𝟙}_\mathrm{}^+(n)I_EP^u\right),$$
where $`\mathrm{}^+=\{1,2,\mathrm{}\}`$. Then $`g\mathrm{}_1(\mathrm{},(E))`$ and the corresponding convolution operator $`R_g(c_0(E))`$ is a right inverse of $`ST_{}`$. Moreover,
$$\mathrm{ker}(ST_{})=\{uc_0(E)u(n)=T^nu(0)n\mathrm{},u(0)E^s\}.$$
Proof. Let $``$ be the operator norm induced by a Banach norm on $`E`$. By the spectral radius theorem
$`\underset{n\mathrm{}}{lim}g(n)^{1/n}=\underset{n\mathrm{}}{lim}T^{n1}P^s^{1/n}=\underset{n\mathrm{}}{lim}T|_{E^s}^{n1}^{1/n}=\mathrm{max}|\sigma (T|_{E^s})|<1,`$
$`\underset{n\mathrm{}}{lim}g(n)^{1/n}=\underset{n\mathrm{}}{lim}T^{(n+1)}P^u^{1/n}=\underset{n\mathrm{}}{lim}T|_{E^u}^{(n+1)}^{1/n}=\mathrm{max}|\sigma (T|_{E^u}^1)|<1.`$
Therefore, $`g(n)`$ tends to 0 exponentially fast for $`|n|\mathrm{}`$, in particular $`g`$ is in $`\mathrm{}_1(\mathrm{},(E))`$.
We have, for any $`uc_0(E)`$ and $`n\mathrm{}`$,
$`\left[(ST_{})R_g(u)\right](n)={\displaystyle \underset{h=0}{\overset{\mathrm{}}{}}}g(n+1h)u(h){\displaystyle \underset{h=0}{\overset{\mathrm{}}{}}}Tg(nh)u(h)`$
$`={\displaystyle \underset{h=0}{\overset{\mathrm{}}{}}}T^{nh}\left[\mathbb{𝟙}_\mathrm{}^+(n+1h)\mathbb{𝟙}_\mathrm{}^+(nh)\right]u(h)={\displaystyle \underset{h=0}{\overset{\mathrm{}}{}}}T^{nh}\mathbb{𝟙}_{\{0\}}(nh)u(h)=u(n),`$
that is, $`(ST_{})R_g=I_{c_0(E)}`$. <sup>4</sup><sup>4</sup>4Here is a more heuristic argument to find a right inverse to the linear operator $`ST_{}`$. First notice that the equation $`(ST_{})u=wc_0(E)`$ is equivalent to $`u(n+1)=Tu(n)+w(n)`$, $`n0`$, that iterated gives $`u(n)=T^nu(0)+_{h=0}^{n1}T^{n1h}w(h)`$. We can split this equation into $`u(n)=T^nP^su(0)+_{h=0}^{n1}T^{n1h}P^sw(h)+T^n\left[P^uu(0)+_{h=0}^{n1}T^{1h}P^uw(h)\right]`$. Now the first and the second term converge as $`n\mathrm{}`$, because the spectral radius theorem implies that $`T^nP^sc\lambda ^n`$, for some $`c1`$ and $`\lambda <1`$. The third term may not converge unless the sequence into square brackets converges to $`0`$, that is, $`P^uu(0)+_{h=0}^{n1}T^{1h}P^uw(h)=_{h=n}^{\mathrm{}}T^{1h}P^uw(h)`$, whence $`u(n)=T^nP^su(0)+(gw)(n)`$. Finally, it is clear that $`u\mathrm{ker}(ST_{})`$ if and only if $`u(n)=T^nu(0)`$ for any $`n\mathrm{}`$, which defines an element of $`c_0(E)`$ if and only if $`u(0)E^s`$. $`\mathrm{}`$
Let us prove that the closed subset $`𝒲`$ is a $`C^k`$ submanifold of $`c_x(M)`$. By Lemma 2.1, $`D𝒮(x)Df_{}(x)`$ is a linear retraction. Since the the space of linear retractions is open, $`D𝒮(u)Df_{}(u)`$ is a linear retraction for every $`u𝒲U`$, for a suitable neighborhood $`U`$ of $`x`$ in $`c_x(M)`$.
On the other hand, $`f_{}`$ and $`𝒮`$ commute. As a consequence if $`u𝒲`$ then $`u_n=f_{}^n(u)`$ is also in $`𝒲`$, and the linear operator $`D𝒮(u)Df_{}(u)`$ is related to $`D𝒮(u_n)Df_{}(u_n)`$ by left and right multiplication by invertible linear operators. Since $`u_n`$ eventually belongs to $`𝒲U`$, $`D𝒮(u_n)Df_{}(u_n)`$ is a linear retraction, and so is $`D𝒮(u)Df_{}(u)`$. Therefore, Proposition 1.1 implies that $`𝒲`$ is a $`C^k`$ submanifold of $`c_x(M)`$, with
$$T_x𝒲=\mathrm{ker}(D𝒮(x)Df_{}(x))=\{vc_0(T_xM)v(n)=Df(x)^nv(0)n\mathrm{},v(0)E^s\}.$$
(1)
The $`C^k`$ map $`\mathrm{ev}_0:c_x(M)M`$ is clearly injective on $`𝒲`$, and it remains to show that for any $`u𝒲`$, $`D\mathrm{ev}_0|_𝒲(u):T_u𝒲T_{u(0)}M`$ is a linear section. This is clearly true if $`u=x`$, for the expression (1) shows that $`D\mathrm{ev}_0(x)`$ is an isomorphism onto $`E^s`$, which splits in $`T_xM`$. Since the space of linear sections is open, the same is true for every $`u`$ in a neighborhood of $`x`$ in $`𝒲`$. The formula $`\mathrm{ev}_0f_{}=f\mathrm{ev}_0`$ implies that $`D\mathrm{ev}_0(u)|_{T_u𝒲}`$ is obtained from $`D\mathrm{ev}_0(u_n)|_{T_{u_n}𝒲}`$ by left and right multiplication by invertible operators. As before, since $`u_n`$ converges to $`x`$, we conclude that $`D\mathrm{ev}_0(u)|_{T_u𝒲}`$ is a linear section for every $`u𝒲`$. The proof of Theorem A is complete.
###### Remark 2.2
The fact that $`D𝒮(u)Df_{}(u)`$ is a linear retraction can also be proved directly by the following generalization of Lemma 2.1: if $`𝒯:c_0(E)c_0(E)`$ is the multiplication operator by a sequence $`(T_n)(E)`$ converging to a hyperbolic operator, then $`S𝒯(c_0(E))`$ is a linear retraction.
###### Remark 2.3
A similar argument allows to prove the stable manifold theorem for a hyperbolic equilibrium point $`x`$ of the local flow determined by a $`C^k`$ vector field $`X`$ on a $`C^{k+1}`$ Banach manifold $`M`$, where $`1k\omega `$. Denote by $`C_x^1([0,+\mathrm{}[,M)`$ the space of $`C^1`$ curves $`[0,+\mathrm{}[M`$ converging to $`x`$ for $`t+\mathrm{}`$, with the first derivative converging to 0. Then one can use the implicit function theorem to prove that the set
$$𝒲=\{uC_x^1([0,+\mathrm{}[,M)u^{}(t)X(u(t))=0\}$$
is a $`C^k`$ submanifold of $`C_x^1([0,+\mathrm{}[,M)`$, and that the evaluation at $`0`$ subordinates a $`C^k`$ immersion of $`𝒲`$ onto the stable manifold of $`0`$. The basic linear step consists in proving that if $`L(E)`$ is infinitesimally hyperbolic (i.e. its spectrum does not meet the imaginary axis), then the operator
$$\frac{d}{dt}L:C_0^1([0,+\mathrm{}[,E)C_0^0([0,+\mathrm{}[,E)$$
is a linear retraction. See \[AM04\] for more details.
## 3 The local stable manifold theorem
### Adapted norms.
An adapted norm for a hyperbolic operator $`T(E)`$ is an equivalent norm $`||`$ on $`E`$ such that
$$|\xi |=\mathrm{max}\{|P^s\xi |,|P^u\xi |\},|T\xi |\lambda |\xi |\xi E^s,|T^1\xi |\lambda |\xi |\xi E^u,$$
(2)
for some $`0<\lambda <1`$. One can always find a norm with these properties, provided that $`\lambda >\mathrm{max}|\sigma (T)\{|z|<1\}|`$ and $`\lambda >1/\mathrm{min}|\sigma (T)\{|z|>1\}|`$. Indeed, the following stronger statement holds: if the spectrum of $`T`$ is contained in the annulus $`\{\alpha <|z|<\beta \}`$, then $`E`$ has an equivalent norm $`||`$ such that, denoting by $``$ the corresponding operator norm, there holds
$$T\beta ,T^1\frac{1}{\alpha }.$$
(3)
If $`||_0`$ is any equivalent norm on $`E`$, a norm $`||`$ satisfying (3) can be defined as
$$|\xi |=\underset{n=1}{\overset{\mathrm{}}{}}\alpha ^n|T^n\xi |_0+\underset{n=0}{\overset{\mathrm{}}{}}\beta ^n|T^n\xi |_0,$$
as shown by the the spectral radius theorem (see for instance \[HPS77\], Proposition 2.8).
### The local stable manifold.
Let $`U`$ be an open neighborhood of 0 in the Banach space $`E`$, and let $`f:Uf(U)E`$ be a $`C^k`$ diffeomorphism, $`1k\mathrm{}`$ or $`k=\omega `$, having $`0`$ as a hyperbolic fixed point. Let $`T=Df(0)`$, let $`E=E^sE^u`$ be the corresponding splitting, and let $`||`$ be an adapted norm on $`E`$ for the hyperbolic operator $`T`$. If $`V`$ is a closed linear subspace of $`E`$, we denote by $`V(r)`$ the closed ball in $`V`$ of radius $`r`$. By the first property of adapted norms (2), $`E(r)=E^s(r)\times E^u(r)`$.
Given $`r>0`$ such that $`E(r)U`$, the local stable manifold of $`0`$ is the set
$$W_{\mathrm{loc},r}^s(0):=\{p\underset{n\mathrm{}}{}f^n(E(r))\underset{n\mathrm{}}{lim}f^n(p)=0\}.$$
This definition depends on $`r`$. However, if $`r_0`$ is small enough
$$W_{\mathrm{loc},r}^s(0)=W_{\mathrm{loc},r_0}^s(0)E(r)rr_0.$$
(4)
Indeed, the first set is contained in the second one by definition. Let us prove that the other inclusion holds when $`r_0`$ is small. The point $`0𝒲`$ is a fixed point of the $`C^k`$ map $`f_{}|_𝒲:𝒲𝒲`$. By (1) and by the second property of adapted norms (2),
$$Df_{}(0)|_{T_0𝒲}=T_{}|_{T_0𝒲}\lambda <1,$$
so a first order approximation shows that $`f_{}|_𝒲`$ is locally a contraction at $`0`$. In particular, there exists $`r_0>0`$ such that $`f_{}(u)_{\mathrm{}}u_{\mathrm{}}`$ for every $`u𝒲`$ with $`u_{\mathrm{}}r_0`$. Since $`f_{}`$ coincides with the shift $`S`$ on $`𝒲`$, this is equivalent to saying that $`|u(n)|=|f^n(u(0))|`$ is a decreasing sequence, if $`u𝒲`$ and $`u_{\mathrm{}}r_0`$. The conclusion follows.
###### Theorem 3.1
(Local stable manifold theorem) Let $`U`$ be an open neighborhood of 0 in the Banach space $`E`$, $`f:Uf(U)E`$ be a $`C^k`$ diffeomorphism, $`1k\omega `$, having $`0`$ as a hyperbolic fixed point, inducing the splitting $`E=E^sE^u`$. If $`r>0`$ is small enough, then $`W_{\mathrm{loc},r}^s(0)`$ is the graph of a $`C^k`$ map $`w:E^s(r)E^u(r)`$ such that $`w(0)=0`$ and $`Dw(0)=0`$.
Proof. Since $`\mathrm{ev}_0|_𝒲`$ is a $`C^k`$ immersion, it is an embedding of an open neighborhood of $`0`$ in $`𝒲`$. Therefore, if $`r_0`$ is small enough the set
$$\{\mathrm{ev}_0(u)u𝒲,u_{\mathrm{}}<r_0\}$$
is a $`C^k`$ submanifold of $`E`$, with tangent space at $`0`$ $`D\mathrm{ev}_0(0)T_0𝒲=E^s`$. Then, if $`r<r_0`$ is small enough, the set
$$W_{\mathrm{loc},r_0}^s(0)E(r)=\{\mathrm{ev}_0(u)u𝒲,u_{\mathrm{}}<r_0\}E(r)$$
is the graph of a $`C^k`$ map $`w:E^s(r)E^u(r)`$ such that $`w(0)=0`$ and $`Dw(0)=0`$, and the conclusion follows from (4). $`\mathrm{}`$
## 4 The diffeomorphism class of the global stable manifold
### Characterization of diffeomorphisms such that $`𝐖^𝐬(𝐱)`$ is an immersed copy of $`𝐄^𝐬`$.
A hyperbolic fixed point $`x`$ of a diffeomorphism $`f:MM^{}f(M)M^{}`$ is said to be a local attractor if $`W^s(x)`$ is a neighborhood of $`x`$, or equivalently<sup>5</sup><sup>5</sup>5Indeed, if $`E^sT_xM`$ the local stable manifold has empty interior, so by Baire theorem the global stable manifold cannot fill an open set. if $`E^s=T_xM`$. It is said to be a global attractor if $`W^s(x)=M`$ (in particular, $`f(M)M`$), in which case $`f`$ is said to be a topological contraction of $`M`$.
###### Definition 4.1
Let $`U`$ be an open neighborhood of $`0`$ in the Banach space $`E`$, and let $`f:Uf(U)E`$ be a diffeomorphism of class $`C^k`$, $`1k\omega `$, such that the hyperbolic fixed point $`0`$ is a local attractor. We say that the germ of $`f`$ at $`0`$ extends to a $`C^k`$ topological contraction of $`E`$ if it can be represented by a global $`C^k`$ diffeomorphism of $`E`$ having $`0`$ as a global attractor.
In other words, we are asking the existence of a global diffeomorphism $`\stackrel{~}{f}:EE`$ of class $`C^k`$ which coincides with $`f`$ in a neighborhood $`VU`$ of $`0`$, and which is a topological contraction. If we ask $`\stackrel{~}{f}`$ to be only Lipschitz, its existence is always guaranteed:
###### Lemma 4.2
Under the assumptions of Definition 4.1, there always exists a homeomorphism $`\stackrel{~}{f}:EE`$ such that $`\stackrel{~}{f}`$ and $`\stackrel{~}{f}^1`$ are globally Lipschitz, $`\stackrel{~}{f}`$ coincides with $`f`$ in a neighborhood $`VU`$ of $`0`$, and such that $`0`$ is a global attractor.
Proof. Let $`T=Df(0)`$ and let $`||`$ be an adapted norm for $`T`$: if $``$ denotes the corresponding operator norm, we have $`T<1`$. Let $`\phi :EE`$ be a $`k`$-Lipschitz bounded map which coincides with the identity in a neighborhood of $`0`$ (for instance, $`\phi (\xi )=\chi (\xi )\xi `$, with $`\chi (s)=1`$ for $`s1`$, $`\chi (s)=2s`$ for $`1s2`$, $`\chi (s)=0`$ for $`s2`$, has the required properties, with $`\mathrm{lip}\phi 3`$). Write $`f`$ as $`f(\xi )=T(\xi +f_0(\xi ))`$. Since $`f_0`$ is at least $`C^1`$ and $`Df_0(0)=0`$, we can find a neighborhood $`U_0`$ of $`0`$ such that
$$\mathrm{lip}f_0|_{U_0}<1/k,\theta :=(1+k\mathrm{lip}f_0|_{U_0})T<1.$$
Up to replacing $`\phi (\xi )`$ by $`\lambda \phi (\xi /\lambda )`$ \- which is still $`k`$-Lipschitz - we may assume that $`\phi (E)U_0`$, and we set
$$\stackrel{~}{f}:EE,\stackrel{~}{f}(\xi )=T(\xi +f_0(\phi (\xi ))).$$
The map $`\stackrel{~}{f}`$ is a Lipschitz diffeomorphism of $`E`$ onto $`E`$ together with its inverse because $`\mathrm{lip}(f_0\phi )k\mathrm{lip}f_0|_{U_0}<1`$. Moreover, $`\stackrel{~}{f}=f`$ in a neighborhood of $`0`$, and
$$|\stackrel{~}{f}(\xi )|T(|\xi |+|f_0(\phi (\xi ))|)T(1+k\mathrm{lip}f_0|_{U_0})|\xi |=\theta |\xi |,$$
so the fact that $`\theta <1`$ implies that $`0`$ is a global attractor. $`\mathrm{}`$
In the analytic category the extension of a germ is unique (and not always possible, even in finite dimensional spaces), so the request of Definition 4.1 is quite strong. A weaker property is given by the following:
###### Definition 4.3
Under the same assumptions of Definition 4.1, we say that the germ of $`f`$ at $`0`$ extends up to conjugacy to a $`C^k`$ topological contraction of $`E`$ if there exists a $`C^k`$ diffeomorphism $`h:Vh(V)`$ of an open neighborhood of $`0`$ with $`h(0)=0`$ such that the germ of $`hfh^1`$ at $`0`$ extends to a topological contraction of $`E`$.
For instance, if an analytic diffeomorphism is analytically linearizable near a hyperbolic local attractor, then it extends up to conjugacy to an analytic topological contraction of $`E`$. This latter definition is relevant also in the smooth and in the finite differentiability category because it extends to diffeomorphisms defined on manifolds:
###### Definition 4.4
If $`U`$ is an open neighborhood of $`x`$ in the Banach manifold $`M`$ \- modeled on the Banach space $`E`$ \- and $`f:Uf(U)M`$ is a $`C^k`$ diffeomorphism, $`1k\omega `$, having $`x`$ as a local attractor, we say that the germ of $`f`$ at $`x`$ extends up to conjugacy to a $`C^k`$ topological contraction of $`E`$ if the germ of diffeomorphism on an open neighborhood of $`0`$ in $`E`$ defined by conjugacy by a local chart (mapping $`x`$ into $`0`$) extends up to conjugacy to a topological contraction of $`E`$.
Clearly, this definition does not depend on the choice of the local chart.
###### Theorem 4.5
Let $`f`$ be a $`C^k`$ diffeomorphism of the $`C^k`$ Banach manifold $`M`$ modeled on the Banach space $`E`$, $`1k\omega `$, and assume that the hyperbolic fixed point $`xM`$ is a global attractor. Then $`M`$ is homeomorphic to $`E`$ by a bi-locally Lipschitz homeomorphism. Furthermore, $`M`$ is $`C^k`$ diffeomorphic to $`E`$ if and only if the germ of $`f`$ at $`x`$ extends up to conjugacy to a $`C^k`$ topological contraction of $`E`$.
Proof. Let $`UM`$ be an open neighborhood of $`x`$ in $`M`$, and let $`\psi :U\psi (U)E`$ be a $`C^k`$ local chart with $`\psi (x)=0`$. Consider the $`C^k`$ diffeomorphism
$$g=\psi f\psi ^1:\psi (Uf^1(U))\psi (Uf(U))E.$$
By Lemma 4.2, there exists an invertible topological contraction $`h:EE`$ which is locally Lipschitz together with its inverse and which coincides with $`g`$ in an open neighborhood $`V\psi (Uf^1(U))`$ of $`0`$. If we are also assuming that the germ of $`f`$ at $`x`$ extends up to conjugacy to a $`C^k`$ topological contraction of $`E`$, up to changing the local chart $`\psi `$ we may assume that $`h`$ is a $`C^k`$ diffeomorphism.
Let us define the global homeomorphism $`\varphi :ME`$. Let $`pM`$. Since $`W^s(x)=M`$, there exists $`n\mathrm{}`$ such that $`f^n(p)\psi ^1(V)`$, and we set
$$\varphi (p)=h^n(\psi (f^n(p))).$$
Since $`\psi `$ conjugates the diffeomorphisms $`f|_{\psi ^1(V)}`$ and $`h|_V=g|_V`$, this definition does not depend on the choice of $`n`$. The map $`\varphi `$ is invertible, its inverse being the map
$$\varphi ^1(\xi )=f^n(\psi ^1(h^n(\xi ))),$$
for $`n=n(p)`$ so large that $`h^n(\xi )V`$. So $`\varphi `$ is the required locally Lipschitz homeomorphism, and when $`h`$ is a $`C^k`$ diffeomorphism, $`\varphi `$ is also a $`C^k`$ diffeomorphism.
Conversely, assume that there is a $`C^k`$ diffeomorphism $`\varphi :ME`$. Up to composition with a translation, we may assume that $`\varphi (x)=0`$. In particular, $`\varphi `$ is a local chart mapping $`x`$ into $`0`$, and $`\varphi f\varphi ^1`$ is a global diffeomorphism of $`E`$ onto $`E`$ which is a topological contraction of $`E`$. This shows that the germ of $`f`$ at $`x`$ extends up to conjugacy to a $`C^k`$ topological contraction of $`E`$. $`\mathrm{}`$
Let us consider the general case $`W^s(x)M`$. Denote by $`W_{\mathrm{loc}}^s(x)`$ the image by $`\mathrm{ev}_0`$ of a neighborhood of $`x`$ in $`𝒲`$, so small that $`\mathrm{ev}_0`$ is an embedding on it. Then $`W_{\mathrm{loc}}^s(x)W^s(x)`$ is a $`C^k`$-submanifold modeled on the Banach space $`E^s`$.
###### Corollary 4.6
Let $`f:MM`$ be a $`C^k`$ diffeomorphism of a Banach manifold, $`1k\omega `$. Let $`x`$ be a hyperbolic fixed point of $`f`$, with associated splitting $`T_xM=E^sE^u`$. Then the $`C^k`$ manifold $`𝒲`$ is homeomorphic to $`E^s`$ by a bi-locally Lipschitz homeomorphism. Furthermore, $`𝒲`$ is $`C^k`$ diffeomorphic to $`E^s`$ if and only if the germ of $`f|_{W_{\mathrm{loc}}^s(x)}`$ at $`x`$ extends up to conjugacy to a $`C^k`$ topological contraction of $`E^s`$.
Indeed, it is enough to apply Theorem 4.5 to the diffeomorphism $`f_{}`$ of the manifold $`𝒲`$.
Since the evaluation at zero defines a local conjugacy between the restriction of $`f_{}`$ to a small neighborhood of $`x`$ in $`𝒲`$ and the restriction of $`f`$ to $`W_{\mathrm{loc}}^s(x)`$, the above corollary can be restated in the following way: $`W^s(x)`$ is always the image of a locally Lipschitz and locally closed injective map $`E^sM`$, which can be chosen to be also a $`C^k`$ immersion if and only if the germ of $`f|_{W_{\mathrm{loc}}^s(x)}`$ at $`x`$ extends up to conjugacy to a $`C^k`$ topological contraction of $`E^s`$.
The statement about the locally Lipschitz homeomorphism in the above corollary is the first part of Theorem B.
### Regularity conditions on the Banach space $`𝐄`$.
If $`1k\mathrm{}`$ and the Banach space is somehow regular, one may expect that every germ of $`C^k`$ diffeomorphism having $`0`$ as a hyperbolic local attractor extends to a $`C^k`$ topological contraction of $`E`$. This is actually true for a large class of Banach spaces. Indeed let us consider the following conditions on $`E`$:
1. $`E`$ is finite dimensional;
2. $`E`$ has a Hilbert structure;
3. there exists a $`C^k`$ norm on $`E`$ (i.e. an equivalent norm $`||`$ such that the function $`\xi |\xi |`$ is $`C^k`$ on $`E\{0\}`$);
4. there exists a bounded $`C^k`$ map $`\phi :EE`$ which coincides with the identity on a neighborhood of $`0`$;
5. there exists a bounded $`C^k`$ map $`\phi :EE`$ such that $`\phi (0)=0`$ and $`D\phi (0)=I`$.
It is readily seen that (E1) $``$ (E2) $``$ (E3) $``$ (E4) $``$ (E5). For instance, if $`E`$ admits a $`C^k`$ norm $`||`$, a $`C^k`$ map $`\phi `$ satisfying (E4) can be defined as
$$\phi (\xi )=\chi (|\xi |)\xi ,$$
where $`\chi :[0,+\mathrm{})\mathrm{}`$ is a smooth function with compact support and such that $`\chi (s)=s`$ for $`s1`$. Condition (E4) was considered by Atkin \[Atk01\], who observed that such a condition may hold also for Banach spaces which do not have regular norms. For instance, the Banach space $`\mathrm{}_{\mathrm{}}=\mathrm{}_{\mathrm{}}(\mathrm{},\mathrm{})`$ does not admit a smooth norm, but it supports a smooth map $`\phi `$ satisfying (E4), namely
$$\phi (u)(n)=\chi (u(n)),n\mathrm{},$$
where $`\chi :\mathrm{}\mathrm{}`$ is a smooth bounded function coinciding with the identity in a neighborhood of $`0`$. A similar construction works for the Banach space $`L^{\mathrm{}}(X,,\mu )`$, $`(X,,\mu )`$ a measure space, and for the Banach space $`C^0(K,\mathrm{})`$, $`K`$ a compact topological space.
The condition (E5) holds for a large class of Banach spaces, even in the case $`k=\omega `$, in which (E4) is obviously never fulfilled. For instance, the spaces $`L^{\mathrm{}}(X,,\mu )`$ and $`C^0(K,\mathrm{})`$ admit the analytic map
$$u\mathrm{sin}u,$$
which satisfies (E5). It is not clear whether (E4) and (E5) are equivalent for $`k<\omega `$.
Even if we do not know any counterexample, it is not likely that every Banach space admits a smooth map satisfying (E4) or (E5). In any case, it would be interesting to characterize the class of Banach spaces which admit such maps. At the moment the situation is unclear even for simple spaces such as $`\mathrm{}_1`$.
### The case of finite differentiability.
The following result, together with Corollary 4.6, implies the (i) part of Theorem B.
###### Proposition 4.7
Let $`1k\mathrm{}`$, and assume that the Banach space $`E`$ satisfies condition (E4) above. Let $`UE`$ be an open neighborhood of $`0`$, and let $`f:Uf(U)E`$ be a $`C^k`$ diffeomorphism with hyperbolic fixed point $`0`$ which is a local attractor. Then the germ of $`f`$ at $`0`$ extends to a $`C^k`$ topological contraction of $`E`$.
Notice that if the map $`\phi `$ appearing in (E4) is also globally Lipschitz, we could argue as in Lemma 4.2, writing $`f=T(\mathrm{id}+f_0)`$ and then using $`\phi `$ to extend $`f_0`$ to a $`C^k`$ map on $`E`$ with small Lipschitz norm. Without this assumption, a natural idea is to see $`\mathrm{id}+f_0`$ as the time-one map obtained by integrating a time dependent small vector field $`X`$, and then use $`\phi `$ to extend $`X`$. We need therefore the following easy:
###### Lemma 4.8
Let $`1k\omega `$, let $`UE`$ be an open neighborhood of $`0`$, and let $`g:UE`$ be a $`C^k`$ map such that $`g(0)=0`$ and $`Dg(0)=I`$. Then there exists a neighborhood $`VU`$ of $`0`$ and a $`C^k`$ map $`X:[0,1]\times VE`$ such that $`X(t,0)=0`$, $`D_2X(t,0)=0`$ for every $`t[0,1]`$, and such that the solution of the Cauchy problem
$$_tG(t,\xi )=X(t,G(t,\xi )),G(0,\xi )=\xi ,$$
satisfies $`G(1,\xi )=g(\xi )`$ for every $`\xi `$ in some neighborhood of $`0`$.
Proof. The differential with respect to the second variable of the $`C^k`$ map
$$G:[0,1]\times UE,G(t,\xi )=tg(\xi )+(1t)\xi ,$$
namely $`tDg(\xi )+(1t)I`$, is uniformly invertible for every $`(t,\xi )`$ in a neighborhood of $`[0,1]\times \{0\}`$. By the parametric inverse mapping theorem, there exist a neighborhood $`V`$ of $`0`$ and a $`C^k`$ map $`H:[0,1]\times VE`$ such that
$$H(t,G(t,\xi ))=\xi (t,\xi )[0,1]\times V.$$
Setting
$$X(t,\eta )=g(H(t,\eta ))H(t,\eta ),$$
we get that $`X`$ has the desired properties. $`\mathrm{}`$
We can now prove Proposition 4.7.
Proof. \[of Proposition 4.7\] Let $`T=Df(0)`$ and let $`||`$ be an adapted norm for $`T`$, so that $`T`$ becomes a contraction. Consider the diffeomorphism $`g=T^1f`$, whose differential at $`0`$ is the identity operator. Let $`XC^k([0,1]\times V,E)`$ and $`G`$ be as in Lemma 4.8. By assumption, there is a $`C^k`$ map $`\phi :EE`$ whose image is contained in $`B_{r_0}(0)`$, which coincides with the identity map on $`B_{s_0}(0)`$, for some $`0<s_0<r_0<+\mathrm{}`$. Since $`X(t,0)=0`$ and $`D_2X(t,0)=0`$, for every $`t[0,1]`$, we can find $`r>0`$ such that
$$|X(t,\xi )|ϵ|\xi |,(t,\xi )[0,1]\times B_r(0),$$
(5)
where $`ϵ>0`$ is so small that $`e^{ϵr_0/s_0}T<1`$. The $`C^k`$ map
$$\psi (\xi )=\frac{r}{r_0}\phi \left(\frac{r_0}{r}\xi \right),$$
takes values in $`B_r(0)`$, and coincides with the identity mapping on $`B_s(0)`$, with $`s=rs_0/r_0`$. The time-dependent $`C^k`$ vector field
$$\stackrel{~}{X}:[0,1]\times EE,\stackrel{~}{X}(t,\xi )=X(t,\psi (\xi )),$$
coincides with $`X`$ on $`[0,1]\times B_s(0)`$, so by (5) $`|\stackrel{~}{X}(t,\xi )|ϵ|\xi |`$ there. On the other hand, if $`|\xi |s`$,
$$|\stackrel{~}{X}(t,\xi )|=|X(t,\psi (\xi ))|ϵ|\psi (\xi )|ϵrϵ\frac{r}{s}|\xi |=ϵ\frac{r_0}{s_0}|\xi |.$$
(6)
We conclude that the above estimate holds for every $`(t,\xi )[0,1]\times E`$. Therefore, the solution $`\stackrel{~}{G}`$ of the Cauchy problem
$$_t\stackrel{~}{G}(t,\xi )=\stackrel{~}{X}(t,\stackrel{~}{G}(t,\xi )),\stackrel{~}{G}(0,\xi )=\xi ,$$
is defined for every $`(t,\xi )[0,1]\times E`$, coincides with $`G`$ in a neighborhood of $`[0,1]\times \{0\}`$, and by (6) it satisfies
$$|\stackrel{~}{G}(t,\xi )|e^{ϵ\frac{r_0}{s_0}t}|\xi |,(t,\xi )[0,1]\times E.$$
(7)
Since $`G(1,\xi )=g(\xi )`$, the global $`C^k`$ diffeomorphism
$$\stackrel{~}{g}:EE,\stackrel{~}{g}(\xi )=\stackrel{~}{G}(1,\xi ),$$
coincides with $`g`$ in a neighborhood of $`0`$. Then the $`C^k`$ diffeomorphism $`\stackrel{~}{f}=T\stackrel{~}{g}`$ coincides with $`f`$ in a neighborhood of $`0`$. Finally, by (7),
$$|\stackrel{~}{f}(\xi )|T|\stackrel{~}{g}(\xi )|Te^{ϵr_0/s_0}|\xi |,$$
and the fact that $`e^{ϵr_0/s_0}T<1`$ implies that the hyperbolic fixed point $`0`$ is a global attractor. $`\mathrm{}`$
### The smooth and the analytic cases.
It remains to prove the (ii) part of Theorem B. Let us start by examining some consequences of assumption (E5) (here by Taylor polynomial we mean Taylor polynomial at $`0`$):
###### Lemma 4.9
Let $`k=\mathrm{}`$ or $`k=\omega `$. The following facts are equivalent:
1. there exists a bounded map $`\phi C^k(E,E)`$ such that $`\phi (0)=0`$ and $`D\phi (0)=I`$;
2. every polynomial map $`p:EE`$, $`\mathrm{deg}pn`$, is the Taylor polynomial of order $`n`$ of some bounded map $`\psi C^k(E,E)`$;
3. if $`ϵ>0`$, $`K`$ is a compact manifold of class $`C^h`$, $`0h\omega `$, $`U`$ is an open neighborhood of $`0`$ in $`E`$, $`F`$ is a Banach space, and $`\varphi :K\times UF`$ is a $`C^h`$ map such that for every $`xK`$ the map $`\varphi (x,):EF`$ is of class $`C^k`$ and $`\varphi (x,0)=0`$, then there exists a map $`\psi :K\times EF`$ with the same regularity, such that $`sup|\psi |<ϵ`$, and such that for every $`xK`$ the Taylor polynomials of order $`n`$ of $`\varphi (x,)`$ and of $`\psi (x,)`$ coincide.
Proof. Statement (i) is a particular case of (ii): take $`p(x)=x`$ and $`n=1`$. Statement (ii) is a particular case of (iii): take $`K`$ a singleton, $`\varphi =pp(0)`$. So it is enough to prove that (i) implies (iii). Given $`\phi `$ satisfying (i), it is easy to construct a bounded map $`\phi _nC^k(E,E)`$ such that $`\phi _n(\xi )=\xi +o(|\xi |^n)`$ for $`\xi 0`$. Indeed, one may define $`\phi _n`$ inductively as
$$\{\begin{array}{c}\phi _1=\phi ,\hfill \\ \phi _{n+1}(\xi )=\phi _n\left(\xi \frac{1}{(n+1)!}D^{n+1}\phi _n(0)\xi ^{n+1}\right).\hfill \end{array}$$
Let $`\varphi :K\times UF`$ be the map appearing in (iii). By replacing $`\phi (\xi )`$ by $`\lambda \phi (\xi /\lambda )`$ in the above construction, we may assume that the image of $`\phi `$ \- hence also the image of $`\phi _n`$ \- is contained in a neighborhood $`VU`$ of $`0`$ which is so small that $`sup_{K\times V}|\varphi |<ϵ`$ (such a neighborhood exists because $`\varphi `$ is continuous, and it vanishes on the compact set $`K\times \{0\}`$). Then the map
$$\psi (x,\xi )=\varphi (x,\phi _n(\xi ))$$
has the required properties. $`\mathrm{}`$
The proof of Proposition 4.11 below relies on the following classical conjugacy result (see for instance \[CFdlL03a\]):
###### Theorem 4.10
Let $`k=\mathrm{}`$ or $`k=\omega `$. Let $`UE`$ be an open neighborhood of $`0`$, and let $`f:Uf(U)E`$ be a $`C^k`$ map with $`f(0)=0`$, $`Df(0)=T`$ an isomorphism with spectral radius $`\rho (T)<1`$. Then $`f`$ is $`C^k`$ locally conjugated to its Taylor polynomial of order $`n`$, provided that $`n`$ is so large that $`\rho (T^1)\rho (T)^{n+1}<1`$.
The (ii) part of Theorem B is a consequence of Corollary 4.6 and of the following:
###### Proposition 4.11
Let $`k=\mathrm{}`$ or $`k=\omega `$, and assume that the Banach space $`E`$ satisfies condition (E5). Let $`UE`$ be an open neighborhood of $`0`$, and let $`f:Uf(U)E`$ be a $`C^k`$ diffeomorphism with hyperbolic fixed point $`0`$, which is a local attractor. Then the germ of $`f`$ at $`0`$ extends up to conjugacy to a $`C^k`$ topological contraction of $`E`$.
Proof. Let $`T=Df(0)`$ and let $`||`$ be an adapted norm for $`T`$, so that $`T`$ becomes a contraction. Consider the diffeomorphism $`g:=T^1f`$, whose differential at $`0`$ is the identity operator. Let $`XC^k([0,1]\times V,E)`$ and $`G`$ be as in Lemma 4.8: $`G`$ is the map obtained by integrating the time dependent vector field $`X`$, and $`G(1,)=g`$ in a neighborhood of $`0`$.
Consider the $`C^k`$ map
$$\varphi :[0,1]\times V(E),\varphi (t,\xi )=D_2X(t,\xi ).$$
Fix $`n\mathrm{}`$ so large that
$$\rho (T^1)\rho (T)^{n+1}<1,$$
(8)
and $`ϵ>0`$ so small that $`e^ϵT<1`$. Since $`\varphi (t,0)=D_2X(t,0)=0`$, by Lemma 4.9 there exists a $`C^k`$ map $`\psi :[0,1]\times EL(E)`$ such that
$$\underset{(t,\xi )[0,1]\times E}{sup}\psi (t,\xi )<ϵ,$$
(9)
and such that for every $`t[0,1]`$ the Taylor polynomials of order $`n1`$ of $`\varphi (t,)`$ and $`\psi (t,)`$ coincide. Equivalently:
$$\psi (t,\xi )\varphi (t,\xi )=o(|\xi |^{n1})\text{for }\xi 0,$$
(10)
uniformly in $`t[0,1]`$. Consider the globally defined time dependent vector field of class $`C^k`$,
$$\stackrel{~}{X}:[0,1]\times EE,\stackrel{~}{X}(t,\xi )=_0^1\psi (t,s\xi )\xi 𝑑s.$$
By (9),
$$|\stackrel{~}{X}(t,\xi )|_0^1\psi (t,s\xi )|\xi |𝑑sϵ|\xi |,(t,\xi )[0,1]\times E.$$
(11)
Since
$$X(t,\xi )=_0^1\frac{d}{ds}X(t,s\xi )𝑑s=_0^1D_2X(t,s\xi )\xi 𝑑s=_0^1\varphi (t,s\xi )\xi 𝑑s,$$
by (10) there holds
$$\stackrel{~}{X}(t,\xi )X(t,\xi )=_0^1(\psi (t,s\xi )\varphi (t,s\xi ))\xi 𝑑s=o(|\xi |^n)\text{for }\xi 0,$$
uniformly in $`t[0,1]`$. Therefore, $`X(t,)`$ and $`\stackrel{~}{X}(t,)`$ have the same Taylor polynomial of order $`n`$. An easy induction argument then shows that $`G(t,)`$ and the solution $`\stackrel{~}{G}(t,)`$ of the Cauchy problem
$$_t\stackrel{~}{G}(t,\xi )=\stackrel{~}{X}(t,\stackrel{~}{G}(t,\xi )),\stackrel{~}{G}(0,\xi )=0,$$
have the same Taylor polynomial of order $`n`$, for every $`t[0,1]`$. In particular, $`g(\xi )=G(1,\xi )`$ and $`\stackrel{~}{g}(\xi ):=\stackrel{~}{G}(1,\xi )`$ have the same Taylor polynomial of order $`n`$. The same happens for $`f=Tg`$ and $`\stackrel{~}{f}:=T\stackrel{~}{g}`$, so by (8) Theorem 4.10 implies that $`f`$ and $`\stackrel{~}{f}`$ are $`C^k`$ locally conjugated at $`0`$.
Since $`\stackrel{~}{X}`$ has linear growth by (11), the map $`G`$ is well defined on $`[0,1]\times E`$, and $`G(t,)`$ is a $`C^k`$ diffeomorphism of $`E`$ onto $`E`$ for every $`t[0,1]`$. By (11),
$$|\stackrel{~}{f}(\xi )|T|\stackrel{~}{g}(\xi )|Te^ϵ|\xi |,$$
so the fact that $`e^ϵT<1`$ implies that $`\stackrel{~}{f}`$ is a topological contraction of $`E`$, concluding the proof. $`\mathrm{}`$
|
warning/0506/math0506293.html
|
ar5iv
|
text
|
# Untitled Document
Note on the rational points of a pfaff curve
Jonathan Pila
Abstract
Let $`X\text{}^2`$ be the graph of a pfaffian function $`f`$ in the sense of Khovanskii. Suppose that $`X`$ is nonalgebraic. This note gives an estimate for the number of rational points on $`X`$ of height $`H`$; the estimate is uniform in the order and degree of $`f`$.
2000 Mathematics Subject Classification: 11D99 (11J99)
1. Introduction
In and I have studied the distribution of rational points on the graph $`X`$ of a transcendental real analytic function $`f`$ on a compact interval. I showed that the number of rational points of $`X`$ of height (see 1.2 below) $`H`$ was $`O_{f,ϵ}(H^ϵ)`$ for all positive $`ϵ`$.
Suppose that $`X`$ is the graph of a function analytic on a noncompact domain, such as or $`\text{}^+`$. To bound the number of rational points of height $`H`$ on $`X`$ requires controlling the implied constant in the above estimate over the enlarging intervals $`[H,H]`$.
In the estimate in , the implied constant depends on a bound for the number of solutions of an algebraic equation in $`P(x,f)`$, where $`P\text{}[x,y]`$ is a polynomial (of degree depending on $`ϵ`$), as well as a bound for the number of zeros of derivatives of $`f`$ (of order depending on $`ϵ`$). In general these quantities may not behave at all well over different intervals.
However these numbers are globally bounded for the socalled pfaffian functions (; see 1.1 below), indeed bounded uniformly in terms of the order and degree of the function (see 1.1). For this class of functions, a uniform estimate on the number of rational points of bounded height may be obtained by adapting the methods of .
Definition 1.1. (\[3, 2.1\]) Let $`U\text{}^n`$ be an open domain. A pfaffian chain of order $`r0`$ and degree $`\alpha 1`$ in $`U`$ is a sequence of real analytic functions $`f_1,\mathrm{},f_r`$ in $`U`$ satisfying differential equations
$$df_j=\underset{i=1}{\overset{n}{}}g_{ij}(𝐱,f_1(𝐱),f_2(𝐱),\mathrm{},f_j(𝐱))dx_i$$
for $`j=1,\mathrm{},r`$, where $`𝐱=(x_1,\mathrm{},x_n)`$ and $`g_{ij}\text{}[x_1,\mathrm{},x_n,y_1,\mathrm{},y_r]`$ of degree $`\alpha `$. A function $`f`$ on $`U`$ is called a pfaffian function of order $`r`$ and degree $`(\alpha ,\beta )`$ if $`f(𝐱)=P(𝐱,f_1(𝐱),\mathrm{},f_r(𝐱))`$, where $`P`$ is a polynomial of degree at most $`\beta 1`$.
The usual elementary functions $`e^x,\mathrm{log}x`$ (but not $`\mathrm{sin}x`$ on all ), algebraic functions, combinations and compositions of these are pfaffian functions: see . In this paper always $`n=1`$, so $`𝐱=x`$. A pfaff curve $`X`$ is the graph of a pfaffian function $`f`$ on some connected subset of its domain. The order and degree of $`X`$ will be taken to be the order and degree of $`f`$.
Definition 1.2. For a point $`P=(a_1/b_1,a_2/b_2,\mathrm{},a_n/b_n)\text{}^n`$, where $`a_j,b_j\text{},b_j1`$ and $`(a_j,b_j)=1`$ for all $`j=1,2,\mathrm{},n`$, define the height $`H(P)=\mathrm{max}\{|a_j|,b_j\}`$. Note this is not the projective height. If $`X\text{}^n`$ let $`X(\text{})=X\text{}^n`$ and $`X(\text{},H)`$ the subset of points $`P`$ with $`H(P)H`$. Finally put
$$N(X,H)=\mathrm{\#}X(\text{},H)=\mathrm{\#}\{PX(\text{}),H(P)H\}.$$
Theorem 1.3. There is an explicit function $`c(r,\alpha ,\beta )`$ with the following property. Suppose $`X`$ is a nonalgebraic pfaff curve of order $`r`$ and degree $`(\alpha ,\beta )`$. Let $`Hc(r,\alpha ,\beta )`$. Then
$$N(X,H)\mathrm{exp}\left(5\sqrt{\mathrm{log}H}\right).$$
Now in certain cases where the polynomials defining the chain have rational (or algebraic) coefficients, results in transcendence theory show that the number of algebraic points of $`X`$ is finite, indeed explicitly bounded (see e.g. ). On the other hand, the example $`X=\{(x,y):y=2^x,x\text{}\}`$ shows that the set $`X(\text{})`$ is not finite in general. For many $`X`$, e.g. the graph of $`y=e^{e^x}`$, finiteness is unknown.
For the example $`X=\{(x,y):y=2^x,x\text{}\}`$ of course $`N(X,H)=O(\mathrm{log}H)`$. I do not know examples where the growth of $`N(X,H)`$ is faster than this, so the above bound might be very far from the truth. Note however that elementary considerations do not suffice to establish better bounds on $`N(X,H)`$ for, e.g., $`X=\{(x,y),y=\mathrm{log}\mathrm{log}(e^{e^x}+e^x)\}`$, for which finiteness of $`X(\text{})`$ is presumably expected.
The methods herein are also applicable to algebraic curves: indeed the fact that pfaffian functions have finiteness properties analogous to algebraic functions was the impetus for applying those methods to them. Since algebraic functions are pfaffian (), it is appropriate to record here the following improvement to the result obtained in .
Theorem 1.4. Let $`b,c2`$ be integers and $`H3`$. Let $`F(x,y)\text{}[x,y]`$ be irreducible of bidegree $`(b,c)`$. Let $`d=\mathrm{max}(b,c)`$ and $`X=\{(x,y)\text{}^2,F(x,y)=0\}`$. Then
$$N(X,H)(6d)^{10}4^dH^{2/d}\left(\mathrm{log}H\right)^5.$$
The improvement over the result of is that the exponent of $`\mathrm{log}H`$ is here independent of $`d`$; the exponent $`2/d`$ of $`H`$ is best possible. I refer to for discussion of related results ().
2. The Main Lemma
Let $`M=\{x^hy^k:(h,k)J\}`$ be a finite set of monomials in the indeterminates $`x,y`$. Put
$$D=\mathrm{\#}M,R=\underset{(h,k)J}{}(h+k),s=\underset{(h,k)J}{\mathrm{max}}(h),t=\underset{(h,k)J}{\mathrm{max}}(k),S=D\left(s+t\right),$$
$$\rho =\frac{2R}{D(D1)},\sigma =\frac{2S}{D(D1)},C=\left(D!D^R\right)^{2/D(D1)}+1.$$
Note that $`SR`$ for any $`M`$. If $`Y`$ is a plane algebraic curve defined by $`G(x,y)=0`$, say $`Y`$ is defined in $`M`$ if all the monomials appearing in $`G`$ belong to $`M`$.
Lemma 2.1. Let $`M`$ be a set of monomials with $`D2`$ and $`S2R`$. Let $`H1,L1/H^2`$ and $`I`$ a closed interval of length $`L`$. Let $`fC^D(I)`$ with $`|f^{}|1`$ and $`f^{(j)}`$ either nonvanishing in the interior of $`I`$ or identically zero for $`j=1,2,\mathrm{},D`$. Let $`X`$ be the graph of $`y=f(x)`$ on $`I`$. Then $`X(\text{},H)`$ is contained in the union of at most
$$\left(4CD\mathrm{\hspace{0.17em}4}^{1/\rho }+2\right)L^\rho H^\sigma $$
real algebraic curves defined in $`M`$. Proof. Fix $`M,H`$. If $`f`$ is a function satisfying the hypotheses on some interval $`I`$, and $`X`$ is the graph of $`f`$ on $`I`$, then the set $`X(\text{},H)`$ is contained in some minimal number $`G(f,I)`$ of algeraic curves of degree $`d`$; Let the $`G(L)`$ be the maximum of $`G(f,I)`$ over all intervals and functions satisfying the hypotheses.
Now suppose $`f`$ is such a function on an interval $`I=[a,b]`$, and $`A1`$. An equation $`f^{(2)}(x)=\pm 2AL^1`$ has at most one solution in the interior $`I`$, unless it is satisfied identically. Suppose $`c`$ is a solution. Since $`f^{(2)},f^{(3)}`$ are onesigned throughout $`I`$, it follows that $`|f^{(2)}(x)|2A^{2/(D1)}L^1`$ in either $`[a,c]`$ or $`[c,b]`$, and $`|f^{(2)}(x)|2A^{2/(D1)}L^1`$ in (respectively) either $`[b,c]`$ or $`[a,c]`$. Now an interval with the latter condition has length $`2A^{1/(D1)}`$ by \[10, 2.6\] (or \[2, Lemma 7\]) applied with $`A=A^{2/(D1)}`$.
Continuing to split the interval at points where $`f^{(\kappa )}=\kappa !A^{\kappa /(D1)}L^{1\kappa },\kappa =2,\mathrm{},D`$ yields a (possibly empty) subinterval $`[s,t]`$ in which $`|f^{(\kappa )}|\kappa !A^{\kappa /(D1)}L^{1\kappa }`$ for all $`\kappa =1,2,3,\mathrm{},D`$, while the intervals $`[a,s],[t,b]`$ comprise $`D`$ subintervals of length $`2A^{1/(D1)}L`$ (by \[10, 2.6\] (or \[2, Lemma 7\]) applied with $`A=A^{\kappa /(D1)}`$), and so have length at most $`2DA^{1/(D1)}L`$. (If $`[s,t]`$ is empty take $`s=t=b`$.)
On $`[s,t]`$, the points of height $`H`$ lie on at most $`CA^{1/(D1)}H^\sigma L^\rho `$ curves in $`M`$ by \[10, 2.4\]. Therefore the function $`G(L)`$ satisfies the recurrence
$$G(L)CA^{1/(D1)}H^\sigma L^\rho +2G(\lambda L)$$
when $`L1/H^2`$, where $`\lambda =2DA^{1/(D1)}`$. Thus, provided $`\lambda ^{n1}L1/H^2`$,
$$G(L)CA^{1/(D1)}H^\sigma L^\rho \left(1+2\lambda ^\rho +\mathrm{}+(2\lambda ^\rho )^{n1}\right)+\mathrm{\hspace{0.17em}2}^nG(\lambda ^nL).$$
Choose $`A`$ such that $`2\lambda ^\rho =1/2`$, that is $`A^{1/(D1)}=2D\mathrm{\hspace{0.17em}4}^{1/\rho }`$ (so $`A1`$) and choose $`n`$ such that
$$\frac{\lambda }{LH^2}\lambda ^n<\frac{1}{LH^2}.$$
Then $`G(\lambda ^nL)1`$, while
$$2^n=\lambda ^{n\rho /2}\left(\frac{LH^2}{\lambda }\right)^{\rho /2}=2\left(LH^2\right)^{\rho /2}2L^\rho H^\sigma .$$
Therefore $`G(L)\left(4CD\mathrm{\hspace{0.17em}4}^{1/\alpha }+2\right)H^\rho L^\sigma `$ as required.
3. Pfaff curves
Since a pfaffian function of order $`r=0`$ is a polynomial, to which Theorem 1.3 is inapplicable, it is convenient now to assume $`r1`$.
Proposition 3.1. Let $`f_1,\mathrm{},f_r`$ be a pfaffian chain of order $`r1`$ and degree $`\alpha `$ on an open domain $`U\text{}`$, and $`f`$ a pfaffian function on $`U`$ having this chain and degree $`(\alpha ,\beta )`$.
(a). Let $`k\text{}`$. Then $`f^{(k)}`$ is a pfaffian function with the same chain as $`f`$ (so of order $`r`$) and degree $`(\alpha ,\beta +k(\alpha 1))`$.
(b). Let $`P(x,y)`$ be a polynomial of degree $`d`$. Suppose $`f`$ is not algebraic. Then the equation $`P(x,f(x))=0`$ has at most
$$2^{r(r1)/2}d\beta \left(r\alpha +d\beta \right)^r$$
solutions.
(c). Let $`VU`$ be an open set on which $`f^{}0`$ and $`k1`$. Then on $`V`$ there is an inverse function $`g`$ of $`f`$, and the number of zeros of $`g^{(k)}`$ on $`V`$ is at most
$$2^{r(r1)/2}\left((k1)(\beta +k(\alpha 1))\right)\left(r\alpha +\left((k1)(\beta +k(\alpha 1))\right)\right)^r.$$
Proof. Part (a) is by \[3, 2.5\]. For part (b), observe that $`P(x,f(x))`$ is a pfaffian function of order $`r1`$ and degree $`(\alpha ,d\beta )`$. Since $`f`$ is not algebraic, all the solutions are nondegenerate and the result is in \[3, 3.1\].
Part (c). By differentiating the relation $`g(f(x))=x`$ and simple induction, for $`k1`$,
$$g^{(k)}(y)=\frac{Q_k(f^{(1)},f^{(2)},\mathrm{},f^{(k)})}{(f^{}(x))^{2k1}}$$
where $`Q_k(z_1,z_2,\mathrm{},z_k)`$ is a polynomial of degree $`\gamma _k=k1`$. Since $`f^{(j)}`$ are pfaffian functions with the same chain, the function $`Q_k(f^{(1)},f^{(2)},\mathrm{},f^{(k)})`$ is a pfaffian function of order $`r`$ and degree $`(\alpha ,\gamma _k(\beta +k(\alpha 1)))`$. The statement now follows from (b).
Proof of 1.3. Let $`d2`$ and let $`M=M(d)`$ be the set of monomials of degree $`d`$ in $`x,y`$. Then, elementarily,
$$D=\frac{d(d1)}{2},\rho =\frac{8}{3(d+3)},\sigma =3\rho ,C6.$$
Subdivide the connected domain $`U`$ into at most
$$2.2^{r(r1)/2}(\beta +\alpha 1)(r\alpha +\beta +\alpha 1)+12^{1+r(r1)/2}\left((r+1)(\alpha +\beta )\right)^{r+1}$$
intervals on which $`f^{}1`$, $`1f^{}1`$ or $`f^{}1`$, and then divide further into subintervals on which the inverse $`g`$ has nonvanishing derivatives up to order $`D`$ in the first and third case, or $`f`$ has nonvanishing derivatives up to order $`D`$ in the second case. The total number of intervals is at most
$$2^{1+r(r1)/2}\left((r+1)(\alpha +\beta )\right)^{r+1}D^2\mathrm{\hspace{0.17em}2}^{r(r1)}(\beta +D(\alpha 1))(r+D(\beta +D(\alpha 1)))^r.$$
Intersecting with the interval $`[H,H]`$ of the appropriate axis, the intervals of length $`2H`$. By 2.1, in each interval the points of $`X(\text{},H)`$ lie on at most
$$(24D4^{1/\rho }+2)(2H)^\rho H^{3\rho }6d^24^{1/\rho }2^\rho $$
real algebraic curves of degree $`d`$; the number of points of $`X`$ on a curve of degree $`d`$ is at most
$$2^{r(r1)/2}d\beta (r\alpha +d\beta )^r.$$
Combining these estimates yields
$$N(X,H)c^{}(r,\alpha ,\beta ,d,D)\mathrm{\hspace{0.17em}4}^{3(d+3)/8}H^{32/(3(d+3)}$$
where $`t=3(d+3)/8`$. Choose $`d`$ so that $`t`$ is as near as possible to (and so within $`1/2`$ of) $`\sqrt{4\mathrm{log}H/\mathrm{log}4}`$. Then $`4\sqrt{\mathrm{log}4}<5`$ and noting that $`d,D`$ appear polynomially in $`c^{}`$ completes the proof.
Remarks 3.2.
1. Note that the constant ‘5’ in 1.3 can be improved by further optimizing the proof. However, a bound of the shape $`\mathrm{exp}(c\sqrt{\mathrm{log}H})`$ seems to be the best obtainable by the present method.
2. A result can be formulated for any real analytic (or even smooth) function $`f`$ with suitable finiteness properties (zeros of derivatives, derivatives of the inverse, and algebraic relations). An example of such a function that is not pfaffian is exhibited in . (Indeed the given example $`e^x+\mathrm{sin}x`$ does not belong to any $`o`$-minimal structure: see ).
3. I expect a similar result should hold in higher dimensions for pfaff manifolds: that is, a uniform (in ‘complexity’) $`H^ϵ`$ bound for rational points that do not lie on some semialgebraic subset of positive dimension (cf the conjectures for subanalytic sets made in ). A similar result should hold for sets definable in an $`o`$-minimal structure.
4. Algebraic curves
For integers $`\beta ,\gamma 2`$ let
$$M(\beta ,\gamma )=\{x^hy^k:0h\beta 1,0k\gamma 1\}.$$
Then (), for $`M=M(\beta ,\gamma )`$,
$$D=\beta \gamma ,R=\frac{D(\gamma +\beta 2)}{2},S=D(\beta 1+\gamma 1)=2R,C2D,$$
and (elementarily)
$$\mathrm{max}(\frac{1}{\beta },\frac{1}{\gamma })\rho \frac{1}{\beta }+\frac{1}{\gamma }.$$
Proof of 1.4. The proof adapts the proof of \[10, 1.4\] using 2.1 instead of \[10, 4.2\].
Consider first a $`C^{\mathrm{}}`$ function $`f`$ on a subinterval of $`[1,1]`$ with $`|f^{}|1`$, with $`f^{(j)}`$ either nonvanishing or identically vanishing for $`j=0,\mathrm{},D`$. Suppose $`f`$ satisfies an irreducible algebraic relation of degree $`(b,c),d=\mathrm{max}(b,c)`$. If $`d=b`$ take $`M=(d,\delta )`$ with $`\delta d`$; if $`d=c`$ take $`M=M(\delta ,d)`$ with $`\delta d`$. Then, by 2.1, $`X(\text{},H)`$ is contained in the union of at most
$$10d^2\delta ^24^d2^\rho H^{2\rho }20d^2\delta ^24^dH^{2/d+2/\delta }$$
curves defined in $`M`$. The intersections are proper, $`X`$ is of degree $`b+c2d`$, the curves in $`M`$ of degree $`2\delta `$, so
$$N(X,H)80d^3\delta ^34^dH^{2/d+2/\delta }.$$
Next consider an algebraic curve $`X`$ defined by $`F(x,y)=0`$ in the box $`B=[1,1]^2`$, where $`F`$ is irreducible of bidegree $`(b,c)`$ and $`d=\mathrm{max}(b,c)`$. Then $`X`$ has at most $`2d(2d1)`$ singular points, and at most $`4d(d1)`$ points with slope $`\pm 1`$. So $`XB`$ consists of at most $`20d^3`$ graphs of $`C^{\mathrm{}}`$ functions $`f`$ with slope $`|f^{}|1`$ relative to one of the coordinate axes.
For each such function, the domain can be divided into at most $`8d^2D^2`$ subintervals (see \[2, Lemmas 5 and 6\]) in which $`f^{(j)}`$ is nonvanishing or identically zero, $`j=1,2,\mathrm{},D`$. So
$$N(X,H)25.2^{10}d^{1o}\delta ^54^dH^{2/d+2/\delta }.$$
Finally, let $`F(x,y)`$ of bidegree $`(b,c),d=\mathrm{max}(b,c)`$, $`X=\{(x,y)\text{}^2:F(x,y)=0\}`$. Let $`P=(x,y)X(\text{})`$ with $`H(P)H`$. Then one of the following holds:
(i) $`|x|,|y|1`$
(ii) $`|x|1,|y|>1`$
(iii) $`|x|>1,|y|1`$
(iv) $`|x|>1,|y|>1`$.
In case (i), $`P`$ lies in the box $`[1,1]^2\text{}^2`$. In case (ii), the point $`Q=(x,1/y)`$ is on the curve $`Y:y^cF(x,1/y)=0`$. This curve is also irreducible and of bidegree $`(b,c)`$ (because $`F`$ must have a term independent of $`y`$). The point $`Q`$ is then in the box $`[1,1]^2`$ and has $`H(Q)H`$. Likewise in cases (iii) and (iv) the corresponding points $`R=(1/x,y),S=(1/x,1/y)`$ lie on irreducible curves $`x^bF(1/x,y)=0,x^by^cF(1/x,1/y)=0`$ of bidegree $`(b,c)`$ in the box $`[1,1]^2`$ and have height $`H`$.
Therefore, up to a factor 4, it suffices to consider the points of $`F`$ inside the box $`[1,1]^2`$, so
$$N(X,H)100(2d)^{10}4^d\delta ^5H^{2/d+2/\delta }.$$
Take $`\delta `$ to be the least integer exceeding $`\mathrm{log}H`$. Then, provided $`He^d`$ (so that $`\delta d`$),
$$N(X,H)100e^22^{15}d^{10}(\mathrm{log}H)^54^dH^{2/d}.$$
However for $`\mathrm{log}Hd`$ the bound is easily seen to hold as well.
Acknowledgements
This paper was written while I was a visitor at the Mathematical Institute, Oxford. I am grateful to the Institute, and in particular to D. R. Heath-Brown and A. Lauder, for their hospitality. My stay in Oxford was supported by my home institution, McGill University, and in part by a grant from NSERC, Canada.
References
1. E. Bombieri, email 2003.
2. E. Bombieri and J. Pila, The number of integral points on arcs and ovals, Duke Math. J. 59 (1989), 337–357.
3. A. Gabrielov and N. Vorobjov, Complexity of computations with pfaffian and noetherian functions, in Normal Forms, Bifurcations and Finiteness problems in Differential Equations, Kluwer, 2004.
4. J. Gwozdziewicz, K. Kurdyka, and A. Parusinski, On the number of solutions of an algebraic equation on the curve $`y=e^x+\mathrm{sin}x,x>0`$, and a consequence for o-minimal structures, Proc. Amer. Math. Soc. 127 (1999), 1057–1064.
5. D. R. Heath-Brown, The density of rational points on curves and surfaces, Ann. Math. 155 (2002), 553–595.
6. A. G. Khovanskii, Fewnomials, Translations of Mathematical Monographs 88, AMS, Providence, 1991.
7. S. Lang, Introduction to transcendental numbers, Addison-Wesley, Reading, 1966.
8. J. Pila, Geometric postulation of a smooth function and the number of rational points, Duke Math. J. 63 (1991), 449–463.
9. J. Pila, Integer points on the dilation of a subanalytic surface, Quart. J. Math. 55 (2004), 207–223
10. J. Pila, Rational points on a subanalytic surface, submitted.
11. A. B. Shidlovskii, Transcendental numbers, translated from the Russian by N. Koblitz with a foreword by W. D. Brownawell. de Gruyter Studies in Mathematics 12, Walter de Gruyter & Co., Berlin, 1989.
Department of Mathematics and Statistics Mathematical Institute McGill University University of Oxford Burnside Hall 24-29 St Giles 805 Sherbrooke Street West Oxford OX1 3LB Montreal, Quebec, H3A 2K6 UK
Canada
pila@math.mcgill.ca
(Submitted for publication on 11 August 2004)
|
warning/0506/math0506282.html
|
ar5iv
|
text
|
# Untitled Document
ANALYTICITY ON TRANSLATES OF A JORDAN CURVE
Josip Globevnik
ABSTRACT Let $`\mathrm{\Omega }`$ be a domain in $`\mathrm{C}\text{ }`$ which is symmetric with respect to the real axis and whose boundary is a real analytic simple closed curve. Translate $`\overline{\mathrm{\Omega }}`$ vertically to get $`K=\{\overline{\mathrm{\Omega }}+it,rtr\}`$ where $`r>0`$ is such that $`(\overline{\mathrm{\Omega }}ir)(\overline{\mathrm{\Omega }}+ir)=\mathrm{}`$. We prove that if $`f`$ is a continuous function on $`K`$ such that for each $`t,rtr`$, the function $`f|(b\mathrm{\Omega }+it)`$ has a continuous extension to $`\overline{\mathrm{\Omega }}+it`$ which is holomorphic on $`\mathrm{\Omega }+it`$ then $`f`$ is holomorphic on $`\text{Int}K`$.
1. Introduction Write $`\mathrm{\Delta }(a,r)=\{\zeta \mathrm{C}\text{ }:|\zeta a|<r\}`$ and let $`\mathrm{\Delta }=\mathrm{\Delta }(0,1)`$. Translate $`b\mathrm{\Delta }`$ vertically to get the strip $`S=b\mathrm{\Delta }+i\mathrm{IR}=\{\zeta \mathrm{C}\text{ }:1\mathrm{}\zeta 1\}`$. Let $`f`$ be a continuous function on $`S`$ such that for each $`t\mathrm{IR}`$ the function $`f|(b\mathrm{\Delta }+it)`$ has a continuous extension to $`\overline{\mathrm{\Delta }}+it`$ which is holomorphic on $`\mathrm{\Delta }+it`$. Must $`f`$ be holomorphic on $`\text{Int}S`$ \[G2\]?
A positive answer was obtained for real analytic functions by M. Agranovsky and the author \[AG\] and independently by L. Ehrenpreis \[E\] and for continuous functions by A. Tumanov \[T1\].
To answer the question above one passes in both \[AG\] and \[T1\] to an associated problem in $`\mathrm{C}\text{ }^2`$. In \[AG\] the authors use semi-quadrics
$$\mathrm{\Lambda }_{a,\rho }=\{(z,w):(za)(w\overline{a})=\rho ^2,0<|za|<\rho \}$$
which are attached to $`\mathrm{\Sigma }=\{(z,\overline{z}):z\mathrm{C}\text{ }\}`$ along the circles $`\{(z,\overline{z}):zb\mathrm{\Delta }(a,\rho )\}`$. They use the property of $`\mathrm{\Lambda }_{a,\rho }`$ that a continuous function on $`b\mathrm{\Delta }(a,\rho )`$ extends holomorphically through if and only if the function $`F`$ defined on $`b\mathrm{\Lambda }_{a,\rho }=\{(z,\overline{z}):zb\mathrm{\Delta }(a,\rho )\}`$ by $`F(z,\overline{z})=f(z)(zb\mathrm{\Delta }(a,\rho ))`$ has a bounded continuous extension to $`\mathrm{\Lambda }_{a,\rho }b\mathrm{\Lambda }_{a,\rho }`$ which is holomorphic on $`\mathrm{\Lambda }_{a,\rho }`$. Thus, when studying holomorphic extensions of a function $`f`$ from circles in the plane one defines $`F(z,\overline{z})=f(z)`$ in a region in $`\mathrm{\Sigma }`$ and then studies bounded holomorphic extensions of $`F`$ from $`b\mathrm{\Lambda }_{a,\rho }`$ through $`\mathrm{\Lambda }_{a,\rho }`$ \[AG, G3\].
Tumanov \[T1\] passes to a problem in $`\mathrm{C}\text{ }^2`$ by adding an extra variable to make the translates of the disc pairwise disjoint and then, on the union of these discs, the (smooth) manifold
$$N_1=\{(\zeta +it,\zeta ):\zeta \overline{\mathrm{\Delta }},t\mathrm{IR}\},$$
he defines a continuous function $`F`$ by letting, for each $`t\mathrm{IR}`$, the function $`\zeta F(\zeta +it,\zeta )(\zeta \overline{\mathrm{\Delta }})`$ be the holomorphic extension of $`\zeta f(\zeta +it)(\zeta b\mathrm{\Delta })`$ through $`\mathrm{\Delta }`$. He then observes that the symmetry
$$F(\zeta +it,\zeta )=F(\zeta +it,1/\zeta )(\zeta b\mathrm{\Delta },t\mathrm{IR})$$
$`(1.1)`$
which follows from the fact that $`F(\zeta +it,\zeta ))=f(\zeta +it)(\zeta b\mathrm{\Delta },t\mathrm{IR})`$, makes possible to extend $`F`$ continuously to a new geometric object, the (smooth) manifold
$$N_2=\{(\zeta +it,1/\zeta ):\zeta \overline{\mathrm{\Delta }}\{0\},t\mathrm{IR}),$$
by using the equality (1.1) for $`\zeta \overline{\mathrm{\Delta }}\{0\},t\mathrm{IR}`$ as a definition. Thus one gets a continuous CR function $`F`$ on $`N_1N_2`$, the union of two manifolds with common boundary
$$N_1N_2=bN_1=bN_2=\{(\zeta +it,\zeta ):\zeta b\mathrm{\Delta },t\mathrm{IR}\}.$$
Tumanov then uses methods of CR theory to show that $`F`$ does not depend on the second variable which means that $`f`$ is holomorphic on $`\text{Int}S`$. He also discovers that this is actually a finite strip problem.
Very recently Tumanov \[T2\] studied a similar problem for a family of circles with centers sliding along a smooth curve and with smoothly changing radii. He used semi-quadrics. He obtained the result by using a classical argument of H. Lewy about holomorphic extensions of CR functions. In particular, he found a very simple proof of the theorem on the strip.
In the present paper we generalize the result of Tumanov \[T1\] from vertical translates of circles to vertical translates of real-analytic simple closed curves which are symmetric with respect to the real axis.
2. The main results Our first result is about analyticity on vertical translates of curves which are symmetric with respect to the real axis. Theorem 2.1 Let $`\mathrm{\Omega }`$ be a domain in $`\mathrm{C}\text{ }`$ which is symmetric with respect to the real axis and whose boundary is a real analytic simple closed curve. Let $`r>0`$ and let $`f`$ be a continuous function on $`K=\{b\mathrm{\Omega }+it,rtr\}`$ such that for each $`t,rtr`$, the function $`f|(b\mathrm{\Omega }+it)`$ has a continuous extension to $`\overline{\mathrm{\Omega }}+it`$ which is holomorphic on $`\mathrm{\Omega }+it`$. Suppose that $`(\overline{\mathrm{\Omega }}ir)(\overline{\mathrm{\Omega }}+ir)=\mathrm{}`$. Then $`f`$ is holomorphic on $`\text{Int}K`$. Note that our assumptions imply that $`K=\{\overline{\mathrm{\Omega }}+it,rtr\}`$.
We will deduce Theorem 2.1 from a more general result below which involves vertical translates of general domains and their images under conjugation.
Let $`D`$ be a domain in $`\mathrm{C}\text{ }`$ bounded by a real-analytic simple closed curve. Let $`S`$ be the vertical strip defined by $`S=\{bD+it,t\mathrm{IR}\}=\{\zeta \mathrm{C}\text{ }:\alpha \mathrm{}\zeta \beta \}`$. Write $`D^{}=\{\overline{\zeta }:\zeta D\}`$. Obviously, $`\{bD^{}+is,s\mathrm{IR}\}=S`$. Theorem 2.2 Let $`\lambda :[\alpha ,\beta ]\mathrm{IR}`$ be a continuous function and let $`a,b,c,d`$ be real numbers such that $`\overline{D}+ia,\overline{D^{}}+ic`$ are both contained in $`\{t+is:s<\lambda (t),\alpha t\beta \}`$ and such that $`\overline{D}+ib,\overline{D^{}}+id`$ are both contained in $`\{t+is:s>\lambda (t),\alpha t\beta \}`$. Let
$$Q_1=\{bD+it:atb\},Q_2=\{bD^{}+is:csd\}$$
and let $`f`$ be a continuous function on $`Q_1Q_2`$ such that
$$\begin{array}{cc}& \text{for each }t,atb,\text{the function }f|(bD+it)\text{ has a conti-}\hfill \\ & \text{nuous extension to }\overline{D}+it\text{ which is holomorphic on }D+it,\hfill \end{array}\}$$
$`(2.1)`$
$$\begin{array}{cc}\hfill \text{for each }s,csd,\text{the function }f|(bD^{}+is)\text{ has a conti-}& \\ \hfill \text{ nuous extension to }\overline{D^{}}+is\text{ which is holomorphic on }D^{}+is.& \end{array}\}$$
$`(2.2)`$
Then the function $`f`$ is holomorphic on $`\text{Int}Q_1\text{Int}Q_2`$.Our assumptions about $`a,b,c,d`$ mean that $`\overline{D}+ia,\overline{D^{}}+ic`$ both lie below the curve $`\mathrm{}=\{t+i\lambda (t):\alpha t\beta \}`$ and $`\overline{D}+ib,\overline{D^{}}+id`$ both lie above the curve $`\mathrm{}`$. Note that this implies that $`Q_1=\{\overline{D}+it:atb\}`$ and $`Q_2=\{\overline{D^{}}+is:csd\}`$. Theorem 2.1 follows from Theorem 2.2 by putting $`\mathrm{\Omega }=D=D^{}`$, $`a=c=r,b=d=r`$ and $`\lambda 0`$, that is, $`\mathrm{}=[\alpha ,\beta ]`$.
3. From circles to general curves
In this section we describe the idea how to pass from circles to general curves. Let $`\mathrm{\Omega }`$ be a domain bounded by a real-analytic simple closed curve which is symmetric with respect to the real axis. With no loss of generality assume that $`\mathrm{\Omega }`$ contains the origin. Let $`f`$ be a continuous function on $`\{b\mathrm{\Omega }+it:t\mathrm{IR}\}`$ such that for each $`t\mathrm{IR}`$, the function $`\zeta f(\zeta +it)(\zeta b\mathrm{\Omega })`$ has a continuous extension to $`\overline{\mathrm{\Omega }}`$ which is holomorphic on $`\mathrm{\Omega }`$.
Semi-quadrics are related to circles so they cannot be used to study the analyticity of functions on a family of translates of a given curve that is not a circle. We look again at the way how Tumanov \[T1\] adds the extra variable in the case of the circles. An important point in his setting is that on $`b\mathrm{\Delta }`$ the conjugation $`z\overline{z}`$ extends to the map $`z1/z`$ which carries $`\mathrm{\Delta }\{0\}`$ biholomorphically onto $`\mathrm{C}\text{ }\overline{\mathrm{\Delta }}`$. This is not the case for general curves so for domains $`\mathrm{\Omega }`$ more general than a disc it seems difficult to work with the manifold $`\{(\zeta +it,\zeta ):\zeta \overline{\mathrm{\Omega }},t\mathrm{IR}\}`$ used in \[T1\] when $`\mathrm{\Omega }`$ is a disc. However, since the reflection (1.1) takes place only in the second variable the idea is to replace the manifold $`\{(\zeta +it,\zeta ):\zeta \overline{\mathrm{\Omega }},t\mathrm{IR}\}`$ with a manifold that is attached to the cylinder $`\{(z,w):|w|=1\}`$. To do this we take the conformal map $`\mathrm{\Phi }:\mathrm{\Omega }\mathrm{\Delta }`$ that satisfies $`\mathrm{\Phi }(\overline{\zeta })=\overline{\mathrm{\Phi }(\zeta )}(\zeta \mathrm{\Omega }),\mathrm{\Phi }(0)=0`$, notice that $`\mathrm{\Phi }`$ extends to a diffeomorphism $`\mathrm{\Phi }`$ from $`\overline{\mathrm{\Omega }}`$ to $`\overline{\mathrm{\Delta }}`$ and define the smooth manifold $`N_1`$ by
$$N_1=\{(\zeta +it,\mathrm{\Phi }(\zeta )):\zeta \overline{\mathrm{\Omega }},t\mathrm{IR}\}.$$
We define a continuous function $`F`$ on $`N_1`$ by letting, for each $`t\mathrm{IR}`$, the function $`\zeta F(\zeta +it,\mathrm{\Phi }(\zeta ))(\zeta \overline{\mathrm{\Omega }})`$ be the holomorphic extension of $`\zeta f(\zeta +it)(\zeta b\mathrm{\Omega })`$ through $`\mathrm{\Omega }`$.
If $`\zeta b\mathrm{\Omega }`$ then $`\overline{\zeta }b\mathrm{\Omega }`$ and if $`t\mathrm{IR}`$ then $`\zeta +it=\overline{\zeta }+is`$ where $`s=t+(\zeta \overline{\zeta })/i\mathrm{IR}`$ so $`F(\zeta +it,\mathrm{\Phi }(\zeta ))=f(\zeta +it)=f(\overline{\zeta }+is)=F(\overline{\zeta }+is,\mathrm{\Phi }(\overline{\zeta }))=F(\overline{\zeta }+is,\overline{\mathrm{\Phi }(\zeta )})=F(\overline{\zeta }+is,1/\mathrm{\Phi }(\zeta ))=F(\zeta +it,1/\mathrm{\Phi }(\zeta ))`$ so
$$F(\zeta +it,\mathrm{\Phi }(\zeta ))=F(\zeta +it,1/\mathrm{\Phi }(\zeta ))(\zeta b\mathrm{\Omega },t\mathrm{IR})$$
$`(3.1)`$
which makes possible to extend $`F`$ continuously to a new geometric object, the smooth manifold
$$N_2=\{(\zeta +it,1/\mathrm{\Phi }(\zeta )):\zeta \overline{\mathrm{\Omega }}\{0\},t\mathrm{IR}\},$$
by using the equality (3.1) for $`\zeta \overline{\mathrm{\Omega }}\{0\}`$ as a definition. Thus we get a continuous CR function $`F`$ on $`N_1N_2`$, the union of two manifolds with the common boundary
$$N_1N_2=bN_1=bN_2=\{(\zeta +it,\mathrm{\Phi }(\zeta )):\zeta b\mathrm{\Omega },t\mathrm{IR}\}.$$
We then show that the classical argument of H. Lewy which Tumanov used with semiquadrics works also in the present situation. This helps us to prove that $`F`$ depends only on the first variable which implies that $`f`$ is holomorphic.
In fact, our main result, Theorem 2.2, is somewhat more general than the one just described. Its proof, although technically a bit complicated, uses essentially the idea above.
4. The manifold $`N`$We now begin with the proof of Theorem 2.2. With no loss of generality we assume that $`0D`$ and that the imaginary axis intersects $`bD`$ transversely.
Let $`\mathrm{\Phi }:D\mathrm{\Delta }`$ be a conformal map such that $`\mathrm{\Phi }(0)=0`$. Since $`bD`$ is real-analytic the map $`\mathrm{\Phi }`$ extends to a biholomorphic map from a neighbourhood of $`\overline{D}`$ to a neighbourhood of $`\overline{\mathrm{\Delta }}`$. Define $`\mathrm{\Psi }:D^{}\mathrm{\Delta }`$ by
$$\mathrm{\Psi }(\zeta )=\overline{\mathrm{\Phi }(\overline{\zeta })}(\zeta D^{}).$$
The map $`\mathrm{\Psi }`$ maps $`D^{}`$ conformally onto $`\mathrm{\Delta }`$ and extends to a biholomorphic map from a neighbourhood of $`\overline{D^{}}`$ to a neighbourhood of $`\overline{\mathrm{\Delta }}`$.
Define
$$\begin{array}{cc}& N_1=\{(\xi +it,\mathrm{\Phi }(\xi )):\xi \overline{D},t\mathrm{IR}\},\hfill \\ & N_2=\{(\zeta +is,1/\mathrm{\Psi }(\zeta )):\zeta \overline{D^{}}\{0\},s\mathrm{IR}\}\hfill \end{array}$$
and set $`N=N_1N_2`$. Write
$$\mathrm{\Phi }^1(w)=p(w)+iq(w)(w\overline{\mathrm{\Delta }})$$
where $`p`$ and $`q`$ are real functions. Then $`N_1=\{(p(w)+iq(w)+it,w):w\overline{\mathrm{\Delta }},t\mathrm{IR}\}`$ $`=\{(p(w)+it,w):w\overline{\mathrm{\Delta }},t\mathrm{IR}\}=\{(p(w),w):w\overline{\mathrm{\Delta }}\}+\mathrm{IR}(i,0)`$. If $`w=1/\mathrm{\Psi }(\zeta )`$ then $`\overline{\mathrm{\Phi }(\overline{\zeta })}=\mathrm{\Psi }(\zeta )=1/w`$ so $`\zeta =\overline{\mathrm{\Phi }^1(1/\overline{w})}=p(1/\overline{w})iq(1/\overline{w})`$ which implies that $`N_2=\{(p(1/\overline{w})iq(1/\overline{w})+is,w):w\mathrm{C}\text{ }\mathrm{\Delta },s\mathrm{IR}\}=\{(p(1/\overline{w})+is,w):w\mathrm{C}\text{ }\mathrm{\Delta },s\mathrm{IR}\}=\{p(1/\overline{w}),w):w\mathrm{C}\text{ }\mathrm{\Delta }\}+\mathrm{IR}(i,0)`$. Define
$$\theta (w)=\{\begin{array}{cc}& p(w)(w\overline{\mathrm{\Delta }})\hfill \\ & p(1/\overline{w})(w\mathrm{C}\mathrm{\Delta }).\hfill \end{array}$$
The function $`\theta `$ is well defined and continuous on $`\mathrm{C}\text{ }`$ since $`w=1/\overline{w}(wb\mathrm{\Delta })`$. The function $`\theta `$ is invariant with respect to $`w1/\overline{w}`$, the reflection across $`b\mathrm{\Delta }`$. Note that $`\theta |\overline{\mathrm{\Delta }}`$ is smooth as it extends to a harmonic function in a neighbourhood of $`\overline{\mathrm{\Delta }}`$. Similarly, $`\theta |(\mathrm{C}\text{ }\mathrm{\Delta })`$ is smooth. We have
$$N=\{(\theta (w)+it,w):w\mathrm{C}\text{ },t\mathrm{IR}\}$$
$`(4.1)`$
which shows that we obtain $`N`$ by taking the graph $`\{(\theta (w),w):w\mathrm{C}\text{ }\}`$ of $`\theta `$ in $`\mathrm{IR}\times \mathrm{C}\text{ }=\mathrm{IR}\times \{0\}\times \mathrm{C}\text{ }`$ and then making the union of all translates of this graph in the extra perpendicular direction $`(i,0)`$, that is,
$$N=\{(\theta (w),w):w\mathrm{C}\text{ }\}+\mathrm{IR}(i,0).$$
Since
$$\begin{array}{cc}& N_1=\{(\theta (w),w):w\overline{\mathrm{\Delta }}\}+\mathrm{IR}(i,0)\hfill \\ & N_2=\{(\theta (w),w):w\mathrm{C}\mathrm{\Delta }\}+\mathrm{IR}(i,0)\hfill \end{array}$$
$`(4.2)`$
we see that $`N`$ is the union of manifolds $`N_1`$ and $`N_2`$ with boundary which meet along the common boundary
$$N_1N_2=bN_1=bN_2=\{(\theta (w),w):wb\mathrm{\Delta }\}+\mathrm{IR}(i,0).$$
The complement of the graph of $`\theta `$ in $`\mathrm{IR}\times \mathrm{C}\text{ }`$ has two components: $`\{(t,w):t>\theta (w),w\mathrm{C}\text{ }\}`$ and $`\{(t,w):t<\theta (w),w\mathrm{C}\text{ }\}`$, which, by (4.1) implies that $`\mathrm{C}\text{ }^2N`$ has two components
$$\begin{array}{cc}& P_1=\{(t+is,w):t>\theta (w),s\mathrm{IR},w\mathrm{C}\}\hfill \\ & P_2=\{(t+is,w):t<\theta (w),s\mathrm{IR},w\mathrm{C}\}.\hfill \end{array}$$
5. Intersecting $`N`$ with complex lines We will apply the reasoning of H. Lewy about holomorphic extensions of CR functions. To this end, we look first at the intersections of $`N`$ with complex lines $`L(z)=\{(z,w):w\mathrm{C}\text{ }\}`$. We shall use the fact that since $`bD`$ is real-analytic and compact there are at most finitely many points $`\zeta bD`$ such that the tangent line to $`bD`$ at $`\zeta `$ is parallel to the imaginary axis.
For $`zS`$ write
$$\stackrel{~}{E}(z)=NL(z),\stackrel{~}{E}_j(z)=N_jL(z),j=1,2$$
and
$$\mathrm{\Lambda }(z)=\{w\mathrm{C}\text{ }:(z,w)N\},\mathrm{\Lambda }_j(z)=\{w\mathrm{C}\text{ }:(z,w)N_j\},j=1,2,$$
so that
$$\stackrel{~}{E}(z)=\{z\}\times \mathrm{\Lambda }(z),\stackrel{~}{E}_j(z)=\{z\}\times \mathrm{\Lambda }_j(z),j=1,2.$$
For each $`t\mathrm{IR}`$ we have $`N+t(i,0)=N,N_j+t(i,0)=N_j,j=1,2`$, so it follows that
$$\mathrm{\Lambda }(z)=\mathrm{\Lambda }(\mathrm{}z),\mathrm{\Lambda }_j(z)=\mathrm{\Lambda }_j(\mathrm{}z)(j=1,2,zS).$$
Thus, it is enough to study $`\mathrm{\Lambda }(\tau ),\mathrm{\Lambda }_j(\tau ),j=1,2`$, where $`\alpha \tau \beta `$.
The set $`\stackrel{~}{E}(\tau )=\{\tau \}\times \mathrm{\Lambda }(\tau )`$, contained in $`\mathrm{IR}^3=\mathrm{IR}\times \{0\}\times \mathrm{C}\text{ }`$, is the intersection of $`\{(\theta (w),w):w\mathrm{C}\text{ }\}`$, the graph of $`\theta `$, with the two-plane (in fact, the complex line), $`\{\tau \}\times \mathrm{C}\text{ }`$. Since $`\theta `$ is invariant with respect to the reflection across $`b\mathrm{\Delta }`$ it follows that we get $`\mathrm{\Lambda }_2(\tau )`$ by reflecting $`\mathrm{\Lambda }_1(\tau )\overline{\mathrm{\Delta }}`$ across $`b\mathrm{\Delta }`$. So it is enough to study $`\mathrm{\Lambda }_1(\tau )`$. Clearly
$$\mathrm{\Lambda }_1(\tau )=\mathrm{\Phi }(K(\tau ))\text{ where }K(\tau )=(\tau +i\mathrm{IR})\overline{D}.$$
If $`\tau [\alpha ,\beta ]`$ is such that $`\tau +i\mathrm{IR}`$ meets $`bD`$ transversely, as happens for all but finitely many $`\tau `$, then $`\mathrm{\Lambda }_1(\tau )`$ consists of finitely many pairwise disjoint closed arcs with endpoints on $`b\mathrm{\Delta }`$ but otherwise contained in $`\mathrm{\Delta }`$ which meet $`b\mathrm{\Delta }`$ transversely. By transversality and by the fact that $`\mathrm{\Phi }`$ extends across $`bD`$ as a conformal map, these arcs change smoothly with $`\tau `$ as long as $`\tau +i\mathrm{IR}`$ meets $`bD`$ transversely. If $`\tau (\alpha ,\beta )`$ is such that $`\tau +i\mathrm{IR}`$ does not meet $`bD`$ transversely then $`\mathrm{\Lambda }(\tau )`$ consists of a finite number of arcs with endpoints on $`b\mathrm{\Delta }`$ and pairwise disjoint interiors plus a possible finite set on $`b\mathrm{\Delta }`$. There may be points on $`b\mathrm{\Delta }`$ which are common endpoints of two (but not more than two ) of these arcs. Since we get $`\mathrm{\Lambda }_2(\tau )`$ by reflecting $`\mathrm{\Lambda }_1(\tau )`$ across $`b\mathrm{\Delta }`$ it follows that if $`\tau 0`$ and if $`\tau +i\mathrm{IR}`$ is transverse to $`bD`$ then $`\mathrm{\Lambda }(\tau )`$ consists of finitely many pairwise disjoint simple closed curves, symmetric with respect to $`b\mathrm{\Delta }`$. If $`\tau 0`$ and $`\tau +i\mathrm{IR}`$ does not meet $`bD`$ transversely then $`\mathrm{\Lambda }(\tau )`$ consists of finitely many pairwise disjoint simple closed curves, symmetric with respect to $`b\mathrm{\Delta }`$ plus a possible finite subset of $`b\mathrm{\Delta }`$. There may be points on $`b\mathrm{\Delta }`$ that are common points of two, but not more than two of these curves. Except for these points, the curves are pairwise disjoint. Clearly $`\mathrm{\Lambda }(\alpha )`$ and $`\mathrm{\Lambda }(\beta )`$ are finite sets.
Since $`i\mathrm{IR}`$ meets $`bD`$ transversely $`\mathrm{\Lambda }_1(0)`$ consists of finitely many pairwise disjoint closed arcs with endpoints on $`b\mathrm{\Delta }`$ but otherwise contained in $`\mathrm{\Delta }`$ which meet $`b\mathrm{\Delta }`$ transversely. One of these arcs passes through the origin so its image under the reflection across $`b\mathrm{\Delta }`$ passes through infinity. Thus, $`\mathrm{\Lambda }(0)\{\mathrm{}\}`$ consists of finitely many pairwise disjoint simple closed curves on the Riemann sphere one of which contains infinity.
For each $`z,\alpha <\mathrm{}z<0`$ the set $`Y(z)=P_1L(z)`$ is a bounded open subset of $`L(z)`$ whose boundary $`E(z)=bY(z)`$ is the part of $`\stackrel{~}{E}(z)=\{z\}\times \mathrm{\Lambda }(z)`$ consisting of curves (recall that in addition to these curves, $`\stackrel{~}{E}(z)`$ may contain an additional finite set contained in $`\{z\}\times b\mathrm{\Delta })`$. The complex line $`L(z)`$ has a natural orientation. We orient $`E(z)`$ as the boundary of $`Y(z)`$ in $`L(z)`$. Similarly, for $`0<\mathrm{}z<\beta `$ we orient $`E(z)`$ as the boundary of $`Y(z)=P_2L(z)`$ in $`L(z)`$. This determines the orientation of $`\mathrm{\Lambda }(z),\alpha <\mathrm{}z<\beta ,\mathrm{}z0`$, or more precisely, the part of $`\mathrm{\Lambda }(z)`$ consisting of curves, and the orientation of $`K(\tau ),\alpha <\tau <\beta ,\tau 0`$, upwards if $`\alpha <\tau <0`$ and downwards if $`0<\tau <\beta `$.
If $`0<\tau <\beta `$ and a point $`(\tau ,w)`$ is above the graph of $`\theta `$, that is, contained in $`P_1`$, then $`[\tau ,\beta ]\times \{w\}`$ is contained in $`P_1`$. A consequence of this is Proposition 5.1 Suppose that $`0<\tau <\beta `$ and let $`q_0L(\tau +i\lambda (\tau ))P_1`$. Then there is a path $`tq(t)(\tau t\beta )`$ such that $`q(\tau )=q_0`$ and $`q(t)L(t+i\lambda (t))P_1(\tau t\beta )`$.Proof. We have $`q_0=(\tau +i\lambda (\tau ),w)`$ where $`\tau >\theta (w)`$. Define $`q(t)=(t+i\lambda (t),w)(\tau t\beta )`$. It is easy to see that $`q`$ has all the required properties. A similar proposition holds for $`\alpha <\tau <0`$ and for $`P_2`$ in the place of $`P_1`$.
6. Continuity of an integralWe shall need Proposition 6.1 Let $`z_0S`$ and suppose that $`G`$ is a continuous function on a neighbourhood of $`\stackrel{~}{E}(z_0)`$ in $`N`$. Then the function
$$\mathrm{\Theta }(z)=_{\mathrm{\Lambda }(z)}G(z,w)𝑑w$$
defined in a neighbourhood of $`z_0`$ in $`S`$, is continuous at $`z_0`$. Proof. We prove the continuity of
$$\mathrm{\Theta }_1(z)=_{\mathrm{\Lambda }_1(z)}G(z,w)𝑑w=_{K(\mathrm{}z)}G(z,\mathrm{\Phi }(\zeta ))\mathrm{\Phi }^{}(\zeta )𝑑\zeta .$$
The proof for $`\mathrm{\Theta }_2(z)=_{\mathrm{\Lambda }_2(z)}G(z,w)𝑑w`$ will be analogous; note that $`\mathrm{\Theta }=\mathrm{\Theta }_1+\mathrm{\Theta }_2`$ since $`\mathrm{\Lambda }_1(z)`$ and $`\mathrm{\Lambda }_2(z)`$ meet in a finite set. Recall that $`\mathrm{\Phi }`$ and $`\mathrm{\Phi }^{}`$ extend holomorphically into a neighbourhood of $`\overline{D}`$, so the continuity of $`\mathrm{\Theta }_1`$ depends on how $`K(t)=\overline{D}(t+i\mathrm{IR})`$ changes with $`t`$ near $`t_0=\mathrm{}z_0`$. Assume for a moment that $`\alpha <t_0<\beta `$. There are at most finitely many $`t,\alpha t\beta `$ such that $`t+i\mathrm{IR}`$ does not intersect $`bD`$ transversely. Thus there is an $`\eta >0`$ such that $`t+i\mathrm{IR}`$ intersects $`bD`$ transversely for every $`t,0<|tt_0|<\eta `$. In particular, for each $`t,t_0\eta <t<\eta `$, $`K(t)`$ is a finite collection of pairwise disjoint closed segments with endpoints varying continously with $`t`$. Since $`bD(t_0+i\mathrm{IR})`$ is a finite set and since $`bD`$ is compact it follows that each of these endpoints has a limit as $`tt_0`$. As $`tt_0`$ some segments may degenerate into points in the limit $`K(t_00)`$ and some pairs of segments may get a common endpoint. Clearly $`K(t_00)K(t_0)`$ and $`K(t_0)K(t_00)bD(t_0+i\mathrm{IR})`$. Since the set $`bD(t_0+i\mathrm{IR})`$ is finite it follows that $`K(t_00)K(t_0)`$ is a finite set. Thus, $`lim_{zz_0,\mathrm{}z\mathrm{}z_0}\mathrm{\Theta }(z)=\mathrm{\Theta }(z_0)`$. Similarly we show that $`lim_{zz_0,\mathrm{}z\mathrm{}z_0}\mathrm{\Theta }(z)=\mathrm{\Theta }(z_0)`$ which proves that $`\mathrm{\Theta }`$ is continuous at $`z_0`$. The same (one sided) reasoning applies if $`z_0bS`$. The proof is complete.
7. The manifold $`M`$ and the function $`F`$ We now define a submanifold of $`N`$ that is more closely related to our problem. Write
$$\begin{array}{cc}& M_1=\{(\xi +it,\mathrm{\Phi }(\xi )):\xi \overline{D},atb\}\hfill \\ & M_2=\{(\zeta +is,1/\mathrm{\Psi }(\zeta )):\zeta \overline{D^{}}\{0\},csd\}\hfill \end{array}$$
and let $`M=M_1M_2`$. Note that $`M_1`$ is a smooth manifold with boundary $`bM_1=\{(\xi +it,\mathrm{\Phi }(\xi )):\xi bD,atb\}\{(\xi +ia,\mathrm{\Phi }(\xi )):\xi \overline{D}\}\{(\xi +ib,\mathrm{\Phi }(\xi )):\xi \overline{D}\}`$. It is a submanifold of $`N_1`$. Similarly, $`M_2`$ is a smooth manifold with boundary $`bM_2=\{(\zeta +is,1/\mathrm{\Psi }(\zeta )):\zeta bD^{},csd\}\{(\zeta +ic,1/\mathrm{\Psi }(\zeta )):\zeta \overline{D^{}}\{0\}\}\{(\zeta +id,1/\mathrm{\Psi }(\zeta )):\zeta \overline{D^{}}\{0\}\}.`$
Suppose that $`f`$ is a continuous function on $`Q_1Q_2`$ which satisfies (2.1) and (2.2). For each $`t,atb`$, let $`g_t`$ be a continuous extension of $`\xi f(\xi +it)(\xi bD)`$ to $`\overline{D}`$ which is holomorphic on $`D`$ and for each $`s,csd,`$ let $`h_s`$ be the continuous extension of $`\zeta f(\zeta +is)(\zeta bD^{})`$ to $`\overline{D^{}}`$ which is holomorphic on $`D^{}`$. Define the function $`G`$ on $`M_1`$ by
$$G(\xi +it,\mathrm{\Phi }(\xi ))=g_t(\xi )(\xi \overline{D},atb)$$
and the function $`H`$ on $`M_2`$ by
$$H(\zeta +is,1/\mathrm{\Psi }(\zeta ))=h_s(\zeta )(\zeta \overline{D^{}}\{0\},csd).$$
In particular, on the part of $`bM_1`$ contained in $`bN_1`$ we have
$$G(\xi +it,\mathrm{\Phi }(\xi ))=f(\xi +it)(\xi bD,atb),$$
and on the part of $`bM_2`$ contained in $`bN_2`$ we have
$$H(\zeta +is,1/\mathrm{\Psi }(\zeta ))=f(\zeta +is)(\zeta bD^{},csd).$$
Suppose that $`(z,w)M_1M_2`$. Then there are $`\xi bD,\zeta bD^{}`$ and $`t,s,atb,csd`$, such that $`(\xi +it,\mathrm{\Phi }(\xi ))=(z,w)=(\zeta +is,1/\mathrm{\Psi }(\zeta ))=(\zeta +is,\overline{\mathrm{\Psi }(\zeta )})=(\zeta +is,\mathrm{\Phi }(\overline{\zeta }))`$ which implies that $`\zeta =\overline{\xi }`$ and $`\xi +it=\overline{\xi }+is`$. Thus, $`G(z,w)=G(\xi +it,\mathrm{\Phi }(\xi ))=f(\xi +it)=f(\overline{\xi }+is)=H(\overline{\xi }+is,1/\mathrm{\Psi }(\overline{\xi }))=H(\zeta +is,1/\mathrm{\Psi }(\zeta ))=H(z,w)`$. It follows that
$$F(z,w)=\{\begin{array}{cc}& G(z,w)((z,w)M_1)\hfill \\ & H(z,w)((z,w)M_2)\hfill \end{array}$$
is a well defined continuous function on $`M_1M_2`$ which is holomorphic on each holomorphic leaf of $`M_1`$ and on each holomorphic leaf of $`M_2`$.
Our aim is to show that $`F`$ depends only on $`z`$ which will imply that $`f`$ is holomorphic on $`\text{Int}Q_1\text{Int}Q_2`$.
8. Integrals of CR functions on $`M`$Denote by $`\pi _1`$ the projection $`\pi _1(z,w)=z`$. With no loss of generality assume that $`\lambda (0)=0`$. Our assumptions imply that there is an $`\eta >0`$ such that if
$$\mathrm{\Sigma }=\{t+is:\lambda (t)\eta <s<\lambda (t)+\eta ,\alpha <t<\beta \}$$
then $`L(z)M=L(z)N`$ for all $`z\mathrm{\Sigma }`$, that is, $`\pi _1^1(\mathrm{\Sigma })M=\pi _1^1(\mathrm{\Sigma })N`$. Put $`\mathrm{\Sigma }_1=\{z\mathrm{\Sigma },\mathrm{}z<0\}`$$`\mathrm{\Sigma }_2=\{z\mathrm{\Sigma },\mathrm{}z>0\}`$. Recall that for $`z\mathrm{\Sigma }_1`$ the set $`E(z)`$ is the boundary of $`Y(z)=P_1L(z)`$ in $`L(z)`$ and for $`z\mathrm{\Sigma }_2`$ the set $`E(z)`$ is the boundary of $`Y(z)=P_2L(z)`$ in $`L(z)`$. Let
$$A_j=\{Y(z):z\mathrm{\Sigma }_j\}=\pi _1(\mathrm{\Sigma }_j)P_j(j=1,2).$$
For each $`j=1,2,A_j`$ is an open subset of $`\mathrm{\Sigma }_j\times \mathrm{C}\text{ }`$ whose relative boundary is $`N(\mathrm{\Sigma }_j\times \mathrm{C}\text{ })`$. Using Proposition 5.1 we see that the complement of $`\overline{A}_j`$ in $`\mathrm{\Sigma }_j\times \mathrm{C}\text{ }`$ is connected, $`j=1,2`$.
We shall prove that the function $`F`$ extends holomorphically into $`A_1`$ and into $`A_2`$. We begin to follow the reasoning of H. Lewy. In \[L\] this was done for smooth functions on smooth manifolds and for more general, including continuous, functions on smooth manifolds this was done in \[R\]. We cannot refer to these results directly since in our case the manifold is not smooth but consists of two smooth pieces. However, these two pieces are both foliated by analytic discs which simplifies the situation. We provide the details to make the proof self contained.
Let $`\mathrm{\Delta }(u,r)`$ be contained in either $`\mathrm{\Sigma }_1`$ or $`\mathrm{\Sigma }_2`$ and assume that $`\mathrm{\Theta }`$ is a continuous function on $`\pi _1^1(\mathrm{\Delta }(u,r))N`$ which is holomorphic on each holomorphic leaf. The function
$$zQ(z)=_{\mathrm{\Lambda }(z)}\mathrm{\Theta }(z,w)𝑑w$$
is, as we know, well defined and by Proposition 6.1 it is continuous on $`\mathrm{\Delta }(u,r)`$. Recall that there are at most finitely many real values $`\tau `$ such that $`\tau +i\mathrm{IR}`$ is not transversal to $`bD`$. So, if we want to prove that $`Q`$ is holomorphic on $`\mathrm{\Delta }(u,r)`$ it is enough to prove that $`Q`$ is holomorphic in a neighbourhood of each $`z_0\mathrm{\Delta }(u,r)`$ such that $`z_0+i\mathrm{IR}`$ intersects $`bD`$ transversely. Let $`z_0`$ be such a point. Let $`\rho >0`$ be such that for each $`z\overline{\mathrm{\Delta }(z_0,\rho )}`$, $`z+i\mathrm{IR}`$ meets $`bD`$ transversely. Passing to a smaller $`\rho `$ if necessary we may assume that there is a $`\gamma >0`$ such that whenever $`U`$ is a closed disc contained in $`\mathrm{\Delta }(z_0,\rho )`$ of radius not exceeding $`\gamma `$, the circle $`bU+it,t\mathrm{IR}`$, either misses $`bD`$ or meets $`bD`$ in one point or in two points. Let $`U`$ be such a disc. By transversality, there is a positive integer $`\nu `$ such that for each $`zU`$ the set $`L(z)N=L(z)M`$ consists of $`\nu `$ pairwise disjoint simple closed curves, each being the union of two smooth arcs with endpoints on $`\{z\}\times b\mathrm{\Delta }`$, one having its interior contained in $`\{z\}\times \mathrm{\Delta }`$, and the other having its interior contained in $`\{z\}\times (\mathrm{C}\text{ }\overline{\mathrm{\Delta }})`$, which change smoothly with $`z`$. So $`\pi _1^1(U)N=\pi _1^1(U)M=\{L(z)N:zU\}`$ is an open subset of $`N`$ which has $`\nu `$ components whose closures are pairwise disjoint; the boundary of this set is $`\pi _1^1(bU)N=\{L(z)N:zbU\}`$. Let $`\mathrm{\Omega }`$ be one of these components. Write $`T=bN_1=bN_2`$. The set $`\mathrm{\Omega }`$ consists of three pairwise disjoint parts: the domains $`\mathrm{\Omega }_1=\mathrm{\Omega }\text{Int}N_1`$, $`\mathrm{\Omega }_2=\mathrm{\Omega }\text{Int}N_2`$ and the two dimensional surface $`\mathrm{\Omega }T`$. For each $`zbU`$, $`L(z)b\mathrm{\Omega }`$ is a simple closed curve so $`b\mathrm{\Omega }`$ is a torus and $`\mathrm{\Omega }`$ is a solid torus in $`N`$.
We now want to show that
$$_{b\mathrm{\Omega }}\mathrm{\Theta }(z,w)𝑑zdw=0.$$
$`(8.1)`$
Note first that $`\mathrm{\Omega }T`$ is the common part of $`b\mathrm{\Omega }_1`$ and $`b\mathrm{\Omega }_2`$ so to prove (8.1) it is enough to prove that
$$_{b\mathrm{\Omega }_1}\mathrm{\Theta }(z,w)𝑑zdw=0.$$
$`(8.2)`$
and
$$_{b\mathrm{\Omega }_2}\mathrm{\Theta }(z,w)𝑑zdw=0.$$
$`(8.3)`$
Consider (8.2). The properties of $`U`$ imply that $`\mathrm{\Omega }_1`$ can be written as the union of a continous family of pairwise disjoint analytic discs
$$A_t=\{(\zeta +it,\mathrm{\Phi }(\zeta )):\zeta D,\zeta +itU\}=\{((\zeta ,\mathrm{\Phi }(\zeta ))+it:\zeta D(it+U)\}$$
and $`b\mathrm{\Omega }_1`$ is the union of their pairwise disjoint boundaries
$$bA_t=\{(\zeta ,\mathrm{\Phi }(\zeta ))+it:\zeta b(D(it+U))\}$$
These analytic discs, if nonempty, are of two sorts: either their boundaries are smooth simple closed curves which meet $`\mathrm{\Omega }T`$ in at most one point (which happens if $`UD+it`$), or their boundaries are simple closed curves consisting of two arcs, one contained in $`\text{Int}N_1`$ and the other contained in $`T`$ (which happens if $`U`$ meets $`D+it`$ but is not contained in $`D+it`$). Recall that $`N_1=\{(\mathrm{{\rm Y}}(w)+it,w):w\overline{\mathrm{\Delta }},t\mathrm{IR}\}`$ where the conformal map $`\mathrm{{\rm Y}}=\mathrm{\Phi }^1`$ extends to a biholomorphic map from $`R\mathrm{\Delta }`$ for some $`R>1`$ to a neighbourhood of $`\overline{D}`$. Define the function $`\rho `$ by $`\rho (z,w)=(1/i)(z\mathrm{{\rm Y}}(w))`$ and notice that $`\rho `$ is real on $`N_1`$. Then $`\mathrm{\Theta }(z,w)dzdw=d\rho \mu `$ where $`\mu =(1/i)(\mathrm{\Theta }(z,w)/\mathrm{{\rm Y}}^{}(w))dz`$ which, by using the Fubini theorem on each of the three smooth pieces of $`b\mathrm{\Omega }_1`$ and adding the results, implies that
$$_{b\mathrm{\Omega }_1}\mathrm{\Theta }(z,w)𝑑zdw=_I\left[_{bA_t}\mu \right]𝑑t$$
$`(8.4)`$
where $`I\mathrm{IR}`$ is the segment of all $`t`$ such that $`\mathrm{\Omega }_1\{(\zeta +it,\mathrm{\Phi }(\zeta )):\zeta D\}`$ is not empty. For each $`tI`$ we have
$$_{bA_t}\mu =\frac{1}{i}_{bA_t}\mathrm{\Theta }(z,w)\frac{1}{\mathrm{{\rm Y}}^{}(w)}𝑑z$$
where the integral on the right vanishes by the by the Cauchy theorem since the function $`(z,w)\mathrm{\Theta }(z,w)/\mathrm{{\rm Y}}^{}(w)`$ is continuous on $`\overline{A}_t`$ and holomorphic on $`A_t`$. This proves that the integral on the left in (8.4) vanishes. We repeat the reasoning for $`N_2`$ to get (8.3). This proves (8.1). Thus,
$$_{bU}\left[_{b\mathrm{\Lambda }(z)}\mathrm{\Theta }(z,w)𝑑w\right]𝑑z=0$$
for every disc $`U\mathrm{\Delta }(z_0,\rho )`$, which, by the Morera theorem, implies that the function $`Q`$ is holomorphic on $`D(z_0,\rho )`$. This proves that $`Q`$ is holomorphic on $`\mathrm{\Sigma }_1\mathrm{\Sigma }_2`$.
9. Holomorphic extensions of $`F`$ and the completion of the proofWe continue to follow the reasoning of H. Lewy. For each $`z\mathrm{\Sigma }_1\mathrm{\Sigma }_2`$ and each $`w\mathrm{C}\text{ }\mathrm{\Lambda }(z)`$ (that is, for each $`w`$ such that $`(z,w)N)`$ define
$$\mathrm{\Xi }(z,w)=\frac{1}{2\pi i}_{\mathrm{\Lambda }(z)}\frac{F(z,\zeta )}{\zeta w}𝑑\zeta .$$
For a fixed $`z`$ the function $`w\mathrm{\Xi }(z,w)`$ is holomorphic on $`\mathrm{C}\text{ }\mathrm{\Lambda }(z)`$. Now, fix $`z_0\mathrm{\Sigma }_1`$ and $`w\mathrm{C}\text{ }\mathrm{\Lambda }(z_0)`$. We show that $`z\mathrm{\Xi }(z,w)`$ is holomorphic in a neighbourhood of $`z_0`$. There is a $`\rho >0`$ such that $`\mathrm{\Delta }(z_0,\rho )\mathrm{\Sigma }_1`$ and $`(z,w)N(z\mathrm{\Delta }(z_0,\rho ))`$ so the function $`(z,\zeta )F(z,\zeta )/(\zeta w)`$ is continuous on $`\{L(z)N:z\mathrm{\Delta }(z_0,\rho )\}`$ and holomorphic on each holomorphic leaf. By the reasoning in Section 8 it follows that $`z\mathrm{\Xi }(z,w)`$ is holomorphic on $`\mathrm{\Delta }(z_0,\rho )`$. This shows that $`\mathrm{\Xi }`$ is holomorphic on $`\pi _1^1(\mathrm{\Sigma }_1)N`$. Fix a large $`w`$. We know that $`z\mathrm{\Xi }(z,w)`$ is continuous on $`\mathrm{\Sigma }_1[\alpha +i(\delta ,\delta )]`$ and holomorphic on $`\mathrm{\Sigma }`$. Since $`\mathrm{\Lambda }(\alpha )`$ is a finite set it follows that $`\mathrm{\Xi }(z,w)`$ approaches $`0`$ as $`z`$ approaches a point of $`\alpha +i(\delta ,\delta )`$. It follows that $`\mathrm{\Xi }(z,w)0(z\mathrm{\Sigma }_1)`$. Since this holds for all sufficiently large $`w`$, the connectedness of $`(\mathrm{\Sigma }_1\times \mathrm{C}\text{ })\overline{A_1}`$ implies that $`\mathrm{\Xi }0`$ on $`(\mathrm{\Sigma }_1\times \mathrm{C}\text{ })\overline{A_1}`$ so
$$\frac{1}{2\pi i}_{\mathrm{\Lambda }(z)}\frac{F(z,\zeta )}{\zeta w}𝑑\zeta 0(z\mathrm{\Sigma }_1,w\mathrm{C}\text{ }\overline{Y(z)}).$$
It follows, in particular, that for all $`z\mathrm{\Sigma }_1`$ the function $`wF(z,w)`$, defined on $`E(z)=bY(z)`$, has a continuous extension into $`Y(z)bY(z)`$ which is holomorphic on $`Y(z)`$. The same reasoning shows this for all $`z\mathrm{\Sigma }_2`$.
Recall that $`\mathrm{\Lambda }(0)\{\mathrm{}\}`$ consists of finitely many pairwise disjoint simple closed curves on the Riemann sphere one of which passes through infinity. By transversality, $`\mathrm{\Lambda }(t)`$ changes continuously with $`t`$ near $`0`$ and contains $`\mathrm{}`$ only if $`t=0`$. As $`t0`$ the open sets $`Y(t)`$ and their oriented boundaries converge to a domain $`Y^{}`$ and its oriented boundary $`bY^{}`$ which, as a set, coincides with $`\mathrm{\Lambda }(0)`$. As $`t0`$ the open sets $`Y(t)`$ and their oriented boundaries $`\mathrm{\Lambda }(t)`$ converge to a domain $`Y^+`$ and its oriented boundary $`bY^+`$ which, as a set, coincides with $`\mathrm{\Lambda }(0)`$. We have $`Y^{}\mathrm{\Lambda }(0)Y^+=\mathrm{C}\text{ }`$. Since the function $`F`$ is bounded and continuous on $`M`$ it follows that as $`t0`$, the holomorphic extensions of $`F|ML(t+i\lambda (t))`$ converge to the holomorphic extension of $`F|ML(0)`$, a bounded continuous function on $`\{0\}\times \overline{Y^{}}`$, holomorphic on $`\{0\}\times Y^{}`$. In particular, $`F|ML(0)`$ extends to a bounded continuous function on $`\{0\}\times \overline{Y^{}}`$ which is holomorphic on $`\{0\}\times Y^{}`$. Similarly, $`F|ML(0)`$ extends to a bounded continuous function on $`\{0\}\times \overline{Y^+}`$ which is holomorphic on $`\{0\}\times \overline{Y^+}`$. Consequently $`F|ML(0)`$ extends to a bounded continuous function on $`\{0\}\times \mathrm{C}\text{ }`$ which is holomorphic on $`\{0\}\times [Y^+Y^{}]`$. This function extends holomorphically across $`\{0\}\times \mathrm{\Lambda }(0)`$ to a bounded entire function on $`L(0)`$, which, by the Liouville theorem, must be constant. This implies that $`F|ML(0)`$ is a constant. In the same way we show that $`F|ML(is)`$ is constant for each $`s,\eta <s<\eta `$. It follows that there is an $`r>0`$ such that on $`r\mathrm{\Delta }`$ the holomorphic extensions of all functions $`f|(it+bD)`$ to $`it+D`$, for all $`t`$ such that $`r\mathrm{\Delta }it+D`$, coincide, and that on $`r\mathrm{\Delta }`$ the holomorphic extensions of all functions $`f|(it+bD^{})`$ to $`it+D^{}`$, for all $`t`$ such that $`r\mathrm{\Delta }it+D^{}`$, coincide. This implies first that $`f`$ is holomorphic on $`\{it+bD:r\mathrm{\Delta }it+D,a<t<b\}\text{Int}S`$, and then, by translating $`bD`$ further along $`i\mathrm{IR}`$, that $`f`$ is holomorphic on $`\{it+bD:a<t<b\}\text{Int}S=\text{Int}Q_1`$. Similarly, we show that $`f`$ is holomorphic on $`\text{Int}Q_2`$. The proof is complete.
10. Remarks It remains an open problem to prove Theorem 2.1 without the assumption that $`\mathrm{\Omega }`$ is symmetric with respect to the real axis. It is known that one cannot drop the assumption that $`(\overline{\mathrm{\Omega }}ir)(\overline{\mathrm{\Omega }}+ir)=\mathrm{}`$. In fact the function
$$f(z)=\{\begin{array}{cc}& z^2/\overline{z}(z0)\hfill \\ & 0(z=0)\hfill \end{array}$$
is continuous on $`\mathrm{C}\text{ }`$ and extends holomorphically from each circle which either surrounds the origin or contains the origin, yet $`f`$ is not holomorphic.
Acknowledgements The author is indebted to Mark Agranovsky for several very stimulating discussions. After these discussions the author found a way to pass from circles to general curves.
This work was supported in part by the Ministry of Higher Education, Science and Technology of Slovenia through the research program Analysis and Geometry, Contract No. P1-0291
References
\[AG\] M. L. Agranovsky and J. Globevnik: Analyticity on circles for rational and real-analytic functions of two real variables.
J. d’Analyse Math. 91 (2003) 31-65 \[E\] L. Ehrenpreis: Three problems at Mount Holyoke.
Contemp. Math. 278 (2001) 123-130. \[G1\] J. Globevnik: Analyticity on rotation invariant families of curves.
Trans. Amer. Math. Soc. 280 (1983) 247-257 \[G2\] J. Globevnik: Analyticity on familes of curves.
Talk at Bar Ilan University, November 1987 \[G3\] J. Globevnik: Holomorphic extensions from open families of circles.
Trans. Amer. Math. Soc. 355 (2003) 1921-1931 \[L\] H. Lewy: On the local character of the solutions of an atypical linear differential equation in three variables and a related theorem for regular functions of two complex variables.
Ann. of Math. 64 (1956) 514-522 \[R\] H. Rossi: A generalization of a theorem of Hans Lewy.
Proc. Amer. Math. Soc. 19 (1968) 436-440 \[T1\] A. Tumanov: A Morera type theorem in the strip.
Math. Res. Lett. 11 (2004) 23-29 \[T2\] A. Tumanov: Testing analyticity on circles.
Preprint \[ArXiv:math.CV/0502139\]
Institute of Mathematics, Physics and Mechanics
University of Ljubljana, Ljubljana, Slovenia
e-mail: josip.globevnik@fmf.uni-lj.si
|
warning/0506/quant-ph0506179.html
|
ar5iv
|
text
|
# Untitled Document
Velocity quantization approach of the one-dimensional
dissipative harmonic oscillator
G. López and P. López
Departamento de Física de la Universidad de Guadalajara
Apartado Postal 4-137
44410 Guadalajara, Jalisco, México
PACS: 03.20.+i, 03.30.+p, 03.65.-w,03.65.Ca
June, 2005
ABSTRACT
Given a constant of motion for the one-dimensional harmonic oscillator with linear dissipation in the velocity, the problem to get the Hamiltonian for this system is pointed out, and the quantization up to second order in the perturbation approach is used to determine the modification on the eigenvalues when dissipation is taken into consideration. This quantization is realized using the constant of motion instead of the Hamiltonian.
1. Introduction
To find the Lagrangian and the Hamiltonian from the equations of motion of a given system (so called ”inverse problem of the calculus of variations” or ”the inverse problem of the mechanics”) is not so trivial as one could think at first sight, even for one-dimensional systems and despite their existence is guaranteed here. If the system is autonomous (the forces do not depend explicitly on time) or nonautonomous (total force depend explicitly on time), it is natural to try to find Lagrangians and Hamiltonians which do no depend explicitly on time or which depend explicitly on time for the latter. One possible approach to find these quantities for a one-dimensional system is to find first a constant of motion of the system, and then to obtain the Lagrangian and Hamiltonian . This constant of motion, of course, is chosen to be independent explicitly on time for autonomous systems, and explicitly depending on time, for nonautonomous systems. It has been shown that when one tries to get quantities (constant of motion, Lagrangian or Hamiltonian) which are explicitly depending on time for autonomous systems, there may be a concern about ambiguities or about consistence on the mathematical quantities . However, one common procedure that has been done is to guess an explicitly time depending Hamiltonian for the harmonic oscillator with exponential time dependence of its mass which brings about the one-dimensional harmonic oscillator with linear dissipation in the velocity. Indeed, there has been a lot of studies on the quantization of this Hamiltonian and decoherence of this quantum system . However, since the one-dimensional harmonic oscillator with linear dissipation on the velocity is an autonomous system, one should get time independent dynamical quantities (constant of motion, Lagrangian or Hamiltonian) to describe it. After this, one should proceed to make the quantization of this system in a consistent way. In fact, an explicitly time independent constant of motion for this system has been found already . So, in this paper we will study the classical damping behavior of the system with this constant of motion, and we point out the difficulties to get its associated Hamiltonian. Finally, we study its quantization using the same constant of motion and the velocity operator. This last study is done up to second order in perturbation theory and for weak dissipation.
2. Constant of motion and Lagrangian
The one-dimensional harmonic oscillator with linear dissipation on the velocity is described by the equation
$$m\ddot{x}+\alpha x+kx=0,$$
$`\left(1\right)`$
where $`m`$ is the mass of the particle, $`\alpha `$ is the dissipative constant, $`k`$ is the spring constant, $`x`$ is the particle position, and $`\dot{x}`$ and $`\ddot{x}`$ are its first and second differentiation with respect the time. By defining the new variable $`\dot{x}=v`$, Eq. (1) can be written as the following autonomous dynamical system
$$\dot{x}=v,\text{and}\text{ }\dot{v}=\omega ^2x2\omega _\alpha v,$$
$`\left(2\right)`$
where $`\omega =\sqrt{k/m}`$ is the natural angular frequency of the harmonic oscillator without dissipation, and $`\omega _\alpha =\alpha /2m`$ is the new dissipation parameter. A constant of motion for the system (2) is a function $`K=K(x,v)`$ which satisfies the following partial differential equation
$$v\frac{K}{x}\left(\omega ^2x+2\omega _\alpha v\right)\frac{K}{v}=0.$$
$`\left(3\right)`$
The solution, $`K_\alpha `$, of this equation which satisfies the following limit $`lim_{\alpha 0}K_\alpha =mv^2/2+m\omega _\alpha x^2/2`$ is given by
$$K_\alpha (x,v)=\frac{m}{2}\left(v^2+2\omega _\alpha xv+\omega ^2x^2\right)e^{2\omega _\alpha G(v/x,\omega ,\omega _\alpha )},$$
$`\left(4a\right)`$
where the function $`G`$ has been defined as
$$G(v/x,\omega ,\omega _\alpha )=\{\begin{array}{cc}\frac{1}{2\sqrt{\omega _\alpha ^2\omega ^2}}\mathrm{ln}\left[\frac{\omega _\alpha +v/x\sqrt{\omega _\alpha ^2\omega ^2}}{\omega _\alpha +v/x+\sqrt{\omega _\alpha ^2\omega ^2}}\right]\hfill & \text{if }\omega ^2<\omega _\alpha ^2\hfill \\ & \\ \frac{1}{\omega _\alpha +v/x}\hfill & \text{if }\omega ^2=\omega _\alpha ^2\hfill \\ & \\ \frac{1}{\sqrt{\omega ^2\omega _\alpha ^2}}\mathrm{arctan}\left(\frac{\omega _\alpha +v/x}{\sqrt{\omega ^2\omega _\alpha ^2}}\right)\hfill & \text{if }\omega ^2>\omega _\alpha ^2\hfill \end{array}$$
$`\left(4b\right)`$
For weak dissipation, one has the following expression for the constant of motion
$$K=\frac{1}{2}mv^2+\frac{1}{2}m\omega ^2x^2+\frac{m\omega _\alpha }{\omega }\left[xv\omega \left(v^2+\omega ^2x^2\right)\mathrm{arctan}\left(\frac{v}{wx}\right)\right].$$
$`\left(5\right)`$
The Lagrangian for the system (1) can now be constructed from the known expression
$$L(x,v)=v^v\frac{K(x,\xi )d\xi }{\xi ^2}.$$
$`\left(6\right)`$
For the general case (4), it is no possible to get a close expression for the Lagrangian, but for the weak dissipation case (5), one gets
$$L=\frac{1}{2}mv^2\frac{1}{2}m\omega ^2x^2+\frac{m\omega _\alpha }{\omega }\left[\left(\omega ^2x^2v^2\right)\mathrm{arctan}\left(\frac{v}{\omega x}\right)+\omega xv\mathrm{ln}\left(\frac{\omega ^2x^2+v^2}{\omega ^2x^2}\right)\right].$$
$`\left(7\right)`$
The generalized linear momentum ($`p=L/v`$) is
$$p=mv+\frac{m\omega _\alpha }{\omega }\left[\omega x+x\omega \mathrm{ln}\left(\frac{\omega ^2x^2+v^2}{\omega ^2x^2}\right)2v\mathrm{arctan}\left(\frac{v}{\omega x}\right)\right].$$
$`\left(8\right)`$
To get the Hamiltonian, it is necessary from (8) to express the variable $`v`$ as a function of the variables $`x`$ and $`p`$, $`v=v(x,p)`$. In this way, one makes the substitution of this variable on the Legendre transformation, $`H(x,p)=v(x,p)pL(x,v(x,p))`$, or in the constant of motion, $`H(x,p)=K(x,v(x,p))`$. However, one notices immediately form (8) that it is not possible to do this. Thus, the Hamiltonian can not be given explicitly but implicitly through the constant of motion (5).
A trajectory in the phase space ($`x,v`$) can be seen on Fig. 1, where the constant of motion (5) has been used. Of course, in order to keep the continuity at ($`v=0,x<0`$), the arctan function changes its value due to the multivalue functions. This means that our constant of motion is really a local constant of motion (which is valid on the half plane $`v<0`$ or $`v>0`$), and it changes its value every time the trajectory crosses the line $`v=0`$. This number of crossing is a numerable set. Therefore, this set has measure zero . In this way, one can say that (4) or (5) represents a constant of motion almost everywhere in the phase space.
Let us now change the variables ($`x,v`$) by a new variables ($`\varphi ,J`$) defined as
$$\varphi =\mathrm{arctan}\left(\frac{v}{\omega x}\right),\text{and}\text{ }J=\frac{m}{2\omega }\left(v^2+\omega ^2x^2\right),$$
$`\left(9a\right)`$
where the inverse transformation is given by
$$x=\sqrt{\frac{2J}{m\omega }}\mathrm{cos}\varphi ,\text{and}\text{ }v=\sqrt{\frac{2\omega J}{m}}\mathrm{sin}\varphi .$$
$`\left(9b\right)`$
The dynamical system (2) is then transformed to the system
$$\dot{\varphi }=\left(\omega +\frac{2\omega _\alpha \mathrm{tan}\varphi }{1+\mathrm{tan}^2\varphi }\right)$$
$`\left(10a\right)`$
and
$$\dot{J}=4\omega _\alpha J\mathrm{sin}^2\varphi .$$
$`\left(10b\right)`$
These equations are readily solved, and their solutions are
$$\varphi \left(t\right)=\mathrm{arctan}\left[\mathrm{tan}\left(\omega t+a\right)\frac{\omega _\alpha }{\omega }\right]$$
$`\left(11a\right)`$
and
$$J\left(t\right)=J_oe^{\omega _\alpha f\left(t\right)},$$
$`\left(11b\right)`$
where $`a`$ and $`J_o`$ are constants determinate by the initial conditions, and $`f\left(t\right)`$ si defined as
$$f\left(t\right)=\frac{4\left(\mathrm{tan}\omega t\omega _\alpha /\omega \right)^2t}{1+\left(\mathrm{tan}\omega t\omega _\alpha /\omega \right)^2}.$$
$`\left(11c\right)`$
To know the trajectory in the new phase space ($`\varphi ,J`$), one can express $`J`$ as a function of $`\varphi `$ through the integration of $`\dot{J}=\left(dJ/d\varphi \right)\dot{\varphi }`$. This brings about the following expression
$$J\left(\varphi \right)=\frac{\stackrel{~}{J}_o}{\omega +\omega _\alpha \mathrm{sin}\varphi }e^{\frac{2\omega _\alpha }{\omega }\mathrm{arctan}\left(\mathrm{tan}\varphi +\omega _\alpha /\omega \right)},$$
$`\left(12\right)`$
where $`\stackrel{~}{J}_o`$ is another constant. Fig. 2 shows Trajectories in this phase space for $`\omega _\alpha =0`$ and for $`\omega _\alpha 0`$, where the expected jumps at $`\pi /2`$ and $`3\pi /2`$ are clearly seen at the scale shown.
To finish the classical analysis, let us write the constant of motion, the Lagrangian and the generalized linear momentum for the weak dissipative case in terms of the variables $`\varphi `$ and $`J`$. These are given by
$$K=\omega J+\omega _\alpha J\left[\mathrm{sin}\left(2\varphi \right)2\varphi \right],$$
$`\left(13a\right)`$
$$L=\omega J\mathrm{cos}\left(2\varphi \right)+2\omega _\alpha J\left[\mathrm{cos}\left(2\varphi \right)4\mathrm{sin}\left(2\varphi \right)\mathrm{ln}\left(\mathrm{cos}\varphi \right)\right]$$
$`\left(13b\right)`$
and
$$p=\sqrt{2m\omega J}\mathrm{sin}\varphi +\omega _\alpha \sqrt{\frac{2mJ}{\omega }}\left[\mathrm{cos}\varphi \left(12\mathrm{ln}\left(\mathrm{cos}\varphi \right)\right)2\varphi \mathrm{sin}\varphi \right]$$
$`\left(13c\right)`$
Once again, one see the impossibility to get the Hamiltonian $`H(\varphi ,J)`$ due to complexity of the expression (13c).
3. Quantization of the constant of motion
Due to the impossibility of getting the Hamiltonian explicitly for the autonomous dynamical system (2), one may propose to extend the Shrödinger quantization to a dissipative system through the quantization of the constant of motion associated to it. This can be made by associating an Hermitian operator to the velocity as
$$\widehat{v}=\frac{i\mathrm{}}{m}\frac{}{x}.$$
$`\left(14\right)`$
In this way, if $`\widehat{K}`$ is the Hermitian operator associated to the constant of motion (which must have units of energy), the associated Shrödinger equation of the classical autonomous system would be
$$i\mathrm{}\frac{\mathrm{\Psi }}{t}=\widehat{K}(x,\widehat{v})\mathrm{\Psi },$$
$`\left(15\right)`$
where $`\mathrm{\Psi }=\mathrm{\Psi }(x,t)`$. Of course, the whole quantum mechanics structure is exactly the same but with the velocity operator instead of the linear momentum operator as the main operator of the quantum system. Since (15) represents an stationary problem, one just has to solve the eigenvalue problem
$$\widehat{K}(x,\widehat{v})\varphi =E\varphi ,$$
$`\left(16\right)`$
where $`\varphi =\varphi \left(x\right)`$ and one has used $`\mathrm{\Psi }(x,t)=\varphi \left(x\right)\mathrm{exp}\left(iEt/\mathrm{}\right)`$ in (15). If $`\widehat{K}`$ can be written as $`\widehat{K}=\widehat{K}_o+\widehat{K}_I`$, where the solution of the problem $`\widehat{K}_o|n=E_n^{\left(0\right)}|n`$ is known, the eigenvalues of (16) are given at first order in perturbation theory by
$$E_n=E_n^{\left(0\right)}+n\left|\widehat{K}_I\right|n,$$
$`\left(17\right)`$
where one has used Dirac notation . Our constant of motion (5) or (13a) can be expressed of the form $`K=K_o+K_I`$, where $`K_o=\omega J`$ and$`K_I=\omega _\alpha J\left(\mathrm{sin}\left(2\varphi \right)2\varphi \right)`$. It is well known that the eigenvalues of the harmonic oscillator without dissipation are given by $`E_n^{\left(0\right)}=\mathrm{}\omega \left(n+1/2\right)`$. Thus, $`\widehat{K}_o`$ is diagonal in the basis $`\left\{|n\right\}`$ and can be expressed in terms of ascent , $`a^+`$, and descent, $`a`$, operators as
$$\widehat{K}_o=\mathrm{}\omega \left(a^+a+1/2\right),$$
$`\left(18\right)`$
with $`a`$ and $`a^+`$ defined in terms of $`\widehat{x}`$ and $`\widehat{v}`$ as
$$a=\sqrt{\frac{m}{2\mathrm{}\omega }}\left(\omega \widehat{x}+i\widehat{v}\right),\text{and}\text{ }a^+=\sqrt{\frac{m}{2\mathrm{}\omega }}\left(\omega \widehat{x}i\widehat{v}\right).$$
$`\left(19\right)`$
These operator have the following commutation relations
$$[a,a^+]=1,[a,a]=[a^+,a^+]=0,[a,a^+a]=a,[a^+,a^+a]=a^+.$$
Additionally, $`\widehat{N}=a^+a`$ is called the number operator and is diagonal in the basis $`\left\{|n\right\}`$, $`\widehat{N}|n=n|n`$, This, in turns, implies that the operator associated to the variable $`J`$ is given by
$$\widehat{J}=\mathrm{}\left(\widehat{N}+1/2\right).$$
$`\left(20\right)`$
Thus, our main problem is to assign a Hermitian operator to the function$`K_I=2\omega _\alpha J\left(\mathrm{cos}\varphi \mathrm{sin}\varphi \varphi \right)`$. According to refence , one has the following assignments
$$\mathrm{sin}\varphi \widehat{S},\mathrm{cos}\varphi \widehat{C},\varphi \widehat{\varphi },$$
$`\left(21\right)`$
where $`\widehat{S}`$ and $`\widehat{C}`$ are Hermitian operators such that
$$\ddot{\widehat{S}}+\omega ^2\widehat{S}=\ddot{\widehat{C}}+\omega ^2\widehat{C}=0,[\widehat{C},\widehat{N}]=i\widehat{S},[\widehat{S},\widehat{N}]=i\widehat{C}$$
$`\left(22a\right)`$
and
$$\widehat{S}|n=\frac{i}{2}\left(|n+1|n1\right),\widehat{C}|n=\frac{1}{2}\left(|n+1+|n1\right),[\widehat{C},\widehat{S}]=\frac{\pi _o}{2i},$$
$`\left(22b\right)`$
with $`\pi _o`$ is the projector on the ground state of the quantum harmonic oscillator, $`\pi _o=|00|`$. Therefore, $`\widehat{C}`$ and $`\widehat{S}`$ commute for any exited state. The operators $`\widehat{\varphi }`$ is defined as
$$\widehat{\varphi }=\frac{\pi }{2}\underset{k=0}{\overset{\mathrm{}}{}}\frac{\left(1\right)^k}{2k+1}\left(\begin{array}{c}1/2\\ k\end{array}\right)\widehat{C}^{2k+1}$$
$`\left(23\right)`$
where $`\left(\begin{array}{c}a\\ b\end{array}\right)`$ si the combinatorial factor. In this way, the operator associated to the function $`K_I`$ can be written as
$$\widehat{K}_I=\frac{\omega _\alpha }{3}\left[\widehat{J}\widehat{C}\widehat{S}+\widehat{C}\widehat{J}\widehat{S}+\widehat{C}\widehat{S}\widehat{J}+\widehat{J}\widehat{S}\widehat{C}+\widehat{S}\widehat{J}\widehat{C}+\widehat{S}\widehat{C}\widehat{J}\right]\omega _\alpha \left[\widehat{J}\widehat{\varphi }+\widehat{\varphi }\widehat{J}\right].$$
$`\left(24\right)`$
The first term has not contribution at first order in perturbation theory because of (22b), the action of the operators $`\widehat{C}`$ or $`\widehat{S}`$ increases or decreases the state number such that the expected value $`n|..|n`$ is always zero. Due to the same reason, the contribution of the second term of Eq. (24) will come only from the term $`\pi /2`$ of (23). Thus, at first order in perturbation theory, one has the following correction of the eigenvalues
$$\delta E_n^{\left(1\right)}=\mathrm{}\omega _\alpha \pi \left(n+1/2\right).$$
$`\left(25a\right)`$
Thus, one gets
$$E_nE_n^{\left(0\right)}+\delta E_n^{\left(1\right)}=\mathrm{}\omega \left(n+1/2\right)\left[1\frac{\omega _\alpha \pi }{\omega }\right].$$
$`\left(26a\right)`$
That is, there is a small shift on the frequency of oscillation given by
$$\omega ^{}=\omega \pi \omega _\alpha .$$
$`\left(26b\right)`$
At second order in perturbation theory,$`E_n=E_n^{\left(0\right)}+\delta E_n^{\left(1\right)}+\delta E_n^{\left(2\right)}`$, the correction on the energy will come from the expression
$$\delta E_n^{\left(2\right)}=\underset{kn}{}\frac{\left|k\left|\widehat{K}_I\right|n\right|^2}{E_n^{\left(0\right)}E_k^{\left(0\right)}}.$$
$`\left(27\right)`$
Using (22b), (23) and (24), one has
$`k\left|\widehat{K}_I\right|n`$ $`=`$ $`{\displaystyle \frac{\mathrm{}\omega _\alpha }{12}}\left[\left(2k+4n+5\right)\delta _{k,n+2}\left(2k+4n+1\right)\delta _{k,n2}\right]`$
$`+\mathrm{}\omega _\alpha \left(n+k+1\right){\displaystyle \underset{l=0}{\overset{\mathrm{}}{}}}{\displaystyle \underset{s=0}{\overset{2l+1}{}}}{\displaystyle \frac{\left(1\right)^l}{2l+1}}\left(\begin{array}{c}1/2\\ l\end{array}\right)\left(\begin{array}{c}2l+1\\ s\end{array}\right)\delta _{k,n2l1s}`$
$`\left(28\right)`$
Therefore, the correction at second order can be written as
$$\delta E_n^{\left(2\right)}=\frac{\mathrm{}\omega _\alpha ^2}{\omega }\left[\left(\frac{2}{3}n+\frac{1}{4}\right)\underset{l=0}{\overset{\mathrm{}}{}}\underset{s=0}{\overset{2l+1}{}}\left(\begin{array}{c}1/2\\ l\end{array}\right)^2\left(\begin{array}{c}2l+1\\ s\end{array}\right)^2\frac{\left(2n2ls\right)^2}{\left(2l+1\right)^2\left(2l+1+s\right)}\right]$$
$`\left(28\right)`$
Conclusions
We have used the constant of motion for the one-dimensional dissipative harmonic oscillator to study the classical trajectories in the phase space and to point out the difficulty to get its Hamiltonian explicitly. Due to this problem, we have proposed the quantization of the constant of motion directly , via the association of the velocity operator. Then, the eigenvalues where calculated up to second order within perturbation theory to see the first effect of the dissipation on them. This effect at first order corresponds to have a shift by the quantity $`\pi \omega _\alpha `$ on the frequency of oscillation of the nondissipative harmonic oscillator.
Figure captions
Fig. 1 Trajectory on the phase space ($`x,v`$) as determinate by (4), where $`m=1Kgr`$, $`\omega =1sec^1`$ and $`\omega _\alpha =0.001sec^1`$.
Fig. 2 Trajectory on the phase space ($`\varphi ,J`$) as determinate by (12). The straight horizontal line corresponds to $`\omega _\alpha =0`$, solid line corresponds to $`\omega _\alpha =0.001sec^1`$, $`\omega =1sec^1`$ and $`\stackrel{~}{J}_o=1`$ Joules-sec.
References
1. D. Darboux,Leçons sur la théorie général des surfaces et les applications
géométriques du calcul infinitésimal, IViéme partie,
Gauthoer-Villars, Paris, 1984.
2. J.A. Kobussen, Act. Phys. Austr., 51 (1979) 193.
C. Leubner, Phys. Rev. A 86 (1987) 9.
C.C. Yan, Amer. J. Phys., 49 (1981) 296.
G. López, Ann. Phys., 251, 2(1996) 363.
3. G. López, Int. Jour. Theo. Phys., 37,5 (1998) 1617.
4. G. López and J.I. Hernández, Ann. of Phys., 193, 1 (1989) 1.
5. P. Caldirola, Nuovo Cimento, 18 (1941) 393.
E. Kanai, Prog. Theo. Phys., 3 (1948) 440.
K.A. Yean and C.I. Um, Phys. Rev. A, 36 (1987) 5287.
6. V.V. Dodonov and M.S. Man’ko, Phys. Rev. D, 20 (1979) 550.
R.K. Colegrave and M.S. Abdalla, J. Phys. A, 14 (1981) 2269.
P.G.L. Leach, J. Phys. A, 16 (1983) 3261.
H. Dekker, Phys. Rev. A, 16, 5 (1977) 2126.
H. Dekker, Phys. Rep. 80,1 (1981) 1-112.
7. M. Resenau da Costa, A.O. Caldeira, S.M. Deutra and H. Westfohl J.
Phys. Rev. A, 61 (2000) 022107.
8. G. López, Ann. of Phys., 251,2 (1996) 372.
9. F. John, Partial Differential Equations, Springer-Verlag, N.Y. 1974.
G. López, Partial Differential Equations of First Order and Their
Applications to Physics, World Scientific, 1999.
10. E. Hewitt and K. Stromberg,Real and Abstrac Analysis,
Springer-Verlag N.Y., chapter III, 1965.
11. A. Messiah, Quantum Mechanics, vol.I, John Wiley and Sons, 1958.
12. P.A.M. Dirac, The Principles of Quantum Mechanics, IV edition,
Oxford Science Publications, 1992.
|
warning/0506/quant-ph0506214.html
|
ar5iv
|
text
|
# Adventures of the Coupled Yang-Mills Oscillators: I. Semiclassical Expansion
## 1 Introduction
The discovery of the chaoticity of the classical Yang-Mills (YM) equations (see for a review) has attracted broad interest to the system of two (three) coupled quartic oscillators with the potential $`x^2y^2`$ ($`x^2y^2+y^2z^2+z^2x^2`$), where $`x,y(z)`$ are functions of time $`t`$. These systems are the simplest limiting cases for the homogeneous YM equations (sometimes called YM classical mechanics) with $`n=2`$ ($`n=3`$) degrees of freedom, respectively.The $`x^2y^2`$ model, the central object of the present paper, exhibits a rich chaotic behavior despite its extreme simplicity. Not surprisingly, this model has been encountered in various fields of science, including chemistry, astronomy, astrophysics, and cosmology (chaotic inflation).
Quantum mechanically, this model (YM quantum mechanics, YMQM), despite possessing a logarithmically divergent volume of energetically accessible phase space (we set $`m=1`$ throughout) <sup>1</sup><sup>1</sup>1This is in violation of Weil’s famous theorem, which states that the average number $`N(E)`$ of energy levels with energy less than $`E`$ is asymptotically proportional to $`_0^E\mathrm{\Gamma }_E^{}𝑑E^{}`$.
$$\mathrm{\Gamma }_E=_{\mathrm{}}^{\mathrm{}}𝑑x𝑑y𝑑p_x𝑑p_y\delta \left(\frac{1}{2}(p_x^2+p_y^2)+\frac{g^2}{2}x^2y^2E\right),$$
(1)
can be shown to have a discrete spectrum . Physically, it is clear why this is so: Quantum fluctuations, e.g. zero-point fluctuations, forbid the “particle” to escape along the $`x`$ or $`y`$ axis where the potential energy vanishes. The system is thus confined to a finite volume, and this implies the discreteness of the energy levels. Classically, of course, the particle can always escape along one of the axes without increasing its energy.
In this article we calculate the partition function (heat kernel) $`Z(t)`$ for the YMQM with the above given potentials beyond the well-known Thomas-Fermi (TF) approximation, which takes into account only the discreteness of the quantum mechanical phase space, but treats the Hamiltonian classically. There are several interesting articles devoted to the approximate calculation of $`Z(t)`$ in the TF approximation for these potentials. They are based on the adiabatic separation of the dependence of $`Z`$ on $`x`$ and $`y`$ in the narrow channels of the equipotential surface $`xy=\mathrm{const}.`$. The range of the integration over the coordinates $`x,y`$ and the momenta $`p_x,p_y`$ was divided into two regions: the central region ($`|x|,|y|Q`$) and the channels ($`Q|x|,Q|y|`$), which are governed by quite different physics: The system is essentially classical in the central region and intrinsically quantum mechanical in the channels performing oscillatory motion with $`x`$-dependent frequency in the “fast” variable $`y`$ (in the channels along the $`x`$-axis), but quasi-free motion in the “slow” variable $`x`$.
The dependence on the artificial boundary Q dividing the central region from the channels disappears in the final answer for $`Z(t)`$. Below we show that this property survives beyond the TF approximation up to the order $`\mathrm{}^8`$. In the paper an alternative to this approach of calculating $`Z(t)`$ was proposed, starting from the Yang-Mills-Higgs quantum mechanics (YMHQM) and then passing to the limit $`v=0`$, where $`v`$ is the vacuum expectation value of the “Higgs field” defining the strength of the harmonic potential.
Here we study the $`x^2y^2`$ potential and postpone the investigation of YMHQM to a separate publication . We here follow the method of separation of the domain of motion into two regions, the central square $`x,y[Q,Q]`$ and the hyperbolic channels, introduced by Tomsovic and Whelan . The motion in the channels can be treated by adiabatic separation of the motion in the slow variable (in the direction of the channel) and the motion in the fast variable (perpendicular to the channel). We will go beyond the Thomas-Fermi approximation used in by applying the well-known Wigner-Kirkwood (WK) method (see for a review of the WK approach).
We already mentioned that the motion in the central region, where the variables $`x`$ and $`y`$ are treated on an equal footing, is quite different from the motion in the channels, where the motion in the perpendicular direction (for now taken as the variable $`y`$) performs quantum oscillations with $`x`$-dependent frequency, but the motion in the $`x`$ variable is not treated as free as it was done in . We apply the WK method in the central region to both variables, but only to the slow variable $`x`$ in the channels treating the motion in the fast $`y`$ variable fully quantum mechanically. We improve the method of adiabatic separation introduced in by taking into account the effect of the $`x`$-dependence of the oscillation frequency onto the motion in the $`x`$-direction. This is especially important, as we shall see, for the higher-order quantum corrections. We show that, up to eighth-order in $`\mathrm{}`$ and in the leading terms in $`(tQ^4)^11`$, the calculated partition function $`Z(t)`$ does not depend on the boundary $`Q`$ dividing the two regions and the quantum corrections from both regions cancel.
An interesting phenomenon occurs in the channel motion due to the presence of terms suppressed by powers of $`(tQ^4)^1`$, which are independent of $`Q`$. Each term of this kind is negligible in comparison with the leading (but canceling) terms, but their series forms an asymptotic series in the variable $`g^2\mathrm{}^4t^3`$, which does not involve $`Q`$. If we postulate that the entire $`Q`$-dependence of the partition function disappears, which we prove for the leading terms in $`(tQ^4)^1`$, only this series will be left as the contribution from the channels. We may express this phenomenon as the “transmutation” of the small expansion parameter $`(tQ^4)^1`$ governing the approach of adiabatic separation of variables into the small parameter $`g^2\mathrm{}^4t^3`$ characteristic of quantum mechanics. The quasiclassical expansion is shown to have the nature of an asymptotic series.
In the next two sections we present the YMQM system and the WK method of calculating $`Z(t)`$ beyond the TF approximation.
## 2 Yang-Mills-Higgs classical and quantum mechanics: The Thomas-Fermi approximation
For spatially homogeneous fields (long wave length limit of the Yang-Mills field) the classical Hamiltonian for $`n=2`$ is given by the expression
$$H=\frac{1}{2}(p_x^2+p_y^2)+\frac{g^2}{2}x^2y^2.$$
(2)
The quantized counterpart of (2) is
$$\widehat{H}=\frac{\mathrm{}^2}{2}_{x,y}^2+\frac{g^2}{2}x^2y^2.$$
(3)
A brief word on units: All quantities are given below in units of the energy $`E`$ with dimensions $`[H]=1,[t]=1,[x],[y]=1/4,[g]=0,[\mathrm{}]=3/4`$. The operator (3) has a discrete spectrum . The TF approximation to the heat kernel or partition function $`Z(t)=\mathrm{Tr}[\mathrm{exp}(t\widehat{H})]`$ is the standard lowest-order semiclassical approximation valid for small $`\mathrm{}t^{3/4}1`$. It is obtained by substituting the classical Hamiltonian for its quantum counterpart and replacing the trace of the heat kernel by the integral over the phase-space volume normalized by ($`2\pi \mathrm{})^n`$, where $`2n`$ is the phase-space dimension. In other words, the TF approximation takes into account only the discreteness of the quantum mechanical phase space, but considers momenta and coordinates (in our case, the field amplitudes $`x`$ and $`y`$) as commuting variables. This method was used in numerous papers (see e.g. ). For the calculation of the energy level density $`\rho (E)=dN(E)/dE`$ at asymptotic energies, the TF approximation is a consistent approach since, as we shall see below, all corrections to the TF term are structures with factors $`\mathrm{}^kt^{\mathrm{}}`$ with $`k,\mathrm{}`$ positive integers. For the asymptotic energy level density $`\rho (E)`$ or $`N(E)`$ these corrections are negligible according the Karamata-Tauberian theorems relating the most singular part of $`Z(t)`$ to the asymptotic level density: $`N(E)=𝑑E\rho (E)=L^1(Z(t)/t)`$ where $`L^1`$ denotes the inverse Laplace transform.
For the Hamiltonian (3) the naive TF approximation is divergent, because the classical phase space is infinite. As explaines in the Introduction, finite results for $`Z(t)`$ and $`N(E)`$ can be obtained by including certain quantum corrections to the channel motion by means of the method of adiabatic separation of the motion along and perpendicular to the channels . We here give the expression for the partition function obtained in this improved TF approximation in our notation:
$$Z_0=\frac{1}{\sqrt{2\pi }g\mathrm{}^2t^{3/2}}\left(\mathrm{ln}\frac{1}{g^2\mathrm{}^4t^3}+9\mathrm{ln}2+C\right),$$
(4)
where $`C`$ is the Euler constant. We shall return to (4) again below. Because we shall often encounter the pre-factor appearing in (4), we introduce the special symbol for it:
$$K(2\pi g^2\mathrm{}^4t^3)^{1/2}.$$
(5)
## 3 Beyond the TF approximation: The Wigner-Kirkwood expansion
For non asymptotic energies and also for the fluctuating part of the level density one needs to go beyond the TF approximation and calculate the quantum corrections to the TF term in $`Z(t)`$. We remark that this problem is interesting not only from this practical point of view, but also because it provides insight into some problems of perturbation theory in quantum mechanics. Here we use the Wigner-Kirkwood (WK) method. Before we proceed, we emphasize the dual role of the variables $`x(t),y(t),z(t)`$. They formally play the role of the coordinates, but at the same time they stand for the homogeneous gauge field amplitudes. The trace in the heat kernel may be taken with respect to any complete set of states. For the semiclassical expansion involving integration over the phase space for the quantum operators in the Wigner representation it is convenient to use the plane waves as a complete set:
$$Z(t)=\frac{1}{(2\pi \mathrm{})^n}\underset{i=1}{\overset{n}{}}dx_idp_ie^{i\stackrel{}{p}\stackrel{}{r}/\mathrm{}}e^{t\widehat{H}}e^{i\stackrel{}{p}\stackrel{}{r}/\mathrm{}}$$
(6)
with $`\stackrel{}{r}=(x_1,x_2,\mathrm{},x_n),\stackrel{}{p}=(p_1,p_2,\mathrm{},p_n)`$. The kinetic energy term in the exponent of $`\mathrm{exp}(t\widehat{H})`$ does not commute with the potential energy $`V(\stackrel{}{x})`$; the WK expansion provides a convenient method of calculating the noncommuting terms. Following we set
$$e^{t\widehat{H}}e^{i\stackrel{}{p}\stackrel{}{r}/\mathrm{}}=e^{tH}e^{i\stackrel{}{p}\stackrel{}{r}/\mathrm{}}W(\stackrel{}{r},\stackrel{}{p};t)=u(\stackrel{}{r},\stackrel{}{p};t)$$
(7)
where the function $`W(\stackrel{}{r},\stackrel{}{p};t)`$ is to be determined. $`H(\stackrel{}{p},\stackrel{}{r})`$ (as opposed to $`\widehat{H}`$) is the classical Hamiltonian (2). The function $`u(\stackrel{}{r},\stackrel{}{p};t)`$ satisfies the Bloch equation:
$$\frac{u}{t}+\widehat{H}u=0$$
(8)
with the boundary condition
$$\underset{t0}{lim}u(\stackrel{}{r},\stackrel{}{p};t)=e^{i\stackrel{}{p}\stackrel{}{r}/\mathrm{}},$$
(9)
corresponding to the initial condition $`W(\stackrel{}{r},\stackrel{}{p};0)=1`$. From (7) and (8) we obtain an exact equation for $`W`$:
$`{\displaystyle \frac{W}{t}}`$ $`=`$ $`{\displaystyle \frac{\mathrm{}^2}{2}}[\mathrm{\Delta }t(\mathrm{\Delta }V){\displaystyle \frac{2it}{\mathrm{}}}(\stackrel{}{p}V)+t^2(V)^2`$ (10)
$`+{\displaystyle \frac{2}{\mathrm{}}}(i\stackrel{}{p}\mathrm{}tV)]W,`$
where $`\mathrm{\Delta }=^2`$ is the Laplacian. Next we expand $`W`$ in powers of $`\mathrm{}`$:
$$W=\underset{k=0}{\overset{\mathrm{}}{}}\mathrm{}^kW_k$$
(11)
and equate the terms with the same power of $`\mathrm{}`$ on both sides. We thus obtain a recurrence relation of differential equations for the $`W_k`$:
$`{\displaystyle \frac{W_k}{t}}`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left[\mathrm{\Delta }t(\mathrm{\Delta }V)+t^2(V)^22tV\right]W_{k2}`$ (12)
$`+i\stackrel{}{p}\left[t(V)\right]W_{k1},`$
with the initial conditions $`W_k=0`$ for $`k<0,W_0=1`$. Since $`Z(t)`$ and $`W`$ are linearly related, the expansion of $`W`$ in powers of $`\mathrm{}`$ immediately translates into an expansion of $`Z(t)`$ in powers of $`\mathrm{}`$. To obtain the term $`Z_k(t)`$, one needs to calculate $`W_k`$ from (12) and integrate over $`\stackrel{}{p}`$ and $`\stackrel{}{x}`$:
$$Z_k(t)=\frac{\mathrm{}^k}{(2\pi \mathrm{})^n}_{\mathrm{}}^{\mathrm{}}\underset{i=1}{\overset{n}{}}dx_idp_iW_k(\stackrel{}{r},\stackrel{}{p};t)\mathrm{exp}\left[t\left(\frac{\stackrel{}{p}^2}{2}+V(\stackrel{}{x})\right)\right].$$
(13)
The expressions (10) and (12) may be written in more compact form, if we introduce the “covariant derivative” $`\stackrel{}{D}tV`$:
$$\frac{W}{t}=\frac{\mathrm{}^2}{2}\left[D^2+\frac{2i}{\mathrm{}}\stackrel{}{p}\stackrel{}{D}\right]W=\frac{1}{2}\left[(\mathrm{}\stackrel{}{D}+i\stackrel{}{p})^2+\stackrel{}{p}^2\right]W$$
(14)
or its recursive form
$$\frac{W_k}{t}=\frac{1}{2}\left[D^2W_{k2}+2i\stackrel{}{p}\stackrel{}{D}W_{k1}\right].$$
(15)
We emphasize that the symbol $`\stackrel{}{p}`$ here denotes a classical phase-space variable and not an operator. As far as we know, the form (14) of the equation (10) has not previously been presented in the literature.
In terms of the operator $`\stackrel{}{D}`$, eq. (14) resembles a Fokker-Planck equation for $`W(\stackrel{}{r},\stackrel{}{p};t)`$ with the diffusion constant $`\sigma =\mathrm{}^2`$ and the drift vector $`\stackrel{}{\gamma }=i\mathrm{}\stackrel{}{p}`$ (see text below eq. (16)). The relation to the Fokker-Planck equation can be further elucidated by noting that the “vector potential” $`\stackrel{}{A}=tV`$ is a complete gradient and thus can be “gauged” away by means of the transformation $`We^{tV}W^{}`$, yielding an alternative form of (14):
$$\frac{W^{}}{t}=\frac{\mathrm{}^2}{2}\left[^2+\frac{2i}{\mathrm{}}\stackrel{}{p}\right]W^{}VW^{}.$$
(16)
If we interpret $`W^{}(\stackrel{}{r},\stackrel{}{p};t)`$ as a one-time probability density and introduce the probability current (sometimes called the probability flux in the literature)
$$\stackrel{}{J}^{}=i\mathrm{}\stackrel{}{p}W^{}\frac{\mathrm{}^2}{2}W^{},$$
(17)
we may write (16) in the form of a continuity equation:
$$\frac{W^{}}{t}+\stackrel{}{J}^{}=VW^{},$$
(18)
where the potential term acts as a source term and violates the local conservation law associated with the Fokker-Planck equation.
The recurrence relation (15) clearly shows that the expansion in $`\mathrm{}`$ introduced by Kirkwood is also an expansion in powers of the gradient operator as emphasized by Uhlenbeck and Beth . In the general case one needs to expand in powers of $`\mathrm{}`$ or, equivalently, in powers of the gradient operator using (12) or (15), as we will do here. In special cases, however, the compact form of the equations (14) or (16) may be the starting point of other effective approximation schemes.
## 4 Testing the method of separation beyond the TF approximation for the $`x^2y^2`$ potential
We now want to check whether the method of the separation $`x`$\- and $`y`$\- motions in the channels works beyond the TF approximation. In the TF approximation, where the integrand in $`Z(t)`$ is simply $`\mathrm{exp}(tH)`$, the boundary $`Q`$ appears in the argument of a logarithm. Higher WK corrections to the TF term lead to a power-like dependence on $`Q`$, and it is important to confirm that all dependence on $`Q`$ is cancelled in the final answer, at least for the leading terms in the parameter $`tQ^41`$. Here we consider this problem up to the second-order corrections. In the later sections we will consider corrections up to the order $`\mathrm{}^8`$.
As we stressed before, we apply the WK method to the motion in the central region ($`|x|,|y|Q`$) in its full scope, whereas for the motion in the channels ($`|x|,|y|Q`$) this method will be used only to study the quantum effects on the “slow” longitudinal motion, which is adiabatically separated from the quantized oscillatory motion in the transverse coordinate.
Integrating (12) with respect to $`t`$, we have:
$`W_1`$ $`=`$ $`{\displaystyle \frac{it}{2}}\stackrel{}{p}V`$
$`W_2`$ $`=`$ $`{\displaystyle \frac{t^2}{2}}\left[{\displaystyle \frac{1}{2}}\mathrm{\Delta }V+{\displaystyle \frac{t}{3}}(V)^2+{\displaystyle \frac{t}{3}}{\displaystyle \underset{i,k}{}}p_ip_k{\displaystyle \frac{^2V}{x_ix_k}}{\displaystyle \frac{t^2}{4}}(\stackrel{}{p}V)^2\right]`$ (19)
with $`V(x,y)=\frac{1}{2}g^2x^2y^2`$. Integration over $`p_x,p_y`$ and making use of the symmetry of the Hamiltonian with respect to the interchange $`xy`$, we find that the contribution from $`W_1`$ vanishes and
$$𝑑\mathrm{\Gamma }W_2e^{tV}=\frac{\pi tg^2}{3}\left[I_{10}+\frac{tg^2}{2}I_{21}\right],$$
(20)
where we introduced the abbreviations
$$d\mathrm{\Gamma }=dxdydp_xdp_y$$
(21)
and (for $`mn`$)<sup>2</sup><sup>2</sup>2Note that the case $`m=n`$ needs to be calculated separately from the case $`m>n`$ for the leading terms; see below.:
$$I_{mn}=4_0^Q𝑑x_0^Q𝑑yx^{2m}y^{2n}e^{tg^2x^2y^2/2}.$$
(22)
These integrals can be evaluated straightforwardly after the substitution $`x=w`$ and $`y=\sqrt{2/t}(u/gw)`$ and with the help of the condition for the validity of the adiabatic approximation $`Qt^{1/4}1`$. Note that this inequality does not contradict the condition permitting the use of the Wigner representation $`\mathrm{}Qt1`$ if $`\mathrm{}t^{3/4}1`$. We obtain
$$I_{10}\frac{\sqrt{2\pi }}{gt^{1/2}}Q^2,I_{21}\frac{\sqrt{2\pi }}{g^3t^{3/2}}Q^2.$$
(23)
and finally for $`Z_2(t)`$ in the square:
$$Z_2=K\frac{1}{12}(g\mathrm{}tQ)^2.$$
(24)
For the sake of completeness and later use, we note the systematic structure of the integrals $`I_{mn}`$. For a given value of $`mn`$ there are $`mn+1`$ such expressions: $`I_{mn,0},I_{mn+1,1},\mathrm{},I_{2(mn),mn}`$. Applying the same method as above yields the following result for these expressions at fixed $`mn`$:
$$I_{mn}=\frac{\sqrt{2\pi }}{(gt^{1/2})^{2n+1}}\frac{(2n1)!!}{mn}Q^{2(mn)}.$$
(25)
For the second region (the four channels $`|x|[Q,\mathrm{}]`$ or $`|y|[Q,\mathrm{}]`$), the integration is more involved. WK corrections to the TF term introduce pre-exponential functions of $`p_x,p_y,x,y`$ and $`t`$ in $`W_2`$. Because of the fourfold symmetry of the Hamiltonian (3) it is sufficient to consider the channel $`xQ`$. Closely following the method used in we consider the motion in the $`x`$ variable as free and the motion in the $`y`$ variable as that of a harmonic oscillator with an $`x`$-dependent frequency with the Hamiltonian
$$H_y=\frac{1}{2}p_y^2+\frac{1}{2}\omega _x^2y^2$$
(26)
with $`\omega _x=gx`$ and eigenvalues $`e_n(x)=(n+\frac{1}{2})\mathrm{}gx`$.
In order to reduce the integrals over $`p_y^2`$ and $`y^2`$ to derivatives of the well-known exponentiated sum-rule for the harmonic oscillator, we rescale the kinetic energy term in $`H_y`$ by a factor $`b`$ and the potential energy term by a factor $`a`$. We denote the rescaled Hamiltonians by $`H_y^{}`$. We then can generate any pre-exponential powers of $`p_y^2`$ and $`y^2`$ by differentiating with respect to $`a`$ and $`b`$ and setting $`a=b=1`$ in the final result. Integrating $`W_2`$ from (19) over $`p_x`$ and $`x`$ and performing the described substitutions and differentiations we obtain :
$$_{\mathrm{}}^{\mathrm{}}𝑑p_x_Q^{\mathrm{}}𝑑xW_2\mathrm{Tr}\left(e^{tH_y^{}}\right)=2\sqrt{\frac{2\pi }{t}}(gt)^2_Q^{\mathrm{}}𝑑xD(x)$$
(27)
with
$`D(x)`$ $`=`$ $`[{\displaystyle \frac{x^2}{2}}{\displaystyle \frac{2x^2}{3}}{\displaystyle \frac{d}{da}}+{\displaystyle \frac{1}{3g^2tx^2}}{\displaystyle \frac{d^2}{da^2}}`$ (28)
$`{\displaystyle \frac{2x^2}{3}}{\displaystyle \frac{d}{db}}+x^2{\displaystyle \frac{d^2}{dadb}}\left]\mathrm{Tr}\right(e^{tH_y^{}})_{a=b=1}.`$
In (27) we may discard the third term as it is $`𝒪(1/x^4t)1`$ with respect to other terms ($`x>Q,Q^4t1`$ in the channel).<sup>3</sup><sup>3</sup>3One may notice that contributions arising from the derivative with respect to the “fast” variable $`y`$ in $`W_2`$ are not small, whereas derivatives with respect to $`x`$ are kinematically negligible in the channel where $`Q^4t1`$. We note that the algebraic trick dealing with terms including the “fast” variables $`y,p_y`$ effectively reduce the problem in the channels to the level of TF terms with the rescaled Hamiltonians $`H_y^{}`$. Now we can use the exponentiated sum-rule for the harmonic oscillator (see ) to obtain
$`D(x)`$ $`=`$ $`[{\displaystyle \frac{x^4}{4\mathrm{sinh}\xi }}{\displaystyle \frac{2x^2}{3}}{\displaystyle \frac{d}{da}}{\displaystyle \frac{1}{\mathrm{sinh}(\xi \sqrt{a})}}`$ (29)
$`{\displaystyle \frac{x^2}{2}}{\displaystyle \frac{d^2}{dadb}}{\displaystyle \frac{1}{\mathrm{sinh}(\xi \sqrt{ab})}}]_{a=b=1}.`$
where $`\xi =\mathrm{}gtx/2`$. Performing the differentiations, setting $`a=b=1`$, and multiplying by 4 to account for all four channels, we obtain for the channel contribution to $`Z_2(t)`$:
$`Z_2(t)`$ $`=4K{\displaystyle _{\xi _0}^{\mathrm{}}}\xi ^2𝑑\xi `$ $`[{\displaystyle \frac{1}{\mathrm{sinh}\xi }}+{\displaystyle \frac{11}{6}}\xi {\displaystyle \frac{\mathrm{cosh}\xi }{\mathrm{sinh}^2\xi }}`$ (30)
$`+{\displaystyle \frac{\xi ^2}{2\mathrm{sinh}\xi }}{\displaystyle \frac{\xi ^2\mathrm{cosh}\xi }{\mathrm{sinh}^3\xi }}],`$
with $`\xi _0=\mathrm{}gtQ/2(1)`$. Performing the integrations in (30) (see , integrals 2.477.3, 3.523.1, 2.479.4, 2.477.1, 3.523.1), we obtain for the channel contribution to $`Z_2`$:
$$Z_2(t)=K\left[\frac{1}{12}(g\mathrm{}tQ)^221\zeta (3)\right].$$
(31)
where $`\zeta (z)`$ is the Riemann zeta function ($`\zeta (3)1.202`$). From (24) and (31) one sees that the $`Q`$-dependence of $`Z_2(t)`$ cancels, and the final answer is
$$Z_2(t)25.2K.$$
(32)
We achieved the $`Q`$-independence of the second-order correction to the partition function of $`x^2y^2`$ model using the method of . The huge renormalization of the TF term in (32) looks suspicious.
## 5 Improved quantum motion in the channels
The unexpectedly large coefficient in (31) suggests that we need to find a better treatment for the motion in the channel. The previous study of the quantum motion in the channel neglected the quantum nature of the “slow” motion along the channel axis, which was assumed to be free in . We already stressed in the Introduction that this motion is by no means free, rather, it is influenced by an effective linear potential created by the quantum fluctuations in the direction(s) orthogonal to the channel axis. Indeed, in the region $`|x||y|`$, where the derivatives with respect to $`x`$ are small relative to the ones with respect to $`y`$, we may first average the motion over the quantum fluctuations of $`y`$ described by the Hamiltonian $`H_y`$ (26) and by the corresponding wave function
$$\psi _n(y)=\frac{1}{\sqrt{2^nn!}}\left(\frac{gx}{\pi \mathrm{}}\right)^{1/4}e^{gxy^2/2\mathrm{}}H_n(y\sqrt{gx/\mathrm{}}),$$
(33)
where $`H_n(z)`$ are the Hermite polynomials. The corresponding average value of $`H_y`$
$$n|H_y|n=(n+\frac{1}{2})\mathrm{}gx$$
(34)
then becomes an effective potential for the description of the motion in the “slow” variable $`x`$:
$$\left(\frac{\mathrm{}^2}{2}\frac{^2}{x^2}+(n+\frac{1}{2})\mathrm{}gx\right)\varphi _n(x)=E\varphi _n(x).$$
(35)
This is the well-known Schrödinger equation for a linear potential with solutions in terms of Airy functions. Equation (35) clearly shows that, quantum mechanically, the “particle” is linearly confined along the channel axis and its motion is not free.<sup>4</sup><sup>4</sup>4Note that the phenomenon called “confinement” here is not the same as the phenomenon commonly referred to as quark confinement. In our case, the potential depends linearly on the field amplitude $`x(t)`$, not on a spatial coordinate. We already emphasized the dual role of the coordinates $`x,y,z`$ earlier, but we stress this point again here to avoid misunderstandings. One may also have called the phenomenon discussed here “self-confinement”, as the fields themselves “prepare” the effective potential barrier prohibiting the escape to infinity. The eigenvalue problem (35) has a discrete spectrum. Note that this argument constitutes a sixth proof, in addition to the five proofs listed in , for the discreteness of the spectrum of the Hamiltonian (3). It explains the large number in (31) and (32) as an artefact of the assumption of free motion in the $`x`$-direction. This assumption becomes increasingly poor for the higher-order quantum corrections, leading to poor convergence or even divergence of the expansion in powers of $`\mathrm{}`$. Treating the “slow” $`x`$-motion in the channel adiabatically, we apply the WK expansion and eq. (12) to the effective Hamiltonian
$$H_x^{(n)}=\frac{1}{2}p_x^2+(n+\frac{1}{2})\mathrm{}gx$$
(36)
describing the motion of the $`n`$th quantum mode in the region $`xQ`$. For the partition function of the one-dimensional motion of the $`n`$th mode we have (denoting $`p_x`$ simply by $`p`$):
$$Z^{(n)}(t)=\frac{1}{2\pi \mathrm{}}_{\mathrm{}}^{\mathrm{}}𝑑p_Q^{\mathrm{}}𝑑xe^{\frac{1}{2}p^2t(n+\frac{1}{2})\mathrm{}gxt}W(x,p;t).$$
(37)
Expanding $`W(x,p;t)`$ in powers of $`\mathrm{}`$ for the linear potential in (36) and using eq. (12), we easily obtain:
$`W_2`$ $`=`$ $`{\displaystyle \frac{1}{32^3}}a_n^2t^3(43p^2t);`$ (38)
$`W_4`$ $`=`$ $`{\displaystyle \frac{1}{3^22^7}}a_n^4t^6(1624p^2t+3p^4t^2)`$
$`W_6`$ $`=`$ $`{\displaystyle \frac{1}{2^{10}3^35!!}}a_n^6t^9(320720p^2t+180p^4t^29p^6t^3),`$
$`W_8`$ $`=`$ $`{\displaystyle \frac{1}{2^{15}3^47!!}}a_n^8t^{12}(896026880p^2t+10080p^4t^21008p^6t^3+27p^8t^4),`$
where $`a_n=(n+\frac{1}{2})\mathrm{}g`$. The terms with odd indices contain odd powers of $`p`$ and give vanishing contributions after integration over $`p`$. Carrying out the integration over $`p`$ in (37), summing over $`n`$, and retaining all terms with derivatives up to order $`\mathrm{}^8`$, we obtain for the contribution from a single channel:
$`Z_{\mathrm{ch}}(t)`$ $`={\displaystyle _Q^{\mathrm{}}}{\displaystyle \frac{dx}{\sqrt{2\pi t}\mathrm{}}}`$ $`(1+{\displaystyle \frac{\mathrm{}^2t}{24}}{\displaystyle \frac{^2}{x^2}}+{\displaystyle \frac{\mathrm{}^4t^2}{1152}}{\displaystyle \frac{^4}{x^4}}`$
$`+{\displaystyle \frac{\mathrm{}^6t^3}{82944}}{\displaystyle \frac{^6}{x^6}}+{\displaystyle \frac{\mathrm{}^8t^4}{7962624}}{\displaystyle \frac{^8}{x^8}}+\mathrm{}){\displaystyle \frac{1}{2\mathrm{sinh}\xi }},`$
where again $`\xi =\mathrm{}gxt/2`$. The expansion parameter $`\mathrm{}^2t/Q^21`$ in (5) is the squared ratio of the “diffusion” length mentioned in Section 3 and the separation scale $`Q`$. After performing the differentiations and integrating over $`x`$ we finally have for all four channels:
$`Z(t)=K[4\mathrm{ln}\mathrm{coth}{\displaystyle \frac{\xi }{2}}`$ $``$ $`({\displaystyle \frac{\lambda ^2}{2^33}}_\xi +{\displaystyle \frac{\lambda ^4}{2^93^2}}_\xi ^3+{\displaystyle \frac{\lambda ^6}{2^{14}3^4}}_\xi ^5`$ (40)
$`+{\displaystyle \frac{\lambda ^8}{2^{21}3^5}}_\xi ^7\mathrm{}){\displaystyle \frac{1}{\mathrm{sinh}\xi }}]_{\xi =u}`$
with $`u=\mathrm{}gtQ/2`$ and $`\lambda ^2=g^2\mathrm{}^4t^3`$. This expression, with the exception of the logarithm, can be written as a power series in $`u`$ by substituting the Bernouilli expansion of the hyperbolic cosecans:
$$\frac{1}{\mathrm{sinh}\xi }=\frac{1}{\xi }2\underset{n=1}{\overset{\mathrm{}}{}}\frac{2^{2n1}1}{(2n)!}B_{2n}\xi ^{2n1}.$$
(41)
where $`B_{2n}`$ denotes the Bernoulli numbers ($`B_2=1/6,B_4=1/30,B_6=1/42,B_8=1/30,\mathrm{}`$). It is then easy to see that there are three kinds of terms in the series involving derivatives $`_\xi ^{2n1}`$ in (40):
1. terms with a positive power of $`u^2`$; there is an infinite number of such terms;
2. terms with an inverse power of $`u^2`$; there is one such term at for each derivative;
3. terms independent of $`u`$, and thus independent of $`Q`$; again there is only one such term for each derivative.
The terms of type (i) are of the order $`\lambda ^{2n}(\mathrm{}Qt)^{k4n}`$, where $`k`$ denotes the overall power of $`\mathrm{}`$. The leading term in $`Z(t)`$ at the same power of $`\mathrm{}`$ being of order $`(\mathrm{}Qt)^k`$, arising from the expansion of the logarithmic term in (40), the contribution from the series is suppressed by a factor $`(tQ^4)^n1`$. For the terms of type (ii) one can show that they are of the order $`(\mathrm{}Qt)^{m+1}(\mathrm{}^4t^3)^n`$. For a given power $`k=4nm1`$ of $`\mathrm{}`$ their ratio to the leading contribution $`(\mathrm{}Qt)^k`$ is again of order $`(tQ^4)^n1`$. Finally, for the terms of the kind (iii), their ratio to the leading terms at a given power $`k=4n`$ is $`(\mathrm{}^4t^3)^n(\mathrm{}Qt)^n=(tQ^4)^n=(tQ^4)^{k/4}1`$. We thus conclude that each term arising from the series of derivatives in (40) is smaller than the contribution from the leading term, and the relative suppression increases with $`k`$ or $`n`$. However, it would be wrong to conclude that these subdominant terms can all be neglected, because there are certain terms among those of type (iii), which are independent of the cutoff $`Q`$ and which are not canceled by similar contributions from the channels.
Having established that the terms arising from the logarithmic term in (40) dominate in the expansion of $`Z(t)`$ in powers of $`\mathrm{}`$, we now proceed to give those explicitly. The integrated form of the series (41) is:
$$\mathrm{ln}\mathrm{coth}\frac{u}{2}=\mathrm{ln}\frac{u}{2}+\underset{k=1}{\overset{\mathrm{}}{}}\frac{(2^{2k1}1)}{k(2k)!}B_{2k}u^{2k},$$
(42)
This series converges for $`u^2<\pi ^2`$ or $`(\mathrm{}gtQ)<2\pi `$, which is consistent with the condition permitting the use of the Wigner representation in the square $`[Q,Q]`$ and with the inequality $`tQ^41`$. We thus get for the partition function inside the four channels, up to terms involving powers of $`(tQ^4)^11`$:
$`Z_{0+2+4+6+8}^{[Q,\mathrm{}]}(t)`$ $`=`$ $`K[4\mathrm{ln}{\displaystyle \frac{4}{\mathrm{}gtQ}}+{\displaystyle \frac{1}{12}}(\mathrm{}gtQ)^2{\displaystyle \frac{7}{2^735!!}}(\mathrm{}gtQ)^4`$
$`+{\displaystyle \frac{31}{2^939!!}}(\mathrm{}gtQ)^6{\displaystyle \frac{127}{2^{16}59!!}}(\mathrm{}gtQ)^8+\mathrm{}].`$
We retained in (5) the higher-order terms for future use.
The contributions $`Z_0`$ and $`Z_2`$ for the central square are given by (24) and by eq. (5) from . Their sum is:
$$Z_{0+2}^{[Q,Q]}(t)=K\left[2\mathrm{ln}\sqrt{2}t^{1/2}g^2Q^2+C\frac{1}{12}(\mathrm{}gtQ)^2\right].$$
(44)
Adding the first two terms in (5) and (44) we finally obtain for $`Z`$ up to the order of $`\mathrm{}^2`$ the expression (4) found in , i. e. the $`Q`$-dependent corrections to the TF term vanish at the second-order of $`\mathrm{}`$. Below we shall see that this statement is correct even up to the order $`\mathrm{}^8`$. We conclude that the unusually large number renormalizing the TF term in (32) is an artefact of the neglect of the quantum character of the motion along the axis in the hyperbolic channels.
## 6 The corrections to the TF term vanish up to order $`\mathrm{}^8`$
In this Section we show that the leading $`Q`$-dependent quantum corrections to the TF term up to the order $`\mathrm{}^8`$ cancel if we treat the quantum mechanical motion in the channels correctly. Actually, we already gave the corrections to $`Z(t`$) up to $`\mathrm{}^8`$ for the correct motion in the hyperbola channels (see formula (5)). It remains for us to calculate these corrections for the square $`x,y[Q,Q]`$, a straightforward though cumbersome task. First of all,we need to know $`W_4(t)`$ using (12).There are two types of contribution to $`W_4(t)`$: With $`mn=2`$ and $`m=n`$. Terms with $`mn=2`$ give the main contribution of the order $`tQ^4`$, terms with $`m=n`$ contain only logarithms of $`tQ^4`$ and may be neglected with the precision $`\mathrm{ln}(tQ^4)/tQ^41`$. Here we give the expression for $`W_4(t)`$ after the integration over $`p_x`$ and $`p_y`$:
$`{\displaystyle 𝑑\mathrm{\Gamma }W_4e^{Vt}}`$ $`=`$ $`{\displaystyle \frac{2\pi g^2t^3}{2^4t}}({\displaystyle \frac{1}{5}}g^2tI_{20}{\displaystyle \frac{11}{45}}(g^2t)^2I_{31}`$ (45)
$`+{\displaystyle \frac{1}{36}}(g^2t)^3I_{42}{\displaystyle \frac{4}{15}}I_{00}+g^2tI_{11}`$
$`{\displaystyle \frac{17}{45}}(g^2t)^2I_{22}+{\displaystyle \frac{1}{36}}(g^2t)^3I_{33})`$
Integration of $`I_{mn}`$ over $`x`$ and $`y`$ for $`mn`$ was already done before (see (25)); for $`m=n`$ we obtain:
$`I_{00}`$ $`=`$ $`{\displaystyle \frac{\sqrt{2\pi }}{(g^2t)^{1/2}}}\left[\mathrm{ln}(g^2Q^4t)+C+\mathrm{ln}2\right]`$
$`I_{11}`$ $`=`$ $`{\displaystyle \frac{\sqrt{2\pi }}{(g^2t)^{3/2}}}\left[\mathrm{ln}(g^2Q^4t)+C+\mathrm{ln}22\right]`$
$`I_{22}`$ $`=`$ $`{\displaystyle \frac{3\sqrt{2\pi }}{(g^2t)^{5/2}}}\left[\mathrm{ln}(g^2Q^4t)+C+\mathrm{ln}2{\displaystyle \frac{8}{3}}\right]`$
$`I_{33}`$ $`=`$ $`{\displaystyle \frac{15\sqrt{2\pi }}{(g^2t)^{7/2}}}\left[\mathrm{ln}(g^2Q^4t)+C+\mathrm{ln}2{\displaystyle \frac{46}{15}}\right].`$ (46)
Collecting all terms together we have from (45) for the corrections to $`Z(t)`$ of the order of $`\mathrm{}^4`$ in the square $`x,y[Q,+Q]`$:
$`Z_4^{[Q,Q]}(t)`$ $`=`$ $`K[{\displaystyle \frac{7}{2^735!!}}g^4\mathrm{}^4t^4Q^4`$ (47)
$`+g^2\mathrm{}^4t^3[\mathrm{ln}(g^2Q^2\sqrt{t/2})8C16\mathrm{ln}2]]`$
where the first term comes from $`I_{mn}`$ with $`mn=2`$, and the logarithmic terms, as we remarked before, arise from the contributions $`I_{mm}`$. We see that the second term in (47) is of order $`\mathrm{ln}(tQ^4)/Q^4t1`$. Discarding it and adding the $`\mathrm{}^4`$ correction from the channels in (5) we find that with the precision $`\mathrm{ln}(tQ^4)/tQ^41`$,
$$Z_4^{[Q,+Q]}+Z_4^{[Q,\mathrm{}]}=0$$
(48)
as it was for the second-order corrections (with the precision $`1/tQ^4`$).
Consider now the $`\mathrm{}^6`$-order corrections to the partition function. In (5) we included the dominant contribution from the channels $`[Q,\mathrm{}]`$ (fourth term). For $`W_6`$ in the square we have, after integration over $`p_x`$ and $`p_y`$ and using the notation of (22):
$`{\displaystyle 𝑑\mathrm{\Gamma }W_6e^{tV}}`$ $`=`$ $`{\displaystyle \frac{2\pi (gt)^6}{t9!!2^6}}[61I_{30}+{\displaystyle \frac{249}{2}}(g^2t)I_{41}`$ (49)
$`{\displaystyle \frac{119}{4}}(g^2t)^2I_{52}+{\displaystyle \frac{35}{24}}(g^2t)^3I_{63}],`$
Using (25) for the evalulation of the $`I_{mn}`$ and collecting all terms, we obtain for $`Z_6^{[Q,Q]}`$ in the square:
$$Z_6^{[Q,Q]}(t)=K\frac{31}{32^99!!}(\mathrm{}gtQ)^6.$$
(50)
Once more, the contribution from the square is exactly canceled against the one from the four channels given in (5), as it happened for the second and fourth order:
$$Z_6^{[Q,Q]}(t)+Z_6^{[Q,\mathrm{}]}(t)=0$$
(51)
with the precision $`\mathrm{ln}(tQ^4)/(tQ^4)1`$.
Finally, we give here the order $`\mathrm{}^8`$ corrections to $`Z_8^{[Q,Q]}`$ in the square:
$`{\displaystyle 𝑑\mathrm{\Gamma }W_8e^{tV}}`$ $`=`$ $`{\displaystyle \frac{2\pi (gt)^8}{t5!!2^7}}[{\displaystyle \frac{1261}{7560}}I_{40}{\displaystyle \frac{259}{540}}g^2tI_{51}+{\displaystyle \frac{893}{5040}}(g^2t)^2I_{62}`$ (52)
$`{\displaystyle \frac{23}{1296}}(g^2t)^3I_{73}+{\displaystyle \frac{5}{10368}}(g^2t)^4I_{84}].`$
where we again discarded terms proportional to $`I_{mm}`$, which are of the order $`\mathrm{ln}(tQ^4)/(tQ^4)1`$ with respect to the terms containing $`I_{mn}`$ with $`m>n`$. Using now the general expression (25) for $`I_{mn}`$ we obtain:
$$𝑑\mathrm{\Gamma }W_8e^{Vt}=\frac{(2\pi )^{3/2}127}{gt^{3/2}2^95!!32^77!!}(gtQ)^8$$
(53)
and get the result
$$Z_8^{[Q,Q]}(t)=K\frac{127}{52^{16}9!!}(\mathrm{}gtQ)^8$$
(54)
which miraculously cancels with the contribution from the channels in (5).
Although we cannot prove such a cancellation in general, to all orders, we have no doubt that all higher-order corrections to the partition function containing powers of $`(\mathrm{}gtQ)`$ cancel in the limit $`tQ^41`$. Of course, there are other corrections involving powers of $`(tQ^4)^1`$, but these are suppressed due to the classical condition of adiabaticity $`tQ^41`$. We shall consider this issue in Section 7.
In anticipation of later applications, we remark here that there is a strong correlation between the power of $`\mathrm{}`$, denoted by $`k2`$, and power of the dominant terms in the limit $`tQ^41`$. A systematic analysis of the higher-order corrections using Mathematica leads to the conclusion that the difference between $`m`$ and $`n`$ in the leading integral $`I_{mn}`$ from (22) is $`mn=\frac{1}{2}k`$. Besides the terms with the largest difference $`mn`$, which give the leading contribution to $`W_k`$ and which we retain in (49) and (52), there are also terms involving $`I_{mn}`$ with $`mn=\frac{1}{2}k2\mathrm{}`$ with $`\mathrm{}=1,2,\mathrm{}<\frac{1}{4}k`$. There is also a correlation between the powers of $`t`$ and $`m,n`$. The analysis shows that for the terms with $`mn=\frac{1}{2}k`$ the power of $`t`$ is $`2mn1`$, and for terms with $`mn=\frac{1}{2}k2\mathrm{}`$ the power is $`2mn13\mathrm{}`$. The factor of $`g`$ for $`I_{mn}`$ is $`g^{2m}`$. Straightforward calculations similar to the above show that the ratio of the contributions to the partition function for the terms $`mn=\frac{1}{2}k2\mathrm{}`$ to the ones for $`mn=k/2`$ is of the order $`1/(tQ^4)^{\mathrm{}}1`$. Indeed, we find:
$`Z_k^{(m,n)}(t)=K(\mathrm{}gtQ)^k{\displaystyle \frac{(2n1)!!}{2^{n1}}}\left(mn={\displaystyle \frac{k}{2}}\right);`$ (55)
$`Z_k^{(m,n)}(t)=K(g^4tQ^4)^{\mathrm{}}(\mathrm{}gtQ)^k{\displaystyle \frac{(2n1)!!}{2^{n1}(k4\mathrm{})}}\left(mn={\displaystyle \frac{k}{2}}2\mathrm{}\right).`$
This justifies neglecting these subdominant terms in our calculations here.
Finally, at any power $`k`$ of $`\mathrm{}`$ terms with logarithms ($`m=n`$) may be discarded with respect to the terms with powers of $`Q^2`$. Indeed, the last terms with $`m>n`$ are of the order of $`(\mathrm{}gtQ)^k`$, terms with $`m=n`$ are of the order of $`(g^2\mathrm{}^4t^3)^{k/4}\mathrm{ln}(tg^4Q^4)`$ and may be neglected with the precision $`\mathrm{ln}(g^4tQ^4)/(g^2tQ^4)1`$. Again, with increasing $`k`$ the precision improves.
## 7 $`Q`$-independent terms in the channels and central region
We already mentioned that the derivative expansion for the channel contribution (40) has one $`Q`$-independent term for each power of of $`\lambda ^2=g^2\mathrm{}^4t^3`$. From the expansion (41) it is evident that these terms have the form
$$\frac{2^{2n1}1}{n(2n)!}(2n1)!!B_{2n}\lambda ^{2n}.$$
(56)
Each terms is of the order $`(tQ^4)^11`$ compared with the leading term from the expansion of the logarithic term in (40). Collecting these terms we obtain a series of $`Q`$-independent terms contributing to $`Z(t)`$:
$$\frac{1}{2^33}B_2\lambda ^2+\frac{2^31}{2^{10}3^2}B_4\lambda ^4+\frac{2^51}{2^{14}3^5}B_6\lambda ^6+\frac{2^71}{2^{23}3^5}B_8\lambda ^8+\mathrm{}$$
(57)
We can rewrite this series in the form
$$\underset{n=1}{\overset{\mathrm{}}{}}\frac{2^{2n}(2^{2n1}1)(2n1)!!}{2^{2(n1)}n(2n)!}B_{2n}\left(\frac{\lambda }{4\sqrt{3}}\right)^{2n}.$$
(58)
The factor $`(2n1)!!`$, which arises from the $`(2n1)`$-th derivative in (40), spoils the convergence of this series for $`Z(t)`$. Using the identity $`2^nn!(2n1)!!=(2n)!`$, we obtain for the series of $`Q`$-independent terms:
$$Z^{[Q,\mathrm{}]}=K\underset{n=1}{\overset{\mathrm{}}{}}\frac{4(2^{2n1}1)}{2^nn!n}B_{2n}\left(\frac{\lambda }{4\sqrt{3}}\right)^{2n}.$$
(59)
This series is only asymptotic despite the smallness of $`\lambda ^2`$, because the Bernouilli numbers grow factorially at large $`n`$:
$$B_{2n}\frac{2(2n)!}{(2\pi )^{2n}}.$$
(60)
Next we turn to the $`Q`$-independent contributions to $`Z(t)`$ from the central region. We start with the investigation of the corrections to the dominant terms. A straightforward calculation yields the following correction factor to $`I_{mn}`$ from (25):
$$1\frac{(2m1)!!}{(2n1)!!}(g^2Q^4t)^{nm}.$$
(61)
This factor plays an important part making the corrected $`I_{mn}`$ well behaved in the limit $`m=n`$. Indeed, we note that with $`\epsilon =2(mn)`$:
$$\frac{(2m1)!!}{(2n1)!!}1+\epsilon \underset{\mathrm{}=1}{\overset{m}{}}\frac{1}{2\mathrm{}1}$$
(62)
the diagonal elements ($`m=n`$) become
$$I_{mm}=\underset{\epsilon 0}{lim}I_{mn}=\frac{\sqrt{2\pi }(2m1)!!}{(g^2t)^{m+\frac{1}{2}}}\left[\mathrm{ln}(g^2Q^4t)2\underset{\mathrm{}=1}{\overset{m}{}}\frac{1}{2\mathrm{}1}\right],$$
(63)
demonstrating that the corrections to $`I_{mm}`$ are independent of $`Q`$, with the exception of the logarithmic term. One easily confirms this general result by explicit calculations. Substituting $`x=w`$ and $`y=\sqrt{2/t}(u/gw)`$ in (22) we get:
$$I_{mm}=4\left(\frac{2}{g^2t}\right)^{\frac{2m+1}{2}}_0^Q\frac{dw}{w}_0^w𝑑uu^{2m}e^{u^2}.$$
(64)
We may write this expression as a derivative of the error function:
$$I_{mm}=4\left(\frac{2}{g^2t}\right)^{\frac{2m+1}{2}}\frac{\sqrt{\pi }}{2}(1)^m\frac{d^m}{da^m}_0^{w_0}\frac{dw}{\sqrt{a}w}\mathrm{erf}(\sqrt{a}w)|_{a=1},$$
(65)
where $`w_0^2=g^2Q^4t/2`$. Integrating by parts, discarding the exponentially small terms in $`w_0`$, then differentiating with respect to $`a`$ and finally setting $`a=1`$, we obtain:
$$I_{mm}=\frac{\sqrt{2\pi }}{(g^2t)^{m+\frac{1}{2}}}\left[(2m1)!!\mathrm{ln}w_0^2\frac{2^m}{\sqrt{\pi }}_0^{\mathrm{}}𝑑tt^{m\frac{1}{2}}e^t\mathrm{ln}t\right].$$
(66)
For the integral in the second term of this expression one finds (see ref. formula 4.352.2):
$$_0^{\mathrm{}}𝑑tt^{m\frac{1}{2}}e^t\mathrm{ln}t=\frac{\sqrt{\pi }}{2^m}(2m1)!!\left[2\underset{\mathrm{}=1}{\overset{m}{}}\frac{1}{2\mathrm{}1}C2\mathrm{ln}2\right]$$
(67)
which yields
$$I_{mm}=\frac{\sqrt{2\pi }(2m1)!!}{(g^2t)^{m+\frac{1}{2}}}\left[\mathrm{ln}(g^2Q^4t)+C+\mathrm{ln}22\underset{\mathrm{}=1}{\overset{m}{}}\frac{1}{2\mathrm{}1}\right].$$
(68)
Next we use a trick, first introduced by Euler, replacing the finite sum by an asymptotic series (see e.g. ):
$$\underset{\mathrm{}=1}{\overset{m}{}}\frac{1}{2\mathrm{}1}=\frac{1}{2}\left[C+\mathrm{ln}(2m)+\underset{\mathrm{}=1}{\overset{\mathrm{}}{}}\frac{2^{2\mathrm{}1}1}{(8m^2)^{\mathrm{}}}B_2\mathrm{}\right].$$
(69)
We are now ready to combine the contribution from the central region with the asymptotic series (59) obtained earlier for the $`Q`$-independent contribution for the channels. A special case is the case $`m=n=0`$, for which there is no infinite sum, giving
$$I_{00}=\sqrt{\frac{2\pi }{g^2t}}\left[\mathrm{ln}(g^2Q^4t)+C+\mathrm{ln}2\right].$$
(70)
Taken together with the logarithmic term from the channels, this leads to the $`Q`$-independent expression (4) obtained in the improved TF approximation . Our detailed analysis has shown that the structures $`I_{mm}`$ containing $`Q`$-independent terms appear only in $`W_{4k}`$ with $`k=1,2,\mathrm{}`$. This means that such terms appear only at the orders $`\mathrm{}^{4k}`$ and implies that the expansion parameter of $`Z(t)`$ is $`\lambda ^2=g^2\mathrm{}^4t^3`$.
For a given power $`\lambda ^{2n}`$ there are $`3n`$ quantities $`I_{mm}`$ ($`m=1,2,\mathrm{},3n`$) and $`I_{00}`$, which we collect separately. The results given above allow us to write the $`Q`$-independent asymptotic series from the central region. Including also the TF term (4), we thus obtain:
$`Z^{[Q,Q]}(t)`$ $`=`$ $`K[\mathrm{ln}{\displaystyle \frac{1}{g^2\mathrm{}^4t^3}}+9\mathrm{ln}2+C`$ (71)
$`+{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}\lambda ^{2n}(a_0^{(n)}(C+\mathrm{ln}2)`$
$`{\displaystyle \underset{m=1}{\overset{3n}{}}}a_m^{(n)}(2m1)!![\mathrm{ln}m+{\displaystyle \underset{\mathrm{}=1}{\overset{\mathrm{}}{}}}{\displaystyle \frac{2^{2\mathrm{}1}1}{(8m^2)^{\mathrm{}}}}B_2\mathrm{}])].`$
Here the $`a_m^{(n)}`$ ($`m=0,1,2,\mathrm{},3n`$) denote the numerical coefficients in $`(g^2t)^mI_{mm}`$ occurring in $`Z_{4n}`$. For example, for $`n=1`$ we have four such numbers with alternating signs:
$$a_0^{(1)}=\frac{1}{60},a_1^{(1)}=\frac{1}{16},a_2^{(1)}=\frac{17}{720},a_3^{(1)}=\frac{1}{576}.$$
(72)
Summing up all these results, we may surmise that in the limit $`tQ^41`$ all $`Q`$-dependence is canceled, and we are left with two asymptotic series (59) and (71) from the two regions contributing together to the partition function of the two-dimensional YMQM model. We have shown explicitly how the small parameter $`(tQ^4)^1`$, artificially introduced in the adiabatic separation of the degrees of freedom in the channels, transmutes into the genuine quantum mechanical parameter $`\lambda ^2=g^2\mathrm{}^4t^3`$.
## 8 Conclusions
We have shown that the richness of the classical YM mechanics with a $`x^2y^2`$ potential translates into, and even gets amplified by, the quantum mechanical properties of the system. The YM quantum mechanics exhibits a confinement property, which strongly influences the quantum mechanical motion in the $`x^2y^2`$ potential. At higher order in $`\mathrm{}`$ (up to $`\mathrm{}^8`$) this results in the vanishing of the leading quantum corrections (for $`tQ^41`$), when we correctly take into account this property for the motion in the hyperbolic channels. We believe that this result may survive to even higher order. We also derived a novel form of the equation for the Uhlenbeck-Beth function $`W(\stackrel{}{r},\stackrel{}{p};t)`$, which is the basis of the WK expansion. We believe that our expression can be a starting point for new approximation schemes using techniques from diffusion theory. We hope that the lessons derived from the present study of the higher-order quantum corrections to the homogeneous limit of the Yang-Mills equations will be useful for an improved understanding of the internal dynamics of the Yang-Mills quantum field theory.
|
warning/0506/physics0506228.html
|
ar5iv
|
text
|
# A double junction model of irradiated silicon pixel sensors for LHC
## 1 Introduction
The CMS experiment, currently under construction at the Large Hadron Collider (LHC) will include a silicon pixel detector to allow tracking in the region closest to the interaction point. The detector will be a key component for reconstructing interaction vertices and heavy quark decays in a particulary harsh environment, characterized by a high track multiplicity and heavy irradiation. The innermost layer, located at only 4 cm from the beam line, is expected to be exposed to an equivalent fluence of $`3\times 10^{14}`$ n<sub>eq</sub>/cm<sup>2</sup>/yr at full luminosity.
In these conditions, the response of the silicon sensors during the detector operation is of great concern. It is well understood that the intra-diode electric fields in these detectors vary linearly in depth reaching a maximum value at the p-n junction. The linear behavior is a consequence of a constant space charge density, $`N_{\mathrm{eff}}`$, caused by thermodynamically ionized impurities in the bulk material. It is well known that the detector characteristics are affected by radiation exposure, but it is generally assumed that the same picture is valid after irradiation. In fact, it is common to characterize the effects of irradiation in terms of a varying effective charge density. In we have proved that this picture does not provide a good description of irradiated silicon pixel sensors. In addition, it was shown that it is possible to adequately describe the charge collection characteristics of a heavily irradiated silicon detector in terms of a tuned double junction model which produces a double peak electric field profile across the sensor. The modeling is supported by the evidence of doubly peaked electric fields obtained directly from beam test measurements and presented in . In this paper we apply our model to sensors irradiated to lower fluences demonstrating that a doubly peaked electric field is already visible at a fluence of $`0.5\times `$10<sup>14</sup> n<sub>eq</sub>/cm<sup>2</sup>. In addition, the dependence of trap concentrations upon fluence is established by comparing the measured and simulated profiles at several fluences and bias voltages.
This paper is organized as follows: Section 2 describes the experimental setup, Section 3 describes the carrier transport simulation used to interpret the data. The tuning of the double junction model is discussed in Section 4 with the results of the fit procedure. The conclusions are given in Section 5.
## 2 Experimental setup
The measurements were performed in the H2 line of the CERN SPS in 2003/04 using 150-225 GeV pions. The beam test apparatus is described in . A silicon beam telescope consisted of four modules each containing two 300 mm thick single-sided silicon detectors with a strip pitch of 25 mm and readout pitch of 50 mm. The two detectors in each module were oriented to measure horizontal and vertical impact coordinates. A pixel hybrid detector was mounted between the second and third telescope modules on a cooled rotating stage. A trigger signal was generated by a silicon PIN diode. The analog signals from all detectors were digitized in a VME-based readout system by two CAEN (V550) and one custom built flash ADCs. The entire assembly was located in an open-geometry 3T Helmholtz magnet that produced a magnetic field parallel or orthogonal to the beam. The temperature of the tested sensors was controlled with a Peltier cooler that was capable of operating down to -30C. The telescope information was used to reconstruct the trajectories of individual beam particles and to achieve a precise determination of the particle hit position in the pixel detector. The resulting intrinsic resolution of the beam telescope was about 1 mm.
The prototype pixel sensors are so-called “n-in-n” devices: they are designed to collect charge from n<sup>+</sup> structures implanted into n–bulk silicon. All test devices were 22$`\times `$32 arrays of 125$`\times `$125 $`\mu `$m<sup>2</sup> pixels having a sensitive area of 2.75$`\times `$4 mm<sup>2</sup>. The substrate was 285 $`\mu `$m thick, n-doped, diffusively-oxygenated silicon of orientation $`111`$, resistivity of about 3.7 k$`\mathrm{\Omega }`$cm and oxygen concentration in the order of $`10^{17}`$ cm<sup>-3</sup>. Individual sensors were diced from fully processed wafers after the deposition of under-bump metalization and indium bumps. A number of sensors were irradiated at the CERN PS with 24 GeV protons. The irradiation was performed without cooling or bias. The delivered proton fluences scaled to 1 MeV neutrons by the hardness factor 0.62 were $`0.5\times `$10<sup>14</sup> n<sub>eq</sub>/cm<sup>2</sup>, $`2\times `$10<sup>14</sup> n<sub>eq</sub>/cm<sup>2</sup> and $`5.9\times `$10<sup>14</sup> n<sub>eq</sub>/cm<sup>2</sup>. All samples were annealed for three days at 30C. In order to avoid reverse annealing, the sensors were stored at -20C after irradiation and kept at room temperature only for transport and bump bonding. All sensors were bump bonded to PSI30/AC30 readout chips which allow analog readout of all 704 pixel cells without zero suppression. The PSI30 settings were adjusted to provide a linear response to input signals ranging from zero to more than 30,000 electrons.
## 3 Sensor simulation
A detailed sensor simulation was implemented, including a physical modeling of irradiation effects in silicon. Our simulation, pixelav , incorporates the following elements: an accurate model of charge deposition by primary hadronic tracks (in particular to model delta rays); a realistic 3-D electric field map resulting from the simultaneous solution of Poisson’s Equation, continuity equations, and various charge transport models; an established model of charge drift physics including mobilities, Hall Effect, and 3-D diffusion; a simulation of charge trapping and the signal induced from trapped charge; and a simulation of electronic noise, response, and threshold effects. A final step reformats the simulated data into test beam format so that it can be processed by the test beam analysis software.
The effect of irradiation was implemented in the simulation by including two defect levels in the forbidden silicon bandgap with opposite charge states and trapping of charge carriers. The model, similar to one proposed in , is based on the Shockley-Read-Hall statistics and produces an effective space charge density $`\rho _{\mathrm{eff}}`$ from the trapping of free carriers in the leakage current. The effective charge density is related to the occupancies and densities of traps as follows,
$$\rho _{\mathrm{eff}}=e\left[N_Df_DN_Af_A\right]+\rho _{\mathrm{dopants}}$$
(1)
where: $`N_D`$ and $`N_A`$ are the densities of donor and acceptor trapping states, respectively; $`f_D`$ and $`f_A`$ are the occupied fractions of the donor and acceptor states, respectively, and $`\rho _{\mathrm{dopants}}`$ is the charge density due to ionized dopants. Each defect level is characterized by an electron and hole trapping cross section, $`\sigma _{e/h}^D`$ and $`\sigma _{e/h}^A`$, for the donor and acceptor trap, respectively, and by an activation energy, $`E_D`$ and $`E_A`$ for the donor and acceptor trap, respectively.
An illustrative sketch of the model is shown in Fig. 1. Trapping of the mobile carriers from the generation-recombination current produces a net positive space charge density near the p<sup>+</sup> backplane and a net negative space charge density near the n<sup>+</sup> implant as shown in Fig. 1(a). Since positive space charge density corresponds to n-type doping and negative space charge corresponds to p-type doping, there are p-n junctions at both sides of the detector. The electric field in the sensor follows from a simultaneous solution of Poisson’s equation and the continuity equations. The resulting $`z`$-component of the electric field is shown in Fig. 1(b). It varies with an approximately quadratic dependence upon $`z`$ having a minimum at the zero of the space charge density and maxima at both implants. A more detailed description of the double junction model and its implementation can be found in .
## 4 Data analysis
Charge collection across the sensor bulk was measured using the “grazing angle technique” . As is shown in Fig. 2, the surface of the test sensor is oriented by a small angle (15) with respect to the pion beam. A large sample of data is collected with zero magnetic field and at a temperature of $`10^{}`$C. The charge measured by each pixel along the $`y`$ direction samples a different depth $`z`$ in the sensor. Precise entry point information from the beam telescope is used to produce finely binned charge collection profiles.
The charge collection profiles for a sensor irradiated to a fluence of $`\mathrm{\Phi }=0.5\times 10^{14}`$ n<sub>eq</sub>/cm<sup>2</sup> and $`\mathrm{\Phi }=2\times 10^{14}`$ n<sub>eq</sub>/cm<sup>2</sup> and operated at several bias voltages are presented in Fig. 3(a-c) and Fig. 3(d-g), respectively. The measured profiles, shown as solid dots, are compared to the simulated profiles, shown as histograms. The two trap model has six free parameters ($`N_D`$, $`N_A`$, $`\sigma _e^D`$, $`\sigma _h^D`$, $`\sigma _e^A`$, $`\sigma _h^A`$) that can be adjusted. The activation energies are kept fixed to the values of . Additionally, the electron and hole trapping rates, $`\mathrm{\Gamma }_e`$ and $`\mathrm{\Gamma }_h`$, are uncertain at the 30% level due to the fluence uncertainty and possible annealing of the sensors. They are treated as constrained parameters. The donor concentration of the starting material is set to $`1.2\times 10^{12}`$ cm<sup>-3</sup> corresponding to a full depletion voltage of about 70 V for an unirradiated device. The parameters of the double junction model were systematically varied and the agreement between measured and simulated charge collection profiles was judged subjectively. The procedure was repeated at the each fluence and the optimal parameter set was chosen when agreement between measured and simulated profiles was achieved for all bias voltages.
The simulation describes the measured charge collection profiles well both in shape and normalization. In particular,the “wiggle” observed at low bias voltages is also nicely described. The relative signal minimum near $`y=700\mu `$m (see Fig. 3) corresponds to the minimum of the electric field $`z`$-component, $`E_z`$, where both electrons and holes travel only short distances before trapping. This small separation induces only a small signal on the n<sup>+</sup> side of the detector. At larger values of $`y`$, $`E_z`$ increases causing the electrons drift back into the minimum where they are likely to be trapped. However, the holes drift into the higher field region near the p<sup>+</sup> implant and are more likely to be collected. The net induced signal on the n<sup>+</sup> side of the detector therefore increases and creates the local maximum seen near $`y=900\mu `$m. The $`z`$-component of the simulated electric field, $`E_z`$, is plotted as a function of $`z`$ in Fig. 4 and Fig. 4 for $`\mathrm{\Phi }=0.5\times 10^{14}`$ n<sub>eq</sub>/cm<sup>2</sup> and $`\mathrm{\Phi }=2\times 10^{14}`$ n<sub>eq</sub>/cm<sup>2</sup>, respectively. The field profiles have minima near the midplane of the detector and maxima at the detector implants as discussed in Section 3. Figure 4 shows that a double peak electric field is necessary to describe the measured charge collection profiles even at the lowest measured fluence, usually referred to as close to the “type inversion point”. The dependence of the space charge density upon the $`z`$ coordinate is shown in Fig. 4. Before irradiation the sensor is characterized by a constant and positive space charge density of $`1.2\times 10^{12}`$ cm<sup>-3</sup> across the sensor bulk. After a fluence of $`0.5\times 10^{14}`$ n<sub>eq</sub>/cm<sup>2</sup> the device shows a negative space charge density of about $`1\times 10^{12}`$ cm<sup>-3</sup> for about 70% of its thickness, a compensated region corresponding to the $`E_z`$ minimum and a positive space charge density close to the backplane. The increase of the space charge density upon $`z`$ is not linear due to the varying charge carrier mobilities across the bulk and to the requirement of a constant current density.
The model parameters obtained with the best fit procedure are shown in Table 1<sup>1</sup><sup>1</sup>1The comparison of the measured and simulated profiles at $`\mathrm{\Phi }=6\times 10^{14}`$ n<sub>eq</sub>/cm<sup>2</sup> can be found in .. We observe that the donor trap concentration increases more rapidly with fluence than does the acceptor trap concentration. The ratio between acceptor and donor trap concentrations is 0.76 at the lowest fluence and decreases to 0.40 at $`6\times `$10<sup>14</sup> n<sub>eq</sub>/cm<sup>2</sup>. In addition, the fits exclude a linear dependence of the trap concentrations with the irradiation fluence. At $`\mathrm{\Phi }=6\times `$10<sup>14</sup> n<sub>eq</sub>/cm<sup>2</sup> the cross section ratio $`\sigma _h/\sigma _e`$ is set to 0.25 for both donor and acceptor traps while at lower fluences we find $`\sigma _h^A/\sigma _e^A=0.25`$ and $`\sigma _h^D/\sigma _e^D=1`$ for the acceptor and donor traps, respectively.
## 5 Conclusions
In this paper we show that a model of irradiated silicon sensors based on two defect levels with opposite charge states and trapping of charge carriers can be tuned using charge collection measurements and provides a good description of the measured charge collection profiles in the fluence range from $`0.5\times `$10<sup>14</sup> n<sub>eq</sub>/cm<sup>2</sup> to $`6\times `$10<sup>14</sup> n<sub>eq</sub>/cm<sup>2</sup>.
The model produces an electric field profile across the sensor that has maxima at the implants and a minimum near the detector midplane. This corresponds to negative space charge density near the n<sup>+</sup> implant and and positive space charge density near the p<sup>+</sup> backplane. We find that it is necessary to decrease the ratio of acceptor concentration to donor concentration as the fluence increases. This causes the electric field profile to become more symmetric as the fluence increases.
Given the extracted electric field and space charge density profiles we suggest that the correctness and the physical significance of effective doping densities determined from capacitance-voltage measurements are quite unclear. In addition, we remark that the notion of partly depleted silicon sensors after irradiation is inconsistent with the measured charge collection profiles and with the observed doubly peaked electric fields.
The charge-sharing behavior and resolution functions of many detectors are sensitive to the details of the internal electric field. A known response function is a key element of any reconstruction procedure. A working effective model will permit the detailed response of these detectors to be tracked as they are irradiated in the next generation of accelerators.
## Acknowledgments
We gratefully acknowledge Silvan Streuli from ETH Zurich and Fredy Glaus from PSI for their immense effort with the bump bonding, Federico Ravotti, Maurice Glaser and Michael Moll from CERN for carrying out the irradiation, Kurt Bösiger from the Zürich University workshop for the mechanical construction, György Bencze and Pascal Petiot from CERN for the H2 beam line support and, finally, the whole CERN-SPS team.
|
warning/0506/cs0506014.html
|
ar5iv
|
text
|
# The Equivalence Problem for Deterministic MSO Tree Transducers is Decidable
## String and Tree Transductions
Clearly, Theorem 0.1 also holds if we restrict the input graphs to strings or trees. In particular, deterministic MSO $`X`$-to-$`Y`$ transducers have decidable equivalence for all $`X,Y\{\text{string},\text{tree}\}`$. For string transducers this reproves the decidability result of \[Gur82\] (through \[EH01\]). For trees we obtain the following new decidability result.
###### Corollary 1
The equivalence problem is decidable for deterministic MSO tree transducers.
Of course, even stronger statements hold; namely, given an NR set $`D`$ of strings or trees, it is decidable if two deterministic MSO $`X`$-to-$`Y`$ transducers are equivalent when restricted to $`D`$. For string transducers this means the following.
###### Corollary 2
It is decidable whether two deterministic two-way finite state transducers are equivalent on an NR set of strings.
As discussed in Section 6 of \[Eng97\], the NR sets of strings are the same as the ranges of deterministic tree-walking tree-to-string transducers. They properly contain, for instance, the context-free languages and the ranges of deterministic two-way finite state transducers. Since the NR sets of strings form a full AFL of Parikh languages, Corollary 2 is in fact a special case of the general decidability result for deterministic two-way finite state transducers in Theorem 5 of \[Iba82\]. It is incomparable to the decidability of equivalence of two such transducers on an NPDT0L language \[CK87\].
The two statements of the next corollary follow from the characterizations of deterministic MSO definable tree transductions in \[BE00\] and \[EM03b\], respectively. Note that a tree transducer is of linear size increase if the size of the output tree is at most linear in the size of the input tree.
###### Corollary 3
The equivalence problem is decidable
(1) for single-use restricted attributed tree transducers and
(2) for deterministic macro tree transducers of linear size increase.
This result is incomparable with the decidability of the equivalence problem for nonnested separated attributed/macro tree transducers proved in \[CF82\]. It remains open whether the equivalence problem is decidable for attributed tree transducers and for deterministic macro tree transducers.
In \[MSV03\] the $`k`$-pebble tree transducer was introduced, and claimed to subsume (the tree translation core of) all known XML query languages. Hence, we call deterministic pebble tree transducers deterministic XML queries. Such queries can be simulated by compositions of macro tree transducers \[EM03a\]. If such compositions are of linear size increase, then they are MSO definable \[Man03\].
###### Corollary 4
The equivalence problem is decidable for deterministic XML queries of linear size increase.
|
warning/0506/quant-ph0506013.html
|
ar5iv
|
text
|
# Information Complexity of Quantum Gates
### Introduction
In this paper, we consider complexity and realizability of quantum gates from the point of view of information theory. A gate is a physical system that is controlled by varying some input variables, which are classical. In principle, such a physical system could implement a variety of operators based on the control variables. The gate functions may be also implemented by a single physical system that operates sequentially on the qubits in the quantum register. The complexity of the gate will be defined in terms of the entropy associated with its control. From a practical point of view, one is interested in asking how easy it is to control a gate.
As no analog system can have infinite precision, we investigate what happens if the precision levels at the input and the output are different. The complexity of the gate, defined in terms of entropy, will be examined for the rotation and cnot gates in certain circuits.
### Information processing by gate
One aspect of gate performance is its accuracy. Researchers on quantum information science have given much attention to the question of errors and their correction \[1-3\] by drawing upon parallels with classical information. Quantum error-correction coding works like classical error-correction to correct some large errors.
But the framework of quantum information is distinct from that of classical information. In the classical case, it is implicitly assumed that there occurs an automatic correction of errors that are smaller than a threshold by means of clipping or by the use of a decision circuit. In the case of quantum information, the input data is nominally discrete, but in reality its precision cannot be absolute in any actual realization. Furthermore, unknown small errors in quantum information cannot be corrected \[4-5\]. Consequently, proposals for error correction and fault tolerance (such as \[6-8\]) remain unrealistic.
Classical analog computation and quantum processing do have parallels. In general, fixed errors in gate operation could become irreversible due to actual small nonlinearity of nominally linear elements. Analog computing is not practical to implement because noise cannot be separated from useful signal and it accumulates, degrading the system performance in an uncorrectable manner.
If there were no noise, the practicality of analog computing would depend on the feasibility of the gate implementation over the expected input-output range. This feasibility must be checked in the context of the limitations on information processing by the gate.
Consider the gate G of Figure 1. It may be assumed that it is a physical system which is controlled by means of some variable. This control is implemented by choosing a setting on an instrument, and this choice is associated with random error. If one views the circuit operations to be implemented by the same device transitioning through various states in sequence, then one can determine the distribution of the control variable states, and compute its entropy. This entropy, when determined for the entire computing circuit, may be taken to represent its complexity.
Information is preserved, therefore one can define the following relationship for the entropy expressions for the input $`X`$, the gate control information $`C`$, and the output $`Y`$:
$$H(Y)=H(X)+H(C).$$
(1)
Although it is assumed that the variable $`X`$ is discrete, in reality the lack of perfect precision at the state preparation state makes it a continuous variable \[9-11\]. The lack of precision may not affect the measurement variables, but it would introduce continuous phase error.
Similarly, the output variable $`Y`$ has discrete measurement associated with it, but it may come with additional component states and many unknown, continuous phase terms. This has implications for quantum amplitudes and, consequently, with the probabilities associated with the states.
As an aside, equation (1) provides an explanation for the no-cloning theorem. A gate cannot clone a state since this would require the gate to supply information equal to that of the unknown state, which, by virtue of its being unknown, is impossible.
As the variables $`X`$ and $`Y`$ (defined together with associated continuous phase terms) are continuous, the classical variable $`C`$ must also be continuous. The entropy associated with a continuous variable $`Z`$ is given by the expression:
$$H(Z)=h(Z)\underset{\mathrm{\Delta }z0}{lim}log_2\mathrm{\Delta }z$$
(2)
where $`h(Z)`$ is the differential entropy:
$$h(Z)=_{\mathrm{}}^{\mathrm{}}f_Z(z)log_2\left[\frac{1}{f_Z(z)}\right]𝑑z$$
(3)
and $`\mathrm{\Delta }z`$ is the precision associated with the variable.
If the precision is the same at both input and output, the term $`lim_{\mathrm{\Delta }z0}log_2\mathrm{\Delta }z`$ will cancel out and the differential entropies would be a proper measure of the entropy of $`X`$ and $`Y`$. In other words,
$$H(C)=h(Y)h(X).$$
(4)
The entropy associated with H(C) is the information lost in the computation process and it may be converted to heat according to thermodynamic laws \[12-14\]. If $`H(C)`$ is non-zero, error-free quantum computation is impossible, since this is associated with loss of information.
### Multiplication by constant
Example 1. Consider a gate which multiples the inputs by a fixed constant $`k>1`$. If the input $`X`$ is distributed uniformly over the interval $`(0,a)`$, then the output $`Y`$ is distributed uniformly over $`(0,ka)`$. The differential entropy values of the input and the output are:
$$h(X)=log_2a$$
(5)
$$h(Y)=log_2ka$$
(6)
Assuming the same precision at input and output, the gate needs to supply entropy equal to $`H(C)=log_2kalog_2a=log_2k`$, which become large as $`k`$ increases. This supply of entropy will have to be done in terms of interpolation or other processing which cannot be perfect.
If $`k<1`$, then the output entropy is smaller than input entropy and, therefore, $`H(C)`$ represents loss of information in the output. In effect, the assumption of fixed amplification of a variable with the same absolute precision at the output amounts to a nonlinear, irreversible process. For example, when a picture is compressed, one cannot obtain the original to the earlier precision by amplifying it back. In practical terms, the precision needed for the realization of a universal gate will be unattainable for a variety of reasons: one cannot have perfectly linear behavior in an electrical circuit over an unrestricted range. Unrestricted multiplication of a continuous variable is not implementable if the precision remains unchanged.
In quantum computing, problems that somewhat parallel this above example are the implementation of rotation and cnot gates, two operators that are basic to the computation process . The necessarily classical control of the gate is marred by random errors as well as calibration errors.
### Rotation
Example 2. Consider a quantum gate that rotates the input qubit by a fixed angle. Since the input $`X`$ and the output $`Y`$ will be distributed uniformly over the same interval $`(0,a)`$, the entropy associated with this gate will be $`0`$ (as per equation 4) as is required by the reversible nature of the assumed quantum evolution.
But if the precision associated with the measurement and initialization processes at the input and the output is different, then lossless (or, equivalently, error-free) evolution cannot be assumed.
### cnot and Hadamard gates
Consider the cnot gate together with a companion Hadamard gate. The errors in the device implementation of the cnot gate may make the gate effectively nonlinear and hence nonunitary. The matrix values that the device embodies may be different from the nominal ones below:
$$\left[\begin{array}{cccc}1& 0& 0& 0\\ 0& 1& 0& 0\\ 0& 0& 0& 1\\ 0& 0& 1& 0\end{array}\right]$$
(7)
For simplicity, we consider a very straightforward situation which does not affect the cnot gate, but where its companion Hadamard gate is off the correct value, stuck in the state
$$H_s=\left[\begin{array}{cc}cos\theta & sin\theta \\ sin\theta & cos\theta \end{array}\right]$$
(8)
where $`\theta 45^o`$.
#### Stuck Hadamard gate before a cnot
Example 3. Consider the arrangement of Figure 2, where the stuck gate $`H_s`$ ($`\theta \pi /4`$, but its value is known) is to the left of the cnot gate; this circuit demonstrates that quantum processing can compute a global property of a function by a single measurement .
It will be seen that at the output of the cnot gate, the state is:
$`\frac{1}{\sqrt{2}}(cos\theta |0sin\theta |1)(|0|1)`$
The state $`|a=(cos\theta |0sin\theta |1)`$, which is in error, may be passed through the gate $`\left[\begin{array}{cc}1& 0\\ 0& 1\end{array}\right]`$ followed by another $`H_s`$ to yield $`|0`$, which can be transformed to the correct $`|a=\frac{1}{\sqrt{2}}(|0|1)`$. In this example, the state $`|b=\frac{1}{\sqrt{2}}(|0|1)`$ was not affected by the stuck gate $`H_s`$.
When the stuck gate is the lower Hadamard gate, as in Figure 3, the state at the output of the cnot gate is:
$`\frac{1}{\sqrt{2}}(sin\theta |00cos\theta |01+sin\theta |11cos\theta |10)`$
Corresponding to this we have the density function $`\rho ^{ab}`$ given below:
$`\rho ^{ab}=\frac{1}{2}\left[\begin{array}{cccc}sin^2\theta & sin\theta cos\theta & sin\theta cos\theta & sin^2\theta \\ sin\theta cos\theta & cos^2\theta & \mathrm{cos}^2\theta & sin\theta cos\theta \\ sin\theta cos\theta & \mathrm{cos}^2\theta & \mathrm{cos}^2\theta & sin\theta cos\theta \\ sin^2\theta & sin\theta cos\theta & sin\theta cos\theta & sin^2\theta \end{array}\right]`$
It follows that the reduced density matrix for the state $`|a`$ is:
$`\rho ^a=\frac{1}{2}\left[\begin{array}{cc}1& sin2\theta \\ sin2\theta & 1\end{array}\right]`$
Therefore, when $`\theta \pi /4`$, $`\rho ^{ab}`$ is a mixture, and we cannot perform any local correction to $`|a`$ to obtain the correct product state, for a unitary transformation on a mixture will keep it as a mixture. In other words, this error is not locally correctable.
### Stuck Hadamard gate in the teleportation protocol
Example 4. In the teleportation protocol, an unknown quantum state (of particle $`X`$) is teleported to a remote location using two entangled particles ($`Y`$ and $`Z`$) and classical information. Here, for convenience, we use the variant teleportation protocol which requires only one classical bit in its classical information link (Figure 4). But instead of the Hadamard operator, we consider $`H_s`$ to be the rotation operator with angle $`\theta `$. We assume that the receiver has a copy of $`H_s`$ available for local processing, and we would like to estimate what would happen if this copy is not identical to the one used at the transmitting end.
The state $`X`$ is $`|\varphi =\alpha |0+\beta |1`$, where $`\alpha `$ and $`\beta `$ are unknown amplitudes, and $`Y`$ and $`Z`$ are in the pure entangled state $`\frac{1}{\sqrt{2}}(|00+|11)`$. The initial state of the three particles is:
$`\frac{1}{\sqrt{2}}(\alpha |000+\beta |100+\alpha |011+\beta |111`$
The sequence of steps in Figure 4 is as follows:
1. Apply chained transformations: cnot on $`X`$ and $`Y`$, followed by cnot on $`Y`$ and $`Z`$.
2. Apply $`H_s`$ on the state of $`X`$.
3. Measure the state of $`X`$ and transfer information regarding it.
4. Apply appropriate operator $`G`$ to complete teleportation of the unknown state.
A simple calculation will show that the state before the measurement is:
$`\frac{1}{\sqrt{2}}|0(|0+|1)(\alpha cos\theta |0+\beta sin\theta |1)+\frac{1}{\sqrt{2}}|1(|0+|1)(\alpha sin\theta |0\beta cos\theta |1)`$
Therefore, after the measurement, we get either
$`X^+=\alpha cos\theta |0+\beta sin\theta |1`$
or
$`X^{}=\alpha sin\theta |0\beta cos\theta |1`$
based on whether the measurement was $`0`$ or $`1`$. Assuming that the value of $`\theta `$ is also communicated to it, the receiver can recover the unknown $`X`$ probabilistically; when the value of $`\theta `$ is $`45^o`$, then the inversion is trivially simple.
For simplicity, assume that the receiver needs to invert $`X^+`$. He will replicate Figure 4 at his end which means that the Hadamard gate that he would use would have identical characteristics (the same precision) to the one used during the earlier operation. He would now obtain either
$`X^{++}=\alpha cos^2\theta |0+\beta sin^2\theta |1`$
or
$`X^+=\alpha |0+\beta |1`$
Similarly, $`X^{}`$ will, in the next iteration, lead to:
$`X^+=\alpha |0+\beta |1`$
or
$`X^{}=\alpha sin^2\theta |0+\beta cos^2\theta |1`$
This procedure may be extended, and the probability of recovering the unknown state $`X`$ can be shown to be given by the tree diagram of Figure 5.
In the first pass, there is a fifty percent probability of getting the correct state, and this probability reduces in further passes (Figure 5). The probability of recovering the state $`X`$ is thus:
$$\frac{1}{2}+\frac{1}{2}\frac{1}{4}+\frac{1}{2}\frac{3}{4}\frac{1}{6}+\frac{1}{2}\frac{3}{4}\frac{5}{6}\frac{1}{8}+\frac{1}{2}\frac{3}{4}\frac{5}{6}\frac{7}{8}\frac{1}{10}+\mathrm{}$$
(9)
The ability of the receiver to implement the needed transformation will depend on the precision available in its gate control mechanism. If the value of $`\theta `$ at the sending point is smaller than the precision available to the receiver, then the state $`X`$ cannot be recovered.
It is interesting that as long as the receiver possesses a rotation operator $`H_s`$ that is identical to the one used at the sending point, there is no need to know the value of $`\theta `$ and still obtain the unknown state $`X`$ probabilistically, as in expression (9).
### Conclusion
We have considered the problem of gate complexity in quantum systems. The control of the gate – a physical device – is by modifying some classical variable, which is subject to error. Since one cannot assume infinite precision in any control system, the implications of varying accuracy emongst different gates becomes an important problem.
We have shown that in certain arrangements a stuck fault cannot be reversed down the circuit stream using a single qubit operator, for it converted a pure state into a mixed state.
We considered the case of the teleportation circuit with the rotation gate stuck at $`\theta `$. When $`\theta =0^o`$, the state $`X`$ collapses to $`0`$ or $`1`$. When $`\theta 0^o`$ or $`90^o`$, one may obtain the unknown state back probabilistically by passing $`X^+`$ or $`X^{}`$ back through the circuit of Figure 4 iteratively.
Consider two parties, $`A`$ and $`B`$, who are both presented with the state $`X^+`$ or $`X^{}`$. If the precision available to one of them is greater than or equal to that of the sender, and that of the other is less, then one of them can recover the state, whereas the other cannot.
It is essential that the entropy rate associated with the quantum circuit be smaller than what can be implemented by the information capacity of the controller. This perspective may be useful in evaluating proposals for quantum computing with noisy components.
## References
M.A. Nielsen and I.L. Chuang, Quantum Computation and Quantum Information. Cambridge University Press, 2000.
A.Y. Kitaev, “Quantum computations: algorithms and error correction.” Russ. Math. Surv., 52, 1191-1249 (1997).
E. Knill and R. Laflamme, “A theory of quantum error-correcting codes.” Phys. Rev. A, 55, 900-906 (1997).
S. Kak, “General qubit errors cannot be corrected.” Information Sciences, 152, 195-202 (2003); quant-ph/0206144.
S. Kak, “The initialization problem in quantum computing.” Foundations of Physics, 29, 267-279 (1999); quant-ph/9805002.
A.M. Steane, “Efficient fault-tolerant quantum computing.” Nature, 399, 124-126 (1999).
E. Knill, “Fault tolerant post-selected quantum computation.” Physics Arxiv: quant-ph/0404104.
K. Svore, B.M. Terhal, D. P. DiVincenzo, “Local fault-tolerant quantum computation.” Physics Arxiv: quant-ph/0410047.
S. Kak, “Rotating a qubit.” Information Sciences, 128, 149-154 (2000); quant-ph/9910107.
S. Kak, “Statistical constraints on state preparation for a quantum computer.” Pramana, 57, 683-688 (2001); quant-ph/0010109.
S. Kak, “Are quantum computing models realistic?” Physics Arxiv: quant-ph/0110040.
R. Landauer, “Irreversibility and heat generation in the computing process.” IBM J. Res. Dev., 5, 183 (1961).
C.H. Bennett, “The thermodynamics of computation – a review.” Int. J. Theor. Phys., 21, 905-940 (1982).
S. Kak, “Quantum information in a distributed apparatus.” Foundations of Physics 28, 1005 (1998); Physics Archive: quant-ph/9804047.
(1998); quant-ph/9804047.
D.P. DiVincenzo, “Two-bit gates are universal for quantum computation.” Phys. Rev. A, 51, 1015-1022 (1995).
S. Kak, “Teleportation protocols requiring only one classical bit.” Physics Arxiv: quant-ph/0305085.
E. Knill, “Quantum computing with very noisy devices.” Physics Arxiv: quant-ph/0410199.
|
warning/0506/astro-ph0506314.html
|
ar5iv
|
text
|
# The ACS Virgo Cluster Survey V: SBF Calibration for Giant and Dwarf Early-type Galaxies
## 1 Introduction
The Advanced Camera for Surveys (ACS; Ford et al. 1998) Virgo Cluster Survey is a project based on observations with the ACS on the Hubble Space Telescope (HST), aimed at the exploration of the properties of 100 early–type galaxies in the Virgo Cluster, the study of their globular cluster systems, and the reconstruction of Virgo’s three dimensional structure. To measure the galaxy distances needed to realize these goals, we have used measurements of Surface Brightness Fluctuations (SBF). The SBF method was first proposed by Tonry & Schneider (1988) and it is based on the fact that the variance of normalized poissonian fluctuations of the galaxy stellar population depends on galaxy distance (for reviews see Jacoby et al. 1992 and Blakeslee et al. 1999). At present, SBF measurements have been used to determine galaxy distances out to $``$ 7000 km s<sup>-1</sup>, using data from ground–based telescopes and HST (Ajhar et al. 1997, 2001; Tonry et al. 1997, 2001; Jensen et al. 1999, 2003; Blakeslee et al. 2001, 2002; Mei et al. 2001, 2003; Liu et al. 2001, 2002; Mieske et al. 2003; and references in Mei et al. 2005, hereafter Paper IV).
For the Virgo Cluster in particular, SBF measurements have been published by several groups using data from both ground–based telescopes (Tonry et al. 1990, 2000, 2001; Pahre & Mould 1994; Jensen et al. 1998; Jerjen et al. 2004) and HST (Ajhar et al. 1997; Neilsen & Tsvetanov 2000; Jensen et al. 2003). West & Blakeslee (2000) have used published ground-based SBF data in a first examination of the three-dimensional structure of this cluster, particularly the cluster’s “principal axis” and its relations to the surrounding large-scale structure. Our new ACS dataset (for details, see Côté et al. 2004; Paper I) constitutes the largest, most complete, and most homogeneous SBF sample of Virgo galaxies yet assembled.
Since the absolute magnitude of the SBF varies as a function of the stellar population age and metallicity, SBF distance measurements in a given observational bandpass must be carefully calibrated in terms of stellar population observables, typically the galaxy color. Using an extensive ground-based $`I`$-band SBF survey, Tonry et al. (1997, 2001) have established that it is possible to calibrate the SBF method using the integrated galaxy colors. In particular, they find that for bright ellipticals: 1) the absolute SBF magnitude $`\overline{M}_I`$ is a linear function of the galaxy $`(VI)_0`$ color over the range $`1<(VI)_0<1.3`$ with an internal scatter $`\mathrm{}<0.1`$ mag; 2) the zero-point of this relation is universal over this color range; 3) the slope of the relation is consistent with theoretical predictions from recent stellar population models. For observational bands other than the $`I`$-band, the relation between SBF absolute magnitude and galaxy colors has to be calibrated in that particular band. Ajhar et al. (1997) and Jensen et al. (2003) have calibrated SBF absolute magnitudes as a linear function of galaxy color for use, respectively, with the Wide Field and Planetary Camera-2 (WFPC-2) and NICMOS on the HST.
This paper provides the first calibration for SBF using the F850LP filter with the ACS, i.e, calibration of the absolute F850LP SBF magnitude $`\overline{M}_{850}`$, as a function of dereddened color in the F475W and F850LP filters. Observations and SBF reductions have already been described in Paper I, Jordán et al. (2004; Paper II) and Paper IV. In § 2 the main steps of the SBF reduction are briefly summarized. In § 3 the calibration method is discussed and the results are presented and analyzed. In § 4 the theoretical predictions from Bruzual & Charlot (2003) stellar population models are discussed. We summarize and conclude in § 5. The calibration presented here will be used in future papers to derive SBF distances for the galaxies in our sample and to explore the three-dimensional structure of the Virgo Cluster.
## 2 SBF measurements
### 2.1 Observations and SBF reduction
Our observations and data reduction procedures have been discussed in Paper I and Paper II, respectively. In Paper IV, we have addressed the specific issues related to SBF measurements with the ACS. Since the ACS camera presents significant geometrical distortions, the correction of these distortions involves image resampling by an interpolation kernel. For our SBF measurements we use images geometrically corrected with the drizzle software (Fruchter & Hook 2002) using the Lancsoz3 interpolation kernel. We have shown in Paper IV that, when using this kernel, the distortion correction does not add significant error to the SBF distance measurements. The main steps of SBF measurements are also described in Paper IV. We summarize here the main points.
We measure SBF in the F850LP bandpass of the ACS Wide Field Camera. The process involves fitting a smooth model to each galaxy in both the F850LP and the F475W band (Papers II and IV) and subtracting the models from the original images to produce “residual images”. The residual images in each band are then used to construct catalogs of external sources, i.e. globular clusters and background galaxies. From these catalogs we create a ’source mask’ to remove the contamination from the sources to the surface brightness variance. All sources brighter than a completeness magnitude $`m_{cut}`$, that corresponds to a completeness of $`90\%`$ at a given radius, are masked from the image. The power spectrum of the residuals is then calculated in multiple concentric annuli centered on the galaxy center, by multiplying successive annular mask functions with the ’source mask’ (i.e., the total mask function). The image power spectrum is the sum of two components: a flat, white noise power spectrum and the combined power spectrum of the SBF and undetected external sources, both convolved by the PSF (Point Spread Function) in the spatial domain. In the Fourier domain, this sum can be written as:
$$P(k)=P_0\times E(k)+P_1,$$
(1)
where the “expectation power spectrum” $`E(k)`$ is the convolution of the power spectra of the normalized PSF and of the total mask function of the annular region being analyzed. In Paper IV we have shown that this power spectrum is not significantly modified by pixel correlation induced by our interpolation kernel (see Paper IV for details). Armed with a clearer understanding of the systematic errors arising from the data interpolation, we have measured unbiased values of $`P_0`$ and $`P_1`$ for multiple annuli in each galaxy as described in Paper IV. As the minimum innermost radius, we have taken $`1.5\mathrm{}`$. However when dust is present in the galaxy center, the dust regions are masked (see Paper I), and the minimum radius without dust contamination has been taken as the minimum annular radius. For the maximum outermost annular radius, we have chosen the radius where the galaxy brightness falls to half that of the sky (the range of maximum outermost radii being between $`10\mathrm{}`$ and $`40\mathrm{}`$). The next step is then to extract the SBF signal from $`P_0`$ by subtracting off the residual variance $`P_r`$ from faint sources including undetected globular clusters and distant background galaxies. Because of the exquisite resolution and depth of our ACS images, these correction are very small (see the following section).
### 2.2 Contribution to the SBF from external sources
The contribution to the power spectrum from these external sources is indistinguishable from the SBF contribution, since both are spatially convolved with the PSF. Following Tonry et al. (1990) and Blakeslee & Tonry (1995), we can write
$$P_0=P_{SBF}+P_r$$
(2)
where
$$P_r=\frac{\sigma _{gc}^2+\sigma _{bg}^2}{\text{galaxy}},$$
(3)
$`\sigma _{gc}^2`$ is the contribution to the fluctuations due to globular clusters, and $`\sigma _{bg}^2`$ is the contribution due to background galaxies. The mean galaxy brightness in the denominator is chosen for compatibility with the $`P_0`$ definition above.
To estimate the $`P_r`$ contribution, sources were extracted from the same catalog used for masking external sources from the residual images (see above and Paper IV). Source counts were fitted to a composite globular cluster and external source luminosity function using a maximum likelihood approach. For the former, we adopted a Gaussian luminosity function, where $`N_{gc}`$ is a function of both radius $`r`$ (distance from the galaxy center) and magnitude $`m`$:
$$N_{gc}(m,r)=\frac{N_{gc}^0(r)}{\sqrt{2\pi }\sigma }\mathrm{exp}\left(\frac{(mm_{peak}^{gc})^2}{2\sigma ^2}\right).$$
(4)
For the Virgo Cluster in the F850LP filter, we expect $`m_{peak}^{gc}`$ 23 mag and $`\sigma 1.35`$ Harris (1991); Whitmore et al. (1995); Ferrarese et al. (2000, 2003); pen05. We use these values as our initial estimates but iterate on $`m_{peak}^{gc}`$ to improve the fitted luminosity functions.
For the background galaxies, we assumed a power-law luminosity function:
$$N_{bg}(m)=N_{bg}^010^{\gamma m}$$
(5)
with $`\gamma =0.35`$. From the ACS faint galaxy counts of Benitez et al. (2004) $`\gamma =0.33`$ in the F814W filter. At present, a value for $`\gamma `$ in F850LP is unavailable in the literature. However, the F814W and F850LP bandpasses are quite similar so we expect no strong differences in $`\gamma `$ between these filters. A fit of faint galaxy counts in our ACS blank fields gives us a slope of $`\gamma =0.37`$ (Peng et al. 2005). If we vary $`\gamma `$ in the range 0.33 to 0.37, our fits produce differences of only a few thousandths of a magnitude in the final SBF magnitudes.
In fitting the total source counts ($`N_{ogc}`$ and $`N_{obg}`$), we kept as fixed parameters $`\sigma `$ and $`\gamma `$, and iterated on $`m_{peak}^{gc}`$, as a function of the galaxy distance. Sources identified as bright foreground stars were removed from the images but not included in the fits. From the estimated $`N_{ogc}`$ and $`N_{obg}`$ per pixel, $`P_r`$ was calculated as the sum of Blakeslee & Tonry (1995)
$`\sigma _{gc}^2`$ $`=`$ $`{\displaystyle \frac{1}{2}}N_{ogc}10^{0.8[m_1m_{peak}^{gc}+0.4\sigma ^2\mathrm{ln}(10)]}`$
$`\times `$ $`\text{erfc}[{\displaystyle \frac{m_{cut}m_{peak}^{gc}+0.8\sigma ^2\mathrm{ln}(10)}{\sqrt{2}\sigma }}]`$
and
$$\sigma _{bg}^2=\frac{N_{obg}}{(0.8\gamma )\mathrm{ln}(10)}10^{0.8(m_1m_{cut})+\gamma (m_{cut})}.$$
(7)
where $`m_1`$ is the photometric zero-point on the AB system (24.862 mag for the F850LP; Sirianni et al. 2005) plus reddening corrections (Schlegel et al. 1998) as given in Paper II. An example of the external source luminosity function is shown in Fig 1 for NGC 4472 (VCC 1226). Finally, the $`z`$-band SBF magnitudes are given by
$$\overline{m}_{850}=2.5\mathrm{log}(P_0P_r)+m_1,$$
(8)
The uncertainty on $`\overline{m}_{850}`$ in each galaxy has been calculated by propagating the measurement errors in Eq. 8 for the different annuli. This means that we have added in quadrature the errors due to the uncertainty in the fit of $`P_0`$, the uncertainty in the galaxy flux, and the uncertainty in $`P_r`$ ($``$ 25% $`P_r`$, calculated as the variance in $`P_r`$ when different initial fit parameter are chosen). We neglect the $``$1% systematic uncertainty in $`m_1`$ (Sirianni et al. 2005) for the calibration, since it is common to all galaxies. The total uncertainty on the average $`\overline{m}_{850}`$ for each galaxy has been calculated as the quadrature sum of the errors for the different annuli which we are averaging, divided by the number of annuli.
Finally, since $`\overline{m}_{850}\overline{M}_{850}=5log(d)`$, where $`d`$ is the galaxy distance, two more sources of uncertainty must be added in quadrature to the measurement error in $`\overline{m}_{850}`$ for the purpose of this calibration. First, we include an expected “cosmic scatter” of 0.05 mag in $`\overline{M}_{850}`$ (see Tonry et al. 1997), representing the intrinsic dispersion in $`\overline{M}_{850}`$ for galaxies at the same observed color. Second, an uncertainty of 0.11 mag is introduced to account for the actual distance dispersion from depth effects in the Virgo Cluster. This value has been derived from the rms scatter of $`3^{}`$ in position on the sky of the 100 galaxies in our sample.
## 3 Calibration of the ACS $`\overline{M}_{850}`$
### 3.1 Color measurements
We seek to calibrate $`\overline{M}_{850}`$ as a function of the galaxy color (F475W$``$F850LP), hereafter referred to $`(g_{475}z_{850})`$, since the ACS $`F475W`$ approximates the SDSS $`g`$ and F850LP the SDSS $`z`$. The galaxy colors are measured in the same annuli, and with the same masks, as those used for measuring the SBF. The sky values in the two filters for our final calibration sample are the same as in Paper II. Galaxies for which the sky estimates were especially difficult (due to strong gradients from bright neighboring galaxies) have been excluded for the purposes of this calibration (see below). The errors on the colors have been calculated by summing in quadrature the error in galaxy flux and the error due to sky subtraction. When the colors have been averaged on more annuli, the errors in each one have been added in quadrature, squared and divided by number of annuli. An additional uncertainty of 0.01 mag in color is added in quadrature as an estimate of the error due to imperfect flatfielding (see, e.g., Sirianni et al. 2005).
The colors have been dereddened as explained in Paper II, with the reddening for each galaxy position from Schlegel et al. (1998). The mean reddening for the sample is $`E(BV)=0.029`$ mag, with a standard deviation of 0.008 mag. The adopted extinction corrections are appropriate for elliptical galaxies, $`A_g=3.634`$ $`E(BV)`$ and $`A_z=1.485`$ $`E(BV)`$. Hereafter, we use $`(g_{475}z_{850})`$<sub>0</sub> to refer to the reddening-corrected galaxy color.
### 3.2 Calibration Slope
For the calibration of the $`I`$-band SBF survey, Tonry et al. (1997) divided their sample into galaxy groups and clusters, and assumed that, since the galaxies within an individual group or cluster are at nearly the same distance, SBF magnitudes will simultaneously fit the relation:
$$\overline{m}_I=\overline{m}_I^0+\beta [(VI)_01.15]$$
(9)
where $`\overline{m}_I^0`$ is the group mean value at the chosen fiducial color $`(VI)_0=1.15`$ mag, that was fitted for each group to establish the universality of the SBF calibration. To obtain the absolute SBF magnitude $`\overline{M}_I`$, Tonry et al. (1997) calibrated the zero-point in Eq. 9 with Cepheid distances to galaxies in their groups. Thus, they obtained the following calibration, valid in the color range $`1.0(VI)_01.3`$:
$$\overline{M}_I=(1.74\pm 0.07)+(4.5\pm 0.25)[(VI)_01.15].$$
(10)
The Ajhar et al. (1997) HST/WFPC2 sample spans a range in galaxy $`(VI)`$<sub>0</sub> color between 1.15 and 1.3 mag. Their calibration for the WFPC2 F814W bandpass is
$$\overline{M}_{814}=(1.73\pm 0.07)+(6.5\pm 0.7)[(VI)_01.15].$$
(11)
The Jensen et al. (2003) HST/NICMOS sample in the F160W ($`1.6\mu `$m) filter was calibrated over the range $`1.05(VI)_01.24`$ as:
$$\overline{M}_{160}=(4.86\pm 0.03)+(5.1\pm 0.5)[(VI)_01.16].$$
(12)
These three samples were heavily weighted towards bright, red, early–type galaxies. The Ajhar et al. and Jensen et al. samples (due the small field sizes) were mainly confined to the central regions of bright galaxies, which tend to be quite red. Blakeslee et al. (2001) concluded that this fact explained the steep slope found by Ajhar et al., since the stellar population models predict a steepening of the SBF–color relation for very red populations.
Calibrations of bluer samples of dwarf early-type galaxies have been proposed by Jerjen et al. (1998, 2000, 2003, 2004) and Mieske et al. (2003) using ground–based observations of $`10`$ dwarfs coupled with the predicted color-dependence from stellar population models. However, until now, there has not been a sufficient sample of high-quality SBF measurements in dwarf ellipticals to permit a dependable empirical calibration.
Our ACS VCS sample is the most extensive and homogeneous early–type galaxy sample available for SBF measurements. The galaxies span a range in color of $`0.8(g_{475}z_{850})_01.6`$ mag, corresponding to a range in $`(VI)`$<sub>0</sub> between 0.94 and 1.3 mag (based on the color transformations from Sirianni et al. 2005). We adopt here the same technique that Tonry et al. (1997) have used for galaxy groups, fitting the relation:
$$\overline{m}_{850}=\overline{m}_{F850LP}^0+\beta [(g_{475}z_{850})_01.3]$$
(13)
for the Virgo Cluster.
Among the 100 galaxies in our full sample, 14 have been omitted from our calibration sample because they are edge–on disk or barred galaxies which are especially challenging for SBF measurements, or because of difficult sky subtraction. These galaxies include: VCC 571, VCC 654, VCC 685, VCC 1025, VCC 1125, VCC 1192, VCC 1199, VCC 1327, VCC 1535, VCC 1627, VCC 1720, VCC 1857, VCC 2048, and VCC 2095. One galaxy in the sample, VCC 1499, has $`(g_{475}z_{850})_0<0.8`$, and is too blue to be reliably included in a calibration. There are also several outliers that do not appear to lie on the average SBF magnitude versus color relation for the cluster. Fig. 2 shows the SBF and color measurements for the remaining 85 galaxies.
To calibrate our SBF measurements in the F850LP filter as a function of the average galaxy $`(g_{475}z_{850})_0`$ color, we calculate average SBF measurements for each galaxy and fit the relation in Eq. 13 to average galaxy colors, as Tonry et al. (1997) have done for each group in their sample. All galaxies included in the fit have an uncertainty on $`\overline{m}_{850}`$ less than 0.15 mag (not considering the uncertainty due to the ‘cosmic’ scatter on the SBF calibration; see above). Errors on average galaxy colors are typically $``$0.01 mag. Both uncertainties on magnitude and color are included in the $`\chi ^2`$ calculations below.
To exclude the outliers from the final calibration, we first make a linear fit to Eq. 13 to the overall sample with a robust least absolute deviation technique (Press et al. 1992). We obtain $`\overline{m}_{F850LP}^0=29.09\pm 0.02`$ mag and $`\beta =1.25\pm 0.10`$. We identify four outlier galaxies from iterative 3-$`\sigma `$ clipping. The four 3–$`\sigma `$ outliers are: VCC 21, VCC 538, VCC 575, and VCC 731. Of these, VCC 21 contains numerous bright young star clusters, indicative of recent star formation, while VCC 731 (NGC 4365) is known from the ground-based SBF survey to be a Virgo member which lies on the far side of the cluster; we suspect that the other two may be similar cases. After excluding these four outliers and the very blue VCC 1499, our final sample consists of 82 galaxies.
In Fig. 3 our final galaxy sample and an overall fit (the continuous straight line) are shown. Both uncertainties on magnitude and color have been included in the fit $`\chi ^2`$ estimation. We obtain:
$$\overline{m}_{850}=(29.09\pm 0.02)+(1.3\pm 0.1)\times [(g_{475}z_{850})_01.3]$$
in the overall color range $`0.9(g_{475}z_{850})_01.6`$. This color range includes both bright and faint galaxies. Various lines of evidence have suggested that the slope of the $`I`$-band SBF–color relation is steeper for redder galaxies: for instance, the steep empirical slope found by Ajhar et al. (1997) for the red centers of bright ellipticals, which was also reproduced by the reddest composite stellar population models of Blakeslee et al. (2001). Conversely, the few dwarf galaxies in the Tonry et al. (1997) sample indicated a significantly shallower slope at the blue end, and the SBF measurements for Galactic globular clusters (Ajhar & Tonry 1994) showed no dependence on color. Most recent models also show a flattening of the relation for bluer populations.
Our sample, more clearly than any previous study, suggests that changes in color range affect the fitted slope, so we have studied the stability of the above relation as a function of the lower color limit, $`(g_{475}z_{850})_{cut}`$. When we repeat the fit in different ranges $`(g_{475}z_{850})_{cut}<(g_{475}z_{850})_01.6`$, we obtain the results shown in Fig 4. On the top of the figure the fitted slope $`\beta `$ is shown as a function of $`(g_{475}z_{850})_{cut}`$. On the bottom is shown the corresponding value of $`\chi ^2`$. Red galaxies tend to have slopes that are steeper ($``$ 2) than those obtained by fitting to the overall sample. This suggests that our sample would be better represented by two distinct samples of red and blue galaxies, each one having its own calibration slope.
We fit Eq. 13 in two complementary color ranges : $`0.9<(g_{475}z_{850})_0(g_{475}z_{850})_{div}`$ (blue) and $`(g_{475}z_{850})_{div}<(g_{475}z_{850})_01.6`$ (red), with $`1.15(g_{475}z_{850})_{div}1.35`$ (as suggested from Fig. 4). In Fig. 5 the fitted slopes (top) and the respective $`\chi ^2`$ (bottom) are shown as a function of $`(g_{475}z_{850})_{div}`$. Dot-dashed lines correspond to the fit in the color range $`(g_{475}z_{850})_{div}<(g_{475}z_{850})_01.6`$ and dashed lines to the fit in the color range $`0.9<(g_{475}z_{850})_0(g_{475}z_{850})_{div}`$. The continuous line is the total $`\chi ^2`$ from both fits. The minimum total $`\chi ^2`$ is obtained at $`(g_{475}z_{850})_{div}=1.26`$, where the red slope is $`1.95\pm 0.24`$ and the blue slope is $`0.90\pm 0.23`$. The two slopes cross at $`(g_{475}z_{850})_{div}1.3`$, which we take to be the color at which the two regimes separate. The $`\chi ^2`$ for the total fit is minimized for both slope and zero-point, and the errors in both magnitude and color have been taken into account.
As in Tonry et al. (1997), we find the SBF in S0 and ellipticals to follow the same trend as a function of $`(g_{475}z_{850})_0`$. Interestingly, as Figs. 2 and 5 show, the dividing color of $`(g_{475}z_{850})_0=1.3`$ is close to that which separates the two most basic galaxy types in our survey: giant and dwarf early-type galaxies. In fact, in our final sample, of 37 galaxies with color $`(g_{475}z_{850})_01.3`$, only one is classified as dwarf. For $`1.2(g_{475}z_{850})_01.3`$, 10 galaxies are dwarfs and 8 are giants. For $`(g_{475}z_{850})_01.2`$, of 26 galaxies, 20 are dwarfs. Physically then, the two different calibration slopes correspond to different galaxy types (giants corresponding to the red end, and dwarfs to the blue end).
Jerjen et al (2000, 2004) adopted a theoretical calibration to dwarf elliptical galaxies, based on a two branch prediction in the R–band SBF versus the $`(BR)_0`$ galaxy color from Worthey et al. (1994) theoretical stellar population models using the Padova isochrones. While the scatter becomes larger for the bluest dwarfs, we see no evidence for distinct branches in our large data set; nor do the Bruzual & Charlot (2003) models predict such branches in $`\overline{z}_{850}`$ vs $`(g_{475}z_{850})_0`$.
Our sample is sufficiently complete at the blue end that we are able to provide, for the first time in SBF studies, an empirical calibration of SBF in early–type dwarf galaxies.
Splitting the galaxy sample at $`(g_{475}z_{850})_0=1.3`$, we derive the following calibration for $`\overline{m}_{850}`$:
$$\overline{m}_{850}=29.03\pm 0.03+(2.0\pm 0.2)\times [(g_{475}z_{850})_01.3]$$
in the range $`1.3<(g_{475}z_{850})_0<1.6`$, and
$$\overline{m}_{850}=29.03\pm 0.03+(0.9\pm 0.2)\times [(g_{475}z_{850})_01.3]$$
in the range $`0.9<(g_{475}z_{850})_01.3`$. The reduced $`\chi ^2`$ values for the red and blue subsamples are 0.99 and 0.8, respectively. The value of the total reduced $`\chi ^2`$ is calculated as the appropriately weighted sum of the $`\chi ^2`$ values from the two independent fits. When normalized by the total number of objects used in the fits, its value is 0.9. The reduced $`\chi ^2`$ for the blue subsample is lower than that for the red sample. This might be due to the difficulty in accounting for the intrinsic distance dispersion within Virgo, or to overly conservative error estimates for the dwarfs. One of the outliers rejected above, VCC 21, is no longer an outlier with respect to the final, shallower, blue-end calibration, and we have added it to the final calibration above, by iteration.
We show in Fig. 6 the average and standard deviation of $`\overline{m}_{850}`$ in $`(g_{475}z_{850})_0`$ color bins (top) and the color histogram (bottom) of our final calibration sample (82 galaxies, plus VCC 21). From the scatter in $`\overline{m}_{850}`$, we are aware that our calibration has a large scatter for colors $`(g_{475}z_{850})_0<1`$. We will limit our absolute magnitude calibration to the color range $`1.0<(g_{475}z_{850})_0<1.6`$.
We observe a similar behavior of $`m_{850}`$ as a function of the $`(g_{475}z_{850})_0`$ color for the multiple annuli within individual galaxies. However, the precise value of the slope is less robust and more sensitive to errors in the sky.
### 3.3 Calibration Zero-Point
Empirical calibration of the zero-point of the SBF relation requires computing the SBF absolute magnitude $`\overline{M}_{F850LP}`$ from independent distance measurements. The ideal situation would be to have Cepheid distances for multiple galaxies in our sample, however there are no galaxies in our sample with Cepheid distances. We thus calibrate our zero-points to the Virgo Cluster distance modulus found in the $`I`$-band SBF survey of Tonry et al. (2001), but revised in accordance with the final set of HST Key Project Cepheid distances from Freedman et al. (2001). Tonry et al. (2001; Table 4) reported $`31.15\pm 0.03`$ mag as the Virgo SBF distance modulus, where the SBF zero-point was calibrated from SBF measurements for individual early-type spiral galaxies with Cepheid distances given by Ferrarese et al. (2000).
However, using the final set of Key Project distances from Freedman et al. (2001), who adopted the Udalski et al. (1995) period-luminosity relation and added a metallicity correction, the Tonry et al. (2001) distance modulus for Virgo has to be corrected with a shift of $``$0.06 mag, and becomes $`31.09\pm 0.03`$ mag (see discussions by Ajhar et al. 2001; Blakeslee et al. 2002). Here, the errorbar reflects the internal error on the mean Virgo SBF distance; the uncertainties in tying the ground-based SBF distances to the Cepheid distance scale, and in the zero-point of the Cepheid distance scale itself, contribute another $``$0.15 mag systematic error. If the metallicity corrections to the Cepheid distances tabulated by Freedman et al. are not applied, then the Tonry et al. (2001) Virgo distance modulus would shift instead by $`0.16`$ mag to $`30.99\pm 0.03`$ mag (see Jensen et al. 2003). For our SBF calibration, we will adopt $`31.09\pm 0.03`$ mag as the Virgo distance modulus.
Thus, scaling the zero-points from Eq.3.2, we obtain
$$\overline{M}_{850}=2.06\pm 0.04+(2.0\pm 0.2)\times [(g_{475}z_{850})_01.3]$$
in the range $`1.3<(g_{475}z_{850})_01.6`$, and
$$\overline{M}_{850}=2.06\pm 0.04+(0.9\pm 0.2)\times [(g_{475}z_{850})_01.3]$$
in the range $`1.0(g_{475}z_{850})_01.3`$.
## 4 Predictions from stellar population models
We now compare our results to predictions from the Bruzual & Charlot (2003) stellar population models, which we have calculated from the models and stellar luminosity functions kindly provided by S. Charlot. In Fig. 7 we show our data as compared to Bruzual & Charlot (2003) model predictions for different stellar populations. The lines connect models of the same metallicity: the red line is for solar metallicity, the green for 40% solar, the blue for 20% solar, the light blue for 2% solar, and the orange for 2.5 times solar metallicity. The dots on the lines represent, respectively from left to right, ages of 1, 3, 5, 8, 12 and 15 Gyr.
The Bruzual & Charlot models predict $`\overline{M}_{850}`$. To compare these predictions with the data, we add to the theoretical $`\overline{M}_{850}`$ a distance modulus such that the predicted average $`\overline{m}_{850}`$ for solar and 40% solar metallicity at $`(g_{475}z_{850})_0=1.2`$ mag is equal to the average observed $`\overline{m}_{850}`$ at the same $`(g_{475}z_{850})_0`$ for our calibration sample. This procedure yields a theoretically predicted SBF Virgo Cluster distance modulus of $`\overline{M}_{850}\overline{m}_{850}=30.94`$ mag, which is entirely independent of the Cepheid distance scale, but we caution that it depends on the color range we have chosen for shifting the models into agreement with the data.
This theoretically calibrated SBF Virgo Cluster distance modulus differs by $``$0.15 mag with respect to the revised Tonry et al. (2001) Virgo distance modulus of 31.09 mag adopted above. However, the two are consistent within the uncertainties from the zero-point of the Cepheid distance scale, the anchoring of the ground-based SBF measurements to the Cepheids, and the various uncertainties in the stellar population models (see Blakeslee et al. 2001 for a discussion). It is also worth noting that if we had adopted the Virgo distance modulus of $`30.99\pm 0.03`$ obtained by calibrating SBF to the Freedman et al. (2001) Cepheid distances that use the new period–luminosity calibration but omit the metallicity correction, then the difference with respect to the model-predicted distance modulus would be only $`0.05`$ mag.
Our sample of early-type galaxies spans a wide range in metallicity and, presumably, age. Like the observations, the theoretical predictions also show a steeper slope at the red end, and a shallower at the blue end. The slopes predicted from linear fits to the overall model predictions are $`0.0\pm 0.5`$ for $`0.9<(g_{475}z_{850})_01.25`$ mag, $`2.6\pm 0.9`$ for $`1.25<(g_{475}z_{850})_01.6`$ mag, and $`1.26\pm 0.3`$ for $`0.9<(g_{475}z_{850})_01.6`$ mag (in very good agreement with our empirical slope over the full range). If we fit only metallicities less than or equal to solar, at the red end the slope will be $`2\pm 1`$, compared to $`0.4\pm 0.4`$ in the blue. The agreement between theoretical predictions and our empirical calibration is somewhat better at the red end. This is likely due to better accuracy in the modeling of highly evolved populations. The SBF predictions for young to intermediate-age populations is less certain due mainly to the uncertainties in the modeling of the asymptotic giant branch contributions. Furthermore, models of old stellar populations are better constrained by observations of Galactic globular clusters. For these reasons, we prefer our empirical blue-end slope, although additional SBF measurements of large homogeneous samples of dwarf galaxies (e.g., in the Fornax cluster) would be helpful in refining the SBF calibration for blue early-type galaxies.
Considering the uncertainty in the Virgo distance modulus, we conclude that our $`z_{850}`$ SBF data for the giant early-type galaxies are well reproduced by the models for stellar populations ages greater than 3 Gyr and metallicities in the range between 40% and 250% solar. The data for the dwarf galaxies are consistent with either much younger (less than 3 Gyr) solar metallicity populations or intermediate-to-old stellar populations with metallicities in the range 20%–40% solar.
## 5 Conclusions
As part of the ACS Virgo Cluster Survey, we have observed 100 early–type galaxies in the Virgo Cluster. We have measured SBF magnitudes in these galaxies with the aim of determining their distances and the three-dimensional structure of Virgo. We have presented the first calibration of the SBF method using the ACS instrument, giving the F850W SBF magnitude as a function of galaxy $`(g_{475}z_{850})_0`$ color. Obvious outliers, most likely due to a real dispersion in distance for galaxies in the Virgo Cluster region, have been omitted from this calibration, leaving 82 galaxies with $`(g_{475}z_{850})_0`$ colors ranging between 0.9 and 1.6 mag. Over this color range, the blue (mainly dwarf) and red (mainly giant) galaxies follow different linear relations, with the slope being shallower at the blue end. For the first time in SBF studies, our sample allows us to empirically calibrate SBF for dwarf early–type galaxies.
By comparing with stellar population models from Bruzual & Charlot (2003), we find that the slope of the model prediction for $`\overline{m}_{850}`$ with $`(g_{475}z_{850})`$<sub>0</sub> color is highly consistent with the observed relation for the red galaxy subsample. There is also general consistency at the blue end, though the stellar population models with young ages and low-metallicities predict a somewhat shallower slope than observed. This underscores the importance of empirical calibrations, as well as more realistic composite populations in any effort to model the behavior of the SBF method theoretically, especially for dwarf galaxies, for which the scatter in stellar population properties is larger. A forthcoming paper will present the distances to the individual galaxies derived with this calibration, and examine the issue of Virgo’s three dimensional structure and its implications for galaxy and globular cluster properties.
We thank Stéphane Charlot for providing theoretical stellar population models in the ACS filters and Gerhard Meurer for useful discussions. Support for program GO-9401 was provided through a grant from the Space Telescope Science Institute, which is operated by the Association of Universities for Research in Astronomy, Inc., under NASA contract NAS5-26555. ACS was developed under NASA contract NAS 5-32865. S.M. and J.P.B. acknowledge additional support from NASA grant NAG5-7697 to the ACS Team. P.C. acknowledges support provided by NASA LTSA grant NAG5-11714. M.J.W. acknowledges support through NSF grant AST-0205960. D.M. acknowledges support provided by NSF grants AST-0071099, AST-0206031, AST-0420920 and AST-0437519, by NASA grant NNG04GJ48G, and by grant HST-AR-09519.01-A from STScI. M.M. acknowledges support from the Sherman M. Fairchild foundation. This research has made use of the NASA/IPAC Extragalactic Database (NED) which is operated by the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration.
|
warning/0506/astro-ph0506362.html
|
ar5iv
|
text
|
# On the binary nature of 1RXS J162848.1-415241
## 1 INTRODUCTION
X-ray binaries are binary systems where a neutron star or a black hole accretes matter from its companion star. Of $`280`$ known X-ray binaries, fifteen show persistent or episodic relativistic radio jets and are called microquasars as they are reminiscent of AGN (see e.g. Paredes 2005 for an inventory and Mirabel & Rodríguez 1999, Fender 2004a for a review of their properties). Microquasars are of great interest because they may provide a unique opportunity to gain insight into the mechanisms ruling the formation and evolution of relativistic outflows coupled to accretion phenomena. In this regard, the variations observed in microquasars are perceptible on human timescales, which allows studies impossible to perform for the more plentiful (but slowly varying) extragalactic relativistic jet sources. The sample of currently known microquasars is however small, limiting any robust conclusions of how the jet formation and its properties depend on parameters such as the accretion rate, accretion geometry or the mass and spin of the compact object (see e.g. Garcia et al. 2003). In an attempt to enlarge the sample, two long-term programs have been carried out to identify microquasar candidates in the Galaxy by cross-correlating the ROSAT All-Sky Survey (RASS) Bright Source catalogue (Voges et al. 1999) with the NVSS, GB6 and/or PMN radio surveys (Condon et al. 1998, Gregory et al. 1996, Griffith et al. 1994). Both programs share similar selection criteria: positionally coincident X-ray and radio sources that are not extended and that have hard X-ray spectra typical of X-ray binaries (see Tsarevsky et al. 2002 and Paredes, Ribó & Martí 2002 for more details). Once the optical counterpart is identified, spectroscopic observations are carried out to establish the nature of the source, and further follow-up observations are initiated to confirm its candidacy.
1RXS J162848.1-415241 (hereafter J1628) was detected with the Position Sensitive Proportional Counter (PSPC) during the ROSAT All-Sky Survey at a count rate of 0.119 c s<sup>-1</sup> (0.1-2.4 keV). This X-ray source was reported to be the best microquasar candidate found during one of the surveys for new microquasars in the Galactic plane (Tsarevsky et al. 2001, 2002). Tsarevsky et al. selected this ROSAT object following the criteria outlined above, i.e: J1628 is a point-like X-ray source with an X-ray spectral hardness typical of X-ray binaries (Motch et al. 1998) and it is associated with a variable radio source (Tsarevsky et al. 2001; Rupen, Mioduszewski & Dhawan 2002, 2004; Slee et al. 2002). The optical counterpart to J1628 was identified with the apparently variable stellar-like source GSC 07861-01088 (V=13.4). Low resolution optical spectroscopy showed a K-type spectrum with a variable H$`\alpha `$ emission profile (Tsarevsky et al. 2001). The photometric variability of J1628 was confirmed by Buxton et al. (2004) who observed a 0.30 mag amplitude variation in the V-band light curve. On the basis of all the gathered photometry, they found a periodicity of $`4.9364\pm 0.0018`$ d (Tsarevsky; private communication). Buxton et al. interpreted the photometric variations as arising from a double-humped ellipsoidal light curve (as is typical in X-ray novae in quiescence) or from a pulsating star.
During the 2002-2004 seasons, J1628 was observed sporadically during the spectroscopic programs conducted by several CfA astronomers observing at Las Campanas Observatory (LCO). The aim of these observations was to confirm the suggested microquasar nature of J1628 by deriving its binary parameters from the analysis of the spectroscopic data accrued during these three years. The spectra obtained during 2002 showed clear radial velocity variations from night to night.
The binary nature of J1628 was confirmed one year later upon the acquisition of echelle spectra which showed highly broadened photospheric lines due to the stellar rotation: it is well-known that rapid rotation in G and K-type stars is a sign of a close binary where tidal effects have synchronized stellar rotation and orbital revolution. Of course, FK Comae stars (Bopp & Stencel 1981) and young single stars that have not yet been spun down by magnetic braking are also fast rotators; however, they do not show radial velocity variations. A pulsating nature for J1628 is ruled out because the only Population I pulsating stars with low-amplitude ($`\mathrm{\Delta }V<1`$ mag) light curves are Classical Cepheids with long periods; although these stars do show a K-type spectra when they reach luminosity minimum (Efremov 1975, Petit 1985), they are known to be slow rotators (Kraft 1966).
In this paper we make a detailed analysis of the optical spectroscopy of J1628, updating the preliminary results presented in Torres et al. (2004a). The paper is structured as follows: Section 2 presents the observations and the data reduction procedure. In Section 3 we estimate the orbital ephemeris and in Section 4 discuss the spectrum of the secondary star. Note that hereafter, we will call the visible star in the binary the secondary and the invisible stellar component the primary, in analogy with the terminology used in the study of binaries harboring compact objects. In Section 5 we derive an upper limit to the extinction toward J1628. Section 6 describes the emission lines observed in the spectrum. Finally, in Section 7 we discuss our results and a summary is given in Section 8.
## 2 OBSERVATIONS AND DATA REDUCTION
The observations of J1628 were obtained using different spectrographs mounted on the 6.5m Magellan Baade and Clay telescopes at LCO.
The Boller and Chivens (B&C) spectrograph was used to acquire low resolution spectra. A 0.7 arcsec slit width and a 600 (1200) line mm<sup>-1</sup> grating yielded a spectral resolution of 2 pixels and a dispersion of 1.56 (0.80) Å pix<sup>-1</sup>. The spectral interval covered by the B&C was set depending on the requirements of the observer’s scientific program and can be found in Table 1.
The Inamori-Magellan Areal Camera and Spectrograph (IMACS; Bigelow & Dressler 2003) was employed in short-camera mode to obtain intermediate resolution spectra dispersed along the long-axis of two of the 8 SITe CCDs in the IMACS detector. Using a 600 line mm<sup>-1</sup> grism and a 0.5 arcsec slit width yielded a dispersion of 0.48 Å pix<sup>-1</sup> (CCD #2) and 0.58 Å pix<sup>-1</sup> (CCD #5) in the spectral intervals 5670-7620 Å and 7695-10025 Å respectively. The spectral resolution was $`2`$ pixels FWHM.
High-resolution spectra were acquired with the Magellan Inamori Kyocera Echelle (MIKE; Bernstein et al. 2003). During May 2002, data were obtained using a 0.7 arcsec slit and the CCD detector binned by 2 in the spatial direction. In the blue, the useful wavelength range covered from 3360 to 4700 Å over 32 echelle orders. The dispersion was of 0.017-0.023 Å pix<sup>-1</sup> and the spectral resolution $`5.4`$ pixels FWHM. In the red, the useful wavelength range covered the 4775-8500 Å interval over 31 echelle orders with a dispersion varying between 0.034 and 0.060 Å pix<sup>-1</sup>. The spectral resolution was $`4.5`$ pixels FWHM. In July 2004 a new dichroic and CCD were available. High-resolution spectra were obtained using again a 0.7 arcsec slit, but this time the CCD detector was binned by 2 both in the spatial and spectral direction. With this configuration the wavelength interval 3325-5070 Å was covered over 35 orders and the interval 4705-7260 Å over 25 orders. The 0.7 arcsec slit yielded a dispersion of 0.033-0.050 Å pix<sup>-1</sup> and a spectral resolution of 2.7 pixels FWHM in the blue and 0.066-0.10 Å pix<sup>-1</sup> and 2.2 pixels FWHM in the red.
Table 1 provides a detailed journal of the J1628 observations. In addition to the spectra of J1628, spectra of several radial velocity standards, flux standards and other stars were acquired during the observations.
The B&C and IMACS images were bias and flat-field corrected with standard iraf<sup>1</sup><sup>1</sup>1iraf is distributed by the National Optical Astronomy Observatories. routines. The spectra were extracted from each CCD frame with the iraf kpnoslit package. The pixel-to-wavelength calibration was derived from cubic spline fits to HeNe or HeNeAr arc lines. The root-mean square deviation of the fit was $`0.07`$ Å and $`<0.02`$ Å for the data acquired with the B&C and IMACS respectively. Checks for the stability of the wavelength calibration were made using the strongest atmospheric emission lines present in the spectrum. For the spectra acquired with B&C we made use of the \[O i\] 5577.34 Å line and estimated an accuracy in the wavelength calibration $`0.30`$ Å (600 line mm<sup>-1</sup> grating) and $`<0.08`$ Å (1200 line mm<sup>-1</sup> grating). In the case of IMACS, we made use of the \[O i\] 6300.3 Å line and the OH emission blend at 7316.3 Å (Osterbrock et al. 1996). The accuracy estimated in this way was $`0.08`$ Å for the wavelength calibration of the spectra recorded in CCD #2.
The MIKE data were processed with iraf and the spectra extracted with the iraf echelle package and the aid of iraf tasks developed and kindly provided by Professor Jack Baldwin. A dispersion solution was derived from a two-dimensional fit to the ThAr comparison lamp spectra. More than 400 lines were included in the fit, and residuals were typically $`<0.0012`$ Å in the blue and $`<0.0023`$ Å in the red. No corrections for terrestrial lines were applied to the data. The atmospheric NaD emission at $`\lambda \lambda 5889.95,5895.92`$ and the atmospheric O<sub>2</sub> bands with wavelengths provided in Pierce & Breckinridge (1973) were used to establish an accuracy of the fit $`<0.02`$ Å.
## 3 RADIAL VELOCITY MEASUREMENTS
The first task was to measure the radial velocities of the optical counterpart. These were measured from the spectra by the method of cross-correlation with a template star (Tonry & Davis 1979). Because no common template star was observed with all setups, we decided to use for the cross-correlation the spectrum of a template star from the Indo-US library of Coudé feed stellar spectra (Valdes et al. 2004). This library contains spectra of 1273 stars at a dispersion of $`0.4`$ Å pix<sup>-1</sup> and 1 Å FWHM resolution. On basis of the preliminary results for the spectral class of J1628 (Torres et al. 2004a), we selected the spectrum of a K3III star (HD169191) for the cross-correlation. Prior to the cross-correlation, the B&C and IMACS spectra were resampled onto a logarithmic wavelength scale and normalized by dividing with the result of fitting a low order spline to the continuum. The archival K3III spectrum was resampled onto the same logarithmic wavelength scale as for the B&C and IMACS target spectra, broadened to match the IMACS/B&C spectral resolution using a Gaussian function with the appropriate width and normalized to the continuum. In the case of the MIKE spectra, we chose two orders covering the wavelength intervals $`\lambda \lambda 60556225`$ and $`\lambda \lambda 64006580`$ to measure the radial velocities. These spectral ranges contain a number of moderately strong lines and blends useful not only for the radial velocity measurements, but also for the spectral-type/luminosity classification and rotational broadening measurement (see next section). Both orders were rebinned to match the template logarithmic wavelength scale and rectified to the continuum as explained above.
Individual velocities were extracted by cross-correlation with the archival template star in the range $`\lambda \lambda 49005560`$ (B&C; Oct 2002), $`\lambda \lambda 49906540`$ after masking the interstellar NaD lines (B&C; 12 May 2003), $`\lambda \lambda 64006545`$ (IMACS) and after masking the H$`\alpha `$ emission line in the MIKE data. To obtain an estimation of the systematic errors caused by the use of the archival template we proceeded as follows: first, we obtained the radial velocities for each night (when more than one target spectrum was available) using now a template spectrum observed during the respective night. Next, we calculated for each night the radial velocity differences and compared the values (night per night) with the radial velocity differences obtained using the archival template. We found in this way rms errors of 2 km s<sup>-1</sup> and a maximum deviation of 5 km s<sup>-1</sup>.
We performed least-squares sine fits to our radial velocity data using the photometric period P<sub>ph</sub>=4.9364 d and 2$`\times `$P<sub>ph</sub> as initial guesses for the orbital period. A spectroscopic period of 4.93956 $`\pm `$ 0.00025 d provided the lowest value of the $`\chi ^2`$ per degree of freedom, being 5 times lower than that obtained for 2$`\times `$P<sub>ph</sub>. This value is consistent with the deepest minimum at $`0.2`$ cycle d<sup>-1</sup> observed in the $`\chi ^2`$-periodogram of the radial velocities. Furthermore, this period agrees with the photometric period within 1.8-$`\sigma `$. Nevertheless, the orbital period is not determined unambiguously because the spectroscopic data set is also consistent with several alias periods. In this work we adopt the following parameters of the radial velocity curve:
* $`K_2=33.3\pm 0.6`$ km s<sup>-1</sup>
* $`\gamma =14.7\pm 0.6`$ km s<sup>-1</sup>
* P<sub>sp</sub>=4.93956 $`\pm `$ 0.00025 d
* $`T_0=HJD2452575.11\pm 0.03`$
The zero phase is defined as the time of closest approach of the secondary to the observer. All quoted uncertainties are 1-$`\sigma `$ and were obtained after increasing the error in the radial velocities in order to give $`\chi _\nu ^2=1`$. The phase-folded radial velocity curve is shown in Figure 1. Orbital phases of the observations, radial velocity measurements with their associated statistical errors and residuals to the circular orbit solution are given in Table 2. Adopting the values of $`K_2`$ and the orbital period, a mass function of $`f(M{}_{1}{}^{})=0.019\pm 0.001`$ M is obtained.
A 4.9 d orbital period rules out an ellipsoidal modulation of the optical light curve (Buxton et al. 2004) and suggests a chromospherically active binary scenario for J1628 where an evolved stellar component is covered with starspots that modulate the photospheric light with stellar rotation. The 0.30 mag modulation observed in the V-band light curve is at the high end for chromospherically active binaries: only thirteen of 206 binaries listed in Strassmeier et al. (1993) have shown a modulation in the V-band with an amplitude $`0.3`$ mag.
## 4 Spectral Type, Luminosity Class and Rotational Broadening
In a preliminary report based on the analysis of the first B&C and MIKE data sets (Torres et al. 2004a), the spectra of J1628 were compared visually with the spectra of K subgiants and main sequence stars acquired during the observations and also compared with the spectra of G and K-type V/IV/III stars from the Indo-Us Library of Coudé feed stellar spectra. Based on the depth of the TiO bands at $`\lambda \lambda 60806390`$, a spectral type in the range K3$`\pm `$1 was supported. A luminosity class III-IV was suggested because of the strength of the Cai $`\lambda 6450`$ line with respect to the metallic lines in the interval $`\lambda \lambda 64206530`$ and by comparing the relative intensities of the multiplet 33 Tii lines $`\lambda \lambda 8379,8382`$ and Fei $`\lambda 8388`$.
To verify our visual classification and determine the rotational broadening of the secondary star, we made a more detailed analysis of the spectra acquired with MIKE during 2004 as they have a higher signal-to-noise in the red than the 2003 MIKE data due to the binning in the dispersion direction and the use of a new dichroic. We focused our analysis in the wavelength interval $`\lambda \lambda 63506530`$ where there are several temperature and gravity sensitive lines for F, G and K stars (Strassmeier & Fekel 1990; Strassmeier & Schordan 2000). We used the technique outlined in Marsh, Robinson & Wood (1994) that is based on the search of the lowest residual obtained when subtracting a set of templates from the Doppler-corrected average spectrum of the target. The template spectra are broadened prior to subtraction to determine the rotational broadening of the lines. This technique also allows the possibility of a continuum contribution from an accretion flow when searching for the parameters of the secondary star. To implement this procedure we used a set of K-dwarf and subgiant template spectra taken with MIKE during 2004 and a set of K-giant spectra observed with MIKE in March 2003 (see Martini & Ho 2004).
We proceeded as follows: first, the target spectra were Doppler-corrected to the rest frame of the secondary star by subtracting the radial velocity obtained from the cross-correlation with the template. Next, we produced an average spectrum after assigning different weights to the individual spectra in order to maximize the signal-to-noise of the sum. The template spectra were then broadened from 37 to 47 km s<sup>-1</sup> in steps of 0.05 km s<sup>-1</sup> through convolution with the rotational profile of Gray (1992). We adopted a linearized limb-darkening coefficient of $`0.65`$. Each broadened version of the template spectrum was multiplied by a factor $`f`$ (representing the fractional contribution of light from the secondary star) and subtracted from the target Doppler-corrected average. Then a $`\chi ^2`$ test on the residuals was performed in the range $`\lambda \lambda 64006548`$ and the optimal values of $`f`$ and $`v\mathrm{sin}i`$ were provided by minimizing $`\chi ^2`$. The results from the $`\chi _\nu ^2`$ minimization are listed in Table 3, with quoted uncertainties corresponding to $`\chi _{\mathrm{min}}^2`$+1 (Lampton, Margon & Bowyer 1976). The minimization of $`\chi _\nu ^2`$ in the V/IV luminosity class templates shows that the spectral type of the secondary star in J1628 is most likely not later than K3. Additionally, for a single-lined chromospherically active binary star the fractional contribution of the secondary must be $`1.0`$. This rules out a main sequence star and constrains the spectral type to be K3IV/III. The high $`f`$ value as determined from the $`\chi ^2`$ minimization using the IV/III templates together with the fact that the primary star is not detected in the nightly averaged spectra (which have signal-to-noise of about 70) suggest that the visual luminosity of the primary star and/or the accretion flow can be at most a few per cent of the flux the secondary star.
A rotational broadening measurement ($`v\mathrm{sin}i`$) of 43 km s<sup>-1</sup> was obtained from the K3III template. To check the systematic errors introduced by the choice of the linearized limb-darkening coefficient (which is only suitable for the continuum; Collins & Truax 1995), we have allowed it to vary in the range 0.0-1.0. This leads to 7 per cent changes in the resulting value of $`v\mathrm{sin}i`$ (i.e. $``$ 3 km s<sup>-1</sup>) and $`f`$. Therefore it is the uncertainty in the limb-darkening coefficient and not the statistical noise which limits our accuracy. We therefore adopt a value of $`v\mathrm{sin}i=43\pm 3`$ km s<sup>-1</sup>, which encompasses all $`v\mathrm{sin}i`$ values obtained for the templates in Table 3. Figure 2 shows a comparison of the spectra over the range $`\lambda \lambda 64006475`$. It is clear that there is some excess absorption in the Cai lines (in particular Cai $`\lambda 6439.1`$). This could be caused by differences in the spectral luminosity or/and metallicity between template and target spectrum.
## 5 Interstellar Extinction Upper limit
An estimate of the upper limit to the color excess of 0.78 mag is derived from the weighted average Hi column within one degree along the line of sight to J1628 ($`N{}_{H}{}^{}=4.51\times 10^{21}`$ cm<sup>-2</sup>; Dickey & Lockman 1990) and the relation between $`N_H`$ and $`E(BV)`$ of Bohlin, Savage & Drake (1978). For this reddening we should expect the $`\lambda \lambda 6196,6203`$ diffuse interstellar bands (DIBs) to have EWs of about 78 and 220 mÅ respectively (Herbig 1975). However, the EWs of the absorption features close to these wavelength positions are about 10 mÅ indicating that the extinction towards J1628 is significantly lower.
We have searched all medium (IMACS) and high-resolution spectra (MIKE) for other DIBs and atomic interstellar lines. The only interstellar features we were able to identify unambiguously in the forest of absorption lines originating in the secondary were the NaD $`\lambda \lambda 5889.95,5895.92`$ and Ki $`\lambda 7699`$ lines blended with the namesake broader photospheric lines (see Figure 3). The profile of these interstellar lines show a single component. Taking into account that double or multiple components in the profile of the Ki $`\lambda 7699`$ line appear when its EW is $`0.15`$ Å (Munari & Zwitter 1997; MZ97), the upper limit to E(B-V) can be reduced to $`0.6`$ mag according to the relation between reddening and EW for this interstellar line (MZ97). We have measured an EW of $`0.08\pm 0.01`$ Å for the Ki interstellar line after fitting the Ki blend with a two-Gaussian model (with one Gaussian to account for the photospheric absorption line and the other to account for the interstellar component). Using the calibration of MZ97 we derived an interstellar extinction of $`E(BV)=0.30\pm 0.04`$. For this reddening, the $`\lambda \lambda 6196,6203`$ DIBs should be stronger than observed (Herbig 1975, see also Figure 5 in Jenniskens & Désert 1994). This discrepancy suggests that our fit is overestimating the EW for the interstellar line. Given the unreliable utility of Herbig’s calibrations for low reddening and our uncertain fit to the Ki blend, we conservatively adopt for the remaining of this paper $`0.0<E(BV)0.6`$.
## 6 Emission Lines in the Spectrum of J1628: Chromospheric Activity Indicators
Observational evidence of a chromosphere in the visible spectrum of active stars relies commonly on the existence of emission in the cores of the Caii H & K lines. Apart from this hallmark active stars can reveal filling-in or strong emission lines in photospheric lines like Hi Balmer lines and the Caii infrared triplet (see e.g. Linsky 1980, Thatcher & Robinson 1993 and references therein). In this regard, there is clear manifestation of stellar activity in the spectra of J1628:
The Caii H & K lines have emission cores with absorption reversal at the top of the emission that give them a double-peak shape (see Figure 4). Both Caii H & K emission profiles exhibit variations in the strength of the violet (V) and red-ward (R) peaks, the violet peak being stronger than the red-ward peak ($`V/R>1`$) except on 2004 Jun 9 when the red peak becomes stronger ($`V/R<1`$). Single K dwarfs and giants hotter than spectral type K3 commonly show $`V/R>1`$ asymmetries as observed in the integrated disk of the Sun. This is often interpreted to be an indication that they have coronae and chromospheres dynamically similar to the Sun. The $`V/R`$ ratios $`<1`$ observed in giants cooler than K4 are considered to be related to mass outflow in the chromosphere (Stencel 1978). The temporal variability of the Caii peak asymmetries is well documented for Arcturus ($`\alpha `$ Boo; K2III). Arcturus has shown transitions from $`V/R>1`$ to $`V/R<1`$ in the Ca ii K emission core ratio (Chiu et al. 1977, Gray 1980). Temporal variations of the Ca ii emission cores have been observed in other late-type stars and are described in Rebolo et al. (1989), García López et al. (1992) and references therein. In the case of binary systems, Baliunas & Dupree (1982) found that the strength of the Ca ii emission profile for the single-lined (G8III-IV) chromospherically active binary $`\lambda `$ Andromedae increases (decreases) at the time of continuum light minimum (maximum) which correspond to the time when spotted (unspotted) regions dominate the stellar disk. Moreover, they found $`V/R<1`$ only at maximum light and $`V/R>1`$ at other phases and suggested an explanation of the variations in the profile asymmetry as due to differential downward and upward motions in the stellar atmosphere. Unfortunately our observations of J1628 are insufficient to corroborate the above correlation and explanation. Clearly, high-resolution spectroscopy with a better sampling of an orbital cycle and a simultaneous light curve are required.
The mean emission line width $`W_0(K)`$ measured for the Ca ii K line in the MIKE spectra is 0.82 Å, which yields $`W_0(K)=62.5`$ km s<sup>-1</sup> after the quadratic correction of the instrumental broadening (18.3 km s<sup>-1</sup>). From $`W_0(K)`$, we estimated an absolute visual magnitude of 2.0 for the secondary of J1628 by using the Wilson-Bappu relation for chromospherically active binaries (Montes et al. 1994): $`M_V=16.01\mathrm{log}W_0(K)+30.79`$, where $`W_0(K)`$ is expressed in km s<sup>-1</sup>. However J1628 may deviate significantly from the Wilson-Bappu law due to the influence of its high rotational broadening. Montes et al. (1994) found that $`M_V`$ was overestimated for large values of $`v\mathrm{sin}i`$, up to 2 mag when $`v\mathrm{sin}i40`$ km s<sup>-1</sup> (see figure 5 in their paper). Therefore $`M_V`$ could be $`4.0`$ for the secondary in J1628. In any case, these estimations of $`M_V`$ are in between the values for a K3 dwarf ($`M_V=6.8`$; Gray 1992) and a K3 giant ($`M_V=0.3`$). This result is in agreement with the expected evolved secondary for J1628.
The H$`\alpha `$ line is in emission above the continuum (see Figure 4) with EW values between 0.3-1.3 Å, except on night 2003 May 6, when the EW increases to 4 Å. This may be due to intrinsic activity variations, like a flare eruption. The H$`\alpha `$ profile obtained from the high resolution spectra has a FWHM$`160250`$ km s<sup>-1</sup> and shows a self-reversal core during some of the nights. Variable broad H$`\alpha `$ emission (see e.g. Byrne et al. 1995 for an insight on the broadening mechanism) sometimes with a self-reversal core has been observed among the most extreme chromospherically active binary stars, for instance UZ Lib (K0III, P<sub>sp</sub>=4.76 d; Bopp et al. 1984), II Peg (K2IV, P<sub>sp</sub>=6.72 d; Byrne et al. 1995), EZ Peg (G5IV-III/K0V, P<sub>sp</sub>=11.6 d; Montes et al. 1998), XX Tri (K0III, P<sub>sp</sub>=24 d; Bopp et al. 1993) and HD6139 (K2IV-III, P<sub>ph</sub>=31.95 d; Padmakar et al. 2000). The EW of the observed H$`\alpha `$ emission line measured in these systems is of the order of 1 Å (see the above references). For comparison, the H$`\alpha `$ EWs measured in X-ray novae in quiescence are of the order of tens to hundreds of Angstroms. For instance, V404 Cyg (K0IV, P<sub>sp</sub>=6.47 d, $`v\mathrm{sin}i=38`$ km s<sup>-1</sup>) and Cen X-4 (K3-5V, P<sub>sp</sub>=0.63 d, $`v\mathrm{sin}i=43`$ km s<sup>-1</sup>) have respectively H$`\alpha `$ EWs of 38 Å (Casares et al. 1993) and 35 Å (Torres et al. 2002).
## 7 DISCUSSION
Tidal theory (Zahn 1977) predicts that late-type stars in close binary systems rotate in synchronization with the orbital motion because tidal interactions are effective in forcing synchronization on time scales shorter than the evolutionary life time of the systems. A purely hydrodynamical mechanism based on the effects of meridional currents in the atmospheres of the stellar components and due to their non-spherical shape also explains the synchronism observed in close binary systems (Tassoul & Tassoul 1992). The estimated spectroscopic (orbital) and photometric (rotational) periods for J1628 are nearly identical implying synchronism. The small difference (if real) could be due to changes in the spot pattern over the years. Using the measured rotational broadening (section 4) and assuming that the secondary star is spherical, we can obtain a lower limit for the radius of the secondary star: $`R{}_{2}{}^{}R{}_{2}{}^{}\mathrm{sin}i(R_{})=P{}_{rot}{}^{}v\mathrm{sin}i/2\pi =\frac{1}{50.6}P{}_{rot}{}^{}(days)v\mathrm{sin}i(km/s)=4.2R_{}`$. A K3V star has a $`0.7R_{}`$ radius (Gray 1992) and a K3III star has a radius of $`20.5\pm 0.6R_{}`$ (van Belle et al. 1999). Assuming the secondary is not larger than a giant, we obtain a minimum value of the binary inclination of $`\mathrm{arcsin}i=R{}_{2}{}^{}\mathrm{sin}i/R{}_{K3III}{}^{}=12^o`$.
When the secondary star in a synchronous binary fills its Roche lobe, the mass ratio $`q=M_2/M_1`$ can be determined through the expression (see e.g. Wade & Horne 1988 and Eggleton 1983):
$$v\mathrm{sin}i=K_2(1+q)\frac{0.49q^{2/3}}{0.6q^{2/3}+\mathrm{ln}\left(1+q^{1/3}\right)}$$
From the values of $`v\mathrm{sin}i`$ and $`K_2`$ found for J1628, we derive a lower limit of $`q>2.0`$, which implies a velocity semi-amplitude of the primary $`K_1=qK_2>67`$ km s<sup>-1</sup>. These are lower limits because the radius of the secondary in J1628 may be smaller than its Roche lobe and the latter increases its size with $`q`$. Assuming mild mass exchange/loss during the evolution of the system (Popper & Ulrich 1972, Dupree 1986), the mass of the secondary should be in the range of masses expected for K3V to K3III stars, i.e. 0.7 to 1.1 M (Lang 1992). Hence $`M_1=M_2/q<0.5M_20.6`$ M. On this basis, the unseen primary of J1628 is probably a dwarf (if not a white dwarf) of spectral type K7 or later. A stringent lower limit for the binary inclination of $`i41^o`$ is derived from the mass function ($`f(M{}_{1}{}^{})=(1+q)^2M_1\mathrm{sin}^3i`$) when using the upper limit for $`M_1`$, the lower limit for $`q`$ and the value of $`f(M{}_{1}{}^{})`$. The radius of the secondary is thereby $`R{}_{2}{}^{}\mathrm{sin}41^o/\mathrm{sin}41^o=6.4R_{}`$. In short, $`4.2R{}_{}{}^{}R{}_{2}{}^{}6.4R_{}`$ for the visible stellar component in J1628. These constraints on the radius can be used in conjunction with the constraint in the spectral type of the secondary (Section 4) to obtain its absolute visual magnitude by applying Equation 2 of Popper (1980). Using the visual absolute flux (surface brightness parameter) for a K3V/III star (Table 1 in Popper 1980) we derive $`2.0<M{}_{V}{}^{}<3.8`$ in harmony with the absolute magnitudes estimated from the Wilson-Bappu relationship (Section 6). From $`M_V`$, the apparent magnitude $`V=13.4`$ and the constraints to the reddening obtained in Section 5 we find 350 pc $`<d<1.9`$ kpc for the distance to J1628.
We made use of PIMMS to convert the ROSAT PSPC count rates to an unabsorbed flux of $`8.5\times 10^{13}<f{}_{x}{}^{}4.4\times 10^{12}`$ erg cm<sup>-2</sup> s<sup>-1</sup> (0.1-2.4 keV) by applying an absorbed Raymond Smith model with $`\mathrm{log}T=7`$, assuming a metal abundance of 0.2 times the solar value and $`0.0<N_H3.48\times 10^{21}`$ cm<sup>-2</sup> (see Yi et al. 1997 regarding the utility of using this 1-T model to obtain X-ray fluxes). Furthermore, for each possible value of $`N_H`$ and $`M_V`$ we can evaluate the distance towards J1628 using the distance modulus and (as above) the corresponding intrinsic X-ray flux. This yielded an intrinsic X-ray luminosity of $`6.6\times 10^{31}<L{}_{x}{}^{}<6.4\times 10^{32}`$ erg s<sup>-1</sup> <sup>2</sup><sup>2</sup>2The range in computed $`L_x`$ is somewhat less than the square of the range in the distance error because of the complex interplay between the assumed $`N_H`$ and the derived distance. High $`N_H`$ yield high A<sub>V</sub> and lower distances (therefore lower computed $`L_x`$), but the X-ray absorption correction factor is higher at high $`N_H`$ which increases the computed $`L_x`$ and somewhat counteracts the distance effect., well above the averaged X-ray luminosity observed in chromospherically active binaries ($`1\times 10^{30}`$ erg s<sup>-1</sup>; Padmakar et al. 2000), but consistent with the X-ray luminosities for RS CVn stars ($`1\times 10^{29.232.2}`$ erg s<sup>-1</sup>). This range of X-ray luminosities for RS CVn stars was derived from observations with Einstein IPC (energy band 0.16-4 keV) and ROSAT PSPC (see Drake et al. 1989, 1992 and Dempsey et al. 1993). The differences in the X-ray flux due to the different bandpasses are expected to be $`10\%`$ (Dempsey et al. 1993; Benz & Güdel 1994).
An upper limit to the quiescent radio flux density of $`0.3`$ mJy was obtained from observations (Slee et al. 2002, Rupen et al. 2004) and can be used to check if the radio flux is consistent with that expected for a quiescent chromospherically active binary. We compared this with the quiescent radio luminosity found for chromospherically active binaries of $`\mathrm{log}L{}_{rad}{}^{}=(1.37\pm 0.09)\mathrm{log}L{}_{x}{}^{}26.38`$ (Padmakar et al. 2000; see also Drake, Simon & Linsky 1989 and Benz & Güdel 1994). In this way we derived 0.2 $`<f{}_{rad}{}^{}<2.0`$ mjy for J1628. While this range is consistent with the observed upper limit, the range is due almost entirely to our uncertainty in the X-ray luminosity - the correlation between $`L_x`$ and $`L_{rad}`$ is rather tight. If the true X-ray luminosity is at the high end of our range, then the quiescent radio flux is inconsistent (the predicted radio flux is larger than the observed upper limit). In this regard, the hard X-ray emission observed for J1628 (ROSAT hardness ratio HR1=1.00) is common for RS CVn stars undergoing an X-ray flare (see e.g. Graffagnino, Wonnacott & Schaeidt 1995). The radio counterpart to J1628 occasionally increases its flux density to $`0.3513.8`$ mJy at 8.6 and 4.8 GHz (Tsarevsky et al. 2001; Rupen et al. 2002,2004; Slee et al. 2003). Transient radio brightenings with similar or larger amplitude to those observed in J1628 are frequent in chromospherically active binary stars (Slee et al. 1987, Drake et al. 1989).
Finally, we considered the possibility that a foreground object could be the source of the radio and X-ray emission. We searched the images obtained with IMACS during the acquisition of the spectra to find a fainter optical source nearby GSC 07861-01088, the proposed (brighter) optical counterpart (see Figure 5). We performed an analysis of optical astrometry from one of the images to check the positional coincidence of GSC 07861-01088 and its radio counterpart (Rupen et al. 2004). The agreement between the optical and radio position is well within the sub-arcsecond astrometric error and we rule out the possibility that the nearby fainter object, which is located 2 arcsec South at RA (J2000)=16:28:47.27 and DEC(J2000)=-41:52:41.0 is the radio source.
Optical spectroscopy obtained for other microquasar candidates has shown that they are mostly extragalactic in nature and, in the few stellar cases, likely chromospherically active stars/binaries (Martí et al. 2004a,b; Torres et al. 2004b, Tsarevsky et al. 2005). Hence the search for microquasars using X-ray and radio surveys has not provided so far any new and secure microquasars. This suggests that there are few microquasars with persistent bright radio/X-ray emission to be found with the RASS Bright Source catalogue. Apart from the discovery of X-ray transient microquasars (e.g. Garcia et al. 2003), the progress in understanding relativistic outflows in X-ray binaries depends largely on observations of known X-ray binaries using current and future instruments with higher sensitivity and angular resolution. Suitable targets are for instance radio emitting X-ray binaries where jets have not been resolved yet (see e.g. Mirabel & Rodríguez 1999, Fender 2004b) and X-ray transients in quiescence where radio jets are expected at a level of a few $`\mu `$Jy (see e.g. Gallo, Fender & Pooley 2003; Gallo, Fender & Hynes 2005).
## 8 CONCLUSIONS
We have presented comprehensive optical spectroscopy of the microquasar candidate 1RXS J162848.1-41524. From the analysis of the absorption line spectrum, we have determined an orbital period of P<sub>sp</sub>=4.93956 $`\pm `$ 0.00025 d and a radial velocity semiamplitude of the secondary of $`K_2=33.3\pm 0.6`$ km s<sup>-1</sup>. The implied mass function is $`f(M{}_{1}{}^{})=0.019\pm 0.001`$ M. We have established the rotational broadening of the secondary star to be $`v\mathrm{sin}i=43\pm 3`$ km s<sup>-1</sup>. This provides an upper limit to the mass ratio of $`q>2.0`$. A $`\chi ^2`$ test applied to the residuals obtained by subtracting different template stars to the high-resolution spectra (Section 4), the limits to the absolute visual magnitude (Sections 6, 7) and the constraints to the stellar radius (Section 7) support a K3IV secondary. The emission lines observed in the spectrum of J1628 are consistent with those observed in chromospherically active binaries. This fact together with the results presented above indicates a chromospherically active binary and not a microquasar nature for J1628, where the X-ray and radio emission are powered by the stellar chromosphere and the periodic photometric variability reported by Buxton et al. (2004) would be due to cool surface spots. No trace of the primary has been found in the photospheric spectrum leading to the conclusion that it is a low luminosity object, possibly a late-type K dwarf or white dwarf. Our analysis is hampered however by the fact that our spectroscopic observations have not the required temporal coverage to determine unambiguously the orbital period of J1628. Therefore a better sampled radial velocity curve is still necessary to confirm the parameters derived in this paper for J1628. Photometric observations at different epochs will be enlightening: if J1628 is a chromospherically active binary system, it should show significant changes in the shape and amplitude of the light curve due to variations in the starspot distribution over the surface of the secondary.
## ACKNOWLEDGMENTS
We thank J. Bloom, M. Holman and S. Laycock for helping during the observations. We also thank P. Martini for providing some of the MIKE templates. Use of molly, doppler and trailer routines developed largely by T. R. Marsh is acknowledged. We also thank the referee for useful comments. This research has made use of the SIMBAD database, operated at CDS, Strasbourg, France. This work was supported by NASA LTSA grant NAG-5-10889 and NASA contract NAS8-39073 to the Chandra X-ray Center. DS acknowledges a Smithsonian Astrophysical Observatory Clay Fellowship.
|
warning/0506/hep-th0506153.html
|
ar5iv
|
text
|
# Viscous dark energy and phantom evolution
## I Introduction
The existence of a exotic cosmic fluid with negative pressure, which constitutes about the $`70`$ per cent of the total energy of the universe, has been perhaps the most surprising discovery made in cosmology. This dark energy is supported by the astrophysical data obtained from Wilkinson Microwave Anisotropy Probe (WMAP) (Map) and high redshift surveys of supernovae.
The dark energy is considered a fluid characterized by a negative pressure and usually represented by the equation of state $`w=p/\rho `$, where $`w`$ lies very close to $`1`$, most probably being below $`1`$. Dark energy with $`w<1`$, the phantom component of the universe, leads to uncommon cosmological scenarios as it was pointed out in Cadwell . First of all, there is a violation of the dominant energy condition (DEC), since $`\rho +p<0`$. The energy density grows up to infinity in a finite time, which leads to a big rip, characterized by a scale factor blowing up in this finite time. These sudden future singularities are, nevertheless, not necessarily produced by a fluids violating DEC. Barrow Barrow 1 has shown, with explicit examples, that exist solutions which develop a big rip singularity at a finite time even if the matter fields satisfy the strong-energy conditions $`\rho >0`$ and $`\rho +3p>0`$. A generalization of Barrow’s model has been realized in Nojiri , giving its Lagrangian description in terms of scalar tensor theory. It was also proved by Chimento et al Chimento that exist a duality between phantom and flat Friedmann-Robertson-Walker (FRW) cosmologies with nonexotic fluids. This duality is a form-invariance transformation which can be used for constructing phantom cosmologies from standard scalar field universes. Cosmological solutions for phantom matter which violates the weak energy condition were found in Dabrowski .
The role of the dissipative processes in the evolution of the early universe also has been extensively studied. In the case of isotropic and homogeneous cosmologies, any dissipation process in a FRW cosmology is scalar, and therefore may be modelled as a bulk viscosity within a thermodynamical approach.
A well known result of the FRW cosmological solutions, corresponding to universes filled with perfect fluid and bulk viscous stresses, is the possibility of violating DEC Barrow 2 . The bulk viscosity introduces dissipation by only redefining the effective pressure, $`P_{_{eff}}`$, according to
$`P_{_{eff}}=p+\mathrm{\Pi }=p3\xi H,`$ (1)
where $`\mathrm{\Pi }`$ is the bulk viscous pressure, $`\xi `$ is the bulk viscosity coefficient and $`H`$ is the Hubble parameter. Since the equation of energy balance is
$`\dot{\rho }+3H(\rho +p+\pi )=0.`$ (2)
the violation of DEC, i.e., $`\rho +p+\mathrm{\Pi }<0`$ implies an increasing energy density of the fluid that fills the universe, for a positive bulk viscosity coefficient. The condition $`\xi >0`$ guaranties a positive entropy production and, in consequence, no violation of the second law of the thermodynamics Pavon .
In the present paper we show that the above results are straightforward to obtain from the exact cosmological solutions already found by Barrow in Barrow 2 . These solutions were obtained using non causal thermodynamics. Nevertheless, we consider a more physical approach like the full Israel-Stewart-Hiscock causal thermodynamics, showing that it is also possible to obtain big rip type solutions.
The organization of the paper is as follows: in Section II we present the field equations for a flat FRW universe filled with a bulk viscous fluid within the framework of the Eckart theory. We indicate that under a constraint for the parameters of the fluid, one of the Barrow’s solutions presents a future singularity in a finite time. In Section III we obtain big rip solutions using the approach of the full Israel-Stewart-Hiscock causal thermodynamics. In Section IV we discuss our results in relation to the nature of the dark energy of the universe.
## II Eckart theory
The FRW metric for an homogeneous and isotropic flat universe is given by
$`ds^2=dt^2+a(t)^2\left(dr^2+r^2(d\theta ^2+sin^2\theta d\varphi ^2)\right),`$ (3)
where $`a(t)`$ is the scale factor and $`t`$ represents the cosmic time. In the following we use the units $`8\pi G=1`$. In the first order thermodynamic theory of Eckart Eckart the field equations in the presence of bulk viscous stresses are
$`\left({\displaystyle \frac{\dot{a}}{a}}\right)^2=H^2={\displaystyle \frac{\rho }{3}},`$ (4)
$`{\displaystyle \frac{\ddot{a}}{a}}=\dot{H}+H^2={\displaystyle \frac{1}{6}}\left(\rho +3P_{_{eff}}\right),`$ (5)
with
$`P_{_{eff}}=p+\mathrm{\Pi },`$ (6)
and
$`\mathrm{\Pi }=3H\xi .`$ (7)
The conservation equation is
$`\dot{\rho }+3H(\rho +p+\mathrm{\Pi })=0.`$ (8)
Assuming that the dark component obey the state equation
$`p=(\gamma 1)\rho ,`$ (9)
where $`0\gamma 2`$, we can obtain from equations (4) to (9) a single evolution equation for $`H`$:
$`2\dot{H}+3\gamma H^2=3\xi H.`$ (10)
This equation may be integrated directly as a function of the bulk viscosity. For $`\gamma 0`$ the solution has the form
$`H(t)={\displaystyle \frac{e^{\frac{3}{2}{\scriptscriptstyle \xi (t)𝑑t}}}{C+\frac{3}{2}\gamma e^{\frac{3}{2}{\scriptscriptstyle \xi (t)𝑑t}}𝑑t}}`$ (11)
where $`C`$ is an integration constant. From this equation we find the following expression for the scale factor
$`a(t)=D\left(C+{\displaystyle \frac{3}{2}}\gamma {\displaystyle e^{\frac{3}{2}{\scriptscriptstyle \xi (t)𝑑t}}𝑑t}\right)^{2/(3\gamma )},`$ (12)
where $`D`$ is a new integration constant. Thus for a given $`\xi (t)`$ we have the expressions for $`a(t)`$, $`\rho (t)`$ and $`p(t)`$.
For the case $`\gamma =0`$ we have from Eq. (10)
$`\xi ={\displaystyle \frac{2}{3}}{\displaystyle \frac{\dot{H}}{H}},`$ (13)
and substituting this expression into Eq. (8) we have
$`\dot{\rho }=6\dot{H}H,`$ (14)
from which we conclude that $`\rho =3H^2+const`$. Comparing this expression with (4) we have that the integration constant is zero.
Thus we have that for the state equation $`p=\rho `$, i.e. for $`\gamma =0`$, the scale factor is not defined by the field equations. So for a given $`a(t)`$ we can write $`H`$ and then obtain the expressions for the energy density from Eq. (4) and the bulk viscosity from Eq. (13). Clearly if $`\xi =0`$ the well known de Sitter scale factor $`a(t)=e^{H__0t}`$ is obtained, where $`p=\rho `$ and both are constants.
Notice that the solution of the field equations may be written through $`\xi (t)`$ or $`a(t)`$ because there are three independent equations for the four unknown functions $`a(t)`$, $`\rho (t)`$, $`\xi (t)`$ and $`p(t)`$.
Now we are interested in the possibility that there are cosmological models with viscous matter which present in its development a big rip singularity.
### II.1 The case for $`\gamma 0`$
Firstly, let us consider the case $`\gamma 0`$. If the viscous fluid satisfies DEC, then the condition $`0\gamma 2`$ must be satisfied. Thus for $`\gamma <0`$ we have a phantom cosmology. Now from the thermodynamics we know that $`\xi >0`$, and if $`\gamma <0`$ the Eq. (12) implies that we can have a big rip singularity at a finite value of cosmic time.
Let us consider some examples to see this more clearly. From Eq. (12) the well known standard case for a perfect fluid, i.e. $`\xi =0`$, takes the form $`a(t)=D(C+(3/2)\gamma t)^{2/(3\gamma )}`$. This scale factor may be rewritten as
$`a(t)=a__0\left(1+{\displaystyle \frac{3}{2}}H__0\gamma t\right)^{2/(3\gamma )},`$ (15)
and the energy density is given by
$`\rho ={\displaystyle \frac{\rho __0}{\left(1+\frac{3}{2}H__0\gamma t\right)^2}},`$ (16)
where $`\rho __0=3H__0^2`$, in order to have $`H(t__0=0)=H__0>0`$. If $`\gamma <0`$ we have a big rip singularity at a finite value of cosmic time $`t_{_{br}}=2/(3H__0\gamma )>t__0=0`$.
In the special case of $`\xi (t)=\xi __0=const`$ we have from Eq. (12) for the scale factor $`a(t)=D(C+(\gamma /\xi __0)e^{(3/2)\xi __0t})^{2/(3\gamma )}`$. We can rewrite it into the form
$`a(t)=a__0\left(1+{\displaystyle \frac{H__0}{\xi __0}}\gamma \left(e^{\mathrm{\hspace{0.17em}3}\xi __0t/2}1\right)\right)^{2/(3\gamma )},`$ (17)
from which we obtain for the energy density
$`\rho (t)=\rho __0{\displaystyle \frac{e^{3\xi __0t}}{\left(1+\frac{H__0}{\xi __0}\gamma \left(e^{\mathrm{\hspace{0.17em}3}\xi __0t/2}1\right)\right)^2}},`$ (18)
where $`\rho __0=3H__0^2`$. As before, for $`\gamma <0`$ we have a big rip singularity at a finite value of cosmic time $`t_{_{br}}=\frac{2}{3\xi __0}ln(1\frac{\xi __0}{H__0\gamma })>t__0=0`$.
Note that any additional condition on the system of the field equations will fix the unknown functions. So for instance, for a variable $`\xi (t)`$ we can take the condition $`\xi (t)=\xi (\rho (t))`$.
Another example in this line is given by the solution obtained by Barrow Barrow 2 for the case $`\xi \rho ^{1/2}`$. Effectively, Barrow Barrow 2 assumed that the viscosity has a power-law dependence upon the density
$`\xi =\alpha \rho ^s,\alpha 0.`$ (19)
where $`\alpha `$ and $`s`$ are constant parameters, and exact cosmological solutions for a variety of $`\xi (\rho )`$ in the form given by equation (19). In particular, for the case $`s=1/2`$, i.e., $`\xi =\alpha \rho ^{1/2}`$, yields a power-law expansion for the scale factor. Nevertheless, none condition was imposed upon the parameters $`\alpha `$ and $`\gamma `$ in order to obtain solutions with big rip.
For the case $`s=1/2`$, the integration of equation (10) yields the following expression for $`H(t)`$
$`{\displaystyle \frac{1}{H}}={\displaystyle \frac{1}{H_0}}{\displaystyle \frac{3}{2}}\left(\sqrt{3}\alpha \gamma \right)\left(tt_0\right),`$ (20)
where $`H_0=H(t=t_0)`$ and $`t_0`$ correspond to the time where dark component begins to become dominant. The scale factor becomes
$`a(t)=a_0\left(1{\displaystyle \frac{tt_0}{t_{br}}}\right)^{\frac{2}{3(\gamma \sqrt{3}\alpha )}},`$ (21)
where $`a_0=a(t=t_0)`$. If we demand to have the occurrence of a big rip in the future cosmic time then we have the following constraint on the parameters $`\alpha `$ and $`\gamma `$
$`\sqrt{3}\alpha >\gamma ,`$ (22)
leading the scale factor blow up to infinity at a finite time $`t_{br}>t_0`$, which expression is
$`t_{br}={\displaystyle \frac{2}{3(\sqrt{3}\alpha \gamma )}}H_0^1.`$ (23)
In terms of time $`t_{br}`$, the Hubble parameter is given by
$`H(t)=H_0\left(1{\displaystyle \frac{tt_0}{t_{br}}}\right)^1.`$ (24)
From the equation (4) and the parameterized equations (21) and (24) for the scale factor and Hubble parameter, respectively, we obtain the expression for the increasing density of the dark component in terms of scale factor
$`\rho (a)=3H_0^2\left({\displaystyle \frac{a}{a_0}}\right)^{3(\sqrt{3}\alpha \gamma )}.`$ (25)
We reproduce completely this solution if we put into the Eq. (12) the bulk viscosity given by
$`\xi (t)=\sqrt{3}\alpha H_0\left(1{\displaystyle \frac{tt_0}{t_{br}}}\right)^1.`$ (26)
### II.2 The case for $`\gamma =0`$
Notice that the structure of equation (10) changes if $`\gamma =0`$ and $`\xi `$ is an arbitrary function of the density, since the quadratic term in $`H`$ dissapears. Nevertheless, if $`\xi \rho ^{1/2}`$, the structure of the equation (10) is the same for any value of $`\gamma `$ in the range $`0\gamma 2`$, except in the case $`\sqrt{3}\alpha =\gamma `$, where equation (10) becomes $`\dot{H}=0`$. Then, the solution with $`\gamma =0`$ can be obtained directly from the general solution given by equation (21). In this case there is a big rip singularity at a finite value of cosmic time $`t_{_{br}}=2/(3\sqrt{3}H__0\alpha )>t__0=0`$.
## III Israel-Stewart-Hiscock theory
We now consider the dissipative process in the universe within the framework of the full causal theory of Israel-Stewart-Hiscock. In this case we have the same Friedmann equations but instead of equation (7), we have an equation for the causal evolution of the bulk viscous pressure, which is given by
$`\tau \dot{\mathrm{\Pi }}+\mathrm{\Pi }=3\xi H{\displaystyle \frac{1}{2}}\tau \mathrm{\Pi }\left(3H+{\displaystyle \frac{\dot{\tau }}{\tau }}{\displaystyle \frac{\dot{\xi }}{\xi }}{\displaystyle \frac{\dot{T}}{T}}\right),`$ (27)
where $`T`$ is the temperature and $`\tau `$ the relaxation time. In order to close the system we have to give the equation specifying $`T`$
$`T=\beta \rho ^r.`$ (28)
The the relaxation time is defined by the expression
$`\tau ={\displaystyle \frac{\xi }{\rho }}=\alpha \rho ^{s1},`$ (29)
where $`\beta 0`$. This model imposes the constraint
$`r={\displaystyle \frac{\gamma 1}{\gamma }},`$ (30)
in order to have the entropy as an state function. Notice that the above constraint exclude the range $`0<\gamma <1`$, which implies that quintessence fluids are not allowed in this approach. With the above assumptions the field equations and the causal evolution equation for the bulk viscosity lead to the following evolution equation for $`H`$ Maartens
$`\ddot{H}+{\displaystyle \frac{3}{2}}(1+(1r)\gamma )H\dot{H}+3^{1s}\alpha ^1H^{22s}\dot{H}(1+r)H^1\dot{H}^2+{\displaystyle \frac{9}{4}}(\gamma 2))H^3+{\displaystyle \frac{1}{2}}3^{2s}\alpha ^1\gamma H^{42s}=0.`$ (31)
As in the non causal case we will choose $`s=1/2`$ and the above equation becomes
$`\ddot{H}+bH\dot{H}\left(2{\displaystyle \frac{1}{\gamma }}\right)H^1\dot{H}^2+aH^3=0,`$ (32)
where $`a`$ is defined by
$`a{\displaystyle \frac{9}{4}}\left(\left(1+{\displaystyle \frac{2}{\sqrt{3}\alpha }}\right)\gamma 2\right),`$ (33)
and $`b`$ by
$`b3\left(1+{\displaystyle \frac{1}{\sqrt{3}\alpha }}\right).`$ (34)
Solutions of equation (32),were obtained in Mak . In this work only was considered $`\gamma `$ in the range $`1\gamma 2`$. Some of these solutions presents an increasing energy density and accelerated expansion.
Inspired in the solution for the Hubble parameter given by equation (24) in the non causal scheme, we use the following Ansatz, where for simplicity we take $`t_0=0`$
$`H(t)=A\left(\tau _{br}t\right)^1,`$ (35)
where $`AH_0\tau _{br}`$. With this Ansatz the scale factor $`a(t)`$ evolutes as
$`a(t)\left(\tau _{br}t\right)^A,`$ (36)
and the energy density, $`\rho `$, of the dark component as a function of the scale factor becomes
$`\rho (a)a^{2/A}.`$ (37)
Using the Ansatz (35) in equation (32) we obtain a second grade equation for $`A`$
$`aA^2+bA+{\displaystyle \frac{1}{\gamma }}=0.`$ (38)
The solutions for $`A`$ are given by
$`2A_\pm ={\displaystyle \frac{b}{a}}\pm \sqrt{\mathrm{\Delta }},`$ (39)
where the discriminant $`\mathrm{\Delta }`$ has the expression:
$`\mathrm{\Delta }\left({\displaystyle \frac{b}{a}}\right)^24\left({\displaystyle \frac{1}{a\gamma }}\right).`$ (40)
Since we are interested only in positive solutions for $`A`$, the coefficient $`a`$ must be negative. We have two cases of interest.
Case $`a<0;\gamma >0`$. In this case only $`A_+`$ correspond to a solution with big rip. The parameters $`\alpha `$ and $`\gamma `$ satisfy the following constraint
$`\sqrt{3}\alpha >\gamma \left(1{\displaystyle \frac{\gamma }{2}}\right)^1.`$ (41)
Notice that there is no big rip solution if the cosmic fluid representing the dark component is stiff matter ($`\gamma =2`$). The factor $`\left(1\frac{\gamma }{2}\right)^1`$ is the correction introduced by the causal thermodynamics to the constraint given by equation (22). The solution for $`A_+`$ is given by
$`A_+={\displaystyle \frac{1}{3}}{\displaystyle \frac{\left(1+\frac{1}{\sqrt{3}\alpha }\right)+\left(\frac{1}{3\alpha ^2}+\frac{2}{\gamma }\right)^{1/2}}{1\left(1+\frac{2}{\sqrt{3}\alpha }\right)\frac{\gamma }{2}}},`$ (42)
which implies that a big rip will occurs at a time
$`\tau _{br}=A_+H_0^1.`$ (43)
The expressions for $`a=a(t)`$ and $`\rho =\rho (a)`$ can be easily evaluate from equations (36) and (37), respectively.
Case $`a<0;\gamma <0`$ . Since we need $`\mathrm{\Delta }0`$ in order to have real solutions, the parameters $`\alpha `$ and $`\gamma `$ must satisfy the following constraint
$`\sqrt{3}\alpha \sqrt{{\displaystyle \frac{\gamma }{2}}}.`$ (44)
If $`\mathrm{\Delta }=0`$, i.e., $`\sqrt{3}\alpha =\sqrt{\gamma /2}`$, the solution for $`A`$, which we shall call $`A_0`$, has the following expression
$`A_0={\displaystyle \frac{2}{3}}{\displaystyle \frac{\left(1+\frac{1}{\sqrt{3}\alpha }\right)}{2+\left(1+\frac{2}{\sqrt{3}\alpha }\right)\gamma }}.`$ (45)
If $`\mathrm{\Delta }>0`$, the solutions for $`A`$ can be written as
$`A_\pm =A_0\pm {\displaystyle \frac{1}{2}}\sqrt{\mathrm{\Delta }}.`$ (46)
## IV Discussion
Within the framework of the non causal thermodynamics we have showed that the power law solution, found by Barrow in Barrow 2 for dissipative universes with $`\xi =\alpha \rho ^{1/2}`$, yields cosmologies which present big rip singularity when the constraint given in equation (22) holds. If we consider that the dark component is quintessence, i.e., $`0\gamma 2/3`$, with a sufficiently large bulk viscosity will make this quintessence behaves like a phantom energy. In the range $`2/3>\gamma 2`$ it is possible, at least from the mathematical point of view, to obtain solutions with big rip even with a matter fluid. It is not clear for us how can be interpreted a radiation fluid, for example, with a large bulk viscosity leading to high negative pressures and increasing densities.
At the boundary between the quintessence sector and the phantom sector, i.e. $`\gamma =0`$ or $`p=\rho `$, also there exist cosmologies with a big rip singularity.
Using a more accurate approach like the the full causal theory of Israel-Stewart-Hiscock, we have also found cosmological solutions with big rip.
If $`1\gamma 2`$ the parameters $`\alpha `$ and $`\gamma `$ satisfy the constraint given in equation (41). Due to the constraint stiff matter is not allowed. As we mentioned above this correspond to matter fluids that can lead to a phantom behavior. Quintessence region are not allowed. If $`\gamma <0`$ the cosmological solutions can be computed directly from equations (45) and (46). The main conclusion, in the context of the full causal thermodynamics, is that in order to obtain physically reasonable big rip solutions, the dark component must be phantom energy.
Note added: While this manuscript was being written we noticed about the work of Brevik and Gorbunova Brevik . The authors also consider the possibility of big rip in viscous fluids with $`p=w\rho `$, by a different formalism. They consider the case where the bulk viscosity is proportional to the scalar expansion. This is equivalent to the Barrow’s choice $`\xi (t)\sqrt{\rho }`$, and then they also reobtained the Barrow’s solution, for $`\gamma 0`$, considered here. In the framework of the standard Eckart theory citeEckart, the authors show that fluids which lie in the quintessence region ($`w>1`$) can reduce its thermodynamical pressure and cross the barrier $`w=1`$, and behave like a phantom fluid ($`w<1`$) with the inclusion of a sufficiently large bulk viscosity. The case for $`\gamma =0`$ was not considered by these authors.
## V acknowledgements
NC and SL acknowledge the hospitality of the Physics Department of Universidad de Concepción where an important part of this work was done last January. SL acknowledges the hospitality of the Physics Department of Universidad de Santiago de Chile. We thanks the suggestion of a new reference given by the referee, in order to improve the presentation of this paper. We acknowledge the partial support to this research by CONICYT through grants N<sup>0</sup> 1051086, 1030469 and 1040624 (MC); N<sup>0</sup> 1040229 (NC and SL); Grant MECESUP USA0108 (NC). It also was supported by the Direccion de Investigación de la Universidad del Bío–Bío (MC), PUCV Grant 123.771/04 (SL).
|
warning/0506/hep-ph0506157.html
|
ar5iv
|
text
|
# Renormalization of a gapless Hartree-Fock approximation to a theory with spontaneously broken 𝑂(𝑁)-symmetry
## I Introduction
Self-consistent $`\mathrm{\Phi }`$-derivable approximations were introduced long ago in the context of the nonrelativistic many-body problem Luttinger ; Baym and then extended to relativistic quantum field theory Cornwall ; Baym-Grin . Recently the interest in this method has been revived in view of its fruitful applications to calculations of the thermodynamic properties of the quark–gluon plasma Blaizot99 and to non-equilibrium quantum-field dynamics IKV98a ; Berges01 ; Cooper03 , in particular in terms of the off-shell kinetic equation IKV99 .
$`\mathrm{\Phi }`$-derivable approximations are preferable for the dynamical treatment of a system, since they fulfill the conservation laws of energy, momentum, and charge Baym ; IKV98a ; IKV99 . Moreover, the $`\mathrm{\Phi }`$-derivable scheme also guarantees the thermodynamic consistency of an approximation Baym , which makes it advantageous also for thermodynamic calculations. However, the $`\mathrm{\Phi }`$-derivable scheme has its generic problems which were also realized long ago Baym-Grin ; HM65 .
The first problem is related to the fact that $`\mathrm{\Phi }`$-derivable Dyson-Schwinger resummation schemes violate Ward–Takahashi identities beyond the one-point level. This, in particular, results in the violation of the Nambu–Goldstone (NG) theorem Baym-Grin ; HM65 ; Aouissat ; HK3 in the phase of spontaneously broken symmetry. On the other hand, so called “gapless” approximations HM65 respect the NG theorem (which is referred to as the Hugenholtz–Pines theorem in physics of Bose–Einstein condensed systems), though violate conservation laws and thermodynamic consistency. In Ref. HK3 it was shown that any $`\mathrm{\Phi }`$-derivable approximation can be corrected in such a way that it respects the NG theorem and becomes gapless. However, such modifications again violate conservation laws and thermodynamic consistency and, hence, leads back to the problems of the gapless scheme. Recently three of us have proposed a phenomenological way to construct a “gapless $`\mathrm{\Phi }`$-derivable” Hartree–Fock (gHF) approximation to the $`\lambda \varphi ^4`$ theory in the phase of spontaneously broken $`O(N)`$ symmetry IRK05 . This approximation simultaneously preserves all the desirable features of $`\mathrm{\Phi }`$-derivable schemes and respects the Nambu–Goldstone theorem in the broken phase. The treatment of Ref. IRK05 was based on a naive renormalization, where all divergent terms were simply omitted. This was done in order to avoid possible confusions between effects of restoring the NG theorem and those related to renormalization. In the present paper we return to the issue of renormalization.
The renormalization of the $`\mathrm{\Phi }`$-derivable approximations is precisely the second main problem. Following Baym and Grinstein Baym-Grin it was believed that renormalization of $`\mathrm{\Phi }`$-derivable approximations is possible only with medium (e.g., temperature) dependent counter terms, which is inconsistent with the goal of renormalization. Great progress in the proper renormalization of such schemes was recently achieved in Refs. HK3 ; HK1 ; Reinosa1 ; Berges04 ; Cooper05 . As the main result it was shown that partial resummation schemes can indeed be renormalized with medium-independent counter terms provided the scheme is generated from a two-particle irreducible (2PI) functional, i.e., a $`\mathrm{\Phi }`$-functional. Still, as we are going to demonstrate below, certain problems remain in the case of spontaneously broken symmetry.
As an example case we investigate the $`O(N)`$ model in the spontaneously broken phase which is a traditional touchstone for new theoretical approaches, well applied to a variety of physical phenomena, such as the chiral phase transition in nuclear matter. Thus we continue the discussion of the “gapless Hartree-Fock approximation started in the previous paper IRK05 and investigate its features towards renormalization in comparison to the standard HF-approximation.
## II Gapless Hartree–Fock (gHF) Approximation
We consider the $`O(N)`$-model Lagrangian
$`\begin{array}{cc}\hfill =& {\displaystyle \frac{1}{2}}(_\mu \varphi _a)^2{\displaystyle \frac{1}{2}}m^2\varphi ^2{\displaystyle \frac{\lambda }{4N}}(\varphi ^2)^2\hfill \\ & +H\varphi ,\hfill \end{array}`$ (1)
where $`\varphi =(\varphi _1,\varphi _2,\mathrm{},\varphi _N)`$ is an $`N`$-component scalar field, $`\varphi ^2=\varphi _a\varphi _a`$, with summation over $`a`$ implied. For $`H=0`$ this Lagrangian is invariant under $`O(N)`$ rotations of the fields. If $`H=0`$ and $`m^2<0`$, the symmetry of the ground state is spontaneously broken down to $`O(N1)`$, with $`N1`$ Goldstone bosons (pions). The external field $`H\varphi =H_a\varphi _a`$ is a term which explicitly breaks the $`O(N)`$ symmetry. It is introduced to give the physical value of $`140`$ MeV to the pion mass.
The effective action $`\mathrm{\Gamma }`$ for this Lagrangian in the real-time formalism is defined as (cf. Ref. IKV98a )
$`\begin{array}{cc}\hfill \mathrm{\Gamma }\{\varphi ,G\}& =I_0(\varphi )+{\displaystyle \frac{\mathrm{i}}{2}}\mathrm{Tr}\left(\mathrm{ln}G^1\right)\hfill \\ & +{\displaystyle \frac{\mathrm{i}}{2}}\mathrm{Tr}\left(D^1G1\right)+\mathrm{\Phi }_{\text{real-time}}\{\varphi ,G\},\hfill \end{array}`$ (2)
where $`\varphi `$ is the expectation value of the field, $`G`$ is the Green’s function, $`D`$ is the *free* Green’s function, $`I_0(\varphi )`$ is the *free* classical action of the $`\varphi `$ field, $`\mathrm{Tr}`$ implies space–time integration and summation over field indices $`a,b,\mathrm{}`$ All the considerations below are performed in terms of the thermodynamic $`\mathrm{\Phi }`$ functional which differs from $`\mathrm{\Phi }_{\text{real-time}}`$ in the factor of $`\mathrm{i}\beta `$, where $`\beta =1/T`$ is the inverse temperature. In the case of a spatially homogeneous thermodynamic system, an additional factor appears: the volume $`V`$ of the system. Thus, the thermodynamic $`\mathrm{\Phi }`$ is
$$\mathrm{\Phi }=(\mathrm{i}T/V)\mathrm{\Phi }_{\text{real-time}}.$$
In terms of the CJT formalism Cornwall , the same effective action $`\mathrm{\Gamma }`$ is given as (e.g., cf. Ref. Berges01 )
$`\begin{array}{cc}\hfill \mathrm{\Gamma }\{\varphi ,G\}& =I(\varphi )+{\displaystyle \frac{\mathrm{i}}{2}}\mathrm{Tr}\left(\mathrm{ln}G^1\right)\hfill \\ & +{\displaystyle \frac{\mathrm{i}}{2}}\mathrm{Tr}\left(D_\varphi ^1G1\right)+\mathrm{\Gamma }_2\{\varphi ,G\},\hfill \end{array}`$ (3)
i.e., already in terms of the *tree-level* Green’s function, $`D_\varphi (1,2)=\delta ^2I/\delta \varphi _2\delta \varphi _1`$, and the *full* classical action of the $`\varphi `$ field, $`I(\varphi )`$. In the thermodynamic limit, the effective potential, $`V_{\text{CJT}}`$, and its interaction part, $`V_2`$, are defined as
$$V_{\text{CJT}}=(\mathrm{i}T/V)\mathrm{\Gamma },V_2=(\mathrm{i}T/V)\mathrm{\Gamma }_2.$$
Naturally, $`V_2`$ is similar to $`\mathrm{\Phi }`$ but is not quite the same. Contrary to $`V_2`$, the $`\mathrm{\Phi }`$ functional includes all 2PI interaction terms, i.e. also those of zero and first loop order which result from interactions with the classical field (first two graphs in (4)).
The gHF approximation to the $`O(N)`$ theory is defined by the $`\mathrm{\Phi }`$ functional IRK05
$`\mathrm{\Phi }_{\text{gHF}}=\text{ }\text{}+\text{ }\text{}+\text{ }\text{}+\mathrm{\Delta }\mathrm{\Phi },`$ (4)
where the diagrams on the r.h.s. constitutes the conventional HF approximation, while *the phenomenological NG-theorem-restoring correction $`\mathrm{\Delta }\mathrm{\Phi }`$* is specified below, see Eq. (11). Here the crossed pins denote the classical fields $`\varphi _a`$, and loops are tadpoles
$`\text{ }\text{}=Q_{ab}={\displaystyle _\beta }d^4kG_{ab}(k)`$ (5)
in terms of $`G_{ab}`$ Green’s functions, where the Matsubara summation
$`{\displaystyle _\beta }d^4qf(q)T{\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}{\displaystyle \frac{d^3q}{(2\pi )^3}f(2\pi \mathrm{i}nT,\stackrel{}{q})}`$ (6)
is implied with $`T`$ being a finite temperature.
Within the $`\mathrm{\Phi }`$-derivable scheme the r.h.s. of the equations of motion for the classical field ($`J`$) and the Green’s function (self-energy $`\mathrm{\Sigma }`$) follow from the functional variation of $`\mathrm{\Phi }_{\text{gHF}}`$ with respect to the classical field $`\varphi `$ and Green’s function $`G`$, respectively
$`\mathrm{}\varphi +m^2\varphi `$ $`=`$ $`J={\displaystyle \frac{\delta \mathrm{\Phi }_{\text{gHF}}}{\delta \varphi }}`$ (7)
$`=`$ $`\text{ }\text{}+\text{ }\text{},`$ (8)
$`G^1D^1`$ $`=`$ $`\mathrm{\Sigma }=2{\displaystyle \frac{\delta \mathrm{\Phi }_{\text{gHF}}}{\delta G}}`$ (9)
$`=`$ $`\text{ }\text{}+\text{ }\text{}+2{\displaystyle \frac{\delta \mathrm{\Delta }\mathrm{\Phi }}{\delta G}},`$ (10)
where $`D`$ is the free propagator.
The $`\mathrm{\Delta }\mathrm{\Phi }`$ correction, introduced in Ref. IRK05 , is unambiguously determined proceeding from the following requirements: (i) it restores the NG theorem in the broken-symmetry phase, (ii) it does not change results in the phase of restored $`O(N)`$ symmetry, because there is no need for it, (iii) it does not change the HF equation for the classical field, since the conventional $`\mathrm{\Phi }`$-derivable and gapless schemes Baym-Grin ; HM65 provide the same classical-field equation already without any modifications. In particular, due to this latter requirement the $`\mathrm{\Delta }\mathrm{\Phi }`$ correction does not contribute to the classical-field equation (7). Proceeding from these requirements, this $`\mathrm{\Delta }\mathrm{\Phi }`$ can be presented in manifestly $`O(N)`$ symmetric form
$`\mathrm{\Delta }\mathrm{\Phi }={\displaystyle \frac{\lambda }{2N}}\left[N(Q_{ab})^2(Q_{aa})^2\right].`$ (11)
Here and below, summation over repeated indices $`a,b,c,\mathrm{}`$ is implied, if it is not pointed out otherwise.
The nature of this correction can be understood as follows. For the full theory, i.e., when all diagrams in the $`\mathrm{\Phi }`$ functional are taken into account, the gapless and $`\mathrm{\Phi }`$-derivable schemes are identical and both respect the NG theorem. The conventional $`\mathrm{\Phi }`$-derivable HF approximation omits an infinite set of diagrams which is necessary to restore its equivalence with the gapless scheme. The $`\mathrm{\Delta }\mathrm{\Phi }`$ correction to the HF approximation takes into account a part of those omitted diagrams (at the level of the actual approximation), and thus restores this equivalence in the pion sector. For the further discussion we switch to the notation in terms of the CJT effective potential, see e.g. Cornwall ; Lenaghan ; Nemoto , in order to comply with previous considerations in the literature.
The manifestly symmetric form of the CJT effective potential in the gHF approximation reads
$`V_{\text{gHF}}(\varphi ,G)`$ $`=`$ $`{\displaystyle \frac{1}{2}}m^2\varphi ^2+{\displaystyle \frac{\lambda }{4N}}(\varphi ^2)^2H\varphi +{\displaystyle \frac{1}{2}}{\displaystyle _\beta }d^4k\mathrm{ln}detG^1(k)`$ (12)
$`+`$ $`{\displaystyle \frac{1}{2}}{\displaystyle _\beta }d^4k\left\{\left[\left(k^2+m^2\right)\delta _{ab}+{\displaystyle \frac{\lambda }{N}}\left(\varphi ^2\delta _{ab}+2\varphi _a\varphi _b\right)\right]G_{ba}(k)1\right\}`$ (13)
$`+`$ $`{\displaystyle \frac{\lambda }{4N}}\left(Q_{aa}Q_{bb}+2Q_{ab}Q_{ba}\right)+\mathrm{\Delta }\mathrm{\Phi },`$ (14)
e.g., cf. Nemoto . All quantities are symmetric with respect to permutations of indices.
Since the self-energy (9) is momentum independent, the general form of the Green’s function can be written as follows
$`G_{ab}^1(k)=k^2\delta _{ab}+M_{ab}^2,`$ (15)
where $`M_{ab}^2`$ is a constant mass matrix. The equations for $`G_{cd}`$, i.e., for the corresponding tadpoles $`Q_{ab}`$, and the fields $`\varphi _c`$ result from variations of $`V_{\text{gHF}}`$ over $`G_{cd}`$ and $`\varphi _c`$, respectively,
$`M_{cd}^2=m^2\delta _{cd}`$ (16)
$`+{\displaystyle \frac{\lambda }{N}}[\varphi ^2\delta _{cd}+2\varphi _c\varphi _d+3Q_{aa}\delta _{cd}+2(1N)Q_{cd}],`$
$`H_c=m^2\varphi _c+{\displaystyle \frac{\lambda }{N}}[\varphi ^2\varphi _c+Q_{aa}\varphi _c+2Q_{cd}\varphi _d].`$ (17)
These are equations in a general nondiagonal representation. Applying projectors
$`\mathrm{\Pi }_{cd}^\pi `$ $`=`$ $`{\displaystyle \frac{1}{N1}}\left(\delta _{cd}\varphi _c\varphi _d/\varphi ^2\right),`$ (18)
$`\mathrm{\Pi }_{cd}^\sigma `$ $`=`$ $`\varphi _c\varphi _d/\varphi ^2,`$ (19)
to Eq. (16), we project it on $`\pi `$ and $`\sigma `$ states. In order to project the mean-field equation (17) on the $`\sigma `$-direction, we just multiply it by $`\varphi _c`$.
In the diagonal representation ($`\varphi _\sigma 0`$, $`H_\sigma =H`$ and $`H_\pi =\varphi _\pi =0`$) these equations take the following form
$`M_\sigma ^2`$ $`=`$ $`m^2+{\displaystyle \frac{\lambda }{N}}\left[3\varphi ^2+(52N)Q_\sigma +3(N1)Q_\pi \right]`$ (20)
$`=`$ $`M_\pi ^2+{\displaystyle \frac{\lambda }{N}}\left[2\varphi ^2+2(N1)(Q_\pi Q_\sigma )\right],`$ (21)
$`M_\pi ^2`$ $`=`$ $`m^2+{\displaystyle \frac{\lambda }{N}}\left[\varphi ^2+3Q_\sigma +(N1)Q_\pi \right],`$ (22)
$`H`$ $`=`$ $`\varphi \left[m^2+{\displaystyle \frac{\lambda }{N}}\left(\varphi ^2+3Q_\sigma +(N1)Q_\pi \right)\right],`$ (23)
where $`M_\pi ^2=\mathrm{\Pi }_{dc}^\pi M_{cd}^2`$ and $`M_\sigma ^2=\mathrm{\Pi }_{dc}^\sigma M_{cd}^2`$. Here we used $`Q_\sigma =Q_{\sigma \sigma }`$ and $`Q_\pi =Q_{\pi \pi }`$ in terms of definition (5). From these equations it is evident that the NG theorem is fulfilled. Indeed, in the phase of spontaneously broken symmetry ($`H=0`$) the square-bracketed term of the field equation (23) equals zero, which is precisely the pion mass, cf. Eq. (22). At the same time, as it has been demonstrated in numerous papers (see, e.g., Refs. Baym-Grin ; HM65 ; Aouissat ; Lenaghan ), the solution of the conventional HF set of equations (7)–(9), i.e., without $`\mathrm{\Delta }\mathrm{\Phi }`$, violates the NG theorem. A detailed analysis of the conventional HF equations in a notation similar to ours has been given in Ref. Lenaghan .
## III Renormalization of the gHF Approximation
Significant progress in proper renormalization of $`\mathrm{\Phi }`$-derivable approximations was recently achieved in Refs. HK3 ; HK1 ; Reinosa1 . Here we follow the renormalization scheme of Ref. HK3 , i.e., that constructed precisely for the conventional HF approximation to $`\lambda \varphi ^4`$ theory in the $`O(N)`$ broken phase. This renormalization is based on the BPHZ formalism.
### III.1 Equations of Motion
The equations of motion (20)–(23) involve tadpole terms which, based on the explicit form of the Green’s function (15), can be written as
$`Q_a={\displaystyle \frac{d^3k}{(2\pi )^3}\frac{1}{ϵ_a(\stackrel{}{k})}\left[n\left(ϵ_a(\stackrel{}{k})\right)+\frac{1}{2}\right]},`$ (24)
where $`ϵ_a(\stackrel{}{k})=(\stackrel{}{k}^2+M_a^2)^{1/2}`$ and
$`n(ϵ)={\displaystyle \frac{1}{\mathrm{exp}(ϵ/T)1}}`$ (25)
is the thermal occupation number. Evidently, the $`Q`$-function consists of two parts
$`Q_a=Q_a^T+Q_a^{\text{(div)}},`$ (26)
where
$`Q_a^T={\displaystyle \frac{d^3k}{(2\pi )^3}\frac{1}{ϵ_a(\stackrel{}{k})}n\left(ϵ_a(\stackrel{}{k})\right)}`$ (27)
is the convergent thermal part of the tadpole, which is finite, and the divergent part
$`Q_a^{\text{(div)}}`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \frac{d^3k}{(2\pi )^3}\frac{1}{ϵ_a(\stackrel{}{k})}}`$ (28)
$`=`$ $`{\displaystyle \frac{M_a^2}{(4\pi )^2}}\left({\displaystyle \frac{1}{ϵ}}+\mathrm{ln}{\displaystyle \frac{M_a^2}{\mu ^2}}1\right)`$ (29)
is regularized within dimensional regularization. Here $`ϵ0`$, and $`\mu `$ is a regularization scale. We apply the same mass independent renormalization conditions as in Ref. HK3 , i.e., that in the *symmetric vacuum* the self-energies vanish Kugo
$`\mathrm{\Sigma }_a(T=0,\varphi =0,m^2=\mu ^2>0)=0,`$ (30)
$`_{m^2}\mathrm{\Sigma }_a(T=0,\varphi =0,m^2=\mu ^2>0)=0,`$ (31)
for all $`a`$. Such a renormalization scheme preserves the O($`N`$) symmetry of the model Kugo . Since $`\mathrm{\Sigma }_a`$ is momentum independent in the approximation under consideration, additional momentum-derivative conditions are not required. Upon application of this scheme, Eqs. (20)–(23) keep their form with the $`Q_a`$ quantities substituted by the renormalized tadpoles.
$`Q_a^{\text{(ren)}}`$ $`=`$ $`Q_a^T`$ (32)
$`+`$ $`{\displaystyle \frac{1}{(4\pi )^2}}\left[M_a^2\left(\mathrm{ln}{\displaystyle \frac{M_a^2}{\mu ^2}}1\right)+\mu ^2\right].`$ (33)
As it was shown in Ref. HK3 , this renormalization description requires only vacuum (temperature independent) counter terms. For the sake of further discussion, note that according to Ref. HK3 the renormalization of the conventional HF approximation results in precisely the same equations as those in the CT scheme of Ref. Lenaghan , in spite of the different approaches used. The renormalized conventional HF approximation was thoroughly studied in Lenaghan . Therefore, those results are very useful for comparison with the present treatment.
### III.2 Effective Potential
The thermodynamic potential (12) is renormalized following the procedure outlined in Ref. HK1 . In the gHF approximation complications arise only in the $`\mathrm{ln}detG^1(k)`$ term. Because of the topology of the diagrams used in the gHF approximation, for all other contributions to the effective action, we only have to insert the already renormalized self-energies, i.e., $`Q_a^{\text{(ren)}}`$ tadpoles of Eq. (32), to renormalize them. This is legitimate, because we do not encounter any additional contributions from subdivergences, cf. Ref. HK1 . In the diagonal representation the remaining part to be renormalized takes the form
$`{\displaystyle \frac{1}{2}}{\displaystyle _\beta }d^4k\mathrm{ln}detG^1(k)=L_\sigma +(N1)L_\pi ,`$ (34)
where we have introduced the brief notation
$`L_a`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle _\beta }d^4k\mathrm{ln}G_a^1(k)`$ (35)
$`=`$ $`{\displaystyle \frac{d^3k}{(2\pi )^3}\left\{\frac{ϵ_a}{2}+T\mathrm{ln}\left[1\mathrm{exp}\left(\frac{ϵ_a}{T}\right)\right]\right\}}.`$ (36)
The $`L_a`$ also consists of two parts: the convergent thermal part
$`L_a^T=T{\displaystyle \frac{d^3k}{(2\pi )^3}\mathrm{ln}\left[1\mathrm{exp}\left(\frac{ϵ_a(\stackrel{}{k})}{T}\right)\right]},`$ (37)
which is finite, and the divergent integral
$`L_a^{\text{(div)}}={\displaystyle \frac{d^3k}{(2\pi )^3}\frac{ϵ_a(\stackrel{}{k})}{2}}.`$ (38)
This expression implicitly depends on temperature through the mass $`M_a`$. We regularize it by means of a momentum cut-off $`\mathrm{\Lambda }`$:
$`L_a^{\text{(reg)}}(M_a)`$ $`=`$ $`{\displaystyle }{\displaystyle \frac{d^3k}{(2\pi )^3}}{\displaystyle \frac{ϵ_a(\stackrel{}{k})}{2}}\mathrm{\Theta }(\mathrm{\Lambda }^2ϵ_a^2(\stackrel{}{k}))={\displaystyle \frac{1}{3(2\pi )^2}}((\mathrm{\Lambda }^2M_a^2)^{3/2}\mathrm{\Lambda }`$ (39)
$``$ $`{\displaystyle \frac{1}{8}}[2\mathrm{\Lambda }^3\sqrt{\mathrm{\Lambda }^2M^2}5\mathrm{\Lambda }M^2\sqrt{\mathrm{\Lambda }^2M^2}3M^4\mathrm{ln}M+3M^4\mathrm{ln}(\mathrm{\Lambda }+\sqrt{\mathrm{\Lambda }^2M^2})]).`$ (40)
To renormalize the effective potential, we use a mass-independent renormalization scheme in order to avoid effects of unphysical IR singularities. We impose the following renormalization conditions on the effective potential
$`V^{\text{(ren)}}(T=0,\varphi =0,m^2)|_{m^2=\mu ^2>0}`$ $`=`$ $`0,`$ (41)
$`_{m^2}V^{\text{(ren)}}(T=0,\varphi =0,m^2)|_{m^2=\mu ^2>0}`$ $`=`$ $`0,`$ (42)
$`_{m^2}^2V^{\text{(ren)}}(T=0,\varphi =0,m^2)|_{m^2=\mu ^2>0}`$ $`=`$ $`0.`$ (43)
We need precisely these three conditions to remove all the divergences. Imposing these conditions, we keep in mind that other parts of the effective potential, except for $`L_a`$, have already been renormalized such that they fulfill these conditions on their own. This leads to
$`L_a^{\text{(ren)}}`$ $`=`$ $`L_a^T+\underset{\mathrm{\Lambda }\mathrm{}}{lim}\left[L_a^{\text{(reg)}}(M_a)L_a^{\text{(reg)}}(\mu )(M_a^2\mu ^2){\displaystyle \frac{L_a^{\text{(reg)}}(\mu )}{\mu ^2}}{\displaystyle \frac{1}{2}}(M_a^2\mu ^2)^2{\displaystyle \frac{^2L_a^{\text{(reg)}}(\mu )}{(\mu ^2)^2}}\right]`$ (44)
$`=`$ $`L_a^T{\displaystyle \frac{1}{128\pi ^2}}\left(3M^44M^2\mu ^2+\mu ^42M^4\mathrm{ln}{\displaystyle \frac{M^2}{\mu ^2}}\right).`$ (45)
In terms of these $`L_a^{\text{(ren)}}`$ and $`Q_a^{\text{(ren)}}`$ the renormalized effective potential reads
$`V_{\text{gHF}}^{\text{(ren)}}(\varphi ,T)={\displaystyle \frac{1}{2}}m^2\varphi ^2+{\displaystyle \frac{\lambda }{4N}}\varphi ^4H\varphi +L_\sigma ^{\text{(ren)}}+(N1)L_\pi ^{\text{(ren)}}`$ (46)
$`+{\displaystyle \frac{1}{2}}\left[m^2\left(Q_\sigma ^{\text{(ren)}}+(N1)Q_\pi ^{\text{(ren)}}\right)M_\sigma ^2Q_\sigma ^{\text{(ren)}}(N1)M_\pi ^2Q_\pi ^{\text{(ren)}}+{\displaystyle \frac{\lambda }{N}}\varphi ^2\left(3Q_\sigma ^{\text{(ren)}}+(N1)Q_\pi ^{\text{(ren)}}\right)\right]`$ (47)
$`+{\displaystyle \frac{\lambda }{4N}}\left[3\left(Q_\sigma ^{\text{(ren)}}+(N1)Q_\pi ^{\text{(ren)}}\right)^22(N1)\left(\left[Q_\sigma ^{\text{(ren)}}\right]^2+(N1)\left[Q_\pi ^{\text{(ren)}}\right]^2\right)\right].`$ (48)
### III.3 Vacuum ($`T=0`$)
At $`T=0`$, the quantities under investigation are “experimentally” known<sup>1</sup><sup>1</sup>1These values are relevant for the case $`N=4`$.: $`M_\pi (T=0)=m_\pi =139`$ MeV, $`M_\sigma (T=0)=m_\sigma =600`$ MeV, and the pion decay constant $`\varphi _0=f_\pi =93`$ MeV. At $`T=0`$ these known quantities should satisfy Eqs. (20)–(23) with renormalized tadpoles $`Q_a^{\text{(ren)}}`$, cf. Eq. (32),
$`m_\sigma ^2`$ $`=`$ $`m_\pi ^2+{\displaystyle \frac{2\lambda }{N}}f_\pi ^2+{\displaystyle \frac{2(N1)\lambda }{(4\pi )^2N}}[m_\pi ^2\left(\mathrm{ln}{\displaystyle \frac{m_\pi ^2}{\mu ^2}}1\right)m_\sigma ^2\left(\mathrm{ln}{\displaystyle \frac{m_\sigma ^2}{\mu ^2}}1\right)],`$ (49)
$`m_\pi ^2`$ $`=`$ $`m^2+{\displaystyle \frac{\lambda }{N}}f_\pi ^2+{\displaystyle \frac{\lambda }{(4\pi )^2N}}[(N+2)\mu ^2+3m_\sigma ^2\left(\mathrm{ln}{\displaystyle \frac{m_\sigma ^2}{\mu ^2}}1\right)+(N1)m_\pi ^2\left(\mathrm{ln}{\displaystyle \frac{m_\pi ^2}{\mu ^2}}1\right)],`$ (50)
$`{\displaystyle \frac{H}{f_\pi }}`$ $`=`$ $`m^2+{\displaystyle \frac{\lambda }{N}}f_\pi ^2+{\displaystyle \frac{\lambda }{(4\pi )^2N}}[(N+2)\mu ^2+3m_\sigma ^2\left(\mathrm{ln}{\displaystyle \frac{m_\sigma ^2}{\mu ^2}}1\right)+(N1)m_\pi ^2\left(\mathrm{ln}{\displaystyle \frac{m_\pi ^2}{\mu ^2}}1\right)].`$ (51)
In order to make this set of equations consistent, we should put
$`H=m_\pi ^2f_\pi ,`$ (52)
i.e., precisely the same as at the tree level. Then Eqs. (50) and (51) become identical. Now we can solve for the remaining equations, (49) and (50), and express the renormalized quantities $`m^2`$ and $`\lambda `$ in terms of physical quantities $`m_\pi `$, $`m_\sigma `$, and $`f_\pi `$
$`\lambda `$ $`=`$ $`N\left(m_\sigma ^2m_\pi ^2\right)\left\{2f_\pi ^2+{\displaystyle \frac{2(N1)}{(4\pi )^2}}\left[m_\pi ^2\left(\mathrm{ln}{\displaystyle \frac{m_\pi ^2}{\mu ^2}}1\right)m_\sigma ^2\left(\mathrm{ln}{\displaystyle \frac{m_\sigma ^2}{\mu ^2}}1\right)\right]\right\}^1,`$ (53)
$`m^2`$ $`=`$ $`m_\pi ^2+{\displaystyle \frac{\lambda }{N}}f_\pi ^2+{\displaystyle \frac{\lambda }{(4\pi )^2N}}\left[(N+2)\mu ^2+3m_\sigma ^2\left(\mathrm{ln}{\displaystyle \frac{m_\sigma ^2}{\mu ^2}}1\right)+(N1)m_\pi ^2\left(\mathrm{ln}{\displaystyle \frac{m_\pi ^2}{\mu ^2}}1\right)\right].`$ (54)
The expressions manifestly show that the performed renormalization is $`\mu `$-scale independent in the vacuum. Indeed, $`m^2`$ and $`\lambda `$ can take any values depending on the choice of the renormalization scale $`\mu `$, while the same observables keep their physical values at any renormalization scale $`\mu `$ (within a certain range of $`\mu `$). In this respect, the gHF approximation is similar to the leading-order $`1/N`$-approximation, where also both the NG theorem and scale-independent renormalization in the vacuum are fulfilled Lenaghan . The restriction to a certain range is related to the conditions $`\lambda >0`$ and $`m^2<0`$ which should be met. At the scale $`\mu _0`$, determined by the equation
$`f_\pi ^2`$ $`+{\displaystyle \frac{(N1)}{(4\pi )^2}}`$ $`(m_\pi ^2\mathrm{ln}{\displaystyle \frac{m_\pi ^2}{\mu _0^2}}m_\pi ^2`$ (56)
$`m_\sigma ^2\mathrm{ln}{\displaystyle \frac{m_\sigma ^2}{\mu _0^2}}+m_\sigma ^2)=0,`$
$`\lambda `$ and $`m^2`$ become singular, cf. Eqs. (53) and (54). Moreover, $`\mu <\mu _0`$ implies $`m^2>0`$ and $`\lambda <0`$, which defers a spontaneously broken phase and makes the theory unstable because of uncompensated attraction ($`\lambda \varphi ^4<0`$). Therefore, the range of scale independent renormalization in the vacuum is restricted from below by
$`\mu >\mu _0.`$ (57)
The situation is completely different within the conventional HF approximation. In that case, physical observables do depend on the renormalization scale and can be reproduced only at a single value of the scale $`\mu ^2=m_\sigma ^2/e`$, as demonstrated in Lenaghan . The reason of this difference is that the conventional HF approximation of Lenaghan violates the NG theorem at $`H=0`$. This stems from the fact that the equations for the pion self-energy and the mean field are not identical in the broken phase and thus leave three nondegenerate equations for three quantities $`m^2`$, $`\lambda `$ and $`\mu `$, from which the $`\mu `$ value is unambiguously determined. In our case, this set of equations is degenerate, i.e., the equations for the pion self-energy and the mean field are identical as a consequence of the fulfilled NG theorem. This gives us freedom for an arbitrary choice of $`\mu `$. Thus, we arrive at an important conclusion which concerns all partial resummation schemes applied to the case of spontaneously broken symmetry: a scale-independent renormalization in the vacuum is possible only if the scheme preserves the NG theorem. This is an important aspect with respect to possible renormalization-group considerations for this type of self consistent approximations.
However, the scale dependence still persists at finite temperature, which is already seen from the analysis of the symmetry-restoration points.
### III.4 Symmetry Restoration Points at $`H=0`$
Starting from the broken phase, where $`M_\pi ^2=0`$ and $`\varphi ^2>0`$, there exists a temperature range $`T_R[T_1,T_2]`$, cf. Fig. 1 below, where the classical field vanishes together with the pion mass
$`M_\pi ^2(T_R)=\varphi ^2(T_R)=0.`$ (58)
This precisely occurs, when the two equations (22) and (23) reduce to a single one. Solving for this single equation with renormalized $`Q_a^{\text{(ren)}}`$ tadpoles (32) gives
$`T_R^2={\displaystyle \frac{12}{(N+2)}}[(1+{\displaystyle \frac{3}{(N1)}}{\displaystyle \frac{M_\sigma ^2(T_R)}{m_\sigma ^2}})f_\pi ^2`$ (59)
$`+{\displaystyle \frac{3m_\sigma ^2}{(4\pi )^2}}(\mathrm{ln}{\displaystyle \frac{m_\sigma ^2}{\mu ^2}}1)(1{\displaystyle \frac{M_\sigma ^2(T_R)}{m_\sigma ^2}})],`$ (60)
where we have used Eqs. (53) and (54) to express $`m^2`$ and $`\lambda `$ in terms of the vacuum mass $`m_\sigma `$. Strictly speaking, this is not a solution for $`T_R`$, since the r.h.s. of (59) still depends on $`T_R`$ through $`M_\sigma ^2(T_R)`$. Nevertheless, this expression is already quite simple to analyze.
The lower bound $`T_1`$ of condition (58) is of central importance for the phase transition in the conventional HF-approximation IRK05 but of minor relevance in the gHF-scheme, since it corresponds to the metastable solution, cf. Sect. IV below. It is determined if simultaneously also $`M_\sigma ^2=0`$ occurs
$`T_1^2`$ $`=`$ $`{\displaystyle \frac{12}{(N+2)}}[f_\pi ^2+{\displaystyle \frac{3m_\sigma ^2}{(4\pi )^2}}\left(\mathrm{ln}{\displaystyle \frac{m_\sigma ^2}{\mu ^2}}1\right)].`$ (61)
It is still $`\mu `$-dependent, in spite of the scale-independent renormalization in the vacuum. At large $`\mu `$, i.e., above some $`\mu _1`$, $`T_1^2`$ can even become negative, which means that then this solution does not exist.
For the stable solution of the gHF approximation it numerically occurs that $`M_\sigma (T_R)m_\sigma `$, cf. Ref. IRK05 and Sect. IV below. This almost removes the $`M_\sigma (T_R)`$ dependence from the r.h.s. of Eq. (59) and makes the corresponding temperature $`T_2`$ almost $`\mu `$-independent
$`T_2^2{\displaystyle \frac{12}{(N1)}}f_\pi ^2.`$ (62)
This value coincides with that of the naive renormalization IRK05 . The solution $`T_2`$ corresponds to a partial symmetry restoration, since here we still have $`M_\sigma (T_2)M_\pi (T_2)`$ in spite of $`\varphi =0`$.
## IV Results for $`𝐍=\mathrm{𝟒}`$
For the numerical calculations we use the following parameters: $`m_\sigma =600`$ MeV and $`f_\pi =93`$ MeV. The pion mass is either zero, $`m_\pi =0`$, in the case of exact symmetry, or $`m_\pi =139`$ MeV for the approximate symmetry. The general structure of the solutions to the renormalized Eqs. (20)–(23) is similar to that obtained with the naive renormalization IRK05 but with extra complications caused by the additional dependence on the renormalization scale $`\mu `$. According to Eq. (56) physically reasonable solutions exits only for $`\mu >\mu _0`$, where $`\mu _0`$ as the solution of Eq. (56) equals $`200`$ MeV for the above specified parameters.
There are several different branches of the solution. Stable and physically meaningful are determined by the principle of maximum pressure, the pressure being given by the effective potential (46)
$`P=V_{\text{gHF}}^{\text{(ren)}}(\varphi ,T)+\mathrm{const}.`$ (63)
Here the constant is determined by the condition that the pressure should vanish for the physical vacuum, i.e., in the spontaneously broken phase, while our renormalization condition (41) determines $`V_{\text{gHF}}^{\text{(ren)}}`$ to be zero in the unphysical, symmetric vacuum.
### IV.1 Exact $`O(4)`$ Symmetry
The actual structure of the solution depends on the renormalization scale $`\mu `$. We start with a moderate scale $`\mu =600`$ MeV, i.e., of the order of $`m_\sigma `$. The results are presented in Figs. 15.
In the narrow temperature range, displayed in Fig. 1, the results are qualitatively similar to those obtained with the naive renormalization IRK05 . The stable branch starts at $`T=0`$ from the physical vacuum values for the masses and the classical field and crosses the metastable branch at $`T_{\text{cross}}440`$ MeV. In terms of the pressure, they are touching rather than crossing (see Fig. 2). Therefore, no transition from one branch to another occurs at $`T_{\text{cross}}`$. In the broken-symmetry phase, the pion mass equals zero. Then a phase transition of the second order occurs at $`T_2180`$ MeV, at which the field becomes zero (see Fig. 3). However, the $`\pi `$ and $`\sigma `$ masses still differ beyond this transition point. They become equal only after a second phase transition, which is also of second order, at $`T_{\text{cross}}`$. Note that the equal-mass solution above $`T_{\text{cross}}`$ is precisely the same as in the conventional HF approximation (cf. Refs. HK3 ; Lenaghan ), since the gapless modification term (11) vanishes in this case. The $`T_1`$ point proves to be irrelevant for the stable branch. Rather it is the starting point for the metastable branch, which in the range of $`T_1<T<T_{\text{cross}}`$ precisely coincides with the solution of the conventional HF approximation HK3 ; Lenaghan . The corresponding field is always zero for this branch. Contrary to the case of the naive renormalization IRK05 , this metastable branch ends at some temperature ($`650`$ MeV).
However, in the wider temperature range, displayed in Fig. 4, we see a significant difference from the results obtained with the naive renormalization IRK05 . Neither stable nor metastable solutions exist above a certain temperature $`T_{\text{end}}`$ which is $`1.15`$ GeV for the considered $`\mu =600`$ MeV. Let us remind that this solution above $`T_{\text{cross}}`$ corresponds to the conventional HF approximation. Therefore, it is precisely the same as in previous conventional HF calculations with renormalization Lenaghan ; HK3 . The occurrence of such an end point in the HF approximation was first pointed out by Baym and Grinstein Baym-Grin .
At $`T_{\mathrm{end}}`$, the stable branch of the solution joins the upper branch. This upper branch at any temperature corresponds to equal masses and zero field, i.e., $`M_\pi =M_\sigma `$ and $`\varphi =0`$, and hence is also a solution to the conventional renormalized HF approximation. It starts with very high values of masses ($`3`$ GeV) in the vacuum and also ends at $`T_{\text{end}}`$. The vacuum pressure for this branch, evaluated according to Eq. (63), is rather high (see Fig. 5).
It is important note that for the upper branch and at temperatures near the endpoint the logarithmic terms in the gap equation, $`\mathrm{ln}(M_{\pi /\sigma }^2/\mu ^2)`$, become large. This indicates that at such points the expansion of the $`\mathrm{\Phi }`$ functional in powers of the renormalized coupling becomes unreliable, because the effective coupling becomes large. Therefore we consider the upper branch not a physically meaningful solution. The same holds true for temperatures close to the endpoint temperature. Such a behavior must be expected for any effective theory and was indeed also observed in Quantum Hadro Dynamics (QHD) in sewal97 .
At larger renormalization scales, $`\mu `$, the global pattern of the solution remains qualitatively similar, as seen from Fig. 6. Only $`T_{\text{cross}}`$ and $`T_{\text{end}}`$ move to higher temperatures. Inspecting the low-temperature region in more detail, cf. Fig. 7, we see that a new metastable solution, which ends already at rather low temperature, appears. This new solution has a nonzero field (Fig. 8) but violates the NG theorem. In addition the metastable solution, which before started at $`T_1`$, now begins at zero temperature, since at $`\mu =`$ 1.2 GeV we have already $`T_1^2<0`$, cf. Eq. (61).
If we take a low value for the scale $`\mu `$, the structure of the stable solution becomes more involved, see Figs. 911. In the broken-symmetry sector we still have a massless pion and a nonzero field, see Fig. 10. However, at higher temperatures the metastable branch, displayed by the dotted line, reveals back-bending. This back-bended part of the branch turns out to be the most stable one, see Fig. 11. As a result we arrive at a complicated structure for the stable solution, where even the pressure turns out to be discontinuous.
A common feature for all scales is that the point of the first phase transition, $`T_2180`$ MeV, is approximately $`\mu `$-independent and has about the same value as that in the naive renormalization scheme IRK05 . At the same time, the point of the second phase transition, $`T_{\text{cross}}`$, and the end point of the stable solution, $`T_{\text{end}}`$, are essentially $`\mu `$-dependent.
### IV.2 Approximate $`O(4)`$ Symmetry
In the case of explicitly broken symmetry ($`m_\pi =139`$ MeV), the structure of solutions at various $`\mu `$ scales is similar to that described above for the chiral limit. Even the behavior of metastable branches remains similar. We illustrate the changes on the example of $`\mu =600`$ MeV, see Figs. 1214 which looks the most physically appealing and is close to results of the naive renormalization IRK05 . The main difference from the $`m_\pi =0`$ case is that the sequence of two phase transitions is transformed here into a smooth cross-over transition.
## V Conclusion
We have studied a renormalized version of the gapless $`\mathrm{\Phi }`$-derivable HF approximation to the $`\lambda \varphi ^4`$ theory with spontaneous breaking of the $`O(N)`$ symmetry, proposed in Ref. IRK05 . This gHF approximation simultaneously preserves all the desirable features of $`\mathrm{\Phi }`$-derivable approximation schemes (i.e., the validity of conservation laws and thermodynamic consistency) and respects the NG theorem in the phase of spontaneously broken symmetry. This is achieved by adding a correction $`\mathrm{\Delta }\mathrm{\Phi }`$ to the conventional $`\mathrm{\Phi }`$ functional. The nature of this correction can be understood as follows. The conventional $`\mathrm{\Phi }`$-derivable HF approximation cuts off an infinite series of diagrams, among which are those providing the NG theorem in the phase of spontaneously broken symmetry. By introducing the $`\mathrm{\Delta }\mathrm{\Phi }`$ correction to the HF approximation we take into account a part of those omitted diagrams (at the level of the actual approximation), which restores the NG theorem.
An advantage of the gHF approximation is that it allows for scale-independent renormalization in the vacuum, unlike the conventional HF approximation Lenaghan . The scale independence in the vacuum is a direct consequence of the NG theorem, which makes the equations for the pion self-energy and classical field degenerated. However, even in the gHF approximation, only renormalization scales higher than a certain value ($`\mu _0`$, cf. Eq. (56)) are allowed, in order to ensure stability of the renormalized approximation.
Nevertheless, the scale dependence still persists at finite temperatures. The violation of renormalization-scale independence of $`\mathrm{\Phi }`$-derivable approximations was shown in BP02 ; Destri05 from the point of view of the renormalization-group $`\beta `$ function. There the $`\beta `$ function, evaluated from the $`\mathrm{\Phi }`$-functional formalism, was shown to deviate from its perturbative expansion, beginning at orders in the expansion parameter, higher than that explicitly taken in the $`\mathrm{\Phi }`$ functional. The reason is the violation of “crossing symmetry” in the sense of HK1 : Solving the self-consistent equations of motion corresponds to a partial resummation of diagrams to any order in the expansion parameter (e.g., the coupling constant $`\lambda `$ or $`\mathrm{}`$, i.e., the order of loops in perturbative Feynman diagrams) which is necessarily incomplete for any truncation of the $`\mathrm{\Phi }`$ functional.
Within our renormalization scheme, it becomes clear that the renormalization-scale dependence at finite temperatures originates from the subtraction of the “hidden subdivergence” of the four-point function inside the self-consistent tadpole loop. As shown in HK1 , this four-point function consists of a resummation in only one channel, and thus the $`\beta `$ function of this resummed four-point function deviates from the correct one at orders higher than contained in the approximation of the $`\mathrm{\Phi }`$-functional, i.e., to $`𝒪(\lambda ^2)`$.
At large scales ($`\mu m_\sigma `$) the chiral phase transition proceeds similar to that in the naive renormalization scheme IRK05 . In the case of the exact $`O(N)`$ symmetry, it proceeds through a sequence of two second-order phase transitions rather than a single one. In the first transition the mean field vanishes but the meson masses still remain different. The temperature of this phase transition, $`T_2180`$ MeV, is approximately $`\mu `$-independent and has approximately the same value as that in the naive renormalization scheme IRK05 . In the second transition also the masses become equal, and the $`O(N)`$ symmetry is completely restored. The corresponding temperature, $`T_{\text{cross}}`$, turns out to be essentially $`\mu `$-dependent. When the $`O(N)`$ symmetry is explicitly violated, the sequence of two phase transitions is transformed into a smooth cross-over transition. Moreover, at $`\mu m_\sigma `$ the results are even qualitatively close to those obtained with the naive renormalization IRK05 . At small scales (say $`\mu m_\sigma /2`$), the phase structure becomes very complicated, however still respecting the NG theorem in the phase of spontaneously broken symmetry.
Another result concerns both the conventional renormalized HF and gHF approximations, which in fact are identical in the phase of restored $`O(N)`$ symmetry. There exists an end point of the solution, i.e., a temperature $`T_{\text{end}}`$ above which there are no solutions to the gap equations. The occurrence of such an end point in the HF approximation was first pointed out by Baym and Grinstein Baym-Grin and is caused by the dominant role of $`\mathrm{ln}(M^2/\mu ^2)`$ terms in the gap equations, which originate from the renormalization procedure. The dominant behavior of the $`\mathrm{ln}(M_{\pi /\sigma }^2/\mu ^2)`$ terms signals a breakdown of the HF approximation and the need to include higher-order corrections into the $`\mathrm{\Phi }`$ functional Baym-Grin . Another interesting question is whether one can find a renormalization-group improved $`\mathrm{\Phi }`$-derivable approximation to cure this problem. We have found that at low scales (like $`\mu m_\sigma /2`$), the end point appears at rather low temperatures, leaving almost no room for the HF (as well as gHF) approximation in the phase of restored $`O(N)`$ symmetry. At the same time, at ($`\mu m_\sigma `$), the end point moves to rather high energies $`T_{\text{end}}1.2`$ GeV, hence allowing this approximation at least at $`T1`$ GeV.
Summarizing, we have found that the gHF approximation is certainly advantageous as compared to the conventional HF one, since it respects the NG theorem and allows scale-independent renormalization in the vacuum. These properties are closely interrelated. They both require that the set of equations of motion are degenerate in the phase of spontaneously broken symmetry. Nevertheless, there still are serious problems with the renormalization of $`\mathrm{\Phi }`$-derivable approximations for a theory with a spontaneously broken symmetry. At finite temperatures the predictions of renormalized gHF approximations essentially depend on the renormalization scale, contrary to the case of the renormalized perturbation theory. In this respect, the gHF approximation becomes similar to the leading-order $`1/N`$-approximation, where also both the NG theorem and scale-independent renormalization in the vacuum hold true Lenaghan . However, in view of the medium-independent renormalization performed in accordance with Refs. HK3 ; HK1 ; Reinosa1 , this scale-dependence at finite temperatures cannot be already interpreted as an artifact of temperature-dependent counter terms. It may turn out that this scale dependence is a consequence of triviality of the $`\lambda \varphi ^4`$ theory Consoli which, when it is renormalized, therefore requires an external scale to serve for a scale, below which it can be used as an effective field theory to describe the low-energy phenomenology.
###### Acknowledgements.
We are grateful to B. Friman and D.N. Voskresensky for useful discussions. One of the authors (Y.I.) acknowledges partial support by the Deutsche Forschungsgemeinschaft (DFG project 436 RUS 113/558/0-2), the Russian Foundation for Basic Research (RFBR grant 03-02-04008) and Russian Minpromnauki (grant NS-1885.2003.2). H.v.H. acknowledges partial support by the U.S. National Science Foundation under grant PHY-0449489 and by the Alexander von Humboldt Foundation as a Feodor-Lynen fellow.
|
warning/0506/hep-th0506088.html
|
ar5iv
|
text
|
# Meromorphic Scaling Flow of 𝑁=2 Supersymmetric 𝑆𝑈(2) Yang-Mills with Matter
## 1 Introduction
In a previous paper the flow of $`N=2`$ SUSY pure $`SU(2)`$ Yang-Mills, with no matter fields, was analysed and a meromorphic $`\beta `$-function was constructed which is finite at both weak and strong coupling. Up to a constant factor this $`\beta `$-function reproduces the correct 1-loop Callan-Symanzik flow at both strong and weak coupling and interpolates between them analytically, although there is no a priori reason to interpret it as Callan-Symanzik $`\beta `$-function away from the region of the fixed points. This analysis modified previous suggestions in the literature concerning the $`\beta `$-functions for $`N=2`$ SUSY and evades the criticisms in .
In the present paper the analysis is extended to include massless matter fields in the fundamental representation of $`SU(2)`$ with $`N_f=1,2,3`$ flavours. The construction uses the gauge invariant flow parameter $`u=tr<\phi ^2>`$, where $`\phi `$ is the Higgs field whose VEV is a free parameter, and the fact that the $`\beta `$-functions are modular forms of weight $`2`$ of a sub-group of the full modular group, $`\mathrm{\Gamma }(1)Sl(2,𝐙)/𝐙_2`$, depending on $`N_f`$. The significance of the parameter $`u`$ was emphasised in where is was shown that $`u`$ is the Legendre transform of the pre-potential.
Following a convenient choice of modular parameter for $`N_f>0`$ is
$$\tau =\frac{\theta }{\pi }+\frac{8\pi i}{g^2},$$
(1)
where $`\theta `$ is the usual topological parameter labelling $`\theta `$-vacua and $`g`$ is the Yang-Mills coupling constant. In terms of $`\tau `$ the relevant sub-groups of $`Sl(2,𝐙)`$ for determining the $`\beta `$-functions are: $`\mathrm{\Gamma }^0(2)`$ for $`N_f=0`$; $`\mathrm{\Gamma }(1)`$ for $`N_f=1`$; $`\mathrm{\Gamma }_0(2)`$ for $`N_f=2`$; and $`\mathrm{\Gamma }_0(4)`$ for $`N_f=3`$.<sup>1</sup><sup>1</sup>1The notation is that of and is summarised in the appendix for ease of reference. For $`N_f=0`$ and $`N_f=2`$ these are larger than the monodromy group. The $`N_f=1`$ case realises the full modular group, so the self-dual point $`\tau =i`$ is a fixed point of the element sending $`\tau \frac{1}{\tau }`$ which is in the monodromy group. All the $`\beta `$-functions discussed here do still have singularities somewhere in the fundamental domain of the relevant sub-group of $`\mathrm{\Gamma }(1)`$, they must do since any modular form of weight $`2`$ must have at least one singularity within, or on the boundary of, the fundamental domain, but these singularities are off the real axis and, with the exception of $`N_f=1`$, correspond to repulsive fixed points in both directions of the flow. For the $`N_f=1`$ $`\beta `$-functions there are two types of singularities off the real axis: one at $`\tau =i`$ and its images under $`\mathrm{\Gamma }(1)`$ which is repulsive in both directions of the flow; and one at $`\tau =e^{i\pi /3}`$ and its images under $`\mathrm{\Gamma }(1)`$ which is attractive in the direction of decreasing Higgs VEV.
The strategy is to use the technique of where the Seiberg-Witten curves describing the various theories are written both in terms of the ‘bare’ $`\tau =i\mathrm{}`$ and the renormalised finite $`\tau `$ and the co-efficients compared to extract $`\tau (u)`$. An important tool in the analysis is the modular symmetry derived in where it was shown that $`N=2`$ SUSY Yang-Mills has an infinite hierarchy of vacua with massless BPS states and the modular group relates these vacua to each other. The value of $`\tau `$ in one vacuum is related to that of another by
$$\tau \gamma (\tau )=\frac{a\tau +b}{c\tau +d}$$
(2)
where $`\gamma =\left(\begin{array}{cc}a& b\\ c& d\end{array}\right)\mathrm{\Gamma }\mathrm{\Gamma }(1)`$ with $`\mathrm{\Gamma }`$ a sub-group of the full modular group $`\mathrm{\Gamma }(1)PSl(2,𝐙)`$. We therefore have, under a variation $`\delta \tau `$ of $`\tau `$,
$$\delta \gamma (\tau )=\frac{1}{(c\tau +d)^2}\delta \tau ,$$
(3)
since $`adbc=1`$, so we expect that
$$\beta (\gamma (\tau ))=\frac{1}{(c\tau +d)^2}\beta (\tau ).$$
(4)
If $`\beta `$ is meromorphic, it will be a modular form of weight -2 and this fact proves to be a powerful analytical tool.
In §2 the $`\beta `$-function for $`N_f=0`$ discussed in is re-derived in terms of (1), which differs from the normalisation in . The cases $`N_f=2`$, $`N_f=3`$ and $`N_f=1`$ are treated in sections 3, 4 and 5 respectively where meromorphic $`\beta `$-functions are proposed which vanish at all the strong coupling fixed points and are constant at the weak coupling fixed point.
In principle the $`\beta `$-functions for different values of $`N_f`$ should be related by holomorphic decoupling but the inclusion of non-zero masses for the matter multiplets makes analytic calculations much harder and cannot be pushed through using the techniques of the present analysis. In section 6 a perturbative approach is presented, using a strong coupling expansion and turning on a mass for one of the matter fields. It is shown that the limits $`\tau _D=1/\tau i\mathrm{}`$ and $`m0`$ do not commute, but nevertheless a $`\beta `$-function for the massive theory with acceptable behaviour near $`\tau _D=i\mathrm{}`$ can be constructed.
Section 7 contains our conclusions. Two appendices give a summary of the conventions concerning Jacobi $`\vartheta `$-functions and the technical aspects of the strong coupling instanton expansion used for the analysis in section 6.
## 2 $`N_f=0`$
This case was treated in using the normalisation appropriate to the adjoint representation of $`SU(2)`$, $`\stackrel{~}{\tau }=\frac{\theta }{2\pi }+\frac{4\pi i}{g^2}`$, but when matter in the fundamental representation of $`SU(2)`$ is included it is better to define $`\tau =\frac{\theta }{\pi }+\frac{8\pi i}{g^2}`$, . In order to set the notation and illustrate the method for $`N_f0`$ the derivation of the $`\beta `$-function in is given here using the original techniques of , adapted to the present notation.
First recall the monodromy of the $`N_f=0`$ theory . It is generated by
$$_0=\left(\begin{array}{cc}1& 0\\ 1& 1\end{array}\right),_{\mathrm{}}=\left(\begin{array}{cc}1& 4\\ 0& 1\end{array}\right)$$
(5)
with
$$_{\mathrm{}}_0=_2^1\text{where}_2=\left(\begin{array}{cc}1& 4\\ 1& 3\end{array}\right),$$
(6)
($`_0`$ leaves $`\tau =0`$ invariant, $`_2`$ leaves $`\tau =2`$ invariant and $`_{\mathrm{}}`$ leaves $`\tau =i\mathrm{}`$ invariant). The two matrices $`_{\mathrm{}}`$ and $`_0^1`$,
$$\tau \tau +4,\tau \frac{\tau }{\tau +1},$$
(7)
generate $`\mathrm{\Gamma }^0(4)`$ and therefore $`\beta `$ will be a modular form for $`\mathrm{\Gamma }^0(4)`$ of weight -2. To determine its explicit form we start by following the method of . The massless $`N_f=4`$ curve can be written
$$y^2=x^42\overline{u}F(\tau )x^2+\overline{u}^2,$$
(8)
where
$$F(\tau )=\frac{\vartheta _3^4(\tau )+\vartheta _2^4(\tau )}{\vartheta _4^4(\tau )}$$
(9)
and the $`N_f=0`$ curve of can be written as
$$y^2=x^42ux^2+u^2\mathrm{\Lambda }_0^4.$$
(10)
Note that $`F(\tau )`$ is an invariant function under $`\mathrm{\Gamma }(2)`$.
Equation (8) is scale invariant in the sense that $`\tau `$ does not depend on $`\overline{u}`$. Equating co-efficients and eliminating $`\overline{u}`$ gives
$$F(\tau )=\frac{u}{\sqrt{u^2\mathrm{\Lambda }_0^4}}$$
(11)
hence
$$u\frac{d\tau }{du}F^{}(\tau )=F(F^21).$$
(12)
where $`F^{}=\frac{dF}{d\tau }`$. Using (101) to evaluate $`F^{}(\tau )`$ from (9) gives
$$F^{}=2\pi i\frac{\vartheta _3^4(\tau )\vartheta _2^4(\tau )}{\vartheta _4^4(\tau )}$$
(13)
finally leading to
$$u\frac{d\tau }{du}=\frac{2i}{\pi }\frac{\vartheta _3^4(\tau )+\vartheta _2^4(\tau )}{\vartheta _4^8(\tau )}$$
(14)
which is the result of , after taking account of the difference of notation (the variable $`\tau `$ here is $`2\tau `$ in ). The asymptotic behaviour of (14) is
$$u\frac{d\tau }{du}\underset{\tau i\mathrm{}}{}\frac{2i}{\pi }$$
(15)
which is the correct asymptotically free behaviour. However it was pointed out in that (14) has a pathology in that it is singular at strong coupling, in particular at $`\tau =0`$ ($`u=\mathrm{\Lambda }_0^2`$) and $`\tau =2`$ ($`u=\mathrm{\Lambda }_0^2`$).
A remedy was proposed in , based on . The idea is to tame the singularities at strong coupling, without disturbing the asymptotic properties at $`u\mathrm{}`$, by defining
$$\beta (\tau )=\frac{(u\mathrm{\Lambda }_0^2)^m(u+\mathrm{\Lambda }_0^2)^n}{u^{m+n}}u\frac{d\tau }{du},$$
(16)
where $`m`$ and $`n`$ are positive integers (the sign is chosen so the direction of flow is the same as in ). The correct behaviour at strong coupling then forces $`m=n=1`$.
The motivation behind (16) relies on a theorem that any modular form of weight -2 for a sub-group $`\mathrm{\Gamma }\mathrm{\Gamma }(1)`$ can always be written as
$$\beta (\tau )=\frac{P(f)}{Q(f)f^{}}$$
(17)
where $`f(\tau )`$ is a particular invariant function for $`\mathrm{\Gamma }`$ (essentially one with the smallest number of zeros plus number of poles, counting multiplicity) and $`P(f)`$ and $`Q(f)`$ are polynomials in $`f`$ (see e.g , page 111, the idea of applying this theorem to $`N=2`$ SUSY was first proposed in ). Now $`u(\tau )`$ is just such an invariant function for $`\mathrm{\Gamma }^0(4)`$ — the explicit form of $`u(\tau )`$ follows from (9) and (11)
$$\frac{u}{\mathrm{\Lambda }_0^2}=\frac{\vartheta _3^4(\tau )+\vartheta _2^4(\tau )}{2\vartheta _3^2(\tau )\vartheta _2^2(\tau )},$$
(18)
and it can be verified explicitly, using (95) and (96), that $`u`$ is invariant under $`\mathrm{\Gamma }^0(4)`$. Choosing the zeros of $`P`$ and $`Q`$ so as to get the correct asymptotic behaviour at $`\tau =0,2`$ and $`i\mathrm{}`$ without introducing unnecessary zeros or poles, one is led to (16) and, using $`m=n=1`$ in conjunction with (14) and (18), this leads uniquely to
$$\beta (\tau )=\frac{(u\mathrm{\Lambda }_0^2)(u+\mathrm{\Lambda }_0^2)}{u}\frac{d\tau }{du}=\frac{2}{\pi i}\frac{1}{\vartheta _3^4(\tau )+\vartheta _2^4(\tau )}.$$
(19)
The asymptotic behaviour is
$`\beta (\tau )`$ $``$ $`{\displaystyle \frac{2}{\pi i}}\text{as}\tau i\mathrm{}`$
$`\beta (\tau )`$ $``$ $`{\displaystyle \frac{1}{\pi i}}\tau ^2\text{as}\tau 0`$ (20)
$`\beta (\tau )`$ $``$ $`{\displaystyle \frac{1}{\pi i}}(\tau 2)^2\text{as}\tau 2.`$
In particular the behaviour $`\beta \frac{i}{\pi }\tau ^2`$ as $`\tau 0`$ implies that $`\beta (\tau _D)\frac{i}{\pi }`$ as $`\tau _Di\mathrm{}`$, where $`\tau _D=1/\tau `$ is the dual coupling. Had any value of $`n`$ been used in (16) other than unity, the result near $`\tau 0`$ would have been
$$\beta (\tau _D)\left(e^{2i\pi \tau _D}\right)^{n1}\frac{i}{\pi }$$
(21)
which is not the correct asymptotic behaviour. An exactly parallel argument applies at $`\tau =2`$ with $`n`$ replaced by $`m`$.
Near both the weak and the strong coupling fixed points this $`\beta `$-function can be interpreted as a Callan-Symanzik $`\beta `$-function for $`N_f=0`$ SUSY $`SU(2)`$ Yang-Mills because $`u`$ is a mass squared at weak coupling, where
$$\beta u\frac{d\tau }{du},$$
(22)
while at strong coupling near $`\tau 0`$, where $`u\mathrm{\Lambda }_0^2\mathrm{\Lambda }_0a_D`$ with $`a_D`$ proportional to the mass of the BPS monopole, we have
$$\beta 2(u\mathrm{\Lambda }_0^2)\frac{d\tau }{du}2a_D\frac{d\tau }{da_D}.$$
(23)
The flow is shown in figure 1. As the Higgs VEV is lowered there are attractive fixed points for all even integral $`\tau `$ and repulsive fixed points for all odd integral $`\tau `$. There are repulsive fixed points in both flow directions at $`\tau =k+i`$ for all odd $`k`$ ($`\beta (\tau )`$ diverges at these points because $`\vartheta _3^4(\tau )=\vartheta _2^4(\tau )`$ there). From the transformation properties of the $`\vartheta `$-functions in the appendix (95) and (96), the $`\beta `$-function in (19) is a modular form of weight -2 for the group generated by
$$\tau \tau +2,\tau \frac{\tau }{\tau +1},$$
(24)
which is the group $`\mathrm{\Gamma }^0(2)`$. $`\mathrm{\Gamma }^0(4)`$ is a sub-group of $`\mathrm{\Gamma }^0(2)`$ and the fact that the $`\beta `$-function has a larger symmetry than that of the monodromy group is due to the $`𝐙_2`$ symmetry of $`\beta `$ under $`uu`$, enforced by choosing $`m=n`$ in (16). This $`𝐙_2`$ is the anomaly-free residue of global $`R`$-symmetry which is not a symmetry of the effective action. Although $`u`$ is not invariant under $`\mathrm{\Gamma }^0(2)`$, it changes sign under $`\tau \tau +2`$, $`u^2`$ is invariant.
There are attractive fixed points at all the images of $`\tau =0`$ under $`\mathrm{\Gamma }^0(2)`$ and repulsive fixed points at all the images of $`\tau =1`$ under $`\mathrm{\Gamma }^0(2)`$: at strong coupling all rational values $`\tau =q/m`$ with $`q`$ even are attractive fixed points and those with $`q`$ odd are repulsive (with $`q`$ and $`m`$ mutually prime) as the Higgs VEV is decreased.
The $`\beta `$-functions in , using $`\stackrel{~}{\tau }=\tau /2`$ rather than $`\tau `$, are the same as (19) above as can be checked using (A) from which
$$\beta (\stackrel{~}{\tau })=\frac{1}{2}\beta (\tau )=\frac{2}{\pi i}\frac{1}{\vartheta _3^4(\stackrel{~}{\tau })+\vartheta _4^4(\stackrel{~}{\tau })}.$$
(25)
These are modular forms for $`\mathrm{\Gamma }_0(2)`$ which is generated by
$$\stackrel{~}{\tau }\stackrel{~}{\tau }+1,\stackrel{~}{\tau }\frac{\stackrel{~}{\tau }}{2\stackrel{~}{\tau }+1},$$
(26)
equivalent to
$$\tau \tau +2,\tau \frac{\tau }{\tau +1}$$
(27)
with $`\tau =2\stackrel{~}{\tau }`$. The details of the flow generated by (25) were analysed in .
## 3 $`N_f=2`$
The results of the last section can be immediately be used to guess the form of the $`\beta `$-function for $`N_f=2`$, which can then be verified explicitly. The monodromy for the $`N_f=2`$ case is generated by
$$_0=\left(\begin{array}{cc}1& 0\\ 2& 1\end{array}\right),_{\mathrm{}}=\left(\begin{array}{cc}1& 2\\ 0& 1\end{array}\right),$$
(28)
with
$$_{\mathrm{}}_0=_1^1\text{where}_1=\left(\begin{array}{cc}1& 2\\ 2& 3\end{array}\right)$$
(29)
($`_0`$ leaves $`\tau =0`$ invariant, $`_1`$ leaves $`\tau =1`$ invariant and $`_{\mathrm{}}`$ leaves $`\tau =i\mathrm{}`$ invariant). The two matrices $`_{\mathrm{}}`$ and $`_0^1`$,
$$\tau \tau +2,\tau \frac{\tau }{2\tau +1},$$
(30)
generate $`\mathrm{\Gamma }(2)`$. If a $`\beta `$-function is constructed which is invariant under the anomaly-free $`𝐙_2`$ acting on the $`u`$-plane, $`uu`$, along the same lines as before then we expect it will have a further symmetry under
$$\tau \tau +1,$$
(31)
and so will be a modular form of $`\mathrm{\Gamma }_0(2)`$ of weight -2. Demanding the correct behaviour at $`\tau =i\mathrm{}`$, $`\tau =0`$ and $`\tau =1`$ leaves only one possibility and that is (25) with $`\stackrel{~}{\tau }`$ replaced by $`\tau `$, namely
$$\beta (\tau )=\frac{2}{\pi i}\frac{1}{\vartheta _3^4(\tau )+\vartheta _4^4(\tau )},$$
(32)
the normalisation being determined by the asymptotic condition
$$\beta (\tau )\underset{\tau i\mathrm{}}{}\frac{1}{\pi i}.$$
(33)
We now verify (32) by explicit calculation using the same technique as in the previous section. The analysis initially parallels that of : the massless $`N_f=2`$ curve is
$$y^2=x^42(u+3\mathrm{\Lambda }_2^2/8)x^2+(u\mathrm{\Lambda }_2^2/8)^2.$$
(34)
Equating co-efficients with (8) gives
$$F(\tau )=\frac{u+3\mathrm{\Lambda }_2^2/8}{u\mathrm{\Lambda }_2^2/8}$$
(35)
or equivalently
$$u(\tau )=\frac{\mathrm{\Lambda }_2^2}{8}\frac{\vartheta _3^4(\tau )}{\vartheta _2^4(\tau )}.$$
(36)
Using (9) and (13), now leads to
$$u\frac{d\tau }{du}=\frac{(F1)(F+3)}{4F^{}}=\frac{i}{2\pi }\frac{\vartheta _3^4(\tau )+\vartheta _4^4(\tau )}{\vartheta _3^4(\tau )\vartheta _4^4(\tau )}$$
(37)
as in . There are singularities at $`\tau =0`$ (where $`u=\mathrm{\Lambda }_2^2/8`$) and $`\tau =1`$ (where $`u=\mathrm{\Lambda }_2^2/8`$) which can be modified, along lines similar to §2, to give a $`\beta `$-function with the correct strong coupling behaviour,
$$\beta (\tau )=\frac{\left(u^2\frac{\mathrm{\Lambda }_2^4}{64}\right)}{u^2}u\frac{d\tau }{du}=\frac{2}{\pi i}\frac{1}{\vartheta _3^4(\tau )+\vartheta _4^4(\tau )},$$
(38)
confirming (32). This flow is essentially the same as that of the $`N_f=1`$ case treated in §2, except that $`\tau `$ is rescaled by a factor of $`1/2`$, and the flow shown in figure 2.
## 4 $`N_f=3`$
The monodromy for $`N_f=3`$ is generated by
$$_0=\left(\begin{array}{cc}1& 0\\ 4& 1\end{array}\right),_{\mathrm{}}=\left(\begin{array}{cc}1& 1\\ 0& 1\end{array}\right)$$
(39)
with
$$_0_{\mathrm{}}=_{1/2}^1\text{where}_{1/2}=\left(\begin{array}{cc}3& 1\\ 4& 1\end{array}\right)$$
(40)
($`_0`$ leaves $`\tau =0`$ invariant, $`_{1/2}`$ leaves $`\tau =1/2`$ invariant and $`_{\mathrm{}}`$ leaves $`\tau =i\mathrm{}`$ invariant), . The two matrices $`_{\mathrm{}}`$ and $`_0^1`$,
$$\tau \tau +1,\tau \frac{\tau }{4\tau +1},$$
(41)
generate $`\mathrm{\Gamma }_0(4)`$. But it is not obvious what the full symmetry of the $`\beta `$-functions might be as $`𝐙_2`$ does not play any role in the $`N_f=3`$ theory . We must therefore perform the explicit calculation. Our starting point this time is the massless $`N_f=4`$ curve in the original form of , namely
$$y^2=x^3\frac{1}{4}g_2(\tau )\overline{u}^2x\frac{1}{4}g_3(\tau )\overline{u}^3$$
(42)
with
$$g_2(\tau )=\frac{2}{3}\left(\vartheta _2^8(\tau )+\vartheta _3^8(\tau )+\vartheta _4^8(\tau )\right)$$
(43)
and
$$g_3(\tau )=\frac{4}{27}\left(\vartheta _3^4(\tau )+\vartheta _4^4(\tau )\right)\left(\vartheta _2^4(\tau )+\vartheta _3^4(\tau )\right)\left(\vartheta _4^4(\tau )\vartheta _2^4(\tau )\right),$$
(44)
the combination
$$\frac{g_2^3(\tau )}{g_2^3(\tau )27g_3^2(\tau )}=\frac{(\vartheta _2^8+\vartheta _3^8+\vartheta _4^8)^3}{54\vartheta _2^8\vartheta _3^8\vartheta _4^8}=J(\tau )$$
(45)
being Klein’s absolute invariant which is invariant under the action of $`\mathrm{\Gamma }(1)`$ on $`\tau `$.
The massless $`N_f=3`$ curve is
$$y^2=\left(x^2\frac{\mathrm{\Lambda }_3^2}{64}(xu)\right)(xu).$$
(46)
Without loss of generality we can set<sup>2</sup><sup>2</sup>2This is equivalent to defining the dimensionless variable $`\stackrel{~}{u}=64u/(\mathrm{\Lambda }_3^2))`$ and then dropping the tilde. $`\mathrm{\Lambda }_3^2=64`$ and then eliminate the quadratic term in (46) by shifting $`xx+(u+1)/3`$ giving
$$y^2=x^3\frac{1}{3}(u^24u+1)x\frac{1}{27}(2u1)(u^2+8u2).$$
(47)
There are singular points were the roots of this equation coincide, at $`u=0`$, $`u=1/4`$ and $`|u|=\mathrm{}`$, corresponding to $`\tau =0`$, $`\tau =1/2`$ and $`\tau =i\mathrm{}`$ respectively.
Equating co-efficients between (42) and (47) and eliminating $`\overline{u}`$ yields
$$J(\tau )=\frac{4}{27}\frac{(u^24u+1)^3}{u^4(14u)}.$$
(48)
To extract the $`\beta `$-function from this equation we will need an explicit expression for $`u(\tau )`$. To this end define
$$Y(u)=\frac{u^2}{14u}\text{and}X(\tau )=\frac{\vartheta _4^4(\tau )}{\vartheta _3^4(\tau )}=\frac{2}{F(\tau )+1}$$
(49)
in terms of which (48) reads
$$\frac{(1+Y)^3}{Y^2}=\frac{(1X+X^2)^3}{(1X)^2X^2}$$
(50)
with the three roots
$$Y_1=X(1X),Y_2=\frac{(1X)}{X^2}\text{and}Y_3=\frac{X}{(1X)^2}.$$
(51)
Comparing with the three asymptotic forms
$`\tau i\mathrm{},`$ $`X1,|u|\mathrm{},Y\mathrm{},`$
$`\tau 0,`$ $`X0,u0,Y0,`$ (52)
$`\tau 1/2,`$ $`X1,u1/4,Y\mathrm{},`$
only $`Y_3`$ has the correct asymptotic behaviour at the three singular points and so we must choose
$$\frac{u^2}{14u}=\frac{X}{(1X)^2}=\frac{\vartheta _3^4\vartheta _4^4}{\vartheta _2^8},$$
(53)
which is an invariant function for $`\mathrm{\Gamma }_0(2)`$ . Solving for $`u`$ the asymptotic conditions pick out the unique solution
$$u(\tau )=\frac{\vartheta _3^2\vartheta _4^2}{(\vartheta _3^2\vartheta _4^2)^2}$$
(54)
which is, of course, an invariant function for $`\mathrm{\Gamma }_0(4)`$.
Differentiating this equation with respect to $`\tau `$, and employing (101), yields
$$u\frac{d\tau }{du}=\frac{2i}{\pi }\frac{1}{\left(\vartheta _3^2(\tau )+\vartheta _4^2(\tau )\right)^2}.$$
(55)
This has the correct asymptotic form as $`\tau i\mathrm{}`$,
$$u\frac{d\tau }{du}\underset{\tau i\mathrm{}}{}\frac{i}{2\pi },$$
(56)
and is well behaved at $`\tau =0`$
$$u\frac{d\tau }{du}\underset{\tau 0}{}\frac{2i}{\pi }\tau ^2,$$
(57)
but diverges at $`\tau =1/2`$.
The method used in the previous sections for $`N_f=0`$ and $`N_f=2`$ is not immediately applicable here since one of the fixed points is at $`u=0`$: to eliminate the singularity at $`u=1/4`$ we would have to multiply by $`(u1/4)^m/u^m`$ for some positive $`m`$ which would introduce another singularity at $`u=0`$. One strategy is to use $`m=1`$ to remove the singularity at $`u=1/4`$ and shift the other fixed point away from the origin by shifting $`u`$: thus let $`u^{}=uϵ`$ for some constant $`ϵ`$ and define
$$\beta (\tau )=\frac{(u^{}+ϵ1/4)(u^{}+ϵ)}{u^{}}\frac{d\tau }{du^{}}.$$
(58)
This does not disturb the behaviour at $`|u|=\mathrm{}`$. Clearly this is equivalent to
$$\beta (\tau )=\frac{(u1/4)}{(uϵ)}u\frac{d\tau }{du}$$
(59)
which preserves the good behaviour at $`\tau =i\mathrm{}`$ and $`\tau =0`$ and gives the correct behaviour at $`\tau =1/2`$.
In terms of $`\vartheta `$-functions
$$\beta (\tau )=\frac{1}{2\pi i}\frac{1}{\vartheta _3^2(\tau )\vartheta _4^2(\tau )+ϵ\left(\vartheta _3^2(\tau )\vartheta _4^2(\tau )\right)^2},$$
(60)
and has the following asymptotic forms
$`\beta (\tau )`$ $`\underset{\tau \mathrm{}}{}{\displaystyle \frac{1}{2\pi i}},`$
$`\beta (\tau )`$ $`\underset{\tau 0}{}{\displaystyle \frac{2}{\pi i}}{\displaystyle \frac{\tau ^2}{ϵ}},`$
$`\beta (\tau )`$ $`\underset{\tau 1/2}{}{\displaystyle \frac{2}{\pi i}}{\displaystyle \frac{(\tau +1/2)^2}{(14ϵ)}}.`$
If $`ϵ`$ is real and $`0<ϵ<1/4`$, the $`\beta `$-function is finite at both $`\tau =0`$ and $`\tau =1/2`$ and flows in the right direction, i.e. in towards the fixed points as the Higgs VEV is lowered.
In fact just such a constant shift of $`u`$ was found to be necessary in instanton calculations performed to check the validity of the Seiberg-Witten curve . These instanton calculations give $`ϵ=4/27`$ when $`\mathrm{\Lambda }_3/8`$ is set to one as we have done here. The point is that, since $`N_f=3`$ has no discrete symmetry acting on the $`u`$-plane, there is no a priori way, using the techniques in , to determine where the origin of the $`u`$-plane should lie and a constant finite shift of $`u`$ does not affect the weak coupling physics. Seiberg and Witten chose $`u=0`$ to correspond to $`\tau =0`$ but the instanton calculations show that $`u^{}=4/27`$ is a more natural choice of origin.
The flow (60) with $`ϵ=4/27`$ is plotted in figure 3. At first glance it looks very like the flow for $`N_f=0`$ and $`N_f=2`$, with $`\tau `$ rescaled, but closer examination reveals subtle differences. Figure 3 is not symmetric under $`\tau \tau +1/2`$, it is slightly distorted and the repulsive fixed point close to $`\tau =(1+i)/4`$ is not exactly at $`\tau =(1+i)/4`$, it is displaced away from the top of the semi-circular arch by a small amount. The flow looks like a distorted version of the $`N_f=2`$ flow with $`\tau `$ rescaled by a factor of $`2`$, $`\tau _{N_f=2}=2\tau _{N_f=3}`$. This is because $`\mathrm{\Gamma }^0(4)`$ acting on $`\tau `$ is equivalent to $`\mathrm{\Gamma }(2)`$ acting on $`2\tau `$: the $`N_f=3`$ $`\beta `$-functions for $`2\tau `$ are therefore modular forms of $`\mathrm{\Gamma }(2)`$. Modular forms of $`\mathrm{\Gamma }(2)`$ can be obtained by distorting modular forms of $`\mathrm{\Gamma }_0(2)`$. Provided $`0<ϵ<1/4`$ the unstable fixed point of the $`N_f=3`$ flow lies on the semi-circle in the upper-half $`\tau `$-plane spanning the two points $`\tau =0`$ and $`\tau =1/2`$ on the real axis. The special case $`ϵ=1/8`$ has a higher symmetry than other values because this corresponds to the unstable fixed point being at the top of the semi-circular arch at $`\tau =(1+i)/4`$ and this gives $`\mathrm{\Gamma }_0(2)`$ symmetry acting on $`2\tau `$. Other values of $`ϵ`$ have the lower symmetry of $`\mathrm{\Gamma }(2)`$, as in figure 3. Flow diagrams like figure 3 have been postulated for the quantum Hall effect when the electron spins are poorly split .
## 5 $`N_f=1`$
The $`N_f=1`$ case has been left till last because it is more involved than the three cases already considered, though paradoxically it has a higher symmetry — the monodromy generates the full modular group $`\mathrm{\Gamma }(1)`$. The monodromy is calculated in : there are four singular points, at $`\tau =0`$, $`\tau =1`$, $`\tau =2`$ and $`\tau =i\mathrm{}`$, with monodromies
$`_0`$ $`=`$ $`\left(\begin{array}{cc}1& 0\\ 1& 1\end{array}\right),_1=\left(\begin{array}{cc}0& 1\\ 1& 2\end{array}\right),`$
$`_2`$ $`=`$ $`\left(\begin{array}{cc}1& 4\\ 1& 3\end{array}\right),_{\mathrm{}}=\left(\begin{array}{cc}1& 3\\ 0& 1\end{array}\right)`$ (62)
respectively. Now
$$_0^2_1=\left(\begin{array}{cc}0& 1\\ 1& 0\end{array}\right)$$
(63)
and
$$_0_1_0^1=\left(\begin{array}{cc}1& 1\\ 0& 1\end{array}\right)$$
(64)
so the two operations
$$\tau 1/\tau \text{and}\tau \tau +1$$
(65)
are in the group generated by (62), which is therefore the full modular group $`\mathrm{\Gamma }(1)`$.
Again the discussion starts along the lines of . We take the massless $`N_f=4`$ curve
$$y^2=x^3\frac{1}{4}g_2(\tau )\overline{u}^2x\frac{1}{4}g_3(\tau )\overline{u}^3$$
(66)
with
$$\frac{g_2^3(\tau )}{g_2^3(\tau )27g_3^2(\tau )}=\frac{(\vartheta _2^8+\vartheta _3^8+\vartheta _4^8)^3}{54\vartheta _2^8\vartheta _3^8\vartheta _4^8}=J(\tau ),$$
(67)
and compare this with the massless $`N_f=1`$ curve
$$y^2=x^2(xu)\mathrm{\Lambda }_1^6/64.$$
(68)
First eliminate the $`x^2`$ term in the $`N_f=1`$ curve by shifting $`xx+u/3`$,
$$y^2=x^3\frac{u^2}{3}x\left(\frac{2}{27}u^3+\frac{\mathrm{\Lambda }_1^6}{64}\right).$$
(69)
Equating co-efficients with (66) and eliminating $`\overline{u}`$ then produces
$$J(\tau )=\frac{1}{4}\frac{u^6}{u^3+1},$$
(70)
where the scale has been set by choosing $`27\mathrm{\Lambda }_1^6/256=1`$. As $`\tau i\mathrm{}`$, $`J\mathrm{}`$ and this corresponds to $`u\mathrm{}`$. But $`J`$ is invariant under $`\mathrm{\Gamma }(1)`$ and so $`J\mathrm{}`$ at the three points $`\tau =0`$, $`\tau =1`$ and $`\tau =2`$ as well and these correspond to the three roots of $`u^3=1`$. Differentiating equation (70) then leads to
$$u\frac{d\tau }{du}=3\left(\frac{u^3+2}{u^3+1}\right)\frac{J}{J^{}}$$
(71)
and (101) gives
$$\frac{J}{J^{}}=\frac{1}{2\pi i}\frac{\vartheta _2^8+\vartheta _3^8+\vartheta _4^8}{(\vartheta _3^4+\vartheta _2^4)(\vartheta _2^4\vartheta _4^4)(\vartheta _3^4+\vartheta _4^4)}.$$
(72)
As $`\tau i\mathrm{}`$,
$$J/J^{}\frac{i}{2\pi }\text{and}u\frac{d\tau }{du}=\frac{3i}{2\pi },$$
(73)
which is the correct asymptotic behaviour. But $`J/J^{}`$ has the same value at $`\tau =0,1`$ and $`2`$ as at $`\tau =i\mathrm{}`$ and $`u^3=1`$ at these 3 points, so (71) diverges at strong coupling. Following the same procedure as before the singularities in (71) can be eliminated, without disturbing the behaviour at $`u\mathrm{}`$, by using
$$\left(\frac{u^3+1}{u^3}\right)u\frac{d\tau }{du}=3\left(\frac{u^3+2}{u^3}\right)\frac{J}{J^{}}.$$
(74)
This is not a modular form for $`\mathrm{\Gamma }(1)`$ however, since $`u`$ is not invariant. Solving (70) for $`u`$ gives
$$u^3=2\left(J\pm \sqrt{J(J1)}\right).$$
(75)
and both roots are necessary: the upper sign for $`u\mathrm{}`$ and the lower sign for $`u^31`$. Eliminating $`u^3`$ from (74) then gives an ambiguity in the direction of flow
$$\left(\frac{u^3+1}{u^3}\right)u\frac{d\tau }{du}=\pm 3\frac{\sqrt{J(J1)}}{J^{}}=\frac{\pm 3}{i\pi \sqrt{2(\vartheta _2^8+\vartheta _3^8+\vartheta _4^8)}}.$$
(76)
The resolution of this problem is that the true $`\beta `$-function, which should be a modular form for $`\mathrm{\Gamma }(1)`$, must have poles or zeros that are not accounted for in (74). This equation was derived by making the minimal modification of (71) that would remove the infinities at $`u^3=1`$ yet not disturb the asymptotic behaviour at $`u=\mathrm{}`$. For $`N_f=1,2`$ and $`3`$ this minimalist approach worked. For $`N_f=1`$ it does not, there must be another pole or zero somewhere. In fact (76) already has a singularity at $`\tau =e^{i\pi /3}`$, where $`\vartheta _2^8+\vartheta _3^8+\vartheta _4^8=0`$ corresponding to $`u=0`$, and its images. This is a fixed point of $`\mathrm{\Gamma }(1)`$, meaning that there is at least one element $`\gamma \mathrm{\Gamma }(1)`$ that leaves it invariant. <sup>3</sup><sup>3</sup>3Actually there two such elements: $`\gamma =\left(\begin{array}{cc}1& 1\\ 1& 0\end{array}\right)`$ and $`\gamma =\left(\begin{array}{cc}0& 1\\ 1& 1\end{array}\right)`$. The point $`\tau =i`$, where $`\vartheta _2^2=\vartheta _4^2`$, is also a fixed point of $`\mathrm{\Gamma }(1)`$ and, if the flow commutes with $`\mathrm{\Gamma }(1)`$, then $`\tau =i`$ should be a fixed point of the flow also. The next simplest assumption that can be made is that the only fixed points above the real axis are at $`\tau =i`$ and $`\tau =e^{i\pi /3}`$ and their images under $`\mathrm{\Gamma }(1)`$, and we shall therefore assume that these are the only places above the real axis where a pole or a zero of the $`\beta `$-function is allowed. We shall further assume that the total number of poles plus zeros (counting multiplicity) is the smallest number compatible with these requirements, since this was our experience for $`N_f=0,1`$ and $`3`$ and was an assumption in . Jacobi’s invariant function takes the values $`J=0`$ at $`\tau =e^{i\pi /3}`$ and $`J=1`$ at $`\tau =i`$. Therefore, with the above assumption, the $`\beta `$-function must be of the form
$$\beta (\tau )=3\frac{J^m(J1)^n}{J^{}}$$
(77)
with $`m`$ and $`n`$ integers. In order that the $`\beta `$-function has the correct asymptotic form as $`\tau i\mathrm{}`$ it must be the case that $`m+n=1`$ and minimising the total number of zeros and poles leaves only two possibilities $`m=0`$, $`n=1`$ or $`m=1`$, $`n=0`$. Examining these two possibilities the first gives
$$\frac{J1}{J^{}}=\frac{2}{\pi i}\frac{(\vartheta _3^4+\vartheta _2^4)(\vartheta _2^4\vartheta _4^4)(\vartheta _3^4+\vartheta _4^4)}{(\vartheta _2^8+\vartheta _3^8+\vartheta _4^8)^2}$$
(78)
which is singular when $`\vartheta _2^8+\vartheta _3^8+\vartheta _4^8=0`$, that is at $`\tau =e^{i\pi /3}`$ and its images, and the second gives
$$\frac{J}{J^{}}=\frac{1}{2\pi i}\frac{\vartheta _2^8+\vartheta _3^8+\vartheta _4^8}{(\vartheta _3^4+\vartheta _2^4)(\vartheta _2^4\vartheta _4^4)(\vartheta _3^4+\vartheta _4^4)}$$
(79)
which is singular when $`(\vartheta _3^4+\vartheta _2^4)(\vartheta _2^4\vartheta _4^4)(\vartheta _3^4+\vartheta _4^4)=0`$, that is at $`\tau =i`$ and its images.<sup>4</sup><sup>4</sup>4$`\vartheta _2^4=\vartheta _4^4`$ at $`\tau =i`$, $`\vartheta _3^4=\vartheta _4^4`$ at $`\tau =(1+i)/2`$ and $`\vartheta _3^4=\vartheta _2^4`$ at $`\tau =1+i`$. The latter has a milder singularity, and clearly has a smaller number of poles plus zeros than the former, and so is the unique choice that fits the criteria. So we conjecture that the analytic $`\beta `$-function for $`N_f=1`$ is
$$\beta (\tau )=\left(\frac{u^3+1}{u^3+2}\right)u\frac{d\tau }{du}=\frac{3}{2\pi i}\frac{(\vartheta _2^8+\vartheta _3^8+\vartheta _4^8)}{(\vartheta _3^4+\vartheta _2^4)(\vartheta _2^4\vartheta _4^4)(\vartheta _3^4+\vartheta _4^4)},$$
(80)
which differs from (74) only by using $`u^3+2`$ in the denominator, rather than $`u^3`$. This flow is plotted in figure 4, the pole at $`\tau =i`$, corresponding to $`u^3=2`$, renders this a repulsive fixed point in both flow directions while the zero at $`\tau =e^{i\pi /3}`$, corresponding to $`u=0`$, is an attractive fixed point in the direction of decreasing Higgs VEV.
## 6 Massive Matter Multiplets
When the matter fields in the fundamental representation of $`SU(2)`$ have non-zero mass the analysis is much harder. For massive matter multiplets the Seiberg-Witten curves determine $`\tau (u,m_i,\mathrm{\Lambda }_{N_f})`$ with $`i=1,\mathrm{},N_f`$ and the $`N_f1`$ case can be determined from the $`N_f`$ case by holomorphic decoupling : set $`\mathrm{\Lambda }_{N_f1}^{5N_f}=m_{N_f}\mathrm{\Lambda }_{N_f}^{4N_f}`$ and send $`m_{N_f}\mathrm{}`$ and $`\mathrm{\Lambda }_{N_f}0`$ keeping $`\mathrm{\Lambda }_{N_f1}`$ finite. In principle therefore it ought to be possible to distort figure 3 for example, by turning on one mass, and recover figure 2 as the mass goes to infinity. This would correspond to a family of $`\beta `$-functions obtained by differentiating $`\tau (u,m_i,\mathrm{\Lambda }_{N_f})`$ with respect to $`u`$ and considering the $`m_i`$ to parameterise different $`\beta `$-functions: near the fixed points this corresponds to defining $`\beta `$-functions by varying the $`W^\pm `$-boson, and therefore also the gluino, masses while keeping the quark masses fixed.
In this section we address the question of quark masses using a strong coupling expansion, taking the $`N_f=3`$ case for illustrative purposes. The details of the analysis are rather technical and so are relegated to appendix B, from which we quote the relevant formulae here.
For $`N_f=3`$ with three different masses $`m_1`$, $`m_2`$ and $`m_3`$ the Seiberg-Witten curve is uniquely determined by $`m_1`$, $`m_2`$, $`m_3`$ and the $`\mathrm{\Gamma }_0(4)`$ invariant parameter $`u`$. With the term quadratic in $`x`$ eliminated the curve is
$`y^2`$ $`=`$ $`x^3{\displaystyle \frac{1}{3}}\left(u^24u\stackrel{~}{\mathrm{\Lambda }}_3^2+\stackrel{~}{\mathrm{\Lambda }}_3^4+3(m_1^2+m_2^2+m_3^2)\stackrel{~}{\mathrm{\Lambda }}_3^26\stackrel{~}{\mathrm{\Lambda }}_3m_1m_2m_3\right)x`$
$``$ $`{\displaystyle \frac{1}{27}}\{(2u\stackrel{~}{\mathrm{\Lambda }}_3^2)(u^2+8u\stackrel{~}{\mathrm{\Lambda }}_3^22\stackrel{~}{\mathrm{\Lambda }}_3^49(m_1^2+m_2^2+m_3^2)\stackrel{~}{\mathrm{\Lambda }}_3^2)`$
$``$ $`18(u+\stackrel{~}{\mathrm{\Lambda }}_3^2)\stackrel{~}{\mathrm{\Lambda }}_3m_1m_2m_3+27\stackrel{~}{\mathrm{\Lambda }}_3^2(m_1^2m_2^2+m_2^2m_3^2+m_3^2m_1^2)\},`$
where, in the notation of , $`\stackrel{~}{\mathrm{\Lambda }}_3=\mathrm{\Lambda }_3/8`$. In principle this curve determines $`\tau (u,m_i)`$ and there are four different masses that could be varied to define $`\beta `$-functions. To connect with the $`\beta `$-functions of §4 we shall keep $`m_i`$ fixed and vary only $`u`$.
In the massless case the strategy was to find an explicit expression for $`u(\tau )`$, in terms of Jacobi $`\vartheta `$-functions, and then use the properties of the $`\vartheta `$-functions to determine $`\frac{du}{d\tau }`$. The problem in the massive case is to find $`u(\tau ,m_i)`$ and this is much harder. We shall first simplify the problem and only consider one non-zero mass $`m_1=m`$, $`m_2=m_3=0`$. Even then one cannot hope for a closed form solution. However it is shown in appendix B that, at strong coupling where $`\tau _D=1/\tau i\mathrm{}`$, it is appropriate to expand in $`\stackrel{~}{q}_D:=e^{i\pi \tau _D/2}`$. The details are somewhat technical and left to the appendix but for one non-zero mass there are singularities at $`\tau =0`$ when $`u=\pm m\stackrel{~}{\mathrm{\Lambda }}_3`$. Near $`u=m\stackrel{~}{\mathrm{\Lambda }}_3`$ the expansion
$$u(\tau _D)=m\stackrel{~}{\mathrm{\Lambda }}_3\left(1+\alpha (m)e^{i\pi \tau _D}+\mathrm{}\right)$$
(82)
is derived in appendix B where $`\alpha (m)`$ is an unknown function of $`m/\stackrel{~}{\mathrm{\Lambda }}_3`$ which cannot be determined without further assumptions, but it diverges like $`1/m^2`$ as $`m0`$ and $`\alpha (m)16`$ as $`m\mathrm{}`$.
For $`u`$ infinitesimally close to $`m`$ the $`\beta `$-function will be of the form
$$\beta (\tau )(um)\frac{\tau }{u},$$
(83)
assuming the BPS monopoles have mass $`um`$. Using
$$\frac{u}{\tau _D}i\pi m\stackrel{~}{\mathrm{\Lambda }}_3\alpha (m)e^{i\pi \tau _D}i\pi (um),$$
(84)
gives, for $`u`$ near $`m`$,
$$\beta (\tau _D)(um)\frac{\tau _D}{u}\frac{i}{\pi },$$
(85)
which is perfectly well behaved, even in the massless limit $`m0`$. Indeed
$$\beta (\tau )\frac{i\tau ^2}{\pi }$$
(86)
which, up to a constant, is the same behaviour as equation (4) even though it is clear from (82) that the limits $`\tau _Di\mathrm{}`$ and $`m0`$ do not commute, since $`\alpha (m)`$ behaves as $`1/m^2`$ as $`m0`$.
## 7 Conclusions
Explicit expressions have been proposed for the $`\beta `$-functions of $`N=2`$ SUSY $`SU(2)`$ Yang-Mills with massless matter fields in the fundamental representation. Asymptotically close to the strong and weak coupling fixed points they co-incide with the 1-loop Callan-Symanzik $`\beta `$-functions, up to a constant factor.
The $`\beta `$-functions are modular forms of sub-groups of $`\mathrm{\Gamma }(1)`$ for each value of $`N_f`$:
$`N_f=0`$ $`\mathrm{\Gamma }^0(2)`$ (87)
$`N_f=1`$ $`\mathrm{\Gamma }(1)`$ (88)
$`N_f=2`$ $`\mathrm{\Gamma }_0(2)`$ (89)
$`N_f=3`$ $`\mathrm{\Gamma }_0(4),`$ (90)
for $`N_f=1`$ and $`N_f=3`$ the group is the same as the monodromy group and for $`N_f=0`$ and $`N_f=2`$ it is larger, due to the $`𝐙_2`$ action on the $`u`$-plane.
The $`\beta `$-functions are determined by demanding that they have the correct asymptotic behaviour at both weak and strong coupling fixed points. The relevant flows, in the direction of decreasing Higgs VEV, are shown in figures 1,4,2 and 3 respectively.
These functions differ from previous expressions in the literature in that they have the correct behaviour at all the strong coupling fixed points. In all cases the $`\beta `$-functions have a singularity in the interior, or on the boundary, of the fundamental domain, corresponding to an unstable fixed point which is repulsive in both directions of flow. This is a necessary consequence of their being modular forms of weight -2. For the $`N_f=1`$ case, for which the monodromy group is the full modular group this repulsive fixed point is at the self-dual point $`\tau =i`$ and its images and in this case there is also a fully attractive fixed point at $`\tau =e^{i\pi /3}`$ and its images. The physical significance, if any, of this attractive fixed point is not clear.
The case of finite masses is probably intractable using the methods developed here, though it may well be possible to make progress with other gauge groups by using the methods in . Nevertheless it has been possible to show, in a strong coupling expansion, that turning on one quark mass in the $`N_f=3`$ case still allows for a $`\beta `$-function with the correct asymptotic behaviour near $`\tau =0`$. Unfortunately it is not possible to say anything about the $`\beta `$-function away from $`\tau =0`$ because of the limitations of the technique.
It is a pleasure to thank the Perimeter Institute, Waterloo, where this work was completed, for hospitality. This work was supported in part by Enterprise Ireland Basic Research Grant no. SC/2003/415.
## Appendix A Appendix: properties of Jacobi $`\vartheta `$-functions
We collect together some useful properties of $`\vartheta `$-functions. The definitions are those of and most of the formulae here are proven in that reference. The three Jacobi $`\vartheta `$-functions used in the text are defined as
$`\vartheta _2(\tau )`$ $`=`$ $`2{\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}q^{(n+\frac{1}{2})^2}=2q^{\frac{1}{4}}{\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}\left(1q^{2n}\right)\left(1+q^{2n}\right)^2,`$ (91)
$`\vartheta _3(\tau )`$ $`=`$ $`{\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}q^{n^2}={\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}\left(1q^{2n}\right)\left(1+q^{2n1}\right)^2,`$ (92)
$`\vartheta _4(\tau )`$ $`=`$ $`{\displaystyle \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}}(1)^nq^{n^2}={\displaystyle \underset{n=1}{\overset{\mathrm{}}{}}}\left(1q^{2n}\right)\left(1q^{2n1}\right)^2,`$ (93)
where $`q:=e^{i\pi \tau }`$.
These three $`\vartheta `$-functions are not independent but are related by
$$\vartheta _3^4(\tau )=\vartheta _2^4(\tau )+\vartheta _4^4(\tau ).$$
(94)
The following relations can be used to determine their properties under modular transformations:
$$\vartheta _2(\tau +1)=e^{i\pi /4}\vartheta _2(\tau ),\vartheta _3(\tau +1)=\vartheta _4(\tau ),\vartheta _4(\tau +1)=\vartheta _3(\tau ),$$
(95)
$`\vartheta _2(1/\tau )`$ $`=`$ $`\sqrt{i\tau }\vartheta _4(\tau ),`$
$`\vartheta _3(1/\tau )`$ $`=`$ $`\sqrt{i\tau }\vartheta _3(\tau ),`$ (96)
$`\vartheta _4(1/\tau )`$ $`=`$ $`\sqrt{i\tau }\vartheta _2(\tau ).`$
The duplication formulae
$`\vartheta _2^2(2\tau )`$ $`=`$ $`{\displaystyle \frac{1}{2}}((\vartheta _3^2(\tau )\vartheta _4^2(\tau )),`$
$`\vartheta _3^2(2\tau )`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left(\vartheta _3^2(\tau )+\vartheta _4^2(\tau )\right),`$ (97)
$`\vartheta _4^2(2\tau )`$ $`=`$ $`\vartheta _3(\tau )\vartheta _4(\tau ),`$
and
$$\vartheta _3(4\tau )=\frac{1}{2}\left(\vartheta _3(\tau )+\vartheta _4(\tau )\right),\vartheta _2(4\tau )=\frac{1}{2}\left(\vartheta _3(\tau )\vartheta _4(\tau )\right)$$
(98)
are also useful.
At the special points $`\tau =e^{i\pi /2}`$ and $`\tau =e^{i\pi /3}`$ the $`\vartheta `$-functions have the values
$`\vartheta _3^2(e^{i\pi /2})`$ $`=`$ $`\sqrt{2}\vartheta _2^2(e^{i\pi /2})=\sqrt{2}\vartheta _4^2(e^{i\pi /2})={\displaystyle \frac{2}{\pi }}K\left(\mathrm{sin}\left({\displaystyle \frac{\pi }{4}}\right)\right),`$ (99)
$`e^{i\pi /4}\vartheta _2^2(e^{i\pi /3})`$ $`=`$ $`e^{i\pi /12}\vartheta _3^2(e^{i\pi /3})=e^{i\pi /12}\vartheta _4^2(e^{i\pi /3})={\displaystyle \frac{2}{\pi }}K\left(\mathrm{sin}\left({\displaystyle \frac{\pi }{12}}\right)\right),`$
where $`K(k)`$ is the complete elliptic of the second kind: $`K\left(\mathrm{sin}(\pi /4)\right)=\frac{1}{4\sqrt{\pi }}(\mathrm{\Gamma }(1/4))^2`$, with $`\mathrm{\Gamma }(1/4)3.6256`$ the Euler $`\mathrm{\Gamma }`$-function evaluated at $`1/4`$, and $`K\left(\mathrm{sin}(\pi /12)\right)1.5981`$.
The $`\vartheta `$-functions have the following asymptotic forms
$`\tau i\mathrm{}`$ $`:\vartheta _2(\tau )2e^{\frac{i\pi \tau }{4}}0,\vartheta _3(\tau )1,\vartheta _4(\tau )1;`$ (100)
$`\tau 0`$ $`:\vartheta _2(\tau )\sqrt{{\displaystyle \frac{i}{\tau }}},\vartheta _3(\tau )\sqrt{{\displaystyle \frac{i}{\tau }}},\vartheta _4(\tau )2\sqrt{{\displaystyle \frac{i}{\tau }}}e^{\frac{i\pi }{4\tau }}0.`$
In addition they satisfy the following differential equations (see , p.231, equation (7.2.17)),
$`{\displaystyle \frac{\vartheta _3^{}}{\vartheta _3}}{\displaystyle \frac{\vartheta _4^{}}{\vartheta _4}}`$ $`=`$ $`{\displaystyle \frac{i\pi }{4}}\vartheta _2^4,`$
$`{\displaystyle \frac{\vartheta _2^{}}{\vartheta _2}}{\displaystyle \frac{\vartheta _3^{}}{\vartheta _3}}`$ $`=`$ $`{\displaystyle \frac{i\pi }{4}}\vartheta _4^4,`$ (101)
$`{\displaystyle \frac{\vartheta _2^{}}{\vartheta _2}}{\displaystyle \frac{\vartheta _4^{}}{\vartheta _4}}`$ $`=`$ $`{\displaystyle \frac{i\pi }{4}}\vartheta _3^4.`$
In the text $`\mathrm{\Gamma }_0(N)`$ consists of matrices $`\gamma =\left(\begin{array}{cc}a& b\\ c& d\end{array}\right)`$ in $`\mathrm{\Gamma }(1)PSl(2,𝐙)`$ with $`c0\text{mod}N`$, sometimes written
$$\gamma \left(\begin{array}{cc}& \\ 0& \end{array}\right)\text{mod}N,$$
(102)
$`\mathrm{\Gamma }^0(N)`$ consists of matrices with $`b0\text{mod}N`$,
$$\gamma \left(\begin{array}{cc}& 0\\ & \end{array}\right)\text{mod}N$$
(103)
and $`\mathrm{\Gamma }(N)`$ consists of matrices with
$$\gamma \left(\begin{array}{cc}1& 0\\ 0& 1\end{array}\right)\text{mod}N.$$
(104)
## Appendix B Appendix: Massive Matter Fields for $`N_f=3`$
The starting point is the curve for four massive matter multiplets in the fundamental representation ,
$`y^2`$ $`=`$ $`(x^2c_2^2\overline{u}^2)(xc_1\overline{u})c_2^2(xc_1\overline{u})^2\overline{A}c_2^2(c_1^2c_2^2)(xc_1\overline{u})\overline{B}`$ (105)
$`+`$ $`2c_2(c_1^2c_2^2)(c_1xc_2^2\overline{u})\overline{C}c_2^2(c_1^2c_2^2)^2\overline{D},`$
with
$$c_1(\tau )=\frac{1}{2}\left(\vartheta _3^4(\tau )+\vartheta _4^4(\tau )\right),c_2(\tau )=\frac{1}{2}\left(\vartheta _3^4(\tau )\vartheta _4^4(\tau )\right),$$
(106)
and
$$\overline{A}=\underset{i}{\overset{4}{}}\overline{m}_i^2,\overline{B}=\underset{i<j}{}\overline{m}_i^2\overline{m}_j^2,\overline{C}=\overline{m}_1\overline{m}_2\overline{m}_3\overline{m}_4\overline{D}=\underset{i<j<k}{}\overline{m}_i^2\overline{m}_j^2\overline{m}_k^2.$$
(107)
First eliminate the quadratic term in $`x`$ by shifting $`xx+(c_1\overline{u}+c_2^2\overline{A})/3`$, giving
$$y^2=x^3\frac{1}{3}P(\tau ,\overline{u},\overline{m}_i)x\frac{1}{27}Q(\tau ,\overline{u},\overline{m}_i),$$
(108)
where
$$P(\tau ,\overline{u},\overline{m}_i)=(c_1^2+3c_2^2)\overline{u}^24c_1c_2^2\overline{A}\overline{u}+3c_2(c_1^2c_2^2)(c_2\overline{B}2c_1\overline{C})+c_2^4\overline{A}^2,$$
(109)
is quadratic in $`\overline{u}`$ and
$`Q(\tau ,\overline{u},m_i)`$ $`=`$ $`2c_1(c_1^29c_2^2)\overline{u}^3+3c_2^2(5c_1^2+3c_2^2)\overline{A}\overline{u}^2`$
$``$ $`2\left(6c_1c_2^4\overline{A}^2+9c_1c_2^2(c_1^2c_2^2)\overline{B}+9c_2(c_1^2c_2^2)(c_1^23c_2^2)\overline{C}\right)\overline{u}`$
$`+`$ $`2c_2^6\overline{A}^3+27c_2^2(c_1^2c_2^2)^2\overline{D}+9c_2^3(c_1^2c_2^2)(c_2\overline{B}2c_1\overline{C})\overline{A}`$
is cubic in $`\overline{u}`$.
Next this curve can be reduced to the $`N_f=3`$ curve using the holomorphic decoupling of : send $`\overline{m}_4\mathrm{}`$ and take the ‘bare’ coupling $`\tau i\mathrm{}`$, so $`c_20`$ and $`c_11`$, keeping $`c_2\overline{m}_4=\mathrm{\Lambda }_3/8:=\stackrel{~}{\mathrm{\Lambda }}_3`$ finite. Taking this limit, and dropping the bars on $`u`$, $`m_1`$, $`m_2`$ and $`m_3`$, gives the massive $`N_f=3`$ curve,
$`y^2`$ $`=`$ $`x^3{\displaystyle \frac{1}{3}}(u^24u\stackrel{~}{\mathrm{\Lambda }}_3^2+\stackrel{~}{\mathrm{\Lambda }}_3^4+3B\stackrel{~}{\mathrm{\Lambda }}_3^26C\stackrel{~}{\mathrm{\Lambda }}_3)x`$
$``$ $`{\displaystyle \frac{1}{27}}(2u\stackrel{~}{\mathrm{\Lambda }}_3^2)(u^2+8u\stackrel{~}{\mathrm{\Lambda }}_3^22\stackrel{~}{\mathrm{\Lambda }}_3^49B\stackrel{~}{\mathrm{\Lambda }}_3^2)+{\displaystyle \frac{2}{3}}(u+\stackrel{~}{\mathrm{\Lambda }}_3^2)C\stackrel{~}{\mathrm{\Lambda }}_3D,`$
with
$$B=m_1^2+m_2^2+m_3^2,C=m_1m_2m_3,D=m_1^2m_2^2+m_2^2m_3^2+m_3^2m_1^2.$$
(112)
This reduces to equation (47) when the three masses are set to zero and $`\stackrel{~}{\mathrm{\Lambda }}_3=1`$. Following this curve is now compared to (108) with $`\overline{m}_4=0`$, $`\overline{u}`$ independent of $`\tau `$, that is equation (108)-(B) with $`\overline{C}=0`$ and
$$\overline{A}=\overline{m}_1^2+\overline{m}_2^2+\overline{m}_3^2,\overline{B}=\overline{m}_1^2\overline{m}_2^2+\overline{m}_2^2\overline{m}_3^2+\overline{m}_3^2\overline{m}_1^2,\overline{D}=\overline{m}_1^2\overline{m}_2^2\overline{m}_3^2.$$
(113)
Equating co-efficients leads to
$$(c_1^2+3c_2^2)\overline{u}^24c_1c_2^2\overline{A}\overline{u}+3c_2^2(c_1^2c_2^2)\overline{B}+c_2^4\overline{A}^2=u^24u\stackrel{~}{\mathrm{\Lambda }}_3^2+\stackrel{~}{\mathrm{\Lambda }}_3^4+3B\stackrel{~}{\mathrm{\Lambda }}_3^26C\stackrel{~}{\mathrm{\Lambda }}_3$$
(114)
$`2c_1(c_1^29c_2^2)\overline{u}^3+3c_2^2(5c_1^2+3c_2^2)\overline{A}\overline{u}^22\left(6c_1c_2^4\overline{A}^2+9c_1c_2^2(c_1^2c_2^2)\overline{B}\right)\overline{u}`$
$`+2c_2^6\overline{A}^3+27c_2^2(c_1^2c_2^2)^2\overline{D}+9c_2^4(c_1^2c_2^2)\overline{B}\overline{A}`$ (115)
$`=(2u\stackrel{~}{\mathrm{\Lambda }}_3^2)(u^2+8u\stackrel{~}{\mathrm{\Lambda }}_3^22\stackrel{~}{\mathrm{\Lambda }}_3^49B\stackrel{~}{\mathrm{\Lambda }}_3^2)18(u+\stackrel{~}{\mathrm{\Lambda }}_3^2)C\stackrel{~}{\mathrm{\Lambda }}_3+27D.`$
In the massless case we can eliminate $`\overline{u}`$ from these two equations and determine $`u(\tau )`$, but now we want to determine $`u(\tau ,m_i)`$ with $`\overline{u}`$, $`\overline{A}`$, $`\overline{B}`$ and $`\overline{D}`$ all unknown functions so there is not enough information to solve the problem completely. Nevertheless we can still get information from these equations. The symmetry group for $`N_f=3`$ is $`\mathrm{\Gamma }_0(4)`$ and $`u`$, $`m_i`$ and $`\stackrel{~}{\mathrm{\Lambda }}_3`$ should be invariants of $`\mathrm{\Gamma }_0(4)`$ so the right hand sides of (114) and (115) are also invariants. Using the transformation properties of the $`\vartheta `$-functions, (95) and (A), this implies that $`\overline{u}`$, $`\overline{A}`$, $`\overline{B}`$, and $`\overline{D}`$ are modular forms of weights -2, -4, -8 and -12 respectively, so $`\overline{m}_i`$ have weight -2.
For simplicity we shall focus on the case of a single mass, $`m_1=m`$, $`m_2=m_3=0`$ so
$$\overline{A}=\overline{m}^2,\overline{B}=0,\overline{D}=0$$
(116)
$$B=m^2,D=0,C=0$$
(117)
and the equations simplify to
$$(c_1^2+3c_2^2)\overline{u}^24c_1c_2^2\overline{m}^2\overline{u}+c_2^4\overline{m}^4=u^24u\stackrel{~}{\mathrm{\Lambda }}_3^2+\stackrel{~}{\mathrm{\Lambda }}_3^4+3m^2\stackrel{~}{\mathrm{\Lambda }}_3^2$$
(118)
$`2c_1(c_1^29c_2^2)\overline{u}^3+3c_2^2(5c_1^2+3c_2^2)\overline{m}^2\overline{u}^212c_1c_2^4\overline{m}^4\overline{u}+2c_2^6\overline{m}^6`$ (119)
$`=`$ $`(2u\stackrel{~}{\mathrm{\Lambda }}_3^2)(u^2+8u\stackrel{~}{\mathrm{\Lambda }}_3^22\stackrel{~}{\mathrm{\Lambda }}_3^49m^2\stackrel{~}{\mathrm{\Lambda }}_3^2).`$
We now want to eliminate $`\overline{u}`$ and $`\overline{m}`$ to get $`u(\tau ,m)`$ but there is still not enough information. However, knowing that $`\overline{u}`$ and $`\overline{m}^2`$ are modular forms of weight -2 and -4 respectively, we can say something about their functional form at strong coupling. To see how this works let us first look at the case $`\overline{m}=0`$, where the explicit solution is given in §4. Using the details there one finds
$$\overline{u}(\tau )=\frac{\stackrel{~}{\mathrm{\Lambda }}_3^2}{\left(\vartheta _3^2(\tau )\vartheta _4^2(\tau )\right)^2}$$
(120)
which is indeed a modular form for $`\mathrm{\Gamma }_0(4)`$ of weight -2 and it vanishes at strong coupling, $`\tau 0`$. Now $`\tau 1/\tau =\tau _D`$ is not in $`\mathrm{\Gamma }_0(4)`$, so $`\overline{u}`$ is not a modular form under this transformation, rather
$$\overline{u}(\tau _D)=\frac{1}{\tau _D^2}\frac{\stackrel{~}{\mathrm{\Lambda }}_3^2}{\left(\vartheta _3^2(\tau _D)\vartheta _2^2(\tau _D)\right)^2}.$$
(121)
Writing $`\stackrel{~}{q}_D=e^{i\pi \tau _D/2}`$ we have the strong coupling expansions
$$\overline{u}(\tau _D)=\frac{\stackrel{~}{\mathrm{\Lambda }}_3^2}{\tau _D^2}\left(1+8\stackrel{~}{q}_D+40\stackrel{~}{q}_D^2+160\stackrel{~}{q}_D^3+\mathrm{}\right).$$
(122)
and, from (54),
$$u(\tau _D)=\frac{\vartheta _3^2(\tau _D)\vartheta _2^2(\tau _D)}{\left(\vartheta _3^2(\tau _D)\vartheta _2^2(\tau _D)\right)^2}=4\stackrel{~}{\mathrm{\Lambda }}_3^2\stackrel{~}{q}_D\left(1+8\stackrel{~}{q}_D+44\stackrel{~}{q}_D^2+192\stackrel{~}{q}_D^3+\mathrm{}\right).$$
(123)
For non-zero $`m`$ it is consistent with all we know to assume a similar form
$$\overline{u}(\tau _D)=\frac{\stackrel{~}{\mathrm{\Lambda }}_3^2}{\tau _D^2}\left(\overline{u}_0+\overline{u}_1\stackrel{~}{q}_D+\overline{u}_2\stackrel{~}{q}_D^2+\overline{u}_3\stackrel{~}{q}_D^3+\mathrm{}\right).$$
(124)
and similarly for $`\overline{m}^2`$
$$\overline{m}^2(\tau _D)=\frac{\stackrel{~}{\mathrm{\Lambda }}_3^2}{\tau _D^4}\left(\overline{a}_0+\overline{a}_1\stackrel{~}{q}_D+\overline{a}_2\stackrel{~}{q}_D^2+\overline{a}_3\stackrel{~}{q}_D^3+\mathrm{}\right),$$
(125)
where $`\overline{u}_k`$ and $`\overline{a}_k`$ are functions of $`m/\stackrel{~}{\mathrm{\Lambda }}_3`$, with $`\overline{a}_k`$ vanishing for $`m=0`$. A similar strong coupling expansion for $`u`$ has no prefactor of $`1/\tau _D^2`$ because $`u`$ has weight zero not -2,
$$\frac{u}{\stackrel{~}{\mathrm{\Lambda }}_3^2}=u_0+u_1\stackrel{~}{q}_D+u_2\stackrel{~}{q}_D^2+u_3\stackrel{~}{q}_D^3+\mathrm{}.$$
(126)
Using these expansions in (118) and (119), together with
$`c_1(\tau )`$ $`=`$ $`c_1(1/\tau _D)={\displaystyle \frac{\tau _D^2}{2}}\left(\vartheta _3(\tau _D)^4+\vartheta _2(\tau _D)^4\right),`$
$`c_2(\tau )`$ $`=`$ $`c_2(1/\tau _D)={\displaystyle \frac{\tau _D^2}{2}}\vartheta _4(\tau _D)^4,`$ (127)
we can equate powers of $`\stackrel{~}{q}_D`$ to obtain recurrence relations between the $`u_k`$ and the $`\overline{a}_k`$. Without making further assumptions there is not enough information to determine $`u(\tau ,m)`$ but we can still extract useful information about the strong coupling $`\beta `$-function. At zeroth order in $`\stackrel{~}{q}_D`$ (118) and (119) (with $`\stackrel{~}{\mathrm{\Lambda }}_3=1`$ for simplicity) yield three possibilities:
$$u_0=\pm m,\overline{u}_0=12m\frac{\overline{a}_0}{4};\text{and}u_0=m^2+\frac{1}{4},\overline{u}_0=m^2\frac{\overline{a}_0+1}{4}.$$
(128)
Only the first two are relevant for the strong coupling fixed point, $`\tau =0`$ at $`u=0`$, of the massless $`N_f=3`$ theory (in the massless case $`u=1/4`$ is associated with $`\tau =1/2`$ where $`\tau _D=2`$). Choosing the root $`u=m`$, at order $`\stackrel{~}{q}_D`$ we find a pair of equations linear in $`u_1`$ and $`\overline{u}_1`$ which are degenerate. Solving for $`\overline{u}_1`$ gives
$$\overline{u}_1:=\frac{(m2)u_1}{(2m1)}\frac{\overline{a}_1}{4}.$$
(129)
At order $`\stackrel{~}{q}_D^2`$ the pair of linear equations for for $`u_2`$ and $`\overline{u}_2`$ are parallel in the $`u_2\overline{u}_2`$ plane and have no solution unless they co-incide, which only happens if
$$\frac{u_1m}{2m1}=0.$$
(130)
For $`m=0`$ this is automatic, but for $`m0`$ it forces $`u_1=0`$. Setting $`u_1=0`$ then gives one linear equation relating $`\overline{u}_2`$ to $`u_2`$,
$$\overline{u}_2=\frac{2(m2)u_2+\overline{a}_0(3\overline{a}_0+8m4)}{2(2m1)}\frac{\overline{a}_2}{4}$$
(131)
At order $`\stackrel{~}{q}_D^3`$ the two linear equations for $`u_3`$ and $`\overline{u}_3`$ are again degenerate giving only one constraint which can be used to solve for $`\overline{u}_3`$
$$\overline{u}_3=\frac{(m2)u_3+\overline{a}_1(3\overline{a}_0+4m2)}{(2m1)}\frac{\overline{a}_3}{4}.$$
(132)
At order $`\stackrel{~}{q}_D^4`$ one again obtains two parallel lines in the $`u_4\overline{u}_4`$ plane which do not intersect unless
$$u_2=\pm \frac{(\overline{a}_0+8m4)^2}{2m}$$
(133)
in which case $`\overline{u}_4`$ can be obtained as a function of $`u_4`$, $`\overline{a}_0`$, $`\overline{a}_2`$ and $`m`$.
One can continue but for the present purposes we have gone as far as necessary. We are only really interested in $`u`$ and we have
$$u(\tau _D)=m\pm \frac{(\overline{a}_0+8m4)^2}{2m}\stackrel{~}{q}_D^2+\mathrm{}$$
(134)
with $`\overline{a}_0`$ and undetermined function of $`m`$, but independent of $`\tau _D`$. To get the dimensions correct we should re-instate $`\stackrel{~}{\mathrm{\Lambda }}_3`$ and write
$$u(\tau _D)=m\stackrel{~}{\mathrm{\Lambda }}_3\left(1+\alpha (m)e^{i\pi \tau _D}+\mathrm{}\right)$$
(135)
where
$$\alpha (m):=\pm \frac{(\overline{a}_0\stackrel{~}{\mathrm{\Lambda }}_3^2+8m\stackrel{~}{\mathrm{\Lambda }}_34\stackrel{~}{\mathrm{\Lambda }}_3^2)^2}{2m^2\stackrel{~}{\mathrm{\Lambda }}_3^2}.$$
(136)
Note that, in order to pin down the co-efficient $`u_2`$ one has to go to order $`\stackrel{~}{q}_D^4`$ and examine $`u_4`$. At every value of $`k`$ in the expansion one gets a pair of linear equations in $`u_k`$ and $`\overline{u}_k`$ in terms of $`m`$ and the $`\overline{a}_k^{}`$ with $`k^{}k`$. For $`k=1`$ these equations are degenerate and $`u_1`$ is not determined; for $`k=2`$ the equations have no solution unless $`u_1=0`$ in which case they are again degenerate and $`u_2`$ is undetermined; for $`k=3`$ the equations are again degenerate and $`u_3`$ is undetermined and for $`k=4`$ the equations have no solution unless $`u_2`$ has one of the two possible values shown in (135).
The explicit from of $`\alpha (m)`$ is not needed in the analysis, but we can fix its asymptotic form as $`m\mathrm{}`$ using holomorphic decoupling . For $`N_f=2`$ equation (36) gives
$$u(\tau _D)=\frac{\mathrm{\Lambda }_2^2}{8}\frac{\vartheta _3^4(\tau _D)}{\vartheta _4^4(\tau _D)}\frac{\mathrm{\Lambda }_2^2}{8}\left(1+16e^{i\pi \tau _D}+\mathrm{}\right),$$
(137)
and this should agree with (135) as $`m\mathrm{}`$, $`\mathrm{\Lambda }_30`$ with $`\mathrm{\Lambda }_2^2=m\mathrm{\Lambda }_3=8m\stackrel{~}{\mathrm{\Lambda }}_3`$ fixed. Hence $`\alpha (m)16`$ as $`m\mathrm{}`$.
The other singularity of the $`N_f=2`$ theory, at $`u=\frac{\mathrm{\Lambda }_2^2}{8}`$, is obtained by holomorphic decoupling in the weak coupling limit of the $`N_f=3`$ theory by expanding around $`u=m`$.
Notice that the $`m=0`$ expansion for $`u`$ in equation (123) contains a term linear in $`\stackrel{~}{q}_D`$ while (135) does not. This is because, when $`m0`$, (130) forces us to set $`u_1=0`$. Since $`\overline{a}_00`$ as $`m0`$, $`\alpha (m)`$ diverges like $`8\stackrel{~}{\mathrm{\Lambda }}_3^2/m^2`$ as $`m0`$ and the limits $`\tau _Di\mathrm{}`$ and $`m0`$ do not commute.
We can perform a similar expansion around $`\tau =1/2`$, using $`u_0=m^2+1/4`$, where $`\tau _D=2+i\epsilon `$ with $`\epsilon `$ small. The analysis is simpler than the $`u_0=\pm m`$ case in that at each order, at least up to order 5 in $`\stackrel{~}{q}_D`$ which is as far as we have gone, one simply finds a pair of linear equations in $`u_k`$ and $`\overline{u}_k`$ which can be solved in terms of $`\overline{a}_k^{}`$ with $`k^{}<k`$. The details are omitted but one finds
$$u_1=u_2=u_3=0,u_4=4\frac{(\overline{a}_04m^2+1)^4}{(4m^21)^2},u_5=16\frac{\overline{a}_1(\overline{a}_04m^2+1)^3}{(4m^21)^2},$$
(138)
and so, for $`\tau _D=2+i\epsilon `$ with $`\epsilon `$ small,
$$u(\tau _D)=m^2+\frac{1}{4}4\frac{(\overline{a}_04m^2+1)^4}{(4m^21)^2}e^{2\pi \epsilon }+16\frac{\overline{a}_1(\overline{a}_04m^2+1)^3}{(4m^21)^2}e^{5\pi \epsilon /2}+\mathrm{}.$$
(139)
Re-instating $`\stackrel{~}{\mathrm{\Lambda }}_3`$ now gives
$$u(\tau _D)=m^2+\frac{\stackrel{~}{\mathrm{\Lambda }}_3^2}{4}+\stackrel{~}{\mathrm{\Lambda }}_3^2\stackrel{~}{\alpha }(m)e^{2\pi \epsilon }\mathrm{}.$$
(140)
where
$$\stackrel{~}{\alpha }(m)=4\frac{(\overline{a}_0\stackrel{~}{\mathrm{\Lambda }}_3^24m^2+\stackrel{~}{\mathrm{\Lambda }}_3^2)^4}{(4m^2\stackrel{~}{\mathrm{\Lambda }}_3^2)^2\stackrel{~}{\mathrm{\Lambda }}_3^4}.$$
(141)
As $`m\mathrm{}`$ this point in the $`u`$-plane goes out to infinity in the $`N_f=2`$ theory.
Equations (135) and (140) were the aim of this appendix and are used in §6 in the discussion of the $`\beta `$-functions for the massive $`N_f=3`$ theory at strong coupling.
Fig. 1: Flow of effective coupling of $`N=2`$ SUSY Yang-Mills with $`N_f=0`$. The arrows indicate the direction of the flow as as the Higgs VEV is reduced. The $`\beta `$-functions are modular forms of $`\mathrm{\Gamma }^0(2)`$ and the pattern repeats under $`\tau \tau +2`$.
Fig. 2: Flow of effective coupling of $`N=2`$ SUSY Yang-Mills with $`N_f=2`$. The arrows indicate the direction of the flow as as the Higgs VEV is reduced. The $`\beta `$-functions are modular forms of $`\mathrm{\Gamma }_0(2)`$ and the pattern repeats under $`\tau \tau +1`$.
Fig. 3: Flow of effective coupling of $`N=2`$ SUSY Yang-Mills with $`N_f=3`$ and $`ϵ=4/27`$. The arrows indicate the direction of the flow as as the Higgs VEV is reduced. The $`\beta `$-functions are modular forms of $`\mathrm{\Gamma }_0(4)`$ and the pattern repeats under $`\tau \tau +1`$.
Fig. 4: Flow of effective coupling of $`N=2`$ SUSY Yang-Mills with $`N_f=1`$. The arrows indicate the direction of the flow as as the Higgs VEV is reduced. The $`\beta `$-functions are modular forms of $`\mathrm{\Gamma }(1)`$ and the pattern repeats under $`\tau \tau +1`$.
|
warning/0506/math0506123.html
|
ar5iv
|
text
|
# A partition theorem for a large dense linear order
## 1 Introduction
For an infinite cardinal $`\kappa `$, the set $`{}_{}{}^{\kappa >}2`$, ordered by end-extension, $``$, is the complete binary tree on $`\kappa `$ with root the empty sequence, $`\mathrm{}`$. By $`st`$ denote the *meet* of $`s`$ and $`t`$, namely the longest initial segment of both $`s`$ and $`t`$. Call two elements $`s`$ and $`t`$ of $`{}_{}{}^{\kappa >}2`$ *incomparable* if neither is an end-extension of the other. The lexicographic order for us will be the partial order $`<_{\text{lex}}`$ on $`{}_{}{}^{\kappa >}2`$ defined by $`s<_{\text{lex}}t`$ if $`s`$ and $`t`$ are incomparable and $`(st){}_{}{}^{}0s`$ and $`(st){}_{}{}^{}1t`$. We also define a linear order $`_Q`$ on $`{}_{}{}^{\kappa >}2`$ by letting $`s<_\text{Q}t`$ if and only if one of the following conditions holds: (1) $`s=t`$; (2) $`t{}_{}{}^{}0s`$; (3) $`s{}_{}{}^{}1t`$; or (4) $`s`$ and $`t`$ are incomparable and $`s<_{\text{lex}}t`$.
Let us recall the definition of a (strongly) $`\kappa `$-dense linear order: it is a linear order $``$ on a set $`L`$ in which for every two subsets $`A,B`$ of $`L`$ of size $`<\kappa `$ satisfying the property that for all $`aA`$ and $`bB`$ the relation $`a<b`$ holds, there is $`cL`$ such that $`a<c<b`$ for all $`aA`$ and $`bB`$. Since $`A`$ or $`B`$ may be taken empty here, the definition in particular implies that there are no endpoints in the order. The adjective ‘strongly’ is used to distinguish this type of ordering from the strictly weaker notion in which for any $`a<b`$ in $`L`$ one is required to have $`\kappa `$ many $`c`$ with $`a<c<b`$. Such orders were first studied by Felix Hausdorff in 1908 () and have been of continuous interest since. A recent paper on the subject, where one can also find a number of further references, is by M. Džamonja and Katherine Thompson, where they give a classification of $`\kappa `$-dense linear orders which are also $`\kappa `$-scattered. In this paper we shall only deal with strongly $`\kappa `$-dense linear orders so we shall omit the adjective ‘strongly’ from our notation.
It is easy to check that for regular $`\kappa `$ the structure $`_\kappa :=({}_{}{}^{\kappa >}2,<_\text{Q})`$ is a $`\kappa `$-dense linear order (see 1.8). In case that $`\kappa ^{<\kappa }=\kappa `$ then of course this order has cardinality $`\kappa `$. All $`\kappa `$ that we shall work with will satisfy this additional cardinal arithmetic assumption. It is well known and easily proved using a back-and-forth argument that in this case the $`\kappa `$-dense linear order of size $`\kappa `$ is unique up to isomorphism, and that it is a $`\kappa `$-saturated homogeneous model of the theory of a dense linear order with no endpoints. (In fact these properties are equivalent to $`\kappa =\kappa ^{<\kappa }`$, as follows from Saharon Shelah’s classification theory, see ).
For $`\kappa =\omega `$, $`_\omega =`$ is a countable dense linear order with no endpoints, so it has the order type $`\eta `$ of the rationals. An unpublished result of Fred Galvin as quoted in is that $`[]_{<\omega ,2}^2`$, see the notation below. Denis C. Devlin proved that
$$()_{<\omega ,t_n}^n\text{ and }()_{<\omega ,t_n1}^n$$
where the value of $`t_n`$ is the $`n`$-th tangent number. The notation here means that for every $`n`$, when one colors $`[]^n`$ with finitely many colors (this is the role of the part $`<\omega `$ in the subscript), there is a copy $`^{}`$ of $``$ with the property that $`[^{}]^n`$ is colored in at most $`t_n`$ colors. At the same time, $`t_n`$ is the smallest number for which such a statement holds.
Tangent numbers may be computed using the power series $`\mathrm{tan}(x)=_1^{\mathrm{}}t_n\frac{x^{2n1}}{(2n1)!}`$. Devlin’s proof used the language of category theory. A proof of this theorem using trees was sketched in in the Farah-Todorcevic book ; a complete proof was given by Vojkan Vuksanovic in which he uses the special case for the complete binary tree $`{}_{}{}^{\omega >}2`$ of Keith R. Milliken’s theorem () about weakly embedded subtrees. Since many of the notions used by Vuksanovic generalize to arbitrary infinite $`\kappa `$ in place of $`\omega `$, one may wonder if his proof may be used to obtain a partition theorem for $`\kappa `$-dense linear orders. Here we use some of these ideas along with new insights to get such a theorem. Our work was also inspired by the strong diagonalization of Norbert Sauer in , the approach to similarities in a triple paper by Claude Laflamme, Sauer and Vuksanovic , and the use of collapses both in the Shelah version of the Halpern-Läuchli Theorem and in another paper of Vuksanovic .
One difficulty of the generalization was that the special case of Milliken’s theorem on binary trees was only known to be valid for $`{}_{}{}^{\omega >}2`$, not for an arbitrary $`\kappa `$. In particular, suppose $`\kappa `$ is uncountable and $``$ is a well-ordering of $`{}_{}{}^{\kappa >}2`$ with the property that whenever $`s`$ is shorter than $`t`$ then also $`st`$. Define a coloring of the height $`2`$ complete binary trees strongly embedded in $`{}_{}{}^{\kappa >}2`$ by $`g(S)=0`$ if and only if the lexicographic order and the $``$-order agree on the leaves of $`S`$. For any strongly embedded copy of $`{}_{}{}^{\kappa >}2`$, this coloring on the binary trees generated by pairs of nodes on the $`\omega `$th level is essentially the Sierpinski partition, so has edges of both colors.
Milliken’s theorem for weakly embedded subtrees follows from his theorem for strongly embedded subtrees, and to prove it, he uses a generalization of the Halpern-Läuchli Theorem due independently to Richard Laver and David Pincus (see ). Here we are able to generalize part of D. Devlin’s theorem starting with a theorem of Shelah from , which is a generalization of the Halpern-Läuchli theorem . We slightly improve Shelah’s Theorem to colorings of antichains rather than only level sets. Let us now state our main result for large dense linear orders.
###### Theorem 1.1.
For every natural number $`m`$ there is a value $`t_m^+<\omega `$ such that for any cardinal $`\kappa `$ which is measurable in the generic extension obtained by adding $`\lambda `$ Cohen subsets of $`\kappa `$, where $`\lambda `$ is some cardinal satisying $`\lambda (\kappa )_{2^\kappa }^{2m}`$, the $`\kappa `$-dense linear order $`_\kappa `$ satisfies
$$_\kappa (_\kappa )_{<\kappa ,t_m^+}^m\text{ and }_\kappa (_\kappa )_{<\kappa ,t_m^+1}^m.$$
In Theorem 3.15, the positive partition relation is shown to hold for $`t_m^+`$ the number of *sparse vip $`m`$-types* (defined in Section 3). The sparse vip $`m`$-types are closely related to those unique $`m`$-element *strongly diagonal* subsets of $`{}_{}{}^{2m2}2`$ that are representatives of the “essential types” in . If we close such a strongly diagonal set under initial segments and add a *vip level order*, we obtain a sparse vip $`m`$-type, and all sparse vip $`m`$-types are obtained in this way.
In Theorem 5.8 we show that the negative partition relation holds for the same value of $`t_m^+`$. In Theorem 5.9 we show that for $`\kappa `$ as in Theorem 1.1 there is a *canonical partition* $`𝒞=\{C_0,C_1,\mathrm{},C_{t_m^+1}\}`$ of $`[_\kappa ]^m`$. That is, a partition whose classes are *persistent* and *indivisible*. We say $`C_j`$ is *persistent* if for every $`\kappa `$-dense $`Q^{}`$ the set $`[Q^{}]^mC_j`$ is non-empty. We say $`C_j`$ is *indivisible* if for every coloring of $`C_j`$ with fewer than $`\kappa `$ many colors, there is a $`\kappa `$-dense subset $`Q^{}`$ on which $`[Q^{}]^mC_j`$ is monochromatic.
Richard Rado constructed a (strongly) universal countable graph in 1964. That is, he constructed a countable graph for which every countable graph is an induced subgraph. By a *$`\kappa `$-Rado graph* we mean a graph $`G`$ of size $`\kappa `$ with the property that for every two disjoint subsets $`A`$, $`B`$ of $`G`$, each of size $`<\kappa `$, there is $`cG`$ connected to all points of $`A`$ and no point of $`B`$. The existence of such $`G`$ follows from the assumption $`\kappa ^{<\kappa }=\kappa `$.
Note that a $`\kappa `$-Rado graph embeds every graph with at most $`\kappa `$-many vertices. That is, it is universal for the family of graphs of size at most $`\kappa `$. For $`\kappa =\omega `$, this graph is also called the infinite random graph.
Interest in Rado graphs and the uncountable continues. Let $`G_\omega `$ denote the $`\omega `$-Rado graph. Recently Gregory Cherlin and Simon Thomas have shown that for any infinite cardinals $`\kappa \lambda `$, the assumption $`\lambda 2^\kappa `$ is equivalent to the existence of a graph $`G^{}`$ of size lambda which is elementarily equivalent to $`G_\omega `$ and which has a vertex whose set of neighbours has size $`\kappa `$.
Here is our main result on $`\kappa `$-Rado graphs.
###### Theorem 1.2.
For every natural number $`m`$ there is a value $`r_m^+<\omega `$ such that for any cardinal $`\kappa `$ which is measurable in the generic extension obtained by adding $`\lambda `$ Cohen subsets of $`\kappa `$, where $`\lambda `$ is some cardinal satisying $`\lambda (\kappa )_{2^\kappa }^{2m}`$, the $`\kappa `$-Rado graph $`𝔾_\kappa `$
$$𝔾_\kappa (𝔾_\kappa )_{<\kappa ,r_m^+}^m\text{ and }𝔾_\kappa (𝔾_\kappa )_{<\kappa ,r_m^+1}^m.$$
In Theorem 9.14, the positive partition relation is shown to hold for $`r_m^+`$ the number of *vip $`m`$-types* (defined in Section 3). The vip $`m`$-types are closely related to those unique $`m`$-element *strongly diagonal* subsets of $`{}_{}{}^{2m2}2`$ that are representatives of the “essential types” used by Laflamme, Sauer and Vuksanovic in and by Vuksanovic in for the countable Rado graph. If we close such a strongly diagonal set under initial segments and add a *vip level order*, we obtain a vip $`m`$-type, and all vip $`m`$-types are obtained in this way.
In Theorem 10.9 we show that the negative partition relation holds for the same value of $`r_m^+`$.
It is not known if the large cardinal assumptions used in the proof of Shelah’s Theorem from are optimal; in Section 8 we comment more on this as well as on the consistency strength of these requirements. We note that in conjunction with a result of András Hajnal and Péter Komjáth from our theorem gives some necessary indestructibility conditions on $`\kappa `$ from Shelah’s Theorem. The paper also includes a section with a proof of the particular variant of Shelah’s Theorem that we need.
For the remainder of the paper an unattributed $`m`$ will mean a natural number with $`2m`$ and $`\kappa `$ a cardinal satisfying the hypotheses of Theorem 1.1 for some number $`m`$. In particular, $`\kappa =\kappa ^{<\kappa }`$ and $`\kappa `$ is a strong limit.
The paper is organized as follows. In Section 2, we define the notion *$``$-similarity*, and state the special variant of Shelah’s Theorem that we will use.
In Section 3, we define the notions of *diagonal*, vip order and sparse vip $`m`$-type and use them together with our variant of Shelah’s Theorem to prove Theorem 3.15.
In Section 4, we show that for any $`S{}_{}{}^{\kappa >}2`$ with $`(S,<_\text{Q})`$, one can build an *almost perfect* $`\kappa `$-dense subtree inside the tree of nodes of $`{}_{}{}^{\kappa >}2`$ with a $`\kappa `$-dense set of extensions in $`S`$.
In Section 5, we use the construction techniques of Section 4 to prove Theorem 5.8 by showing all sparse vip $`m`$-types are embeddable in every sparse diagonal $`\kappa `$-dense subset.
In Section 6, we show the critical numbers $`t_m^+`$ for the $`\kappa `$-dense linear order are bounded below by $`t_m`$ and bounded above by $`t_m(m1)!\left[_{i<m1}(i!)^2\right]`$, and indicate how one may compute the value of $`t_m^+`$ recursively. We conclude the section with a table of small values of $`t_m^+`$ quoted from an upcoming paper by Jean Larson .
In Section 7 we prove the variant of Shelah’s Theorem that we use, formulating it using colorings of antichains. In Section 8 we comment on the necessity of the use of large cardinals in our theorem and in Shelah’s Theorem and the way that the results of this paper shed light on that question. We also give some open questions.
In Section 9, we use nuanced diagonalization to prove Theorem 9.14 giving the upper bound on the critical values $`r_m^+`$ for $`\kappa `$-Rado graphs
In Section 10, we prove a reduction theorem for the translations to tree form of increasing embeddings of the $`\kappa `$-Rado graph $`𝔾_\kappa `$: for every such translation with range $`D`$ there is always a particularly nice form of a diagonalization which has range a subset of $`D`$ and is itself the translation of an increasing embedding of $`𝔾_\kappa `$ into itself. We use this reduction to prove Theorem 10.9. by showing every vip $`m`$-type is embeddable in the range of these particularly nice diagonalizations.
The remainder of this introduction is devoted to background information, including some definitions, notation, and the statements of some theorems that will be used as tools. For any cardinal $`\lambda `$, let $`[A]^\lambda `$ denote the collection of all subsets of $`A`$ of cardinality $`\lambda `$, and let $`[A]^{<\lambda }`$ denote the collection of all subsets of $`A`$ of cardinality less than $`\lambda `$.
For any tree $`T=(T,)`$, a node $`s`$ is a *leaf of $`T`$* or *terminal node of $`T`$* if for all $`tT\{s\}`$, one has $`st`$. The notion of the meet of two nodes has already been defined, in the first paragraph of the Introduction. If $`s`$ is a node of $`T`$ and both $`s{}_{}{}^{}0`$ and $`s{}_{}{}^{}1`$ have extensions in $`T`$, then we call $`s`$ a *splitting node of $`T`$*. For any subset $`ST`$, let $`S^{}`$ denote the *meet closure* of $`S`$, i.e. the set $`\{u:u=st\text{ for some }s,tS\}`$. Note that $`S^{}S`$ and that $`S^{}`$ is closed under meets and has a unique node of the smallest length.
###### Definition 1.3.
For any tree $`T`$ of sequences ordered by end extension, and any node $`sT`$, define the *set of immediate successors of $`s`$ in $`T`$* as
$$\text{IS}(s,T):=\{tT:st(u)(sutu=t)\}.$$
###### Definition 1.4.
For any tree $`T`$ of sequences under end extension, the *$`\alpha `$th level of $`T`$*, in symbols $`T(\alpha )`$, is the set of all nodes $`tT`$ for which $`\alpha `$ is the order type of the set of predecessors of $`t`$, namely $`\{sT:st\}`$. For $`sT`$ the length $`\mathrm{lg}(s)`$ is defined to be $`\alpha `$ if and only if $`sT(\alpha )`$. Call $`T`$ an *$`\alpha `$-tree* if each branch of $`T`$ has order type $`\alpha `$.
The above notion of an $`\alpha `$-tree in the case that $`\alpha `$ is a cardinal differs from the usual notion of an $`\alpha `$-tree (see e.g. ), which is defined as a tree of height $`\alpha `$ all of whose levels have size $`<\alpha `$. We trust that no confusion will arise, but emphasize that our definition is given above. Note that for the complete binary tree $`T={}_{}{}^{\kappa >}2`$, the levels of $`T`$ are $`T(\alpha )={}_{}{}^{\alpha }2`$ for $`\alpha <\kappa `$. Call $`AT`$ a *level set* if $`AT(\alpha )`$ for some $`\alpha `$.
Milliken (see Definition 1.2 of ) defined a notion of strongly embedded tree which can be simplified in the case of a binary tree. The key idea is that there are *splitting levels* (see the range of $`h`$ below); all nodes of a strongly embedded tree split at splitting levels as much as is possible, and no splitting occurs elsewhere. Another way to say this is that a strongly embedded subtree of height $`\alpha `$ is a copy of $`{}_{}{}^{\alpha >}2`$ in $`{}_{}{}^{\kappa >}2`$ where levels are mapped to levels. Note that a strongly embedded subtree $`S`$ of $`{}_{}{}^{\kappa >}2`$ is not required to be an induced subtree of $`{}_{}{}^{\kappa >}2`$, in the sense that it is not required that $`S`$ is closed under $``$.
###### Definition 1.5.
A subset $`S{}_{}{}^{\kappa >}2`$ is *a strongly embedded subtree of $`{}_{}{}^{\kappa >}2`$* if
(1) $`(S,)`$ is an $`\alpha `$-tree for some $`\alpha \kappa `$;
(2) $`S`$ has a root and every non-maximal node $`sS`$ has exactly one extension of $`s{}_{}{}^{}0`$ and one extension of $`s{}_{}{}^{}1`$ in $`\text{IS}(s,S)`$;
(3) there is a *level assignment function* $`h:\alpha \kappa `$ which is a strictly increasing function such that for all $`\beta <\alpha `$, $`S(\beta ){}_{}{}^{h(\beta )}2`$;
(4) for any limit $`\beta <\alpha `$, for any $`stS(\beta )`$, there is $`\delta <sup\{h(\gamma ):\gamma <\beta \}`$ such that $`s\delta t\delta `$.
Note of course that part (4) above is irrelevant for strongly embedded trees of height $`\omega `$, as originally studied by Milliken. As mentioned above, strongly embedded subtrees of $`{}_{}{}^{\kappa >}2`$ are copies of full binary trees of height $`\kappa `$ where levels are mapped to levels. The precise statement of this is the content of Lemma 1.7.
###### Definition 1.6.
For any subsets $`S_0`$ and $`S_1`$ of $`{}_{}{}^{\kappa >}2`$, a function $`e:S_0S_1`$ is a *strong embedding* if it is an injection with the following preservation properties:
1. (extension) $`st`$ if and only if $`e(s)e(t)`$;
2. (level order) $`\mathrm{lg}(s)<\mathrm{lg}(t)`$ if and only if $`\mathrm{lg}(e(s))<\mathrm{lg}(e(t))`$ and $`\mathrm{lg}(s)=\mathrm{lg}(t)`$ if and only if $`\mathrm{lg}(e(s))=\mathrm{lg}(e(t))`$;
3. (passing number) if $`\mathrm{lg}(s)<\mathrm{lg}(t)`$, then $`e(t)(lg(e(s)))=t(\mathrm{lg}(s))`$.
###### Lemma 1.7.
An $`\alpha `$-tree $`S{}_{}{}^{\kappa >}2`$ is strongly embedded in $`{}_{}{}^{\kappa >}2`$ if and only if there is a strong embedding $`e:{}_{}{}^{\alpha >}2{}_{}{}^{\kappa >}2`$ whose range is $`S`$.
###### Proof.
If $`e:{}_{}{}^{\alpha >}2{}_{}{}^{\kappa >}2`$ is a strong embedding, then there is a level assignment function $`h:\kappa \kappa `$ witnessing the strong embedding in that for all $`\beta <\alpha `$, if $`s{}_{}{}^{\beta }2`$, then $`e(s){}_{}{}^{h(\beta )}2`$ and the immediate successors of $`e(s)`$ in the range of $`e`$ are $`e(s{}_{}{}^{}0)`$ and $`e(s{}_{}{}^{}1)`$, which are both in $`{}_{}{}^{h(\beta +1)}2`$. Thus if $`e`$ is a strong embedding its image is a strongly embedded tree.
For the other direction, suppose $`S{}_{}{}^{\kappa >}2`$ is a strongly embedded $`\alpha `$-tree and $`h:\alpha \kappa `$ is the level assignment function that witnesses it. We shall define $`e`$ on $`{}_{}{}^{\alpha >}2`$ such that $`e`$ maps $`{}_{}{}^{\beta }2`$ into $`{}_{}{}^{h(\beta )}2`$ for all $`\beta <\alpha `$, as follows. Let $`e()`$ be the root of $`S`$. Given $`t{}_{}{}^{\alpha >}2`$ non-maximal and of height $`\beta `$, let $`e(t{}_{}{}^{}l)`$ for $`l<2`$ be the unique extension of $`e(t){}_{}{}^{}l`$ in $`\text{IS}(e(t),S)`$. For $`\beta <\alpha `$ limit and $`t{}_{}{}^{\beta }2`$, note that $`_{st}e(s)`$ is an element of $`{}_{}{}^{\kappa >}2`$ and that it must have an extension in $`{}_{}{}^{h(\beta )}2`$, as $`S`$ is an $`\alpha `$-tree. By requirement (4) in 1.5, this extension is unique and we choose it as $`e(t)`$. Now $`e`$ is a strong embedding whose image is $`S`$.
Let us finish the section by proving the above mentioned fact that for regular $`\kappa `$ the order $`<_Q`$ on $`{}_{}{}^{\kappa >}2`$ is $`\kappa `$-dense.
###### Lemma 1.8.
The order $`({}_{}{}^{\kappa >}2,<_Q)`$ is $`\kappa `$-dense if and only if $`\kappa `$ is a regular cardinal.
###### Proof.
Suppose that $`\kappa `$ is regular and $`A,B{}_{}{}^{\kappa >}2`$ are both of size $`<\kappa `$ and such that for all $`aA`$ and $`bB`$ we have $`a<_Qb`$ (we write this as $`A<_QB`$). Let $`B^{}:=\{t{}_{}{}^{\kappa >}2:(bB)t{}_{}{}^{}0b\}`$. If $`aA`$ and $`tB^{}`$ then there is $`bB`$ such that $`t{}_{}{}^{}0b`$, in particular $`b<_Qt`$. By the transitivity of $`<_Q`$ we obtain $`a<_Qt`$, and hence $`A<_QB^{}`$.
Let $`\gamma `$ be the minimal ordinal such that all elements of $`AB`$ have length $`<\gamma `$, so by the regularity of $`\kappa `$ we have $`\gamma <\kappa `$. We define $`c{}_{}{}^{\kappa >}2`$ by defining $`c(\alpha )`$ for $`\alpha <\gamma `$ by recursion on $`\alpha `$. If $`c\alpha BB^{}`$ we set $`c(\alpha )=0`$, and we set $`c(\alpha )=1`$ otherwise. We claim that $`A<_Qc<_QB`$.
If $`bB`$ then the length of $`b`$ is less than the length of $`c`$ and hence either $`bc`$ or $`b`$ and $`c`$ are incomparable. In the first case we have defined $`c`$ so that $`b{}_{}{}^{}0c`$. In the second case, if $`\alpha `$ is the length of $`bc`$ and $`b(\alpha )=0`$, then $`t:=b\alpha B^{}`$ so we have defined $`c(\alpha )=0`$, contradicting the choice of $`\alpha `$. So $`b(\alpha )=1`$ and $`c(\alpha )=0`$. In both cases we have $`c<_Qb`$, and hence $`c<_QB`$.
If $`aA`$ and $`ac`$ then we claim $`a{}_{}{}^{}1c`$. Otherwise, since the length of $`a`$ is strictly less than that of $`c`$ we have $`a{}_{}{}^{}0c`$, so $`a=c\alpha `$ must be a member of $`BB^{}`$, a contradiction. If $`a`$ and $`c`$ are incomparable we can similarly show that $`a<_{\mathrm{lex}}c`$. In any case $`a<_Qc`$ and we have proved that $`A<_Qc`$. This proves that $`_\kappa `$ is $`\kappa `$-dense.
Suppose now that $`\kappa `$ is singular and $`\kappa _i`$ for $`i<\mathrm{cf}(\kappa )`$ is an increasing sequence of limit ordinals with limit $`\kappa `$, and with $`\kappa _0=\kappa `$. For all $`i`$ let $`a_i`$ be the sequence in $`{}_{}{}^{\kappa _i}2`$ which is constantly equal to 1, and let $`A:=\{a_i:i<\mathrm{cf}(\kappa )\}`$. We then have $`|A|<\kappa `$, yet we claim that there is no $`c{}_{}{}^{\kappa }>2`$ with $`A<_Qc`$. Namely, let $`c{}_{}{}^{\alpha }2`$ for some $`\alpha <\kappa `$ and let $`i`$ be such that $`\alpha [\kappa _i,\kappa _{i+1})`$. Then $`c{}_{}{}^{}1a_{i+2}`$, so $`A<_Qc`$ cannot hold. This proves that $`_\kappa `$ is not $`\kappa `$-dense.
An argument similar to the first part of the above proof is presented in the proof of Lemma 3.11.
## 2 Uniformization
In this section we state a particular variant of a theorem of Shelah. His theorem was stated as a generalization of the Laver-Pincus version of the Halpern-Läuchli Theorem, and the variant we state is a generalization of Milliken’s Ramsey theorem for finite weakly embedded subtrees.
Before we can state Shelah’s Theorem, we need the definition of two subsets being $``$-similar, where $``$ is a level ordering of $`{}_{}{}^{\kappa >}2`$:
###### Definition 2.1.
Say that $``$ is a *level ordering* of $`{}_{}{}^{\kappa >}2`$ or alternatively that $``$ is *an ordering of the levels* of $`{}_{}{}^{\kappa >}2`$, if $``$ extends the length order, i.e. $``$ is a linear order of $`{}_{}{}^{\kappa >}2`$ and $`\mathrm{lg}(s)<\mathrm{lg}(t)`$ implies $`st`$.
Milliken’s Theorem for weakly embedded subtrees requires us to recognize subtrees of the same embedding type. We will be counting the number of types of particularly nice trees at a later point in the paper, so it is convenient to relate these types to specific finite examples called *similarity trees*.
###### Definition 2.2.
A *similarity tree* is a finite subtree of $`{}_{}{}^{\omega >}2`$ closed under initial segments and such that every level contains at least one leaf (terminal) node or meet of leaf nodes (split point). An *ordered similarity tree* is a similarity tree $`t`$ with an ordering $`_t`$ of its levels.
For a finite subset $`x`$ of $`{}_{}{}^{\kappa >}2`$ we define as $`\mathrm{clp}(x)`$ the subtree $`y`$ of $`{}_{}{}^{\omega >}2`$ that includes the root and is of minimal possible height such that there is a strong embedding from $`x^{}`$ onto the closure $`z^{}`$ of the set $`z`$ of terminal nodes of $`y`$. If $``$ is a given order of $`{}_{}{}^{\kappa >}2`$ then $`_x`$ is the order on $`\mathrm{clp}(x)`$ induced by the strong embedding from $`\mathrm{clp}(x)`$ to $`x`$ and $``$.
To illustrate the definition, consider a specific antichain:
$$x=\{0,0,0,1,0,0,1,0,1,0,0,0,1\}.$$
Further suppose that $``$ is the following ordering of the levels: $`0,00,1`$; $`0,0,00,1,00,0,1`$; and $`0,0,0,10,1,0,0`$. The corresponding similarity tree is $`(\mathrm{clp}(x),)`$ where
$$\mathrm{clp}(x)=\{\mathrm{},0,1,0,0,0,1,1,0,0,0,1,1,0,0,1,0,0,0\}.$$
The induced ordering of the levels is $`0_x1`$; $`0,0_x1,0_x0,1`$; and $`0,0,1_x1,0,0`$.
###### Definition 2.3.
Suppose that $``$ is an ordering of the levels of $`{}_{}{}^{\kappa >}2`$. Two antichains $`x`$ and $`y`$ in $`{}_{}{}^{\kappa >}2`$ are *similar* if $`\mathrm{clp}(x)=\mathrm{clp}(y)`$, and *$``$-similar* if the ordering $``$ induces the same ordering $`_x=_y`$ on the collapsed trees. In this case we call $`(\mathrm{clp}(x),_x)`$ the *ordered similarity type* of $`(x,)`$ and $`(y,)`$.
We remark that if $`x`$ and $`y`$ are $``$-similar antichains, then $`\mathrm{clp}(x)`$ and $`\mathrm{clp}(y)`$ are subtrees of $`{}_{}{}^{\kappa >}2`$ having the same *weak embedding type*, a notion which appears in various places in the literature, see e.g. and . The notion of $``$-similarity given above is a translation of the definition of similarity by Shelah in Definition 2.2 . It also corresponds to the notion of *strong similarity* from , which coincides with the notion of *similarity* of that paper when restricted to *strongly diagonal sets*.
Theorem 2.5 below is a variant of Shelah’s Lemma 4.1 of together with Conclusion 4.2 where it is called “$`Pr_{ht}^f(\kappa ,m,\sigma )`$ (even with (3))”. In fact, except for the assertion that $`e`$ preserves $``$ and that we are dealing with coloring of antichains rather than level sets (a level set is a subset of $`{}_{}{}^{\alpha }2`$ for some $`\alpha <\kappa `$), our Theorem 2.5 is equivalent to Shelah’s Conclusion 4.2 from that paper with option (3)(b). The fact that $`e`$ preserves $``$ implies that option (3)(a) is automatically satisfied as well. Lemma 7.1, which is the main lemma in the proof of Theorem 2.5, is equivalent to Shelah’s Lemma 4.1 with option (3)(a). In Shelah’s notation the superscript $`f`$ of Lemma 4.1 indicates that one gets a strongly embedded tree $`T`$, and the subscript $`ht`$ means that the coloring restricted to $`m`$-element level sets of $`T`$ is *homogeneous*, i.e. is constant on subsets whose collapses under the level assignment function of $`T`$ are $``$-similar. The star orderings, $`<_\alpha ^{}`$, in this option, are defined from the $`0`$ orderings, $`<_\beta ^0`$. We use $``$ in place of $`<_\beta ^0`$, and will define the order of interest at a later point. We write $`e()`$ for $`\{(e(a),e(b)):ab\}`$. Thus given a strong embedding $`e:{}_{}{}^{\kappa >}2{}_{}{}^{\kappa >}2`$, $`s<_\alpha ^{}t`$ is translated by $`e(s)e()e(t)`$, which holds if and only if $`st`$. We also note that the statement of the theorem as we give it only refers to a dense set of elements $`w`$. A proof of Theorem 2.5 will be given in section 7.
The following notation is introduced for convenience.
###### Definition 2.4.
Write $`\mathrm{Cone}(w)`$ for the set of all extensions $`tw`$ in $`{}_{}{}^{\kappa >}2`$.
###### Theorem 2.5.
\[Shelah \] Suppose that $`m<\omega `$ and $`\kappa `$ is a cardinal which is measurable in the generic extension obtained by adding $`\lambda `$ Cohen subsets of $`\kappa `$, where $`\lambda (\kappa )_{2^\kappa }^{2m}`$. Then for any coloring $`d`$ of the $`m`$-element antichains of $`{}_{}{}^{\kappa >}2`$ into $`\sigma <\kappa `$ colors, and any well-ordering $``$ of the levels of $`{}_{}{}^{\kappa >}2`$, there is a strong embedding $`e:{}_{}{}^{\kappa >}2{}_{}{}^{\kappa >}2`$ and a dense set of elements $`w`$ such that
1. $`e(s)e(t)`$ for all $`st`$ from $`\mathrm{Cone}(w)`$, and
2. $`d(e[a])=d(e[b])`$ for all $``$-similar $`m`$-element antichains $`a`$ and $`b`$ of $`\mathrm{Cone}(w)`$.
## 3 Upper bound for dense linear orders
In this section we prove a limitation of colors result for $`\kappa `$-dense linear orders using Shelah’s Theorem 2.5.
We turn to some ideas from work of Sauer, Laflamme and Vuksanovic and others for ways to guarantee that the $`m`$-element sets of our yet to be chosen transverse set have a minimal number of (unordered) weak embedding types. The notions of *diagonal* set and *strongly diagonal* set are used in , for example. Our definitions below are simplifications of those definitions to the special case of binary trees.
###### Definition 3.1.
Call $`A{}_{}{}^{\kappa >}2`$ *diagonal* if it is an *antichain* (its elements are pairwise incomparable), and its meet closure, $`A^{}`$, is transverse. Call it *strongly diagonal* if, in addition, for all all $`tA`$ and all $`sA^{}\{t\}`$, the following implication holds:
$$(\mathrm{lg}(s)<\mathrm{lg}(t)\text{ and }t(\mathrm{lg}(s))=1)st\text{ or }s\text{ has no extension in }A.$$
###### Lemma 3.2.
For any finite diagonal set $`a{}_{}{}^{\kappa >}2`$, its meet closure has $`|a^{}|=2|a|1`$ elements and $`\mathrm{clp}(a){}_{}{}^{2\left|a\right|2}2`$. Thus for positive $`m<\omega `$, the number of ordered similarity types of $`m`$-element diagonal sets is finite.
In the case of $`\kappa `$-dense linear orders, we are particularly interested in *sparse diagonal* sets.
###### Definition 3.3.
Call $`A{}_{}{}^{\kappa >}2`$ *sparse* if for all $`tA`$ and $`sA^{}\left\{t\right\}`$ the following implication holds:
$$(\mathrm{lg}(s)<\mathrm{lg}(t)\text{ and }t(\mathrm{lg}(s))=1)st.$$
Notice that if $`A`$ is a sparse diagonal set, then it is a strongly diagonal set. However, a strongly diagonal subset need not be sparse.
We are interested in a special collection of level orders. We call them *$`D`$-vip orders* since the elements of $`D`$ are Very Important Points, with special roles to play in the orders, and these roles continue to be played even when finite subsets are collapsed.
###### Definition 3.4.
Suppose $`T`$ is a subtree of $`{}_{}{}^{\kappa >}2`$ and $`D{}_{}{}^{\kappa >}2`$. Call $``$ a *pre-$`D`$-vip order* on $`T{}_{}{}^{\kappa >}2`$ if $`D`$ is transverse and $``$ is a well-ordering of the levels of $`T`$ such that for every $`dD`$, $`d`$ is the $``$-least element of its level, $`T(\mathrm{lg}(d))`$, and for all $`u,vT(\mathrm{lg}(d))\{d\}`$, $``$ satisfies the following condition:
1. if $`dudvd`$, then $`uv`$, and
If $`D`$ is diagonal, call $``$ a *$`D`$-vip order* if it is a pre-$`D^{}`$-vip order which also satisfies the following condition for all $`dD^{}`$ and for all $`u,vD\{d\}`$:
1. if $`du=dvd`$ and $`u(\mathrm{lg}(d))<v(\mathrm{lg}(d))`$, then $`uv`$.
It is not difficult to construct a pre-$`D`$-vip order for a transverse set $`D`$.
###### Lemma 3.5.
If $`D{}_{}{}^{\kappa >}2`$ is transverse, then there is a pre-$`D`$-vip order of $`{}_{}{}^{\kappa >}2`$. If $`D{}_{}{}^{\kappa >}2`$ is a sparse diagonal set and $``$ is a pre-$`S`$-vip order for some $`S`$ with $`D^{}S`$, then $``$ is a $`D`$-vip order.
###### Proof.
Let $``$ be any well-ordering of the levels of $`{}_{}{}^{\kappa >}2`$. Use recursion to define a pre-$`D`$-vip order $``$ by adjusting $``$ separately on each level which has an element of $`D`$.
Note that if $`D{}_{}{}^{\kappa >}2`$ is a sparse diagonal set, and $``$ is a pre-$`S`$-vip order for some $`S`$ with $`D^{}S`$, then the second clause never applies so $``$ is a $`D`$-vip order. ∎
Now that we have at least some idea of the level order we will use, we define a family of ordered similarity types which is rich enough to have all the ordered similarity types of $`m`$-element subsets of sparse diagonal sets $`D{}_{}{}^{\kappa >}2`$ for a $`D`$-vip level ordering $``$.
###### Definition 3.6.
Call $`\tau `$ an *$`m`$-type* if it is a downward closed subtree of $`{}_{}{}^{2m2}2`$ whose set $`L`$ of leaves is an $`m`$-element strongly diagonal set. Call $`(\tau ,)`$ a *vip $`m`$-type* if $`\tau `$ is a $`m`$-type and $``$ is an $`L`$-vip order. If $`L`$ is sparse, then $`\tau `$ is called a *sparse $`m`$-type* and $`(\tau ,)`$ is called a *sparse vip $`m`$-type*.
###### Lemma 3.7.
Assume $`D{}_{}{}^{\kappa >}2`$ is a strongly diagonal set and $``$ is an ordering of the levels of $`{}_{}{}^{\kappa >}2`$ which is a $`D`$-vip order. Then for all $`m`$-element sets $`xD`$, $`(\mathrm{clp}(x),_x)`$ is a vip $`m`$-type, and if $`D`$ is sparse, then $`(\mathrm{clp}(x),_x)`$ is a sparse vip $`m`$-type.
###### Proof.
The collapse of any $`m`$-element strongly diagonal set $`x`$ is a subtree closed under initial segments whose set of leaves, $`L`$, is strongly diagonal. Moreover, since the order $``$ is $`x`$-vip, it follows that $`_x`$ is $`L`$-vip. If, in addition, $`D`$ is sparse, then $`\mathrm{clp}(x)`$ is also sparse, by definition of collapse. ∎
In order to apply Shelah’s Theorem, we shall need an ordering of the levels of $`{}_{}{}^{\kappa >}2`$ and a conveniently chosen subset of $`{}_{}{}^{\kappa >}2`$ the antichains of which will realize the smallest possible number of weak embedding types. Toward that end, we introduce *cofinal transverse* subsets of $`{}_{}{}^{\kappa >}2`$.
###### Definition 3.8.
A subset $`S{}_{}{}^{\kappa >}2`$ is *cofinal above $`w`$* if for all $`t\mathrm{Cone}(w)`$ there is some $`sS`$ with $`ts`$. If $`w=\mathrm{}`$, we say $`S`$ is *cofinal*.
###### Definition 3.9.
Call $`A{}_{}{}^{\kappa >}2`$ *transverse* if distinct elements of $`A`$ have different lengths.
By recursion one can construct a cofinal transverse subset of $`{}_{}{}^{\kappa >}2`$.
###### Lemma 3.10.
There is a cofinal transverse subset of $`{}_{}{}^{\kappa >}2`$.
###### Lemma 3.11.
If $`S{}_{}{}^{\kappa >}2`$ is cofinal above $`w`$ and transverse, then $`(S\mathrm{Cone}(w),<_\text{Q})`$ is $`\kappa `$-dense.
###### Proof.
Suppose $`A,BS\mathrm{Cone}(w)`$ are two disjoint subsets of size less than $`\kappa `$ with $`a<_Qb`$ for all $`aA`$ and $`bB`$. Use the fact that $`\mathrm{Cone}(w)`$ is $`\kappa `$-dense to find $`d\mathrm{Cone}(w)(AB)`$ with $`a<_\text{Q}d<_\text{Q}b`$ for all $`aA`$ and $`bB`$. Let $`\alpha `$ be a limit ordinal larger than the length of any element of $`AB`$. Let $`d^{}`$ be the extension of $`d{}_{}{}^{}1`$ by zeros of length $`\alpha `$ and let $`c`$ be an extension of $`d^{}{}_{}{}^{}0`$ in $`S`$. Since $`d^{}`$ and $`c`$ are longer than any element of $`AB`$, they are not in $`A`$ nor in $`B`$. Then $`d<_\text{Q}c<_\text{Q}d^{}`$ and $`d^{}<_\text{Q}b`$ for all $`bB`$. Thus $`c`$ is the required witness showing $`(S,<_\text{Q})`$ is $`\kappa `$-dense. ∎
###### Definition 3.12.
Assume $`z,w{}_{}{}^{\kappa >}2`$ and $`S{}_{}{}^{\kappa >}2`$ is cofinal and transverse. Call $`g`$ a *sparse diagonalization of $`{}_{}{}^{\kappa >}2`$ into $`S\mathrm{Cone}(w)`$* if $`g:{}_{}{}^{\kappa >}2S\mathrm{Cone}(w)`$ is an injective $`<_\text{Q}`$-preserving map such that $`D:=g[{}_{}{}^{\kappa >}2]`$ is a sparse diagonal subset, $`D^{}S\mathrm{Cone}(w)`$ and the following conditions hold:
1. for all three element diagonal sets $`\{x,u,v\}`$, if $`xu=xv`$, then $`g(x)g(u)=g(x)g(v)`$;
2. for all $`x,y,u,v{}_{}{}^{\kappa >}2`$, if $`\mathrm{lg}(xy)<\mathrm{lg}(uv)`$, then $`\mathrm{lg}(g(x)g(y))<\mathrm{lg}(g(u)g(v))`$;
3. for all sparse diagonal $`E{}_{}{}^{\kappa >}2`$, $`\mathrm{clp}(E)=\mathrm{clp}(g[E])`$.
If $`w=\mathrm{}`$, then we call $`g`$ a *sparse diagonalization of $`{}_{}{}^{\kappa >}2`$ into $`S`$*.
###### Lemma 3.13.
\[First Diagonalization Lemma\] Assume $`\kappa =2^{<\kappa }`$, $`w{}_{}{}^{\kappa >}2`$ and $`S{}_{}{}^{\kappa >}2`$ is cofinal and transverse. Then there is a sparse diagonalization $`\phi `$ of $`{}_{}{}^{\kappa >}2`$ into $`S\mathrm{Cone}(w)`$.
###### Proof.
Let $``$ be a well-ordering of the levels of $`{}_{}{}^{\kappa >}2`$. Let $`t_\alpha :\alpha <\kappa `$ list the elements of $`{}_{}{}^{\kappa >}2`$ in $``$-increasing order (recall that $`\kappa =2^{<\kappa }`$ is assumed).
Define functions $`\phi _0,\phi _1`$ and $`\phi `$ on $`{}_{}{}^{\kappa >}2`$ by recursion on $`\alpha `$. To start the recursion, notice that $`t_0=\mathrm{}`$, and let $`\phi _0(t_0)`$ be an element of $`S\mathrm{Cone}(w)`$ of minimal length.
If $`\phi _0(t_\alpha )`$ has been defined, let $`\phi _1(t_\alpha )`$ be an element of $`S\mathrm{Cone}(w)`$ extending $`\phi _0(t_\alpha ){}_{}{}^{}1`$ and let $`\phi (t_\alpha )`$ be an element of $`S\mathrm{Cone}(w)`$ extending $`\phi _1(t_\alpha ){}_{}{}^{}0`$. Note that $`\phi (t_\alpha )\phi _1(t_\alpha )\phi _0(t_\alpha )`$.
Suppose $`\phi _0(t_\beta )`$, $`\phi _1(t_\beta )`$ and $`\phi (t_\beta )`$ have been defined for all $`\beta <\alpha `$, and $`t_\alpha =s_\alpha {}_{}{}^{}\delta `$ for some $`\delta `$. Let $`\gamma _\alpha `$ be the least ordinal $`\gamma `$ strictly greater than $`\mathrm{lg}(\phi (t_\beta ))`$ for all $`\beta <\alpha `$. Then let $`\phi _0(t_\alpha )`$ be an element of $`S\mathrm{Cone}(w)`$ extending the extension by zeros of $`\phi _\delta (s_\alpha ){}_{}{}^{}\delta `$ of length $`\gamma _\alpha `$.
Suppose $`\phi _0(t_\beta )`$, $`\phi _1(t_\beta )`$ and $`\phi (t_\beta )`$ have been defined for all $`\beta <\alpha `$, and $`\mathrm{lg}(t_\alpha )=\zeta `$ is a limit ordinal. Let $`\phi ^{}(t_\alpha ):=\{\phi _0(t_\alpha \eta ):\eta <\zeta \}`$. Let $`\gamma _\alpha `$ be the least ordinal $`\gamma `$ strictly greater than $`\mathrm{lg}(\phi ^{}(t_\alpha ))`$ and strictly greater than $`\mathrm{lg}(\phi (t_\beta ))`$ for all $`\beta <\alpha `$. Let $`\phi _0(t_\alpha )`$ be an element of $`S\mathrm{Cone}(w)`$ extending the extension by zeros of $`\phi ^{}(t_\alpha )`$ of length $`\gamma _\alpha `$.
By construction, $`\phi _0`$ and hence $`\phi _1`$ and $`\phi `$ are injective. Moreover, the ranges of all three are subsets of $`S\mathrm{Cone}(w)`$, so for $`D:=\phi [{}_{}{}^{\kappa >}2]`$, the meet closure satisfies $`D^{}\mathrm{Cone}(w)`$. Below we shall show that $`D^{}S`$.
For all $`\alpha <\kappa `$, one has $`\mathrm{lg}(\phi _0(t_\alpha ))<\mathrm{lg}(\phi _1(t_\alpha ))<\mathrm{lg}(\phi (t_\alpha ))<\mathrm{lg}(\phi _0(t_{\alpha +1}))`$ and if $`\alpha `$ is a limit then for all $`\beta <\alpha `$ we have $`\mathrm{lg}(\phi (t_\beta ))<\mathrm{lg}(\phi (t_\alpha ))`$. Thus if $`\alpha <\beta `$, then $`\mathrm{lg}(\phi (t_\alpha ))<\mathrm{lg}(\phi _0(t_\beta ))`$, so different elements of the union of the ranges of $`\phi _0`$, $`\phi _1`$ and $`\phi `$ have different lengths.
Consider two distinct elements, $`t_\alpha <_\text{Q}t_\beta `$. By a case analysis below, we show that their images are incomparable, the $`<_\text{Q}`$-order between $`t_\alpha `$ and $`t_\beta `$ is preserved by $`\phi `$ and we show how to express the meet $`\phi (t_\alpha )\phi (t_\beta )`$ as one of $`\phi (t_\alpha t_\beta )`$, $`\phi _0(t_\beta )`$ and $`\phi _1(t_\alpha )`$.
For the first case, suppose $`t_\alpha `$ and $`t_\beta `$ are incomparable. Then $`t_\alpha <_{\text{lex}}t_\beta `$. Let $`\gamma `$ be such that $`t_\gamma =t_\alpha t_\beta `$. Then $`t_\gamma {}_{}{}^{}0t_\alpha `$ and $`t_\gamma {}_{}{}^{}1t_\beta `$. From the definition of $`\phi _0`$, $`\phi _1`$ and $`\phi `$, it follows that $`\phi _0(t_\alpha t_\beta )=\phi _0(t_\gamma )=\phi (t_\alpha )\phi (t_\beta )`$ and $`\phi (t_\alpha )<_{\text{lex}}\phi (t_\beta )`$, so $`t_\alpha <_\text{Q}t_\beta `$ and $`\phi (t_\alpha )<_\text{Q}\phi (t_\beta )`$.
For the second case, suppose $`t_\beta t_\alpha `$. Then, since $`t_\alpha `$ and $`t_\beta `$ are distinct and $`t_\alpha <_\text{Q}t_\beta `$, it follows that $`t_\beta {}_{}{}^{}0t_\alpha `$. Consequently $`t_\alpha t_\beta =t_\beta `$, $`\phi (t_\alpha )\phi (t_\beta )=\phi _0(t_\beta )`$, and $`\phi (t_\alpha )<_{\text{lex}}\phi _1(t_\beta )\phi (t_\beta )`$, so $`\phi (t_\alpha )<_\text{Q}\phi (t_\beta )`$.
For the third case, suppose $`t_\alpha t_\beta `$. Then, since $`t_\alpha `$ and $`t_\beta `$ are distinct and $`t_\alpha <_\text{Q}t_\beta `$, it follows that $`t_\alpha {}_{}{}^{}1t_\beta `$. Consequently $`t_\alpha t_\beta =t_\alpha `$, $`\phi (t_\alpha )\phi (t_\beta )=\phi _1(t_\alpha )`$. Since $`\phi _1(t_\alpha ){}_{}{}^{}0\phi (t_\alpha )`$ and $`\phi _1(t_\alpha ){}_{}{}^{}0<_{\text{lex}}\phi (t_\beta )`$, it follows that $`\phi (t_\alpha )<_\text{Q}\phi (t_\beta )`$.
By the above analysis, $`\phi `$ preserves the $`<_\text{Q}`$-order and sends distinct elements of $`{}_{}{}^{\kappa >}2`$ to incomparable elements. Thus the image of $`\phi `$ is an antichain. Moreover, for $`\eta <\zeta `$, the meet, $`\phi (t_\eta )\phi (t_\zeta )`$, is one of $`\phi _0(t_\eta t_\zeta )`$ and $`\phi _1(t_\eta t_\zeta )`$, and the latter occurs only if $`t_\eta {}_{}{}^{}1t_\zeta `$. Thus different elements of the meet closure of the image of $`\phi `$ have different lengths. It follows that the image of $`\phi `$ is diagonal. Moreover, by construction, the meet closure of the image $`D`$ of $`\phi `$ is a subset of $`S\mathrm{Cone}(w)`$.
To complete the proof of the lemma, we must show that the image $`D`$ is sparse. First use induction on $`\beta <\kappa `$ to show that for all $`\alpha <\beta `$, the following two statements hold:
1. $`\phi (t_\beta )(\mathrm{lg}(\phi _0(t_\alpha )))=0`$;
2. ($`\phi (t_\beta )(\mathrm{lg}(\phi (t_\alpha )))=1`$ or $`\phi (t_\beta )(\mathrm{lg}(\phi _1(t_\alpha )))=1`$) if and only if
($`\phi (t_\beta )(\mathrm{lg}(\phi (t_\alpha )))=1`$ and $`\phi (t_\beta )(\mathrm{lg}(\phi _1(t_\alpha )))=1`$) if and only if
($`t_\alpha {}_{}{}^{}1t_\beta `$ and $`\phi _1(t_\alpha ){}_{}{}^{}1\phi (t_\beta )`$).
Next suppose $`s`$ and $`t`$ in the meet closure of $`\phi [{}_{}{}^{\kappa >}2]`$ are such that $`\mathrm{lg}(t)>\mathrm{lg}(s)`$ and $`t(\mathrm{lg}(s))=1`$. Without loss of generality, we assume that $`t`$ is in the image of $`\phi `$, since we know it has an extension in the image, and let $`t_\beta `$ be such that $`t=\phi (t_\beta )`$. By construction, since $`t(\mathrm{lg}(s))=1`$, either $`s=\phi _0(t_\alpha )`$ or $`s=\phi _1(t_\alpha )`$ for some $`\alpha \beta `$. If $`\alpha <\beta `$, then $`s{}_{}{}^{}1t`$ by the second statement above since $`\phi _0(t_\alpha ){}_{}{}^{}1\phi _0(t_\alpha )`$. If $`\alpha =\beta `$, then $`s=\phi _0(t_\alpha )=\phi _0(t_\beta )`$ since $`\phi (t_\beta )`$ is an extension of $`\phi _1(t_\beta ){}_{}{}^{}0`$. In this case, $`s{}_{}{}^{}1t`$ as well. Thus the image of $`D`$ is sparse. ∎
###### Lemma 3.14.
Suppose $``$ is ordering of the levels of $`{}_{}{}^{\kappa >}2`$. If $`e:{}_{}{}^{\kappa >}2{}_{}{}^{\kappa >}2`$ is a strong embedding which preserves $``$ on $`\mathrm{Cone}(w)`$ and $`A\mathrm{Cone}(w)`$ is an $`m`$-element sparse diagonal subset, then $`(\mathrm{clp}(A),_A)=(\mathrm{clp}(e[A]),_{e[A]})`$.
###### Proof.
Under the hypotheses of the lemma, since $`e`$ is a strong embedding, $`\mathrm{clp}(e[A])=\mathrm{clp}(A)`$. Since $`e`$ preserves order on $`\mathrm{Cone}(w)`$, $`_{e[A]}=_A`$. ∎
###### Theorem 3.15.
Let $`m2`$ and suppose that $`\kappa `$ is a cardinal which is measurable in the generic extension obtained by adding $`\lambda `$ Cohen subsets of $`\kappa `$, where $`\lambda (\kappa )_{2^\kappa }^{2m}`$. Then for $`t_m^+`$ equal to the number of vip $`m`$-types,
$$_\kappa (_\kappa )_{<\kappa ,t_m^+}^m.$$
###### Proof.
Suppose $`d:[{}_{}{}^{\kappa >}2]^m\mu `$ is a fixed coloring for some $`\mu <\kappa `$. Use Lemma 3.10 to find $`S{}_{}{}^{\kappa >}2`$ cofinal and transverse. Lemma 3.5 to find a pre-$`S`$-vip order $``$ of the levels of $`{}_{}{}^{\kappa >}2`$. Apply Shelah’s Theorem 2.5 to the restriction to antichains to obtain a strong embedding $`e`$ and a node $`w`$ such that $`e`$ preserves $``$ on $`\mathrm{Cone}(w)`$ and $`d`$ is constant on $`m`$-element subsets of the same $``$-ordered similarity type.
Apply the First Diagonalization Lemma 3.13 to find a sparse diagonalization $`\phi `$ of $`{}_{}{}^{\kappa >}2`$ into $`S\mathrm{Cone}(w)`$. Let $`D=\phi [{}_{}{}^{\kappa >}2]`$. Note that $``$ is a $`D`$-vip order, since $``$ is a pre-$`S`$-vip order. By Lemma 3.7, all $``$-ordered similarity types of $`m`$-element subsets of $`D`$ are sparse vip $`m`$-types. Let $`Q=e[D]`$. By Lemma 3.14, all $``$-ordered similarity types of $`m`$-element subsets of $`Q`$ are sparse vip $`m`$-types.
Since $`\phi `$ is a sparse diagonalization and $`e`$ a strong embedding, $`(D,<_\text{Q})`$ and $`(Q,<_\text{Q})`$ are both $`\kappa `$-dense. Since $`d`$ is constant on $`m`$-element subsets of $`Q`$ of the same $``$-ordered similarity type, it follows that $`d`$ takes on no more colors than the number $`t_m^+`$ of sparse vip $`m`$-types. ∎
## 4 An almost perfect subtree
In this section, in preparation for computing some small values of $`r_n^+`$, we show that, for an arbitrary $`S{}_{}{}^{\kappa >}2`$ with $`(S,<_\text{Q})`$ $`\kappa `$-dense, the set of all nodes $`w`$ with a $`\kappa `$-dense set of extensions in $`S`$ forms an *almost perfect subtree* (defined later in this section). We use the almost perfect subtree to construct a $`\kappa `$-dense diagonal subset of an arbitrary $`S{}_{}{}^{\kappa >}2`$ with $`(S,<_\text{Q})`$ $`\kappa `$-dense.
###### Lemma 4.1.
Suppose $`\kappa `$ satisfies $`\kappa ^{<\kappa }=\kappa `$ (so $`\kappa `$ is a regular cardinal), $`S{}_{}{}^{\kappa >}2`$, and $`(S,<_\text{Q})`$ is a $`\kappa `$-dense linear order. For all $`w{}_{}{}^{\kappa >}2`$, the set $`S\mathrm{Cone}(w)`$ is either empty, a singleton or $`\kappa `$-dense.
###### Proof.
Fix $`w{}_{}{}^{\kappa >}2`$. If $`S\mathrm{Cone}(w)`$ is empty or a singleton, there is nothing to prove. So suppose $`xS`$ and $`yS`$ are two different extensions of $`w`$. Notice that $`wxy`$. Without loss of generality, assume $`x<_\text{Q}y`$.
If $`x`$ and $`y`$ are incomparable and $`x<_\text{Q}z<_\text{Q}y`$, then either $`x{}_{}{}^{}1z`$ or $`y{}_{}{}^{}0z`$ or $`x<_{\text{lex}}z<_{\text{lex}}y`$. In all three cases $`wz`$.
If $`x`$ and $`y`$ are comparable, then either $`y{}_{}{}^{}0x`$ or $`x{}_{}{}^{}1y`$. Hence for any $`z`$ with $`x<_\text{Q}z<_\text{Q}y`$, one of $`x`$ and $`y`$ is a subset of $`z`$. In either case $`wz`$.
Since $`S`$ is $`\kappa `$-dense, it has a $`\kappa `$-dense subset $`S^{}`$ with $`\left\{x\right\}<S^{}<\left\{y\right\}`$. By the above two paragraphs, every element of $`S^{}`$ is an extension of $`w`$, so $`S\mathrm{Cone}(w)`$ is $`\kappa `$-dense. ∎
###### Definition 4.2.
Suppose $`\kappa ^{<\kappa }=\kappa `$, $`S{}_{}{}^{\kappa >}2`$ and $`(S,<_\text{Q})`$ is $`\kappa `$-dense. Let $`T(S)`$ be the set of all $`t{}_{}{}^{\kappa >}2`$ for which $`S\mathrm{Cone}(t)`$, is $`\kappa `$-dense.
We plan to show that $`T(S)`$ is an almost perfect tree. The first step is to show that arbitrarily high above every node there is a *densely splitting node*.
###### Definition 4.3.
Define $`𝒲:\mathrm{}({}_{}{}^{\kappa >}2)\mathrm{}({}_{}{}^{\kappa >}2)`$ by
$$𝒲(S):=\{wS^{}:S\mathrm{Cone}(w{}_{}{}^{}0)\text{ and }S\mathrm{Cone}(w{}_{}{}^{}1)\text{ are }\kappa \text{-dense}\},$$
and call the elements of $`𝒲(S)`$ *densely splitting nodes of $`S`$*.
###### Lemma 4.4.
Suppose $`\kappa ^{<\kappa }=\kappa `$, $`S{}_{}{}^{\kappa >}2`$ and $`(S,<_\text{Q})`$ is a $`\kappa `$-dense linear order. Then $`𝒲(S)`$ is non-empty.
###### Proof.
Assume toward a contradiction that for all $`u{}_{}{}^{\kappa >}2`$, one or both of $`(S\mathrm{Cone}(u{}_{}{}^{}0,<_\text{Q})`$ and $`(S\mathrm{Cone}(u{}_{}{}^{}1),<_\text{Q})`$ have cardinality less than $`\kappa `$. By Lemma 4.1, it follows that for all $`u{}_{}{}^{\kappa >}2`$, one or both of $`S\mathrm{Cone}(u{}_{}{}^{}0)`$ and $`S\mathrm{Cone}(u{}_{}{}^{}1)`$ have cardinality less than $`2`$.
Define $`t_\alpha :\alpha <\kappa `$ by recursion and prove by induction that $`S\mathrm{Cone}(t_\alpha )`$ is $`\kappa `$-dense for all $`\alpha `$.
To start the recursion, let $`t_0=\mathrm{}`$. Then $`S\mathrm{Cone}(t_0)=S`$ which is $`\kappa `$-dense.
If $`\alpha =\beta +1`$ and $`S\mathrm{Cone}(t_\beta )`$ is $`\kappa `$-dense, then let $`t_\alpha =t_\beta {}_{}{}^{}0`$ if $`S\mathrm{Cone}(t_\beta {}_{}{}^{}0)`$ is $`\kappa `$-dense, and $`t_\alpha =t_\beta {}_{}{}^{}1`$ otherwise. Since $`S\mathrm{Cone}(t_\beta )(S\mathrm{Cone}(t_\beta {}_{}{}^{}0))(S\mathrm{Cone}(t_\beta {}_{}{}^{}1))`$, by Lemma 4.1 and a cardinality argument, $`S\mathrm{Cone}(t_\alpha )`$ is $`\kappa `$-dense.
If $`\alpha `$ is a limit ordinal, let $`t_\alpha =\{t_\beta :\beta <\alpha \}`$. By assumption, for all $`\beta <\alpha `$, if $`t_\beta {}_{}{}^{}\delta t_{\beta +1}`$, then $`S\mathrm{Cone}(t_\beta {}_{}{}^{}\delta )`$ is not $`\kappa `$-dense, so has cardinality less than $`2`$. It follows that $`|S\mathrm{Cone}(t_\alpha )|<\kappa `$. Hence by Lemma 4.1, $`S\mathrm{Cone}(t_\alpha )`$ is $`\kappa `$-dense.
Note that $`b=\{t_\alpha :\alpha <\kappa \}`$ is a branch through $`{}_{}{}^{\kappa >}2`$. Every element $`s`$ of $`S`$ is either an initial segment of this branch or there is some $`\beta <\kappa `$ such that $`st_{\beta +1}=t_\beta `$. Recall our assumption that for all $`\beta <\alpha `$, if $`t_\beta {}_{}{}^{}\delta t_{\beta +1}`$, then $`S\mathrm{Cone}(t_\beta {}_{}{}^{}\delta )`$ is not $`\kappa `$-dense, so it has cardinality less than $`2`$. It follows that $`|S|<\kappa `$, contradicting the assumption that $`S`$ is $`\kappa `$-dense. Thus the lemma follows. ∎
###### Lemma 4.5.
Suppose $`\kappa ^{<\kappa }=\kappa `$, $`S{}_{}{}^{\kappa >}2`$ and $`(S,<_\text{Q})`$ is a $`\kappa `$-dense linear order. Then for every $`w𝒲(S)`$, every $`\delta <2`$, and every $`\alpha <\kappa `$, there is $`u𝒲(S)`$ such that $`w{}_{}{}^{}\delta u`$ and $`\mathrm{lg}(u)\alpha `$.
###### Proof.
Fix $`w𝒲(S)`$, $`\delta <2`$ and $`\alpha <\kappa `$. Without loss of generality, assume $`\mathrm{lg}(w{}_{}{}^{}\delta )<\alpha `$. Since $`|S\mathrm{Cone}(w{}_{}{}^{}\delta )|=\kappa `$, it follows that $`S\mathrm{Cone}(w{}_{}{}^{}\delta )`$ is $`\kappa `$-dense by Lemma 4.1.
Since $`\kappa ^{<\kappa }=\kappa `$, the set of nodes of $`{}_{}{}^{\kappa >}2`$ of length at most $`\alpha `$ has cardinality less than $`\kappa `$, but $`S\mathrm{Cone}(w{}_{}{}^{}\delta )`$ has cardinality $`\kappa `$ because it is $`\kappa `$-dense.
By the pigeonhole principle, there is some $`u_0{}_{}{}^{\alpha }2`$ which has $`\kappa `$ many extensions in $`S\mathrm{Cone}(w{}_{}{}^{}\delta )`$. It follows that $`w{}_{}{}^{}\delta u_0`$ and that $`S\mathrm{Cone}(u_0)`$ is $`\kappa `$-dense by Lemma 4.1. By Lemma 4.4 there is an extension $`uu_0`$ in $`𝒲(S)`$, so the lemma follows. ∎
###### Lemma 4.6.
Suppose $`\kappa ^{<\kappa }=\kappa `$, $`S{}_{}{}^{\kappa >}2`$ and $`(S,<_\text{Q})`$ is $`\kappa `$-dense. Then $`(T(S),)`$ is a rooted tree, $`𝒲(S)T(S)`$, and for all $`tT(S)`$, for all $`\alpha >\mathrm{lg}(t)`$, there is some $`r𝒲(S)`$ with $`tr`$ and $`\mathrm{lg}(r)\alpha `$.
###### Proof.
Since $`T(S)`$ is closed under initial segments, $`(T(S),)`$ is a rooted tree. By definition of $`𝒲(S)`$, it is a subset of $`T(S)`$. By Lemma 4.5, every element of $`T`$ has extensions of arbitrarily large length which are densely splitting nodes. ∎
To prove that $`T(S)`$ is an almost perfect tree, we need to be able to prove that it has certain continuity properties at limit levels. Toward that end, we introduce the notion of a *limit of densely splitting nodes of $`S`$*.
###### Definition 4.7.
Suppose $`\kappa `$ is a regular cardinal, $`S{}_{}{}^{\kappa >}2`$, and $`(S,<_\text{Q})`$ is $`\kappa `$-dense. Say $`t{}_{}{}^{\kappa >}2`$ *is a limit of densely splitting nodes of $`S`$* if $`\mathrm{lg}(t)`$ is a limit ordinal and for unboundedly many $`\beta <\mathrm{lg}(t)`$, $`t\beta `$ is in $`𝒲(S)`$. If $`t`$ is a limit of densely splitting nodes of $`S`$, then say it is *evenhanded in $`S`$* if for $`\delta =0,1`$, the set $`\{\beta <\mathrm{lg}(t):t\beta 𝒲(S)t(\beta )=\delta \}`$ is unbounded in $`\mathrm{lg}(t)`$.
###### Lemma 4.8.
Suppose $`\kappa ^{<\kappa }=\kappa `$, $`S{}_{}{}^{\kappa >}2`$ and $`(S,<_\text{Q})`$ is a $`\kappa `$-dense linear order. Further suppose that $`z`$ is a limit of densely splitting nodes of $`S`$. If $`z`$ is evenhanded in $`S`$, then $`z`$ has an extension in $`𝒲(S)`$.
###### Proof.
For each $`\beta <\mathrm{lg}(z)`$ with $`z\beta S`$, let $`g(\beta )`$ be an element of $`S`$ extending $`z\beta {}_{}{}^{}1z(\beta )`$. Let $`A`$ be the set of $`g(\beta )`$ for $`\beta <\mathrm{lg}(z)`$ with $`z\beta S`$ and $`z(\beta )=1`$, and let $`B=\mathrm{ran}(g)A`$.
Then $`A<_\text{Q}\left\{z\right\}<_\text{Q}B`$.
###### Claim 4.8.a.
If $`wS`$ and $`A<_\text{Q}\left\{w\right\}<_\text{Q}B`$, then $`zw`$.
###### Proof.
Suppose $`wS`$ and $`A<_\text{Q}\left\{w\right\}<_\text{Q}B`$.
Assume toward a contradiction that $`\gamma :=\mathrm{lg}(wz)<\mathrm{lg}(z)`$. Use the fact that $`z`$ is evenhanded in $`S`$ to choose $`\eta `$ and $`\theta `$ strictly greater than $`\gamma :=\mathrm{lg}(wz)`$ such that $`z\eta 𝒲(S)`$, $`z(\eta )=1`$, $`z\theta 𝒲(S)`$ and $`z(\theta )=0`$. Then $`g(\eta )A`$ and $`g(\theta )B`$, so $`g(\eta )<_\text{Q}w<_\text{Q}g(\theta )`$.
By definition of $`g`$, $`z\eta g(\eta )`$ and $`z\theta g(\theta )`$. Consequently $`z(\gamma +1)g(\eta )g(\theta )`$, and $`wg(\eta )=wz=wg(\theta )`$. It follows that both $`g(\eta )`$ and $`g(\theta )`$ are in the same $`<_\text{Q}`$-relationship with $`w`$ as is $`z`$, which is not possible because either $`w<_\text{Q}g(\eta )`$ and $`w<_\text{Q}g(\theta )`$ contradicting the fact that $`g(\eta )<_\text{Q}w<_\text{Q}g(\theta )`$. Thus $`\mathrm{lg}(wz)\mathrm{lg}(z)`$. In other words, $`zw`$. ∎
Since $`S`$ is $`\kappa `$-dense, there is a $`\kappa `$-dense set $`CS`$ with $`A<_\text{Q}C<_\text{Q}B`$. By the claim, $`CS\mathrm{Cone}(z)`$. It follows that $`S\mathrm{Cone}(z)`$ is $`\kappa `$-dense, so by Lemma 4.4, $`z`$ has an extension in $`𝒲(S)`$. ∎
###### Definition 4.9.
Suppose $`\kappa ^{<\kappa }=\kappa `$, $`S{}_{}{}^{\kappa >}2`$ and $`(S,<_\text{Q})`$ is $`\kappa `$-dense. Say $`t{}_{}{}^{\kappa >}2`$ *favors $`\delta `$ above $`s`$ in $`T(S)`$* if $`st`$, either $`tT(S)`$ or $`t`$ is a limit of densely splitting nodes of $`S`$, and for all $`r𝒲(S)`$ with $`srt`$, $`r{}_{}{}^{}\delta t`$.
###### Lemma 4.10.
Suppose $`\kappa ^{<\kappa }=\kappa `$, $`S{}_{}{}^{\kappa >}2`$ and $`(S,<_\text{Q})`$ is $`\kappa `$-dense. Further suppose $`u_0<_{\text{lex}}u_1`$, $`u_0`$ favors $`1`$ above $`u_0u_1`$ in $`T(S)`$, and $`u_1`$ favors $`0`$ above $`u_0u_1`$ in $`T(S)`$. Then at least one of $`u_0`$ and $`u_1`$ is in $`T(S)`$.
###### Proof.
If one of $`u_0`$ and $`u_1`$ is not a limit of densely splitting nodes of $`S`$, then it is in $`T(S)`$ by definition of favor. So assume both are limits of densely splitting nodes of $`S`$.
For each $`r𝒲(S)`$ with $`u_0u_1ru_0`$, let $`r^{}S`$ be an extension of $`r{}_{}{}^{}0`$, and let $`A`$ be the set of these elements. Similarly, for each $`r𝒲(S)`$ with $`u_0u_1ru_1`$, let $`r^+S`$ be an extension of $`r{}_{}{}^{}1`$, and let $`B`$ be the set of these elements. Then $`A<_{\text{lex}}\left\{u_0\right\}<_\text{Q}\left\{u_0u_1\right\}<_\text{Q}\left\{u_1\right\}<_{\text{lex}}B`$.
###### Claim 4.10.a.
If $`A<_\text{Q}\left\{w\right\}<_\text{Q}\left\{u_0u_1\right\}`$ and $`wu_0u_1`$, then $`w`$ extends $`(u_0u_1){}_{}{}^{}0`$ and either $`wu_0`$ or $`u_0w`$ or $`u_0<_{\text{lex}}w`$.
###### Proof.
Suppose $`w`$ satisfies the hypotheses. Then by definition of $`<_\text{Q}`$, $`(u_0u_1){}_{}{}^{}0w`$. Assume toward a contradiction that none of the three conclusions holds. Then $`w<_{\text{lex}}u_0`$. Let $`ru_0`$ be an element of $`𝒲(S)`$ with $`u_0u_1r`$ and $`\mathrm{lg}(r)>\mathrm{lg}(uw)+1`$. Then $`w<_{\text{lex}}r`$, so $`w<_{\text{lex}}r^{}`$, contradicting $`A<_\text{Q}\left\{w\right\}`$. ∎
###### Claim 4.10.b.
The set of all $`wS`$ such that $`A<_\text{Q}\left\{w\right\}<_\text{Q}\left\{u_0u_1\right\}`$, and $`u_0<_{\text{lex}}w`$ has cardinality $`<\kappa `$.
###### Proof.
Otherwise, by the pigeonhole principle and the previous claim, there is some $`ru_0`$ with $`(u_0u_1){}_{}{}^{}0r`$ such that the set of $`wS`$ with $`u_0<_{\text{lex}}w`$ and $`u_0w=r`$ has cardinality $`\kappa `$. It follows that $`u_0(\mathrm{lg}(r))=0`$, and, by Lemma 4.1, that $`r{}_{}{}^{}1T(S)`$. Thus $`r𝒲(S)`$ contradicts the assumption that $`u_0`$ favors $`1`$ above $`u_0u_1`$. ∎
The proofs of the next two claims are similar to those above, so they are left to the reader.
###### Claim 4.10.c.
If $`\left\{u_0u_1\right\}<_\text{Q}\left\{w\right\}<_\text{Q}B`$ and $`wu_0u_1`$, then $`w`$ extends $`(u_0u_1){}_{}{}^{}1`$ and either $`wu_1`$ or $`u_1w`$ or $`w<_{\text{lex}}u_1`$.
###### Claim 4.10.d.
The set of all $`wS`$ such that $`A<_\text{Q}\left\{w\right\}<_\text{Q}\left\{u_0u_1\right\}`$, and $`u_0<_{\text{lex}}w`$ has cardinality $`<\kappa `$.
Let $`CS`$ be a $`\kappa `$-dense subset with $`A<_\text{Q}C<_\text{Q}B`$. Since the inequalities $`A<_\text{Q}\left\{u_0u_1\right\}<_\text{Q}B`$ hold, either $`C^{}:=\{cC:c<_\text{Q}u_0u_1\}`$ or $`C^+:=\{cC:u_0u_1<_\text{Q}c\}`$ has cardinality $`\kappa `$. If $`C^{}`$ has cardinality $`\kappa `$, then $`S\mathrm{Cone}(u_0){}_{}{}^{}C_{}^{}`$ has cardinality $`\kappa `$, so by Lemma 4.1, $`u_0`$ is in $`T`$. Similarly, if $`C^+`$ has cardinality $`\kappa `$, then $`u_1`$ is in $`T`$. Thus the lemma follows. ∎
###### Lemma 4.11.
Suppose $`\kappa ^{<\kappa }=\kappa `$, $`S{}_{}{}^{\kappa >}2`$ and $`(S,<_\text{Q})`$ is $`\kappa `$-dense. For all $`tT(S)`$ there is a minimal extension of $`t`$ in $`𝒲(S)`$. That is, there is $`r𝒲(S)`$ such that $`tr`$ and $`rw`$ for all $`w𝒲(S)\mathrm{Cone}(t)`$.
###### Proof.
By Lemma 4.1, $`𝒲(S)\mathrm{Cone}(t)`$ is non-empty. By definition of $`𝒲(S)`$, the meet of two incomparable elements of it is also in the set. It follows that $`𝒲(S)\mathrm{Cone}(t)`$ has an element of minimum length, and this element is the desired minimal extension. ∎
###### Lemma 4.12.
Suppose $`\kappa ^{<\kappa }=\kappa `$, $`S{}_{}{}^{\kappa >}2`$ and $`(S,<_\text{Q})`$ is $`\kappa `$-dense. For all $`x𝒲(S)`$ and all $`\alpha >\mathrm{lg}(x)`$, there is some extension $`y`$ of $`x`$ with $`\mathrm{lg}(y)=\alpha `$ such that $`y`$ favors one of $`0`$, $`1`$ above $`x`$.
###### Proof.
Fix $`x`$ in $`𝒲(S)`$.
For as long as possible, define a $``$-increasing sequence $`u_\alpha `$ of extensions of $`x{}_{}{}^{}1`$ which favor $`0`$ above $`x`$, with $`\mathrm{lg}(u_\alpha )=\alpha `$. To start the recursions with $`\alpha =\mathrm{lg}(x)+1`$, let $`u_\alpha =x{}_{}{}^{}1T(S)`$. If $`\alpha `$ is a limit ordinal and $`u_\beta `$ has been defined for $`\mathrm{lg}(x)<\beta <\alpha `$, then let $`u_\alpha =\{u_\beta :\mathrm{lg}(x)<\beta <\alpha \}`$. If $`\alpha =\beta +1`$, $`u_\beta `$ has been defined, and $`u_\beta T(S)`$, then let $`u_\alpha =u_\beta {}_{}{}^{}0`$ if $`u_\beta 𝒲(S)`$, and otherwise let $`u_\alpha `$ be the one-point extension of $`u_\beta `$ which is a subset of the extension of $`u_\beta `$ of minimal length in $`𝒲(S)`$. If $`\alpha =\beta +1`$, $`u_\beta `$ has been defined, and $`u_\beta T(S)`$, then $`\beta `$ is a limit ordinal, $`u_\beta `$ is a limit of densely splitting nodes of $`S`$ and the recursion stops with its definition.
Also, for as long as possible, define a $``$-increasing sequence $`v_\alpha `$ of extensions of $`x{}_{}{}^{}0`$ which favor $`1`$ above $`x`$, with $`\mathrm{lg}(v_\alpha )=\alpha `$. To start the recursion with $`\alpha =\mathrm{lg}(x)+1`$, let $`v_\alpha =x{}_{}{}^{}0T(S)`$. If $`\alpha `$ is a limit ordinal and $`v_\beta `$ has been defined for $`\mathrm{lg}(x)<\beta <\alpha `$, then let $`v_\alpha =\{v_\beta :\mathrm{lg}(x)<\beta <\alpha \}`$. If $`\alpha =\beta +1`$, $`v_\beta `$ has been defined, and $`v_\beta T(S)`$, then let $`v_\alpha =v_\beta {}_{}{}^{}0`$ if $`v_\beta 𝒲(S)`$, and otherwise let $`v_\alpha `$ be the one-point extension of $`v_\beta `$ which is a subset of the extension of $`v_\beta `$ of minimal length in $`𝒲(S)`$. If $`\alpha =\beta +1`$, $`v_\beta `$ has been defined, and $`v_\beta T(S)`$, then $`\beta `$ is a limit ordinal, $`v_\beta `$ is a limit of densely splitting nodes of $`S`$ and the recursion stops with its definition.
By construction, for all $`\alpha `$ with $`u_\alpha `$ defined, $`u_\alpha `$ favors $`0`$ above $`x`$ in $`T(S)`$. Similarly, for all $`\alpha `$ with $`v_\alpha `$ defined, $`v_\alpha `$ favors $`1`$ above $`x`$ in $`T(S)`$. If one of the recursions continues for all $`\alpha <\kappa `$, the lemma follows.
Assume toward a contradiction that $`u_\beta `$ is defined but $`u_{\beta +1}`$ is not, and that $`v_\beta `$ is defined but $`v_{\beta +1}`$ is not. Then $`u_\beta T(S)`$ and $`v_\beta T(S)`$, contradicting Lemma 4.10. ∎
###### Definition 4.13.
A subset $`T{}_{}{}^{\kappa >}2`$ is an *almost perfect tree* if it is a rooted induced subtree of $`{}_{}{}^{\kappa >}2`$ closed under initial segments such that the following conditions hold:
1. for all $`tT`$, for all $`\alpha <\kappa `$, there is an extension $`wT`$ of $`t`$ of length at least $`\alpha `$ which is a densely splitting node (both $`w{}_{}{}^{}0`$ and $`w{}_{}{}^{}1`$ are in $`T`$);
2. for all $`s{}_{}{}^{\kappa >}2`$, if $`s`$ is an evenhanded limit of densely splitting nodes of $`T`$, then $`s`$ is in $`T`$;
3. for all $`s,t{}_{}{}^{\kappa >}2`$, if $`\mathrm{lg}(s)=\mathrm{lg}(t)`$, both $`s`$ and $`t`$ are limits of densely splitting nodes of $`T`$, and for some $`x`$, $`s`$ favors $`0`$ above $`x`$ and $`t`$ favors $`1`$ above $`x`$, then one of $`s`$ and $`t`$ is in $`T`$.
###### Lemma 4.14.
Suppose $`\kappa ^{<\kappa }=\kappa `$, $`S{}_{}{}^{\kappa >}2`$ and $`(S,<_\text{Q})`$ is $`\kappa `$-dense. Then $`T(S)`$ is almost perfect.
###### Proof.
Apply Lemmas 4.6, 4.8 and 4.12. ∎
###### Lemma 4.15.
Suppose $`\kappa ^{<\kappa }=\kappa `$ and $`T{}_{}{}^{\kappa >}2`$ is almost perfect. Let $`C`$ be the set of all limit ordinals $`\alpha >0`$ such that every node of $`T`$ of length less than $`\alpha `$ extends to a densely splitting node of $`T`$ of length less than $`\alpha `$. Then $`C`$ is closed unbounded in $`\kappa `$, and for all $`\alpha C`$, for all $`uT{}_{}{}^{\alpha >}2`$, the set $`T_\alpha (u):=\{tT:ut\mathrm{lg}(t)=\alpha \}`$ has cardinality at least $`2^{\mathrm{cf}(\alpha )}`$.
###### Proof.
Use the definition of almost perfect to show that $`C`$ is non-empty and unbounded. It follows immediately from the definition of $`C`$ that it is closed.
Fix attention on $`\alpha C`$ and $`uT{}_{}{}^{\alpha >}2`$. Let $`\lambda =\mathrm{cf}(\alpha )`$, and suppose $`\sigma {}_{}{}^{\lambda }2`$ is a sequence such that if $`\eta =\theta +k<\lambda `$ for $`\theta =0`$ or $`\theta `$ limit, and $`k\delta `$ mod $`3`$ for $`\delta <2`$, then $`\sigma (\eta )=\delta `$.
Define $`r_\sigma (\eta ):\eta <\lambda `$ by recursion and show by induction that every element of it has length $`<\alpha `$.
To start the recursion, let $`r_\sigma (0)`$ be the minimal densely splitting node of $`T`$ extending $`u`$. It has length less than $`\alpha `$ by the definition of $`C`$.
Suppose $`0<\eta <\lambda `$ and $`r_\sigma `$ has been defined on elements smaller than $`\eta `$. If $`\eta =\zeta +1`$, let $`r_\sigma (\eta )`$ be a densely splitting node of $`T`$ which extends $`r_\sigma (\zeta ){}_{}{}^{}\sigma (\zeta )`$ of length less than $`\alpha `$. Such a node exists by the definition of $`C`$.
If $`\eta `$ is a limit ordinal, then $`r_\sigma ^{}(\eta ):=\{r_\sigma (\zeta ):\zeta <\eta \}`$ is a limit of densely splitting nodes of $`T`$ and has length a limit ordinal less than $`\alpha `$ since $`\eta <\lambda =\mathrm{cf}(\alpha )`$. The properties of $`\sigma `$ guarantee that this union is evenhanded, so by the definition of an almost perfect tree, $`r_\sigma ^{}(\eta )`$ is in $`T`$. Let $`r_\sigma (\eta )`$ be a densely splitting node of length less than $`\alpha `$ extending $`r_\sigma ^{}(\eta )`$, which exists by definition of $`C`$.
This completes the definition of $`r_\sigma (\eta ):\eta <\lambda `$. Let $`s_\sigma `$ be the union of this sequence. Then $`s_\sigma `$ is a limit of densely splitting nodes in $`S`$ and is evenhanded, so it is an element of $`T`$. Since $`r_\sigma (0)`$ is an extension of $`u`$, so is $`s_\sigma `$.
If $`s_\sigma `$ has length $`\alpha `$, then let $`t_\sigma =s_\sigma `$ be this union, and notice that it is in $`T`$. Otherwise let $`t_\sigma `$ be an extension of $`s_\sigma `$ of length $`\alpha `$ in $`T`$, which must exist by the definition of an almost perfect tree.
Notice that if $`\sigma ,\tau {}_{}{}^{\lambda }2`$ are two distinct sequences with the property that $`\eta =\theta +k<\lambda `$ for $`\theta =0`$ or $`\theta `$ limit and $`k\delta `$ mod $`3`$ for $`\delta <2`$ implies $`\sigma (\eta )=\tau (\eta )=\delta `$, then $`t_\sigma t_\tau `$.
Since no constraints have been placed on $`\sigma (\theta +k)`$ for $`k2`$ mod $`3`$, there are $`2^\lambda `$ sequences $`\sigma {}_{}{}^{\lambda }2`$ with the special property described above. Thus the set $`T_\alpha (u)`$ has cardinality at least $`2^\lambda `$, and the lemma follows. ∎
###### Lemma 4.16.
Suppose $`\kappa ^{<\kappa }=\kappa `$, $`S{}_{}{}^{\kappa >}2`$ and $`(S,<_\text{Q})`$ is $`\kappa `$-dense. Let $`C(S)`$ be the set of all limit ordinals $`\alpha >0`$ such that every $`tT(S){}_{}{}^{\alpha >}2`$ has proper extensions in both $`S{}_{}{}^{\alpha >}2`$ and $`𝒲(S){}_{}{}^{\alpha >}2`$. Then $`C(S)`$ is closed unbounded in $`\kappa `$.
###### Proof.
Use the definition of $`T`$ and Lemmas 4.1 and 4.4 to show that $`C(S)`$ is non-empty and unbounded. It follows immediately from the definition of $`C(S)`$ that it is closed. ∎
###### Lemma 4.17.
Suppose that $`\kappa `$ is an inaccessible limit of inaccessible cardinals and $`S`$ is a subset of $`{}_{}{}^{\kappa >}2`$ with $`(S,<_\text{Q})`$ $`\kappa `$-dense. Then there is a diagonal set $`DS`$ such that $`(D,<_\text{Q})`$ is $`\kappa `$-dense.
###### Proof.
Let $``$ be a total order on $`{}_{}{}^{\kappa >}2`$ satisfying $`\mathrm{lg}(s)<\mathrm{lg}(t)st`$. Define $`r:T(S)𝒲(S)`$ by setting $`r(t)`$ to be the minimal extension of $`t`$ in $`𝒲(S)`$. By Lemma 4.11, $`r`$ is well-defined. Let $`s:T(S)S`$ be such that $`s(t)`$ is an extension of $`t`$.
By recursion on $``$, define functions $`\mathrm{}:{}_{}{}^{\kappa >}2\kappa `$, $`f_0,f_1:{}_{}{}^{\kappa >}2𝒲(S)`$, and $`g:{}_{}{}^{\kappa >}2S`$ as follows.
To start the recursion, note that the $``$-least element is $`\mathrm{}`$, define $`\mathrm{}(\mathrm{}):=0`$, and let $`f_0(\mathrm{}):=r(\mathrm{})`$. If $`f_0(z)`$ has been defined, let $`f_1(z):=r(f_0(z){}_{}{}^{}1)`$, and set $`g(z):=s(f_1(z){}_{}{}^{}0)`$.
To continue the recursion, suppose $`z{}_{}{}^{\kappa >}2`$ has $`\mathrm{lg}(z)=\alpha `$ and for all $`xz`$, both $`\mathrm{}(x)`$ and $`f_0(x)`$ have been defined, with $`f_1`$ and $`g`$ defined from them as above. Let $`\mathrm{}(z)`$ be the least ordinal greater than $`\mathrm{lg}(g(x))`$ for all $`xz`$.
If $`\alpha =\beta +1`$ is a successor and $`z=y{}_{}{}^{}\delta `$ for some $`y`$, then let $`f_0(z)`$ be an extension in $`𝒲(S)`$ of $`r(f_\delta (y){}_{}{}^{}\delta ){}_{}{}^{}1\delta `$ of length greater than $`\mathrm{}(z)`$. Since $`T(S)`$ is almost perfect and $`𝒲(S)`$ is the set of densely splitting nodes of $`T(S)`$, such a node exists.
If $`\alpha >0`$ is a limit ordinal, then $`f^{}(z):=\{f(z\beta ):\beta <\alpha \}`$ has length a limit ordinal. Since $`f^{}(z)`$ extends $`f_0(z\eta )`$ for all $`\eta <\alpha `$, the definition of $`f_0`$ on nodes of successor length guarantees that $`f^{}(z)`$ is a limit of densely splitting nodes of $`S`$ and evenhanded. Hence by Lemma 4.8, it has an extension in $`𝒲(S)`$. It follows that $`f^{}(z)`$ is in $`T(S)`$. Let $`f_0(z)`$ be an extension of $`f^{}(z)`$ in $`𝒲(S)`$ of length greater than $`\mathrm{}(z)`$.
Let $`D`$ be the range of $`g`$. Then $`D`$ is a subset of $`S`$, and the meet closure of $`D`$ is a subset of union of the ranges of $`f_0`$, $`f_1`$ and $`g`$. Since $`f_0(x)f_1(x)g(x)`$, the lengths of these sequences are strictly increasing. Also, by construction, if $`xz`$, then $`\mathrm{lg}(g(x))<\mathrm{lg}(f_0(z))`$. So different elements of the meet closure of $`D`$ have different lengths.
By induction, one can show that $`f_0`$ preserves $``$ and length order. By definition of $`f_0`$, $`f_1`$ and $`g`$, we have the following properties:
1. if $`x`$ and $`y`$ are incomparable with $`x<_{\text{lex}}y`$, then $`g(x)g(y)=f_0(xy)`$ and $`g(x)<_{\text{lex}}g(y)`$;
2. if $`x{}_{}{}^{}0y`$, then $`g(x)g(y)=f_0(x)`$ and $`g(y)<_{\text{lex}}g(x)`$;
3. if $`x{}_{}{}^{}1y`$, then $`g(x)g(y)=f_1(x)`$ and $`g(x)<_{\text{lex}}g(y)`$.
It follows that any two elements of $`D`$ are incomparable, that is, $`D`$ is an antichain. Hence $`D`$ is diagonal. By construction, $`D`$ is a subset of $`S`$. By the above three properties, $`g`$ preserves $`<_\text{Q}`$, so $`D`$ is $`\kappa `$-dense. ∎
## 5 Lower bound for dense linear orders
In this section, we show that all sparse vip $`m`$-types can be embedded in any sparse diagonal set $`D`$ with $`(D,<_\text{Q})`$ $`\kappa `$-dense, and derive a lower bound result for $`_\kappa `$.
###### Definition 5.1.
Call an ordering $``$ of the levels of $`{}_{}{}^{\kappa >}2`$ *small* if $`({}_{}{}^{\alpha }2,)`$ has order type $`2^\alpha `$ for each cardinal $`\alpha <\kappa `$.
###### Lemma 5.2.
Suppose $``$ is a small ordering of the levels of $`{}_{}{}^{\kappa >}2`$ and $`\alpha <\kappa `$ is a cardinal. For all $`s`$ with $`\mathrm{lg}(s)<\alpha `$, the interval $`\{t{}_{}{}^{\alpha }2:st\}`$ is cofinal in $`({}_{}{}^{\alpha }2,)`$.
###### Proof.
For all $`s`$ with $`\mathrm{lg}(s)<\alpha `$, the interval $`\{t{}_{}{}^{\alpha }2:st\}`$ has cardinality $`2^\alpha `$. Hence such an interval has order type $`2^\alpha `$ under $``$. ∎
###### Lemma 5.3.
Suppose $`\kappa `$ is a limit cardinal, $`^{}`$ is a small ordering of the levels of $`{}_{}{}^{\kappa >}2`$ and $`w{}_{}{}^{\kappa >}2`$. For all $`n<\omega `$ and orderings $``$ of the levels of $`{}_{}{}^{n}2`$, there is an order preserving strong embedding $`j`$ of $`({}_{}{}^{n}2,)`$ into $`(\mathrm{Cone}(w),^{})`$, i.e. $`st`$ implies $`j(s)j(t)`$. Furthermore, $`j`$ may be chosen such that for all $`s`$, the length of $`j(s)`$ is a cardinal.
###### Proof.
Use induction on $`n`$. For $`n=0`$, there is only the empty sequence, and any embedding of this single point to a point $`\mathrm{Cone}(w)`$ whose length is a cardinal works.
Suppose the lemma is true for $`m`$ and $``$ is an ordering of the levels of $`{}_{}{}^{n}2`$ for $`n=m+1`$. Let $`j_0:{}_{}{}^{m}2{}_{}{}^{\kappa >}2`$ be an order preserving strong embedding obtained from the induction hypothesis. We define an extension $`j`$ of $`j_0`$ as follows. Let $`s_i:i<2^{m+1}`$ enumerate in increasing $``$-order the nodes of $`{}_{}{}^{m+1}2`$. Pick a cardinal $`\alpha `$ larger than the level at which $`{}_{}{}^{m}2`$ is embedded, and define by recursion on $`i`$ nodes $`j(s_i)=t_i`$ in $`{}_{}{}^{\alpha }2`$ such that $`j(s_im){}_{}{}^{}s_i(m)t_i`$ and $`t_it_{i1}`$ if $`i>0`$. Lemma 5.2 guarantees that this recursion is possible. ∎
###### Theorem 5.4.
Suppose that $`\kappa `$ is a cardinal which is measurable in the generic extension obtained by adding $`\lambda `$ Cohen subsets of $`\kappa `$, where $`\lambda (\kappa )_{2^\kappa }^6`$. Further suppose $``$ is well ordering of the levels of $`{}_{}{}^{\kappa >}2`$ and $`^{}`$ is a small ordering of the levels of $`{}_{}{}^{\kappa >}2`$. Then there is a strong embedding $`e`$ and a node $`w`$ such that $`e`$ preserves $`^{}`$ and $``$ and $`^{}`$ agree on all pairs in $`e[\mathrm{Cone}(w)]`$.
###### Proof.
Apply Shelah’s Theorem 2.5 to the coloring $`d:[{}_{}{}^{\kappa >}2]^22`$ defined using the Boolean value operator $``$ by
$$d(\{s,t\})=s<_{\text{lex}}tst$$
to get a strong embedding $`e:{}_{}{}^{\kappa >}2{}_{}{}^{\kappa >}2`$ and an element $`w{}_{}{}^{\kappa >}2`$ such that for all $`s,t\mathrm{Cone}(w)`$ with $`s^{}t`$ one has $`e(s)^{}e(t)`$ and the color of $`d(\{s,t\})`$ depends only on the $`^{}`$-ordered similarity type of the pair $`\{s,t\}`$.
If $`\mathrm{lg}(s)<\mathrm{lg}(t)`$, then $`\mathrm{lg}(e(s))<\mathrm{lg}(e(t))`$, so $``$ and $`^{}`$ agree on $`e(s),e(t)`$, since both are level orders. Similarly, if $`\mathrm{lg}(s)>\mathrm{lg}(t)`$, then $``$ and $`^{}`$ agree on $`e(s),e(t)`$.
Next consider pairs $`\{s,t\}`$ with $`\mathrm{lg}(s)=\mathrm{lg}(t)`$, $`s<_{\text{lex}}t`$ and $`s^{}t`$. Note that since Cohen forcing on $`\kappa `$ does not add bounded subsets to $`\kappa `$, our assumptions in particular imply that $`\kappa `$ must be a limit cardinal. Suppose $`\alpha `$ is a cardinal larger than $`\mathrm{lg}(w)`$ but $`<\kappa `$. Then $`e[{}_{}{}^{\alpha }2\mathrm{Cone}(w)]{}_{}{}^{\gamma }2`$ is infinite for some $`\gamma <\kappa `$. Let $`s_n{}_{}{}^{\alpha }2\mathrm{Cone}(w):n<\omega `$ be a sequence which is increasing in both the $`<_{\text{lex}}`$ and $`^{}`$ orders. Since $`e`$ is a strong embedding, the sequence $`e(s_n){}_{}{}^{\gamma }2:n<\omega `$ is $`<_{\text{lex}}`$-increasing, and cannot be decreasing in $``$. Since $`s^{}t`$ implies $`e(s)^{}e(t)`$, it follows that $`^{}`$ and $``$ agree on pairs $`\{e(s),e(t)\}`$ with $`\mathrm{lg}(s)=\mathrm{lg}(t)`$, $`s<_{\text{lex}}t`$ and $`s^{}t`$.
Finally consider pairs $`\{s,t\}`$ with $`\mathrm{lg}(s)=\mathrm{lg}(t)`$, $`t<_{\text{lex}}s`$ and $`s^{}t`$. For $`\alpha `$ as in the previous case, let $`t_n{}_{}{}^{\alpha }2\mathrm{Cone}(w):n<\omega `$ be a sequence which is decreasing in the $`<_{\text{lex}}`$ order and increasing in the $`^{}`$ order. Then $`e(s_n){}_{}{}^{\gamma }2:n<\omega `$ is $`<_{\text{lex}}`$-decreasing, and cannot be decreasing in $``$. Thus by an argument like that above, $`^{}`$ and $``$ agree on pairs $`\{e(s),e(t)\}`$ with $`\mathrm{lg}(s)=\mathrm{lg}(t)`$, $`t<_{\text{lex}}s`$ and $`s^{}t`$. ∎
For the following definition, recall the notation $`T(D)`$ from Definition 4.2.
###### Definition 5.5.
Suppose $`D{}_{}{}^{\kappa >}2`$ is diagonal and $`(D,<_\text{Q})`$ is $`\kappa `$-dense. A function $`f:{}_{}{}^{\kappa >}2T(D)`$ is a *semi-strong embedding* if $`f`$ preserves extension and lexicographic order, maps levels to levels, and for every $`s{}_{}{}^{\kappa >}2`$, there is some $`vD`$ such that
$$f(s)f(s{}_{}{}^{}0)f(s{}_{}{}^{}1)v$$
and $`\mathrm{lg}(v)<\mathrm{lg}(f(s{}_{}{}^{}0))`$.
###### Lemma 5.6.
Suppose $`\kappa `$ is inaccessible and $`D{}_{}{}^{\kappa >}2`$ is diagonal and $`(D,<_\text{Q})`$ is $`\kappa `$-dense. Then there is a semi-strong embedding $`f:{}_{}{}^{\kappa >}2T(D)`$.
###### Proof.
Let $`C\kappa `$ be the set of all limit $`\alpha >0`$ such that for all $`tT(D){}_{}{}^{\alpha }2`$, $`t`$ is a limit of densely splitting points of $`T(D)`$ and for all $`\beta <\alpha `$, there is $`vT(D){}_{}{}^{\alpha }2`$ such that $`t\beta vD`$. By Lemma 4.16, $`C`$ is closed unbounded.
Define $`f`$ on $`{}_{}{}^{\alpha }2`$ by recursion on $`\alpha <\kappa `$, using the assumption that $`\kappa `$ is inaccessible. To start the recursion, let $`f(\mathrm{})`$ be an element of $`T(D)`$ on the $`\gamma _0`$ level of $`{}_{}{}^{\kappa >}2`$, where $`\gamma _0`$ is the least infinite cardinal in $`C`$.
If $`\alpha `$ is a limit and $`f`$ has been defined on $`{}_{}{}^{\alpha >}2`$, then let $`\gamma _\alpha `$ be a cardinal in $`C`$ greater than $`sup_{\beta <\alpha }\gamma _\beta `$, and for each $`s{}_{}{}^{\alpha }2`$, let $`f(s)`$ be an element of $`T(D){}_{}{}^{\gamma _\alpha }2`$ which is an evenhanded limit of densely splitting points of $`T(D)`$ extending $`\{f(s\beta ):\beta <\alpha \}`$.
Suppose $`\alpha =\alpha ^{}+1`$ is a successor and $`f`$ has been defined on $`{}_{}{}^{\alpha >}2`$. Let $`\gamma _\alpha `$ be a cardinal in $`C`$ greater than $`\gamma _\alpha ^{}`$. Recall that by Lemma 4.11 every node of $`T(D)`$ has a minimal extension to a densely splitting node (one in $`𝒲(S)`$). Also, by Lemma 4.8, every evenhanded extension of a node of $`T(D)`$ has an extension to a densely splitting node. For $`s=t{}_{}{}^{}\delta `$ and $`u_t`$ the minimal length densely splitting node properly extending $`t`$, let $`f(s)`$ be an evenhanded extension of $`u_t{}_{}{}^{}\delta `$ of length $`\gamma _\alpha `$.
Use induction to show that $`f`$ preserves extension and lexicographic order. By its construction and the choice of $`C`$, the remaining conditions are satisfied for all $`s`$. ∎
For the following definition, recall the notion of a sparse $`m`$-type from Definition 3.6.
###### Theorem 5.7.
Suppose that $`\kappa `$ is a cardinal which is measurable in the generic extension obtained by adding $`\lambda `$ Cohen subsets of $`\kappa `$, where $`\lambda (\kappa )_{2^\kappa }^6`$. If $`D{}_{}{}^{\kappa }2`$ is a sparse diagonal set with $`(D,<_\text{Q})`$ $`\kappa `$-dense and $``$ is a $`D`$-vip order of the levels of $`{}_{}{}^{\kappa >}2`$, then every sparse vip $`m`$-type $`(\tau ,)`$ is realized as $`(\mathrm{clp}(x),_x)`$ for some $`xD`$.
###### Proof.
The assumptions imply that $`\kappa `$ is inaccessible, so Lemma 5.6 applies. Let $`f:{}_{}{}^{\kappa >}2T(D)`$ be a semi-strong embedding from Lemma 5.6. Define $`f^0:{}_{}{}^{\kappa >}2D^{}`$ by $`f^0(s):=f(s{}_{}{}^{}0)f(s{}_{}{}^{}1)`$. Let $`f^1:{}_{}{}^{\kappa >}2D`$ be defined by $`f^1(s)=v`$ where $`v`$ is the minimal extension of $`f^0(s)`$ in $`D`$ as guaranteed by Definition 5.5.
For $`t{}_{}{}^{\alpha }2`$ and $`i=0,1`$, define well-orderings $`_t^i`$ on $`{}_{}{}^{\alpha }2`$ as follows: let $`\beta _i=\mathrm{lg}(f^i(t))`$ and set $`s_t^is^{}`$ if and only if $`f(s{}_{}{}^{}0)\beta _if(s^{}{}_{}{}^{}0)\beta _i`$.
Let $`^{}`$ be any small well-ordering of the levels of $`{}_{}{}^{\kappa >}2`$. Call a triple $`\{s,s^{},t\}`$ *local* if $`\mathrm{lg}(s)=\mathrm{lg}(s^{})=\mathrm{lg}(t)`$, $`s<_{\text{lex}}s^{}`$, $`t^{}s`$, and $`t^{}s^{}`$, and $`ss^{}t`$. Let $`d`$ be a coloring of the triples of $`{}_{}{}^{\kappa >}2`$ defined as follows: if $`\{s,s^{},t\}`$ is not local, let $`d(\{s,s^{},t\}):=(2,2)`$ and otherwise set
$$d(\{s,s^{},t\}):=(s^{}s^{}s_t^0s^{},s^{}s^{}s_t^1s^{}).$$
Apply Shelah’s Theorem 2.5 to $`d`$ and $`^{}`$ to obtain a strong embedding $`e:{}_{}{}^{\kappa >}2{}_{}{}^{\kappa >}2`$ and a node $`w`$ such that for triples from $`T:=e[\mathrm{Cone}(w)]`$, the coloring depends only on the $`^{}`$-ordered similarity type of the triple. Then two local triples $`\{s,s^{},t\}`$ and $`\{u,u^{},v\}`$ of $`T`$ are colored the same if and only if
$$s_t^0s^{}u_v^0u^{}\text{and}s_t^1s^{}u_v^1u^{}.$$
Hence for $`tT`$ and for both ordered similarity types of incomparable pairs of same length nodes from $`T`$, the orderings $`_t^\delta `$ must always agree with one of $`^{}`$ and its converse on $`T`$. Since $``$ is a well-ordering of the levels of $`{}_{}{}^{\kappa >}2`$, the orderings $`_t^i`$ are well-orderings of $`{}_{}{}^{\alpha }2`$, so they must agree with $`^{}`$ in both cases.
Assume that $`\tau `$ has $`n+1`$ leaves and let $`L`$ be the set of these leaves. Then $`\tau `$ is a subtree of $`{}_{}{}^{2n}2`$ and every level of $`{}_{}{}^{2n}2`$ has exactly one element of $`L^{}`$. Extend $``$ defined on $`\tau `$ to $`^{}`$ defined on all of $`{}_{}{}^{2n}2`$ in such a way that the extension is still a $`L^{}`$-vip order.
Apply Lemma 5.3 to get an order preserving strong embedding $`j`$ of $`({}_{}{}^{2n}2,^{})`$ into $`({}_{}{}^{\kappa >}2,^{})`$.
Let $`t_{\mathrm{}}:\mathrm{}2n`$ enumerate the elements of $`L^{}`$ in increasing order of length. Note that $`\mathrm{lg}(t_{\mathrm{}})=\mathrm{}`$. For $`\mathrm{}2n`$, define $`\beta _{\mathrm{}}:=\mathrm{lg}(f^i(e(j(t_{\mathrm{}}))))`$ where $`i=0`$ if $`t_{\mathrm{}}L`$ and $`i=1`$ if $`t_{\mathrm{}}L`$.
Finally define $`\sigma :\tau T(D)`$ by recursion on $`\mathrm{}2n`$. For $`\mathrm{}=0`$, let $`\sigma (\mathrm{})=f^0(e(j(\mathrm{})))`$. For $`\mathrm{}>0`$, consider three cases for elements of $`\tau {}_{}{}^{\mathrm{}}2`$. If $`t_{\mathrm{}}L`$, let $`\sigma (t_{\mathrm{}})=f^1(e(j(t_{\mathrm{}})))`$. If $`t_{\mathrm{}}L`$, let $`\sigma (t_{\mathrm{}})=f^0(e(j(t_{\mathrm{}})))`$. Note that in both of these cases, $`\beta _{\mathrm{}}=\mathrm{lg}(\sigma (t_{\mathrm{}}))`$. If $`s\tau L^{}`$ has length $`\mathrm{}`$, then there is a unique immediate successor in $`\tau `$, $`s{}_{}{}^{}0`$. In this case, let $`\sigma (s)=f^0(e(j(s)){}_{}{}^{}0)\beta _{\mathrm{}}`$. Since $`j`$ sends $`^{}`$-increasing pairs to $`^{}`$-increasing pairs and $`e`$ is a $`^{}`$ order preserving strong embedding, their composition sends sends $`^{}`$-increasing pairs to $`^{}`$-increasing pairs. Since for $`v_{\mathrm{}}=e(j(t_{\mathrm{}}))T`$, the order $`^{}`$ agrees with $`_v_{\mathrm{}}^i`$ on $`T{}_{}{}^{\gamma }2`$ where $`\gamma =\mathrm{lg}(v_{\mathrm{}})`$, it follows that $`\sigma `$ sends $`^{}`$-increasing pairs to $``$-increasing pairs. Since $`f`$ preserves extension and lexicographic order, $`\sigma `$ does as well. By construction $`\sigma `$ sends levels to levels, meets to meets (split nodes) and leaves to leaves (terminal nodes). Let $`x=\sigma [L]`$ be the image under $`\sigma `$ of the leaves of $`\tau `$. Then $`(\mathrm{clp}(x),_x)=(\tau ,)`$, as required. ∎
###### Theorem 5.8.
Let $`m`$ be a natural number and suppose that $`\kappa `$ is a cardinal which is measurable in the generic extension obtained by adding $`\lambda `$ Cohen subsets of $`\kappa `$, where $`\lambda (\kappa )_{2^\kappa }^{2m}`$. Then for $`r=t_m^+`$ equal to the number of sparse vip $`m`$-types, the $`\kappa `$-dense linear order $`_\kappa `$ satisfies
$$_\kappa (_\kappa )_{<\omega ,r1}^m.$$
###### Proof.
Let $`S{}_{}{}^{\kappa >}2`$ be a cofinal transverse subset obtain from Lemma 3.10. Let $``$ be a pre-$`S`$-vip order on $`{}_{}{}^{\kappa >}2`$ obtained from Lemma 3.5.
Let $`\phi :{}_{}{}^{\kappa >}2S`$ be a $`<_\text{Q}`$-preserving injection such that $`D:=\phi [{}_{}{}^{\kappa >}2]`$ is a sparse diagonal set with $`D^{}S`$. Notice that $`(D,<_\text{Q})`$ is $`\kappa `$-dense, since $`\phi `$ is $`<_\text{Q}`$-preserving.
Let $`(\tau _0,_0)`$, …, $`(\tau _{r1},_{r1})`$ be an enumeration of the sparse vip $`m`$-types. Define $`c:[{}_{}{}^{\kappa >}2]^mr`$ by $`c(a)=i`$ where for $`x:=\phi [a]`$, $`(\mathrm{clp}(x),_x)=(\tau _i,_i)`$.
Suppose $`A{}_{}{}^{\kappa >}2`$ is a subset with $`(A,<_\text{Q})`$ $`\kappa `$-dense and $`i<r`$. Since $`\phi `$ is $`<_\text{Q}`$-preserving, its image $`B:=\phi [A]D`$ is a sparse diagonal set with the property that $`(B,<_\text{Q})`$ is $`\kappa `$-dense. Thus by Theorem 5.7, the sparse vip $`m`$-type $`(\tau _i,_i)`$ is realized as $`(\mathrm{clp}(x),_x)`$ for some $`xB`$. Since $`\phi `$ is injective, there is an $`m`$-element subset $`uA`$ with $`\phi [u]=x`$ and $`c(u)=i`$. Since $`A`$ and $`i`$ were arbitrary, in every $`\kappa `$-dense subset $`A{}_{}{}^{\kappa >}2`$, every color $`i`$ is realized by some $`m`$-element subset. Therefore, the theorem follows. ∎
Recall the definition of canonical partition introduced immediately after the statement of Theorem 1.1.
###### Theorem 5.9.
Let $`m`$ be a natural number and suppose that $`\kappa `$ is a cardinal which is measurable in the generic extension obtained by adding $`\lambda `$ Cohen subsets of $`\kappa `$, where $`\lambda (\kappa )_{2^\kappa }^{2m}`$. For $`t_m^+`$ equal to the number of sparse vip $`m`$-types, there is a canonical partition of the $`m`$-element subsets of $`_\kappa =([{}_{}{}^{\kappa >}2]^m,<_\text{Q})`$ into $`t_m^+`$ parts.
###### Proof.
Use Lemma 3.10 to find $`S{}_{}{}^{\kappa >}2`$ cofinal and transverse. Use Lemma 3.5 to find $``$ a pre-$`S`$-vip order on $`{}_{}{}^{\kappa >}2`$.
For all $`w{}_{}{}^{\kappa >}2`$, let $`\phi _w`$ be a sparse diagonalization of $`{}_{}{}^{\kappa >}2`$ into $`S\mathrm{Cone}(w)`$. Use recursion on $``$ to define $`\pi :{}_{}{}^{\kappa >}2{}_{}{}^{\kappa >}2`$ such that for all $`t{}_{}{}^{\kappa >}2`$, $`\pi (t)`$ is an extension of $`t`$ with $`\mathrm{Cone}(\pi (t))`$ disjoint from the union over all $`st`$ of $`\phi _{\pi (s)}[{}_{}{}^{\kappa >}2]`$. Since the order type of $`\{s{}_{}{}^{\kappa >}2:st\}`$ is less than $`\kappa `$ and each $`\phi _{\pi (s)}[{}_{}{}^{\kappa >}2]`$ is a sparse diagonal subset of $`S`$, it is always possible to continue the recursion.
Define $`h:{}_{}{}^{\kappa >}2{}_{}{}^{\kappa >}2`$ as follows. For $`t`$ with $`\mathrm{otp}\{s{}_{}{}^{\kappa >}2:st\}=\alpha `$, $`z_\alpha ={}_{}{}^{\alpha }\{0\}`$, and $`u\left\{z_\alpha \right\}\mathrm{Cone}(z_\alpha {}_{}{}^{}1)`$, let $`h(u)=\phi _{\pi (t)}(u)`$.
Let $`(\tau _0,_0)`$, $`(\tau _1,_1)`$, …, $`(\tau _{r1},_{r1})`$ enumerate the sparse vip $`m`$-types. Let $`C_0`$ be the set of all $`m`$-element subsets $`A`$ for which $`(\mathrm{clp}(h[A]),_{h[A]})`$ is either $`(\tau _0,_0)`$ or not a sparse vip $`m`$-type. For positive $`j<r`$, let $`C_j`$ be the set of all $`m`$-element subsets $`A`$ of $`{}_{}{}^{\kappa >}2`$ for which $`(\mathrm{clp}(h[A]),_{h[A]})=(\tau _j,_j)`$. Then $`𝒞:=\{C_0,C_1,\mathrm{},C_{r1}\}`$ is a partition of $`[{}_{}{}^{\kappa >}2]^m`$ into $`r`$ sets.
To see that each class of $`𝒞`$ is indivisible, suppose $`d:C_j\mu `$ is a fixed coloring for some $`2\mu <\kappa `$. Extend $`d`$ to all of $`[{}_{}{}^{\kappa >}2]^m`$ by setting $`d(A)=0`$ if $`j>0`$ and $`AC_j`$ or by setting $`d(A)=1`$ if $`j=0`$ and $`AC_j`$. Apply Shelah’s Theorem 2.5 to the restriction to antichains to obtain a strong embedding $`e`$ and a node $`w`$ such that $`e`$ preserves $``$ on $`\mathrm{Cone}(w)`$ and $`d`$ is constant on $`m`$-element subsets of the same $``$-ordered similarity type. Let $`\alpha `$ be the order type of $`\{s{}_{}{}^{\kappa >}2:sw\}`$. Then $`\mathrm{Cone}(z_\alpha {}_{}{}^{}1)`$ is a $`\kappa `$-dense subset. Since $`h`$ agrees with $`\phi _{\pi (w)}`$ on $`\mathrm{Cone}(z_\alpha {}_{}{}^{}1)`$ and $`\phi _{\pi (w)}`$ is $`<_\text{Q}`$-preserving, $`D:=h[\mathrm{Cone}(z_\alpha {}_{}{}^{}1)]`$ is a $`\kappa `$-dense subset of $`\mathrm{Cone}(\pi (w))\mathrm{Cone}(w)`$. Since $`\phi _{\pi (w)}`$ is a sparse diagonalization, the set $`D`$ is a sparse diagonal set with $`D^{}S^{}\mathrm{Cone}(\pi (w))`$. Thus by Lemma 3.7, all $``$-ordered similarity types of $`m`$-element subsets of $`D`$ are sparse vip $`m`$-types. Let $`K:=e[D]`$. By Lemma 3.14, all $``$-ordered similarity types of $`m`$-element subsets of $`K`$ are sparse vip $`m`$-types. It follows that $`[K]^mC_j`$ is $`d`$-monochromatic. Hence each $`C_j`$ is indivisible.
To see that each class of $`𝒞`$ is persistent, suppose $`K{}_{}{}^{\kappa >}2`$ is $`\kappa `$-dense and $`j<r`$. By Lemma 4.4, there is a node $`z^{}`$ in $`𝒲(K)`$, the set of densely splitting nodes of $`K`$. Thus $`z^{}{}_{}{}^{}1`$ has a $`\kappa `$-dense set of extensions in $`K`$. In other words, $`K\mathrm{Cone}(z^{})`$ is $`\kappa `$-dense. Let $`z_\alpha `$ be the longest initial segment of $`z^{}{}_{}{}^{}1`$ consisting only of zeros. Then $`\mathrm{Cone}(z^{})\mathrm{Cone}(z_\alpha {}_{}{}^{}1)`$. Let $`t`$ be the $`\alpha `$th element of $`{}_{}{}^{\kappa >}2`$ in the $``$ order. Since $`h`$ agrees with $`\phi _{\pi (t)}`$ on $`\mathrm{Cone}(z_\alpha {}_{}{}^{}1)`$, it follows that $`h[K\mathrm{Cone}(z^{})]`$ is a sparse diagonal set with the property that $`(h[K\mathrm{Cone}(z^{})],<_\text{Q})`$ is $`\kappa `$-dense. Thus by Theorem 5.7, the sparse vip $`m`$-type $`(\tau _j,_j)`$ is realized as $`(\mathrm{clp}(x),_x)`$ for some $`xh[K\mathrm{Cone}(z^{})]`$. Since $`h`$ is injective, there is an $`m`$-element subset $`uK\mathrm{Cone}(z^{})`$ with $`h[u]=x`$, and $`uC_j`$ as required. ∎
## 6 Sparse vip types
In this section we give closed form upper and lower bounds for the number of sparse vip $`m`$-types that facilitate comparisons with D. Devlin’s theorem for $`(,<)`$. We then describe a recursive procedure for computing the number of sparse vip $`m`$-types.
In Figure 1, we give a picture of a specific example of a sparse vip $`5`$-type, which we will call $`(\tau ^{},<)`$, so we can use it in later examples to illustrate a variety of definitions. To translate the figure into a representation in which each node is a sequence of $`0`$’s and $`1`$’s, note that the root is the empty sequence and only line segments with positive slope represent $`1`$’s. We have circled the nodes in the meet closure of the set $`L`$ of leaves of $`\tau ^{}`$; these nodes are the designated elements for any ordering of the levels which makes $`\tau ^{}`$ a sparse vip $`5`$-type. For only three pairs of nodes does the requirement that $`(\tau ^{},<)`$ be a sparse vip $`5`$-type fail to specify the order, and between each such pair we have indicated the order.
###### Lemma 6.1.
If $`\tau `$ is a sparse $`m`$-type and $`L`$ is the set of its leaves, then $`\tau =\{xi:xLi<2m1\}`$.
###### Proof.
By definition, $`\tau `$ is closed under initial segments. ∎
Before introducing a lemma on properties of the leaves of a sparse $`m`$-type, we list the leaves of $`\tau ^{}`$ (see Figure 1 on page 1) with the lexicographically least one at the top of the stack, and continuing in increasing order down the stack.
> $`0,0,0,0`$
> $`0,1,0`$
> $`1,0,0,0,0,0,0`$
> $`1,0,1,0,0,0`$
> $`1,0,1,0,0,1,0,0`$
###### Lemma 6.2.
Suppose $`\tau `$ is a sparse $`m`$-type whose set of leaves is $`D=\{d_0,d_1,\mathrm{},d_{m1}\}`$ listed in $`<_{\text{lex}}`$-increasing order. Then
1. $`D^{}=D\{d_id_{i+1}:i<m1\}`$;
2. $`d_0`$ is a sequence of all zeros;
3. if $`i<m1`$, then $`d_i(\mathrm{lg}(d_id_{i+1}))=0`$ and $`d_{i+1}(\mathrm{lg}(d_id_{i+1}))=1`$;
4. if $`i<m1`$, then for all $`p`$ with $`\mathrm{lg}(d_id_{i+1})<p<\mathrm{lg}(d_{i+1})`$, $`d_{i+1}(p)=0`$.
###### Proof.
The first item follows from the fact that $`D\{d_id_{i+1}:i<m1\}`$ is a subset of $`D^{}`$ of size $`m+(m1)=2m1=|D^{}|`$. Since $`D`$ is diagonal, all its elements have different lengths from among $`0,1,\mathrm{},2m2`$. Since $`D^{}`$ has $`2m1`$ elements, all of different lengths, it follows that for all $`p<2m1`$, there is some $`x=x_pD^{}`$ with $`\mathrm{lg}(x)=p`$.
Now the second item holds, since for all $`p<\mathrm{lg}(d_0)`$, $`d_0(p)=d_0(\mathrm{lg}(x_p))=0`$, either because $`x_p=d_id_{i+1}d_0`$ and $`d_0<_{\text{lex}}d_{i+1}`$ or because $`\tau `$ is sparse.
Then the third item follows from the definition of lexicographic order in a binary tree.
For the fourth item, fix attention on some $`i<m1`$. Notice that if $`\mathrm{}<i`$ and $`d_{\mathrm{}}d_{\mathrm{}+1}d_{i+1}`$, then $`(d_{\mathrm{}}d_{\mathrm{}+1}){}_{}{}^{}1d_id_{i+1}`$, since $`d_{\mathrm{}+1}<_{\text{lex}}d_i`$. Also, if $`i\mathrm{}`$ and $`d_{\mathrm{}}d_{\mathrm{}+1}d_i`$, then $`(d_{\mathrm{}}d_{\mathrm{}+1}){}_{}{}^{}0d_i`$, since $`d_i<_{\text{lex}}d_{\mathrm{}+1}`$. Now the fourth item follows from the previous two statements and the fact that $`\tau `$ is sparse. ∎
Next we associate with each sparse $`m`$-type a sequence which is characteristic.
###### Definition 6.3.
Suppose $`\tau `$ is a sparse $`m`$-type whose set of leaves is $`D=\{d_0,d_1,\mathrm{},d_{m1}\}`$ listed in $`<_{\text{lex}}`$-increasing order. Define $`P(\tau ):(2m1)(2m1)`$ by
$$P(\tau )(k)=\{\begin{array}{cc}\mathrm{lg}(d_i),\hfill & \text{if }k=2i\text{,}\hfill \\ \mathrm{lg}(d_id_{i+1}),\hfill & \text{otherwise.}\hfill \end{array}$$
###### Lemma 6.4.
If $`\tau `$ and $`\tau ^{}`$ are sparse $`m`$-types and $`P(\tau )=P(\tau ^{})`$, then $`\tau =\tau ^{}`$.
###### Proof.
Let $`D=L(\tau )`$ and $`E=L(\tau ^{})`$ be the sets of leaves of $`\tau `$ and $`\tau ^{}`$. List $`D`$ and $`E`$ in increasing lexicographic order as $`d_0`$, $`d_1`$, …, $`d_{m1}`$ and $`e_0`$, $`e_1`$, …, $`e_{m1}`$. Then for all $`i<m`$, $`\mathrm{lg}(d_i)=\mathrm{lg}(e_i)`$ by definition of $`P`$. Also, for $`i<m1`$, $`\mathrm{lg}(d_id_{i+1})=\mathrm{lg}(e_ie_{i+1})`$. Use induction on $`i<m`$ and the previous lemma to show $`d_i=e_i`$. ∎
###### Definition 6.5.
A function $`P:(2m1)(2m1)`$ is an *alternating permutation* if it is a permutation, for all even $`i<2m2`$, $`P(i)>P(i+1)`$ and for all odd $`i<2m2`$, $`P(i)<P(i+1)`$.
The study of alternating permutations and the alternating group dates back to André in 1881, and continues to be an active area of investigation (see for example).
###### Lemma 6.6.
For all positive integers $`m`$, $`P`$ is a bijection between the collection of sparse $`m`$-types and the the set of alternating permutations on $`(2m1)`$.
###### Proof.
Properties of meets guarantee that $`P(\tau )`$ is an alternating permutation whenever $`\tau `$ is a sparse $`m`$-type. By the previous lemma, $`P`$ is one-to-one.
To see that it is onto, suppose $`p:(2m1)(2m1)`$ is an alternating permutation. Let $`d_0`$ be the sequence of zeros of length $`p(0)`$. Continue the recursion by defining $`d_1`$, $`d_2`$, …, $`d_{m1}`$, by setting $`d_{i+1}`$ to be the sequence of of length $`p(2i+2)`$ which extends $`(d_ip(2i+1)){}_{}{}^{}1`$ with all zeros. Since $`p`$ is a permutation, all elements of $`D:=\{d_0,\mathrm{},d_{m1}\}`$ have different lengths, and $`D`$ is an antichain listed in $`<_{\text{lex}}`$-increasing order. Also by construction, $`\mathrm{lg}(d_jd_{j+1})=p(2j+1)`$, so $`D`$ is diagonal. By construction, $`d_0(p)=0`$ for all $`p<\mathrm{lg}(d_0)`$. Use induction on $`i`$ to show that for all $`i<m1`$ and all $`p<\mathrm{lg}(d_{i+1})`$, if $`d_{i+1}(p)=1`$, then for some $`\mathrm{}i`$, $`p=\mathrm{lg}(d_{\mathrm{}}d_{\mathrm{}+1})`$ and $`d_{\mathrm{}}d_{\mathrm{}+1}d_{i+1}`$. It follows that $`D`$ is sparse. Let $`\tau =\mathrm{clp}(D)`$. Then the set of leaves of $`\tau `$ is $`D`$ and $`\tau `$ is a sparse $`m`$-type. By construction $`P(\tau )=p`$. Since $`p`$ was arbitrary, the mapping $`P`$ is onto. ∎
The above proof was inspired by the counting of *Joyce trees* (named by Ross Street ) in a paper by Joyce applying category theory to physics. The meet closure of an $`m`$-element diagonal subset of $`{}_{}{}^{(2m1)>}2`$ is an example of a Joyce tree. They are counted up to a certain equivalence, and each equivalence class has exactly one example whose closure under initial segment is a sparse $`m`$-type. We also used ideas from a counting argument by Vuksanovic in .
###### Lemma 6.7.
The number of alternating permutations on $`2m1`$ is $`t_m`$, where $`t_m`$ is the $`m`$-th tangent number, defined recursively by $`t_1=1`$ and
$$t_n=\underset{i=1}{\overset{n1}{}}\left(\genfrac{}{}{0pt}{}{2n2}{2i1}\right)t_it_{ni}.$$
###### Proof.
An exponential generating function for the sequence $`a_n`$, where $`a_n`$ is the number of alternating permutations on $`n`$, is $`\mathrm{sec}(x)+\mathrm{tan}(x)`$ (see Stanley’s *Enumerative Combinatorics I* ). Since the terms corresponding to $`\mathrm{sec}(x)`$ in this series have even powers and the terms corresponding to $`\mathrm{tan}(x)`$ have odd powers, an exponential generating function for $`b_m`$, where $`b_m`$ is the number of alternating permutations on $`(2m1)`$ is $`\mathrm{tan}(x)`$.
It is not difficult to show directly that the number of alternating permutations on $`2m1`$ satisfies the above recurrence. For notational convenience, let $`K`$ denote the set $`2m1`$. Given a $`2i1`$ element subset $`IK\left\{\mathrm{\hspace{0.17em}2}i1\right\}`$ and alternating permutations $`p_0`$ on $`2i1`$ and $`p_1`$ on $`2(mi)1=2m2i1`$, use the unique order preserving maps $`e_0`$ from $`(2i1)`$ to $`I`$ and $`e_1`$ from $`(2m2i1)`$ to $`K(I\left\{\mathrm{\hspace{0.17em}2}i1\right\})`$ to $`p=p_0e_0^1\left\{(2i1,0)\right\}p_1e_1^1`$. (See for details). ∎
###### Corollary 6.8.
The set of sparse $`m`$-types has cardinality $`t_m`$, where $`t_m`$ is the $`m`$-th tangent number, and $`t_mt_m^+`$ where $`t_m^+`$ is the number of sparse vip $`m`$-types.
The next goal is to provide a closed form upper bound for the number of sparse vip $`m`$-types as a product of $`t_m`$ times a constant factor. We start by looking at the size of levels of $`\mathrm{clp}(A)`$ for $`A`$ an $`m`$-element diagonal set.
###### Definition 6.9.
Suppose $`A{}_{}{}^{(2m1)>}2`$ is a diagonal set. Enumerate $`A^{}`$ in increasing order of length as $`a_0,a_1,\mathrm{},a_{2m2}`$ and define
$$\mathrm{}_i(A):=i+12\left|\{j<i:a_jA^{}A\}\right|.$$
Note that for $`m`$-element diagonal subsets of $`{}_{}{}^{(2m1)>}2`$, the lengths of the elements of the meet closure are $`|a_0|=0,|a_1|=1,\mathrm{},|a_{2m2}|=2m2`$.
###### Lemma 6.10.
For $`m2`$, and any $`m`$-element diagonal subset $`A{}_{}{}^{(2m1)>}2`$ whose meet closure is listed in increasing order as $`a_0,a_1,\mathrm{},a_{2m2}`$, the cardinality of level $`i`$ of $`\mathrm{clp}(A)`$ is
$$|\{a_ji:i<2m1\}|=\mathrm{}_i(A)=(2m1)i2|\{ji:a_jA^{}A\}|.$$
Moreover, if $`i<m`$, then $`\mathrm{}_i(A)i+1`$ and $`\mathrm{}_{2m2i}i+1`$.
###### Proof.
Use induction to compute the size of $`L_i:=\{a_j|a_i|:j<2m1\}`$ in two different ways. To start the inductions, note that for $`i=0`$ and $`i=2m2`$, $`L_0=\{a_0\}`$ and $`L_{2m2}=\{a_{2m2}\}`$ are singletons, so $`\mathrm{}_0(A)=1`$ and $`\mathrm{}_{2m2}(A)=1`$. Also, since $`a_0`$ is meet of two elements and $`a_0,a_1`$ are the shortest elements, we have $`\mathrm{}_1(A)=|L_1|=2`$. Since $`a_{2m2},a_{2m3}`$ are the longest elements, we have $`L_{2m3}=\{a_{2m2}_{2m3},a_{2m3}\}`$ is also a doubleton, thus $`\mathrm{}_{2m3}(A)=2`$. Counting up from $`i=0`$, we get the values in the formula for $`\mathrm{}_i(A)`$, since the value for $`|L_{i+1}|`$ is one more than the value for $`|L_i|`$ if $`a_iA^{}A`$ and one less if $`a_iA`$. Thus for $`0<i<m`$, $`\mathrm{}_i(A)\mathrm{}_{i1}(A)+1i+1`$. Counting down from $`i=2m2`$, we get the values in the displayed formula, since the value for $`|L_{i1}|`$ is one more than the value for $`|L_i|`$ if $`a_{i1}A`$ and one less if $`a_{i1}A^{}A`$. Thus for $`0<i<m`$, $`\mathrm{}_{2m2i}(A)\mathrm{}_{2m2(i1)}+1i+1`$. ∎
The values $`\mathrm{}_i(A)=i+1`$ and $`\mathrm{}_{2m2i}=i+1`$ are achieved if $`A`$ is a comb whose leaves listed in increasing order as $`a_0,a_1,\mathrm{},a_{m1}`$ satisfy $`\mathrm{lg}(a_i)=m1+i`$ and for $`i<m1`$, $`\mathrm{lg}(a_ia_{i+1})=i`$.
###### Definition 6.11.
For any $`m`$-element diagonal subset $`A{}_{}{}^{(2m1)>}2`$, let $`V(A)`$ be the number of pairs $`(\mathrm{clp}(A),)`$ where $``$ is a $`A^{}`$-vip order of $`\mathrm{clp}(A)`$.
###### Lemma 6.12.
For any $`m`$-element diagonal subset $`A{}_{}{}^{(2m1)>}2`$, the number of $`A`$-vip orders of $`\mathrm{clp}(A)`$ is bounded above:
$$V(A)\underset{0<i<2m2}{}(\mathrm{}_i(A)1)!(m1)!\underset{i<m1}{}(i!)^2.$$
###### Proof.
Recall that every level of $`\mathrm{clp}(A)`$ has an element of $`A^{}`$. For the first inequality, note that in every $`A`$-vip order, the element of $`A^{}`$ on each level is the least element of the level. The righthand side of the first inequality is the count of level orders that satisfy this constraint. For the second inequality, use the estimates of Lemma 6.10 and the fact that $`0!=1`$. ∎
###### Lemma 6.13.
The number of sparse vip $`m`$-types at bounded above by $`t_m^+t_m(m1)!_{i<m}(i!)^2`$.
###### Proof.
By Lemma 6.8 the number of sparse $`m`$-types is $`t_m`$, so the lemma follows from Lemma 6.12. ∎
Using notions from finite combinatorics, such as reverse Raney sequences, and a fine analysis of sparse $`m`$-types, it is possible to calculate a closed form lower bound for the values of $`t_m^+`$. This is done in an upcoming paper by J. Larson, one of the conclusions of which is
###### Theorem 6.14.
\[Larson \] For all $`m2`$, $`t_m+(2^{m1})(1+\mathrm{\Pi }_{i<m}i!)t_m^+`$.
Figure 2 summarizes the calculation from of values of $`t_m^+`$ for $`m5`$. A comparison with $`t_m`$ is also included.
## 7 A proof of Shelah’s Theorem
In this section we proof of Theorem 2.5 based on Shelah’s proof together with ideas from . The major part of the proof is dedicated to the proof of Lemma 7.1 below; following this we show that this lemma implies Theorem 2.5.
###### Lemma 7.1 (End Homogeneity).
Assume $`m2`$ and that $`\kappa `$ is measurable in the model obtained by adding $`\lambda `$ Cohen subsets of $`\kappa `$, where $`\lambda (\kappa )_{2^\kappa }^{2m}`$. Then for any well ordering $``$ of the levels of $`{}_{}{}^{\kappa >}2`$ and coloring $`d:_{\alpha <\kappa }[{}_{}{}^{\alpha }2]^m\sigma `$ of the $`m`$-element level sets of $`{}_{}{}^{\kappa >}2`$ with $`\sigma <\kappa `$ colors there is a strong embedding $`e:{}_{}{}^{\kappa >}2T{}_{}{}^{\kappa >}2`$ such that whenever $`s:=(s_0,\mathrm{},s_{m1})[{}_{}{}^{\alpha }2]^m`$ and $`\beta <\alpha `$ are such that the members of $`s\beta :=(s_0\beta ,\mathrm{},s_{m1}\beta )`$ are distinct, and $`s`$ and $`s\beta `$ are ordered the same way by $``$, then we have $`d(e[s])=d(e[s\beta ])`$.
We actually use a slightly stronger version of Lemma 7.1, although we will indicate how this can be avoided at the cost of a slight strengthening of the hypothesis:
###### Lemma 7.2.
Suppose that $`\kappa `$ and $`m`$ are as in Lemma 7.1, and that for each $`\xi <\kappa `$ we have a coloring $`d_\xi `$ of the $`m`$-element level sets of $`{}_{}{}^{\kappa >}2`$ in fewer than $`\kappa `$ colors. Then there is a strong embedding $`e`$ such that whenever $`s`$ and $`\beta `$ are as in Lemma 7.1 we have $`d_\xi (e[s])=d_\xi (e[s\beta ])`$ for each $`\xi <\kappa `$
We will give the proof of Lemma 7.1, and will indicate in footnotes how this should be modified to prove Lemma 7.2.
###### Proof.
Let $`m,\kappa ,d`$ and $``$ be as in Lemma 7.1, and let $`P`$ be the the forcing notion adding $`\lambda `$ many Cohen subsets of $`\kappa `$. Thus a condition in $`P`$ is a function $`p`$ with domain $`\mathrm{supp}(p)[\lambda ]^{<\kappa }`$ and with values $`p(\nu ){}_{}{}^{\kappa >}2`$. We abuse notation by writing $`p^{}p`$ if $`p^{}`$ is stronger than $`p`$, that is, if $`\mathrm{supp}(p^{})\mathrm{supp}(p)`$ and $`p^{}(\nu )p(\nu )`$ for each $`\nu \mathrm{supp}(p)`$.
For $`i<\lambda `$ let η
~
isubscript𝜂
~
𝑖\mathchoice{\oalign{$\displaystyle\eta$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle\eta$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle\eta$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle\eta$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{i} be a name for the $`i`$th Cohen subset of $`\kappa `$, and let $`𝒟`$ $`\stackrel{~}{}`$ be a name for the $`\kappa `$-complete ultrafilter $`𝒟`$ on $`\kappa `$ assumed to exist in the generic extension. For any $`u[\lambda ]^{<\omega }`$ there is a unique $`j<\sigma `$ such that $`\{\xi <\kappa :d(\{\eta _i\xi :iu\})=j\}𝒟`$; we write $`d(u)`$ for this $`j`$ and let d
~
(u)𝑑
~
𝑢\mathchoice{\oalign{$\displaystyle d$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle d$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle d$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle d$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}(u) be a name for $`d(u)`$.<sup>1</sup><sup>1</sup>1For Lemma 7.2, this becomes d
~
ξ(u)subscript𝑑
~
𝜉𝑢\mathchoice{\oalign{$\displaystyle d$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle d$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle d$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle d$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{\xi}(u) for each $`\xi <\kappa `$.
If $`W\lambda `$ then we write $`P|W`$ for $`\{pP:\mathrm{supp}(p)W\}`$. If $`H:WW^{}`$ is an order preserving map between subsets of $`\lambda `$ then $`h_H:P|WP|W^{}`$ is the map defined by $`h_H(p)(H(i))=p(i)`$, and if $`\mathrm{otp}(W)=\mathrm{otp}(W^{})`$ then we define $`h_{W,W^{}}=h_H`$ where $`H:WW^{}`$ is the unique order preserving map.
###### Claim 7.2.a.
There is a set $`Z[\lambda ]^\kappa `$ and a function $`W:[Z]^m[\lambda ]^\kappa `$ satisfying the following conditions:
1. If $`u[Z]^m`$ then $`uW(u)`$, and $`P|W(u)`$ contains a maximal antichain of conditions $`p`$ deciding the value of d
~
(u)𝑑
~
𝑢\mathchoice{\oalign{$\displaystyle d$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle d$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle d$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle d$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}(u).<sup>2</sup><sup>2</sup>2For Lemma 7.2, $`P|W(u)`$ contains a maximal antichain of conditions deciding the value of d
~
ξ(u)subscript𝑑
~
𝜉𝑢\mathchoice{\oalign{$\displaystyle d$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle d$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle d$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle d$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{\xi}(u) for each $`\xi <\kappa `$.
2. If $`u,u^{}[Z]^m`$ and $`|u|=|u^{}|`$, then $`\mathrm{otp}(W(u))=\mathrm{otp}(W(u^{}))`$, the function $`h_{W(u),W(u^{})}`$ maps $`u`$ to $`u^{}`$, and for all $`pP|W(u)`$ and $`j<\sigma `$,
pd
~
(u)=jhW(u),W(u)(p)d
~
(u)=j.iffforces𝑝𝑑
~
𝑢𝑗forcessubscript𝑊𝑢𝑊superscript𝑢𝑝𝑑
~
superscript𝑢𝑗p\Vdash\mathchoice{\oalign{$\displaystyle d$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle d$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle d$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle d$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}(u)=j\iff h_{W(u),W(u^{\prime})}(p)\Vdash\mathchoice{\oalign{$\displaystyle d$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle d$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle d$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle d$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}(u^{\prime})=j. (1)
3. If $`u^{}u[Z]^m`$ then $`W(u^{})W(u)`$, and if $`u,u^{}[Z]^m`$ then $`W(uu^{})=W(u)W(u^{})`$.
###### Proof.
Because $`P`$ has the $`\kappa ^+`$-chain condition, there are sets $`W^{}(u)[\lambda ]^\kappa `$ such that clause 1 is satisfied when $`W^{}`$ is substituted for $`W`$. We can also arrange that $`W^{}(u^{})W^{}(u)`$ whenever $`u^{}u`$.
Now define an equivalence relation on $`[\lambda ]^{2m}`$ as follows: two sets $`u`$ and $`u^{}`$ in $`[\lambda ]^{2m}`$ are equivalent if they satisfy the following two conditions:
1. $`|u|=|u^{}|`$, and if $`|u|m`$ then clause 2 above holds with $`W^{}(u)`$ and $`W^{}(u^{})`$ substituted for $`W(u)`$ and $`W(u^{})`$.
2. Suppose that $`u_1,u_2u`$, with $`u_1,u_2[\lambda ]^m`$, and that $`u_1^{}`$ and $`u_2^{}`$ are the subsets of $`u^{}`$ such that $`(u,u_1,u_2,<)(u^{},u_1^{},u_2^{},<)`$. Then this isomorphism extends to an isomorphism
$$(W^{}(u),W^{}(u_1),W^{}(u_2),<)(W^{}(u^{}),W^{}(u_1^{}),W^{}(u_2^{}),<).$$
There are only $`2^\kappa `$ equivalence classes, and therefore the assumption that $`\lambda (\kappa )_{2^\kappa }^{2m}`$ implies that there is $`Z^{}[\lambda ]^\kappa `$ such that any pairs $`u,u^{}[Z^{}]^k`$ are equivalent for all $`k2m`$. It follows that clauses 1 and 2 are satisfied when $`W^{}`$ and $`Z^{}`$ are substituted for $`W`$ and $`Z`$.
Now define
$$W(u):=\{\underset{vX}{}W^{}(v):X[Z^{}]^m\&Xu\},$$
(2)
and let $`Z=\{\gamma _{\omega \nu }:\nu <\kappa \}`$ where $`\gamma _\nu :\nu <\kappa `$ is the increasing enumeration of $`Z^{}`$. We claim that $`W`$ and $`Z`$ are as required.
First note that we only need to consider finite sets $`X`$ in (2). To see this, suppose $`Xu`$, fix a set $`u_0X`$, and pick $`X^{}X`$ such that $`u_0X^{}`$ and for each $`\xi u_0u`$ there is a set $`u_\xi X^{}`$ such that $`\xi u_\xi `$. Then $`X^{}u`$, $`|X^{}|1+|u_0||u|1+m`$, and $`_{vX^{}}W^{}(v)_{vX}W^{}(v)`$.
Now note that condition (b) can be extended to arbitrary finite sets: If $`\{u_i:i<k\}`$ and $`\{u_i^{}:i<k\}`$ are subsets of $`[Z^{}]^{<\omega }`$ such that
$$(\underset{i<k}{}u_i,u_0,\mathrm{},u_{k1},<)(\underset{i<k}{}u_i^{},u_0^{},\mathrm{},u_{k1}^{},<)$$
(3)
then
$$\begin{array}{c}(\underset{i<k}{}W^{}(u_i),W^{}(u_0),\mathrm{},W^{}(u_{k1}),<)\hfill \\ \hfill (\underset{i<k}{}W^{}(u_i^{}),W^{}(u_0^{}),\mathrm{},W^{}(u_{k1}^{}),<).\end{array}$$
(4)
To see this, note that condition (b) implies that $`\mathrm{otp}(W_u^{})=\mathrm{otp}(W_u^{}^{})`$ whenever $`u,u^{}[Z^{}]^s`$ for some $`s2m`$. Let $`W^{}(u_i)=\{\gamma _{i,\nu }:\nu <\xi _i\}`$ and $`W^{}(u_i^{})=\{\gamma _{i,\nu }^{}:\nu <\xi _i\}`$ be the increasing enumerations. It follows that $`_{i<k}W^{}(u_i)=\{\gamma _{i,\nu }:i<k\&\nu <\xi _i\}`$; and for each $`i,i^{}<k`$ and $`\nu <\xi _i`$ and $`\nu ^{}<\xi _i^{}`$ we have $`\gamma _{i,\nu }=\gamma _{i^{},\nu ^{}}\gamma _{i,\nu }^{}=\gamma _{i^{},\nu ^{}}^{}`$, and $`\gamma _{i,\nu }<\gamma _{i^{},\nu ^{}}\gamma _{i,\nu }^{}<\gamma _{i^{},\nu ^{}}^{}`$. This easily implies (4).
A consequence of this is that the intersection $`_{vX}W^{}(v)`$ in (2) does not depend on $`X`$, but only on the isomorphism type of $`(X,u,u_0,\mathrm{},u_{k1})`$ where $`X=\{u_i:i<k\}`$. The proof of this proceeds by induction on the size of the set of ordinals in $`X`$ on which two sets $`X`$ and $`X^{}`$ with the same isomorphism type differ. Let $`\xi X`$ and $`\xi ^{}X^{}`$ be the least corresponding pair of ordinals which differ between $`X`$ and $`X^{}`$. Assuming without loss of generality that $`\xi <\xi ^{}`$, let $`X^{\prime \prime }`$ be the set obtained by replacing $`\xi ^{}`$ by $`\xi `$ in $`X^{}`$. Then $`_{vX}W^{}(v)=_{vX^{\prime \prime }}W^{}(v)`$ by the induction hypothesis. To see that $`_{vX^{}}W^{}(v)=_{vX^{\prime \prime }}W^{}(v)`$, pick some $`v_0X^{}`$ with $`\xi v_0`$. Then $`_{vX^{}}W^{}(v)W^{}(v_0)`$, and the members of $`W^{}(v_0)`$ are not moved in the isomorphism (4) between $`_{vX^{}}W^{}(v)`$ and $`_{vX^{\prime \prime }}W^{}(v)`$.
This implies that the union in (2) can be taken to be finite, since there are only finitely many such isomorphism classes, and hence $`|W(u)|\kappa `$. If $`u,u^{}[Z]^s`$ for some $`sm`$ then it follows that $`\mathrm{otp}(W(u))=\mathrm{otp}(W(u^{}))`$, since the fact that $`Z`$ contains only limit members of $`Z^{}`$ implies that for each set $`X`$ contributing an intersection to $`W(u)`$ there is an isomorphic set $`X^{}`$ contributing an intersection to $`W(u^{})`$. Furthermore this isomorphism between $`W(u)`$ and $`W(u^{})`$ preserves the isomorphism between $`W^{}(u)`$ and $`W^{}(u^{})`$, so the equivalence (1) in clause 2 of Claim 7.2.a holds for $`W`$ as well as for $`W^{}`$. Finally, the definition of $`W(u)`$ easily implies clause 3 of Claim 7.2.a. ∎
We are now ready to construct the promised strong embedding $`e`$. For sets $`u[Z]^m`$ define $`W^{}(u)=W(u)_{u^{}u}W(u^{})`$.
For $`\alpha <\kappa `$ let $`R_\alpha `$ be the set of one to one functions $`s`$ with $`\mathrm{dom}(s)[{}_{}{}^{\alpha }2]^m`$ and $`\mathrm{ran}(s)Z`$. If $`sR_\alpha `$, $`\beta <\alpha `$, and the members of $`\{x\beta :x\mathrm{dom}(s)\}`$ are all distinct then we will abuse notation by writing $`s\beta `$ for the function $`s^{}R_\beta `$ with $`\mathrm{dom}(s^{})=\{x\beta :x\mathrm{dom}(s)\}`$ defined by $`s^{}(x\beta )=s(x)`$. The sets $`R_\alpha `$ include the empty function $`\mathrm{}`$ as a member, and we will abuse the notation by taking the function $`\mathrm{}R_\alpha `$ to be different from $`\mathrm{}R_\beta `$ whenever $`\alpha \beta `$.
We use recursion on $`\alpha <\kappa `$ to define an ordinal $`\zeta _\alpha <\kappa `$ and a map $`e{}_{}{}^{\alpha }2:{}_{}{}^{\alpha }2{}_{}{}^{\zeta _\alpha }2`$, along with conditions $`p_sP|W^{}(\mathrm{ran}(s))`$ for each $`sR_\alpha `$. Several times in the following construction we will use the observation that this, together with the fact from Claim 7.2.a that $`W(u)W(u^{})=W(uu^{})`$ for all $`u,u^{}[Z]^m`$, implies that $`p_s`$ and $`p_s^{}`$ are compatible whenever $`ss^{}`$ is a function. Hence $`\{p_s:sR_\alpha \&s\tau \}`$ is a condition for any one to one function $`\tau `$ with $`\mathrm{dom}(\tau ){}_{}{}^{\alpha }2`$ and $`\mathrm{ran}(\tau )Z`$. In particular this is true for all $`\tau R_\alpha `$.
The construction will satisfy the following induction hypotheses:
1. If $`g:\mathrm{ran}(s)Z`$ preserves order, then $`p_{gs}=h_{W^{}(\mathrm{ran}(s)),W^{}(\mathrm{ran}(g))}(p_s)`$.
2. If $`\beta <\alpha `$ and the sets $`\{x\beta :x\mathrm{dom}(s)\}`$ are distinct then $`p_sp_{s\beta }`$.
3. tsptd
~
(ran(s))=d(e[dom(s)])forcessubscript𝑡𝑠subscript𝑝𝑡𝑑
~
ran𝑠𝑑𝑒delimited-[]dom𝑠\bigcup_{t\subseteq s}p_{t}\Vdash\mathchoice{\oalign{$\displaystyle d$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle d$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle d$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle d$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}(\operatorname{ran}(s))=d(e[\operatorname{dom}(s)]) whenever $`s`$ is an order isomorphism between $`(\mathrm{dom}(s),)`$ and $`(\mathrm{ran}(s),)`$. <sup>3</sup><sup>3</sup>3For Lemma 7.2 this should hold for $`d_\xi `$ for all $`\xi <\alpha `$.
4. If $`\mathrm{dom}(s)=\left\{x\right\}`$ and $`s(x)=\eta Z`$ then $`p_s(\eta )=e(x)`$.
Clause 1 asserts that $`p_s`$ depends, up to isomorphism, only on $`\mathrm{dom}(s)`$ and the order which $`s`$ induces on that set.
Clause 3 will imply the required end-homogeneity, since if $`\beta <\alpha `$ then $`_{ts}p_t`$ forces that d
~
(ran(s))=d
~
(ran(sβ))𝑑
~
ran𝑠𝑑
~
ran𝑠𝛽\mathchoice{\oalign{$\displaystyle d$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle d$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle d$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle d$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}(\operatorname{ran}(s))=\mathchoice{\oalign{$\displaystyle d$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle d$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle d$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle d$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}(\operatorname{ran}(s{\upharpoonright}\beta)), and consequently $`d(e[\mathrm{dom}(s)])=d(e[\mathrm{dom}(s\beta )])`$ whenever $``$ orders $`\mathrm{dom}(s)`$ and $`\mathrm{dom}(s\beta )`$ alike.
For $`n`$ small enough that $`2^n<m`$ we can define $`\zeta _n=n`$, along with $`e(s)=s`$ and $`p_s=\mathrm{}`$ for all $`sR_n`$. The construction for $`\alpha `$ with $`2^\alpha m`$ consists of three steps. The first step defines conditions $`\overline{p}_s`$ which will satisfy clauses 1 and 2 (with $`\overline{p}_s`$ instead of $`p_s`$) and satisfies clause 3 to the extent that
(sRα)(js<σ)tsp¯td
~
(ran(s))=js.forcesfor-all𝑠subscript𝑅𝛼subscript𝑗𝑠𝜎subscript𝑡𝑠subscript¯𝑝𝑡𝑑
~
ran𝑠subscript𝑗𝑠(\forall s\in R_{\alpha})(\exists j_{s}<\sigma)\;\bigcup_{t\subseteq s}\bar{p}_{t}\Vdash\mathchoice{\oalign{$\displaystyle d$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle d$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle d$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle d$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}(\operatorname{ran}(s))=j_{s}. (5)
The second step will define $`\zeta _\alpha `$ and define $`e{}_{}{}^{\alpha }2:{}_{}{}^{\alpha }2{}_{}{}^{\zeta _\alpha }2`$ so that if the ordering given to $`\mathrm{dom}(s)`$ by $`s`$ agrees with the given ordering $``$ then $`d(e[\mathrm{dom}s])`$ has the value $`j_s`$ determined in the first step. The final step consists of setting $`p_s=\overline{p}_s`$ except for the adjustments necessary to satisfy clause 4.
The first step is divided into two cases, depending on whether or not $`\alpha `$ is a limit ordinal.
### Case 1: $`\alpha `$ is a limit ordinal.
Let $`\overline{\zeta }:=sup_{\beta <\alpha }\zeta _\beta `$. For $`x{}_{}{}^{\alpha }2`$ let $`\overline{e}(x):=_{\beta <\alpha }(e(x\beta ))`$, and for $`sR_\alpha `$, let
$$\overline{p}_s:=\{p_{s\gamma }:\beta <\alpha \&\{x\beta :x\mathrm{dom}(s)\}\text{ are pairwise distinct}\}.$$
Note that the induction hypothesis implies that $`\overline{p}_sP|W^{}(\mathrm{ran}(s))`$. Furthermore $`\overline{p}_s`$ satisfies clause 2, and by the induction hypothesis this implies that $`\overline{p}_s`$ satisfies (5).
### Case 2: $`\alpha `$ is a successor ordinal.
Let $`\alpha =\beta +1`$, set $`\overline{\zeta }=\zeta _\beta +1`$, and set $`\overline{e}(x)=e(x\beta ){}_{}{}^{}x(\beta ){}_{}{}^{\overline{\zeta }}2`$ for each $`x{}_{}{}^{\alpha }2`$. For each $`sR_\alpha `$ such that the members of $`\{x\gamma :x\mathrm{dom}(s)\}`$ are distinct set $`p_s^0=p_{s\beta }`$. If the members of $`\{x\gamma :x\mathrm{dom}(s)\}`$ are not distinct then set $`p_s^0=\mathrm{}`$.
Let $`s_i:i<\gamma `$ enumerate the set of $`sR_\alpha `$ such that $`\mathrm{ran}(s)`$ is an initial segment of $`Z`$ of length $`m`$. Thus for every $`sR_\alpha `$ there is a unique ordinal $`i<\gamma `$ and order preserving map $`g`$ such that $`s=gs_i`$. Now define $`p_s^i`$ for each $`i\gamma `$ and $`sR_\alpha `$ by recursion on $`i\gamma `$: If $`i`$ is a limit ordinal then $`p_s^i=_{i^{}<i}p_s^i^{}`$. For a successor ordinal $`i+1`$, suppose that $`p_s^i`$ is defined for all $`sR_\alpha `$ and set $`q^{}:=_{ts_i}p_t^i`$. Now choose $`qq^{}`$ in $`W(\mathrm{ran}(s_i))`$ so that $`q`$ decides the value<sup>4</sup><sup>4</sup>4For Lemma 7.2, $`q`$ decides d
~
ξ(ran(si))subscript𝑑
~
𝜉ransubscript𝑠𝑖\mathchoice{\oalign{$\displaystyle d$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle d$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle d$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle d$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}_{\xi}(\operatorname{ran}(s_{i})) for each $`\xi <\alpha `$. of d
~
(ran(si))𝑑
~
ransubscript𝑠𝑖\mathchoice{\oalign{$\displaystyle d$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle d$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle d$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle d$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}(\operatorname{ran}(s_{i})), and for each $`ts_i`$ set $`p_t^{i+1}=qW^{}(\mathrm{ran}(t))`$. To finish up, define $`p_{gt}^{i+1}=h_{W^{}(\mathrm{ran}(s_i)),W^{}(\mathrm{ran}(g))}(p_t)`$, as in clause 1, for all $`ts_i`$ and all order preserving functions $`g:\mathrm{ran}(t)Z`$; and set $`p_s^{i+1}=p_s^i`$ for all other $`sR_\alpha `$.
Now complete the first step of the construction by setting $`\overline{p}_s=p_s^\gamma `$.
For the second step of the construction let $`\tau :{}_{}{}^{\alpha }2Z`$ be a one to one map which preserves the given ordering $``$ on $`{}_{}{}^{\alpha }2`$. Such a map exists since $`|{}_{}{}^{\alpha }2|<\kappa =|Z|`$. Now set $`q:=\{p_{\tau y}:y[{}_{}{}^{\alpha }2]^m\}`$. This is a condition, and $`q`$ decides the value of d
~
(τ[y])𝑑
~
𝜏delimited-[]𝑦\mathchoice{\oalign{$\displaystyle d$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle d$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle d$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle d$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}({\tau[y]}) for each $`y[{}_{}{}^{\alpha }2]^m`$.
Now, since $`𝒟`$ $`\stackrel{~}{}`$ is forced to be a $`\kappa `$-complete ultrafilter in the generic extension, there is a condition $`q^{}q`$ and an ordinal $`\xi \overline{\zeta }`$ such that
qd({ηνξ:ντ[y]})=d
~
(τ[y]))q^{\prime}\Vdash d(\left\{\,\eta_{\nu}{\upharpoonright}\xi:\nu\in\tau[y]\,\right\})=\mathchoice{\oalign{$\displaystyle d$\crcr\vbox to0.86108pt{\hbox{$\displaystyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\textstyle d$\crcr\vbox to0.86108pt{\hbox{$\textstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptstyle d$\crcr\vbox to0.86108pt{\hbox{$\scriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}{\oalign{$\scriptscriptstyle d$\crcr\vbox to0.86108pt{\hbox{$\scriptscriptstyle{\tilde{\mkern-3.0mu}\mkern 3.0mu}{}$}\vss}}}(\tau[y])) (6)
for each $`y[{}_{}{}^{\alpha }2]^m`$. We can assume without loss of generality that for each $`x{}_{}{}^{\alpha }2`$, we have $`\mathrm{dom}(q^{}(\tau (x)))\xi `$. Set $`\zeta _\alpha =\xi `$ and complete the definition of $`e{}_{}{}^{\alpha }2`$ by setting $`e(x)=q^{}(\tau (x))\xi `$. Finally complete the definition of $`p_s`$ by setting $`p_s=\overline{p}_s`$ unless $`s=\tau \left\{x\right\}`$ for some $`x`$, in which case define $`p_s\overline{p}_s`$ by setting $`p_s(\tau (x))=q^{}(\tau (x))\xi +1`$.
This completes the definition of the embedding $`e`$, and hence of the proof of the End Homogeneity Lemma 7.1. ∎
###### Corollary 7.3.
Suppose that $``$ and $`\{_\xi :\xi <\sigma \}`$ are well orderings of $`{}_{}{}^{\kappa >}2`$. Then there is a strong embedding $`e:{}_{}{}^{\kappa >}2{}_{}{}^{\kappa >}2`$ such that for a dense set of nodes $`t{}_{}{}^{\kappa >}2`$, we have $`s_0s_1e(s_0)_\xi e(s_1)`$ for all $`s_0,s_1t`$ and $`\xi <\sigma `$.
###### Proof.
Define $`d:_{\alpha <\kappa }[{}_{}{}^{\alpha }2]^2{}_{}{}^{\sigma }2`$ by setting $`d(s_0,s_1)(\xi )`$ equal to the boolean value $`s_0<_{\text{lex}}s_1s_0_\xi s_1`$.
Apply Lemma 7.1 to get a strong embedding $`e:{}_{}{}^{\kappa >}2{}_{}{}^{\kappa >}2`$ so that if $`s_0<_{\text{lex}}s_1`$ are in $`{}_{}{}^{\alpha }2`$ then $`d(e(s_0),e(s_1))`$ depends only on whether or not $`s_0s_1`$. We want to find, for any given node $`t`$, a node $`t^{}t`$ such that $`d(e(s_0),e(s_1))(\xi )=s_0s_1`$ for all $`s_0,s_1{}_{}{}^{\alpha }2`$ for $`\alpha >\mathrm{lg}(t)`$ such that $`s_0<_{\text{lex}}s_1`$, $`t^{}s_0s_1`$. It will be sufficient to show that we can do this for any one $`\xi <\sigma `$ and for one of the two cases $`s_0s_1`$ or $`s_1s_0`$. We will do it for $`s_0s_1`$, and will indicate the change for $`s_1s_0`$.
Suppose, for the sake of contradiction, that for every $`t^{}t`$ there is a level set $`\{s_0,s_1\}`$ such that $`s_0<_{\text{lex}}s_1`$, $`t^{}s_0s_1`$ and $`s_0s_1`$ but $`e(s_0)_\xi e(s_1)`$. By the end-homogeneity the same will be true of any $`s_0^{}s_1^{}`$ such that $`s_i^{}s_i`$.
Define an infinite sequence of pairs $`(u_n,v_n):n<\omega `$ of nodes as follows: Set $`v_0=t`$. If $`v_n`$ is defined then let $`u_n,v_{n+1}`$ be nodes of the same length, extending $`v_n`$, such that $`u_n<_{\text{lex}}v_{n+1}`$ and $`e(s_0)_\xi e(s_1)`$ for all $`s_0u_n`$ and $`s_1v_{n+1}`$ such that $`s_0s_1`$.
Now fix $`\alpha <\kappa `$ such that $`\alpha >\mathrm{lg}(u_n)`$ for each $`n<\omega `$, and choose $`w_nu_n`$ for each $`n<\omega `$. The nodes $`w_n`$ have the properties that $`w_n<_{\text{lex}}w_n^{}`$ for each $`n<n^{}`$, and $`e(w_n)_\xi e(w_n^{})`$ for each $`n<n^{}`$ such that $`w_nw_n^{}`$. Now apply Ramsey’s theorem to get an infinite subset $`X\omega `$ such that the boolean values $`w_nw_n^{}`$ are constant for $`\{n,n^{}\}[X]^2`$. Since $``$ is a well order we must have $`w_nw_n^{}`$ for all $`n<n^{}`$ in $`X`$, but then $`e(w_n):nX`$ is an infinite descending $`_\xi `$-sequence, contradicting the assumption that $`_\xi `$ is a well order.
The proof for the case $`s_1s_0`$ is the same except that the pair $`\{u_n,v_{n+1}\}`$ satisfies $`v_{n+1}<_{\text{lex}}u_n`$ for $`s_0u_n`$ and $`s_1v_{n+1}`$.
###### Corollary 7.4.
Suppose that $`\kappa `$, $`m`$, $``$, and $`d_\xi `$ are as in Lemma 7.2. Then there is a strongly embedded tree $`T{}_{}{}^{\kappa >}2`$ of height $`\kappa `$ such that $`d_\xi (s)=d_\xi (s^{})`$ whenever $`s`$ and $`s^{}`$ are $``$-similar members of $`[{}_{}{}^{\alpha }2T]^m`$. and $`\xi `$ is smaller than the number of split levels of $`T`$ below $`\alpha `$.
###### Proof.
Let $`e:{}_{}{}^{\kappa >}2{}_{}{}^{\kappa >}2`$ be the strong embedding given by Lemma 7.2. By Corollary 7.3 there is a strong embedding $`e^{}:{}_{}{}^{\kappa >}2{}_{}{}^{\kappa >}2`$ and a $`t{}_{}{}^{\kappa >}2`$ such that, for all $`s,s^{}t`$ we have $`e^{}(s)e^{}(s^{})e^{}(s)e^{}[]e^{}(s^{})se^1[]s^{}`$. Then $`T=e^{}e[{}_{}{}^{\kappa >}2]`$ is as required. ∎
###### Lemma 7.5.
Suppose that $`\kappa `$ is a cardinal which is measurable in the generic extension obtained by adding $`\lambda `$ Cohen subsets of $`\kappa `$, where $`\lambda (\kappa )_{2^\kappa }^{2m}`$. Then for any coloring $`d`$ of the $`m`$-element antichains of $`{}_{}{}^{\kappa >}2`$ into $`\sigma <\kappa `$ colors, and any well-ordering $``$ of the levels of $`{}_{}{}^{\kappa >}2`$, there is a strong embedding $`e:{}_{}{}^{\kappa >}2T{}_{}{}^{\kappa >}2`$ such that $`d(a)=d(b)`$ for all $``$-similar $`m`$-element antichains $`a`$ and $`b`$ of $`T`$.
First we show that this lemma implies Theorem 2.5:
###### Proof of Theorem 2.5 from Lemma 7.5.
Let $`e:{}_{}{}^{\kappa >}2{}_{}{}^{\kappa >}2`$ satisfy the conclusion of Lemma 7.5. By Corollary 7.3 there is a strong embedding $`e^{}:{}_{}{}^{\kappa >}2{}_{}{}^{\kappa >}2`$ and an dense set of nodes $`w{}_{}{}^{\kappa >}2`$ such that $`st`$ if and only if $`e(s)e(t)`$ for all $`s,t_{\alpha <\kappa }\mathrm{Cone}(w)`$. The composition $`e^{}e`$ satisfies the conclusion of Theorem 2.5. ∎
###### Proof of Lemma 7.5.
The maximum height of a similarity tree with $`m`$ terminal nodes is equal to the number of meets plus the number of terminal nodes, which is $`(m1)+m=2m1`$. We will define a sequence $`T_0T_1\mathrm{}T_{2m2}T_{2m1}={}_{}{}^{\kappa >}2`$ of strongly embedded subtrees, using a reverse recursion on $`k<2m1`$ starting with $`T_{2m1}={}_{}{}^{\kappa >}2`$.
For $`k<2m1`$ we assume as a recursion hypothesis that $`T_{k+1}`$ has the following homogeneity property:
> Suppose that $`x,y[T_{k+1}]^m`$ are $``$-similar antichains, with the common collapse $`\mathrm{clp}(x,)=\mathrm{clp}(y,)=(t,_t)`$, and suppose that $`i{}_{}{}^{k}2=j{}_{}{}^{k}2`$ where $`i:(t,_t)(x^{},)`$ and $`j:(t,_t)(y^{},)`$ are the maps witnessing this similarity. Then $`d(x)=d(y)`$.
Note that this puts no constraint on $`T_{2m1}`$, and it implies that $`T_0`$ satisfies the conclusion of Lemma 7.5.
For each $`\alpha <\kappa `$, each antichain $`a[T_{k+1}{}_{}{}^{\alpha >}2]^{<m}`$, and each ordered similarity tree $`(t,_t)`$ consider the following coloring $`d_{k,a,t}`$ of the $`m`$-element level sets $`x[T_{k+1}{}_{}{}^{\alpha }2]^m`$: If $`xa`$ is an antichain, and $`\mathrm{clp}(xa)=(t{}_{}{}^{i}2,_t)`$, then $`d_{k,a,t}(x)=d(y)`$ where $`y[T_{k+1}]^m`$ is any antichain such that $`x=\{\nu \alpha :\nu y\}`$. If $`xa`$ is not an antichain, if $`\mathrm{clp}(xa)(t{}_{}{}^{i}2,_t)`$, or if no such antichain $`y`$ exists then set $`d_{i,a,t}(x)=0`$. Lemma 7.5 implies that there is a strongly embedded subtree $`T_kT_{k+1}`$ on which $`d_{k,a,t}(x)=d_{k,a,t}(\{\nu \beta :\nu x\})`$ whenever the nodes $`\nu \beta `$ for $`\nu x`$ are distinct, and the sets $`x`$ and $`\{\nu \beta :\nu x\}`$ are both in $`T_k`$ and are ordered the same way by $``$. Thus $`T_k`$ satisfies the recursion hypothesis. ∎
We note that the proof of Theorem 2.5 does not require that the ultrafilter $`𝒟`$ be normal, and hence is valid for $`\kappa =\omega `$. This is essentially Harrington’s proof of the Halpern-Läuchli theorem, which may be found in the Farah-Todorcevic book . Harrington’s proof served as a starting point for Shelah in .
## 8 Further remarks on partition theorems
There are a number of remaining open questions suggested by the results presented so far so we comment on some of them.
###### Question 8.1.
Is Theorem 2.5 (or Lemma 7.1) consistent with the GCH? Is the conclusion of Theorem 2.5 compatible with $`L`$, namely can there be an uncountable cardinal $`\kappa `$ in $`L`$ which satisfies that conclusion?
Note that the hypothesis of lemma 7.1 implies that the GCH fails at almost every $`\alpha <\kappa `$. Indeed for almost every $`\alpha <\kappa `$ the power set of $`\alpha `$ in $`V`$ is a generic extension obtained by adding $`\lambda _\alpha `$ Cohen subsets of $`\alpha `$ to some ground model $`M_\alpha `$, where $`M_\alpha \lambda _a(\alpha )_{(2^\alpha )^{M_\alpha }}^{2m}`$, since this is true in the generic extension of $`V`$ obtained by adding $`\lambda `$ Cohen subsets to $`\kappa `$, and this extension does not add bounded subsets to $`\kappa `$.
The latter part of the question was asked by Michael Hrušak. See below for an argument showing that such a $`\kappa `$ must be weakly compact.
###### Question 8.2.
Can Theorem 2.5 be strengthened to require that $`e[]`$?
It would be sufficient to prove this for level sets in $`[{}_{}{}^{\kappa >}2]^2`$, as this would imply the corresponding modification of Corollary 7.3. A positive solution would be likely to also give a positive answer to the next question:
###### Question 8.3.
Is the relation $`{}_{}{}^{\kappa >}2\stackrel{\text{str embedding}}{}({}_{}{}^{\kappa >}2)_{\sigma ,<\omega }^{<\omega }`$ consistent? That is, is it consistent that for any $`d:[{}_{}{}^{\kappa >}2]^{<\omega }\sigma `$ for some $`\sigma <\kappa `$, there is a strongly embedded tree $`T`$ such that $`d[[T]^m]`$ is finite for each $`m<\omega `$?
Shelah observes that the corresponding variation of Lemma 7.1 follows from the assumption that $`\kappa `$ is measurable after adding $`\lambda `$ Cohen subsets of $`\kappa `$, where $`\lambda (\kappa )_{2^\kappa }^{<\omega }`$. This may be easily seen by examining the proof of Lemma 7.1 given here, which requires only minor changes. However the top-down proof of Theorem 2.5 from Lemma 7.1 given here does not work with the superscript $`<\omega `$.
It should be noted that in spite of the use of the ultrafilter $`𝒟`$ in the construction, the set of splitting levels of the homogeneous strongly embedded subtree $`T`$ is not in any sense a member of $`𝒟`$; indeed it is far from being even stationary. This follows from the fact that the set of splitting levels of $`T`$ need not contain any fixed points, even for $`m=1`$, which in turn follows from the observation that $`{}_{}{}^{\alpha >}2`$ contains only $`\alpha `$ many strongly embedded subtrees, each of which has $`2^\alpha `$ many branches. Thus we can color, for each cardinal $`\alpha `$, the members of $`{}_{}{}^{\alpha }2`$ with $`2^\alpha `$ colors in such a way that any strongly embedded subtree of $`{}_{}{}^{\alpha >}2`$ contains branches with every color.
###### Question 8.4.
What is the large cardinal strength of the conclusion of Theorem 2.5?
It is easy to see that a supercompact cardinal whose supercompactness was made indestructible to $`(<\kappa )`$-directed closed forcing using Laver’s method , satisfies the assumptions of the theorem. A much better upper bound is available. Namely, work of Gitik in (mentioned also as in ), together the Erdős-Rado theorem, implies that a model satisfying the hypothesis of Theorem 2.5 can be constructed by forcing over a model of $`\text{GCH}+o(\kappa )=\kappa ^{+2m+2}`$.
On the other hand, it is easy to see that Theorem 2.5 implies that $`\kappa `$ is weakly compact. To show that $`\kappa (\kappa )_2^2`$, let the function $`g:[\kappa ]^22`$ be given and define a coloring $`h`$ of the two element antichains in $`{}_{}{}^{\kappa >}2`$ by by $`h(\{s,t\})=g(\{\mathrm{lg}(s),\mathrm{lg}(st)\})`$ if $`\mathrm{lg}(s)=\mathrm{lg}(t)`$ and $`h(\{s,t\})=0`$ otherwise. If $`T`$ is the strongly embedded tree whose existence is guaranteed by Shelah’s Theorem 2.5 then there are two possible $``$-similarity classes of $`2`$-element level sets, one in which the lexicographic order of the pair agrees with the $``$-order and the other in which these two orders disagree. By a Sierpinski argument, at the $`\omega `$th level of $`T`$, there are pairs of both classes with meets on the same level. Since the coloring of these level sets depends only on the lengths of the pair and their meet, both $``$-similarity classes receive the same color. It follows that $`g`$ is monochromatic on pairs from the set of $`\alpha `$ for which $`\alpha `$ is a splitting level of $`T`$.
As a consequence of Theorem 1.1 and an earlier result of Hajnal and Komjáth () we obtain the following theorem which shows that the conclusion of Theorem 2.5 does not follow from any large cardinal hypothesis. This suggests that the use of an ultrafilter in a generic extension, as opposed to one in $`V`$, is a necessary part of the proof.
###### Theorem 8.5.
The conclusion of Theorem 2.5 does not follow from any large cardinal hypothesis on $`\kappa `$.
###### Proof.
Hajnal and Komjáth show that there is a forcing of size $`\mathrm{}_1`$ which adds an order type $`\theta `$ of size $`\mathrm{}_1`$ with the property that $`\psi \to ̸[\theta ]_{\omega _1}^2`$ for every order type $`\psi `$, regardless of its size. That is, there is a coloring of the pairs of $`\psi `$ into $`\mathrm{}_1`$ many colors such that every suborder of type $`\theta `$ gets all the colors. The conclusion of our Theorem 1.1 states, in contrast, that the ordering $`_\kappa `$ has, for any coloring of its pairs, a subset of the full order type $`_\kappa `$ which gets only finitely many colors. This subset contains subsets of every order type of size less than $`\kappa `$, since $`_\kappa `$ is a $`(<\kappa )`$-universal linear order. Hence Theorem 1.1 cannot hold in the Hajnal-Komjáth extension, and so Shelah’s theorem from which it is derived, cannot either. ∎
## 9 Upper bound for $`\kappa `$-Rado graphs
In this section we prove a limitation of colors result for $`\kappa `$-Rado graphs orders using Shelah’s Theorem 2.5. By a $`\kappa `$-Rado graph we mean a graph $`G`$ of size $`\kappa `$ with the property that for every two disjoint subsets $`A`$, $`B`$ of $`G`$, each of size $`<\kappa `$, there is $`cG`$ connected to all points of $`A`$ and no point of $`B`$. The existence of such $`G`$ follows from the assumption $`\kappa ^{<\kappa }=\kappa `$ is regular.
In order to apply Shelah’s Theorem, we need a method of embedding $`\kappa `$-Rado graphs into $`{}_{}{}^{\kappa >}2`$ and a well-ordering of the levels of $`{}_{}{}^{\kappa >}2`$ that is compatible with that embedding. We observe that any $`\kappa `$-Rado graph is isomorphic to one whose universe is $`\kappa `$, and generalize the approach used by Erdős, Hajnal and Pósa to embed such a graph into $`{}_{}{}^{\kappa >}2`$.
###### Definition 9.1.
Given a $`\kappa `$-Rado graph $`𝔾=(\kappa ,E)`$, the *tree embedding of $`𝔾`$ into $`{}_{}{}^{\kappa >}2`$* is the function $`\sigma _𝔾:\kappa {}_{}{}^{\kappa >}2`$ defined by $`\sigma _𝔾(0)=\mathrm{}`$, and for $`\alpha >0`$, $`\sigma _𝔾(\alpha ):\alpha 2`$ is defined by $`\sigma _𝔾(\alpha )(\beta )=1`$ if and only if $`\{\alpha ,\beta \}E`$.
###### Lemma 9.2.
For any $`\kappa `$-Rado graph $`𝔾=(\kappa ,E)`$, the range of the tree embedding $`\sigma `$ of $`𝔾`$ into $`{}_{}{}^{\kappa >}2`$ is a cofinal transverse subset of $`{}_{}{}^{\kappa >}2`$.
###### Proof.
By definition of $`\sigma `$, for all $`\alpha <\kappa `$, $`\mathrm{lg}(\sigma (\alpha ))=\alpha `$, so the range of $`\sigma `$ is transverse.
To see that the range is cofinal, suppose $`s{}_{}{}^{\kappa >}2`$. Let $`\alpha =\mathrm{lg}(s)`$ and let $`A`$ be the set of all $`\beta <\mathrm{lg}(s)`$ with $`s(\beta )=1`$. Since $`𝔾`$ is a $`\kappa `$-Rado graph, there is an element $`\gamma >\alpha `$ such that $`\{\beta ,\gamma \}E`$ for all $`\beta A`$ and $`\{\beta ,\gamma \}E`$ for all $`\beta \alpha A`$. It follows that $`s\sigma (\gamma )`$. Thus $`\sigma [\kappa ]`$ is cofinal in $`{}_{}{}^{\kappa >}2`$. ∎
Our next goal in this section is a translation of questions about isomorphisms of $`\kappa `$-Rado graphs into themselves to questions about $`{}_{}{}^{\kappa >}2`$. Toward that end, we define *passing number preserving maps*. This notion was used in the proof of the limitation of colors result by Laflamme, Sauer and Vuksanovic for the countable Rado graph, which is also known as the (infinite) random graph.
###### Definition 9.3.
For $`s,t{}_{}{}^{\kappa >}2`$ with $`|t|>|s|`$, call $`t(|s|)`$ the *passing number* of $`t`$ at $`s`$. Call a function $`f:{}_{}{}^{\kappa >}2{}_{}{}^{\kappa >}2`$ *passing number preserving* or a *pnp map* if it preserves
1. length order: $`\mathrm{lg}(s)<\mathrm{lg}(t)`$ implies $`\mathrm{lg}(f(s))<\mathrm{lg}(f(t))`$; and
2. passing numbers: $`\mathrm{lg}(s)<\mathrm{lg}(t)`$ implies $`f(t)(\mathrm{lg}(f(s)))=t(\mathrm{lg}(s))`$.
The first lemma states that any induced subgraph of a $`\kappa `$-Rado graph $`𝔾=(\kappa ,E)`$ has an induced subgraph which is isomorphic to the $`\kappa `$-Rado graph by an isomorphism that also preserves $`<`$.
###### Lemma 9.4.
For any cardinal $`\kappa `$ with $`\kappa ^{<\kappa }=\kappa `$, any $`\kappa `$-Rado graph $`𝔾=(\kappa ,E)`$ and any $`H\kappa `$ with $`𝔾(H,EH)`$ there is a $`<`$-increasing map $`g:\kappa H`$ with $`𝔾(g[\kappa ],Eg[\kappa ])`$.
###### Proof.
Fix attention on a particular $`\kappa `$-Rado graph $`𝔾=(\kappa ,E)`$ and a specific induced subgraph $`(H,EH)`$ isomorphic to $`𝔾`$. Let $`h:\kappa H`$ be the isomorphism.
Since $`\kappa =\kappa ^{<\kappa }`$, by the mapping extension property, for any $`\gamma <\kappa `$ and any subset $`A\gamma `$, there are cofinally many $`\zeta `$ with $`\{\delta <\gamma :\{\delta ,\zeta \}E\}=A`$.
Define $`z:\kappa \kappa `$ and $`g:\kappa H`$ by recursion. Let $`z(\mathrm{})=\mathrm{}`$ and $`g(\mathrm{})=h(\mathrm{})`$. Suppose $`z\alpha `$ and $`g\alpha `$ have been defined such that $`z`$ is increasing, for all $`\beta <\alpha `$, $`g(\beta )=h(z(\beta ))`$, and $`g\alpha `$ is an increasing isomorphism of $`(\alpha ,E\alpha )`$ into $`(H,EH)`$. Let $`\gamma >\alpha `$ be so large that if $`h(\eta )<sup\{\mathrm{lg}(g(\beta ))+1:\beta <\alpha \}`$, then $`\gamma >\eta `$. Let $`A_\alpha :=\{z(\beta ):\beta <\alpha \{\beta ,\alpha \}E\}`$. Let $`z(\alpha )\gamma `$ be such that $`\{\delta <\gamma :\{\delta ,\zeta \}E\}=A_\alpha `$. Let $`g(\alpha )=h(z(\alpha ))`$. Since $`h`$ is an isomorphism, a pair $`\{h(z(\beta )),h(z(\alpha ))\}`$ is in $`E`$ if and only if the pair $`\{z(\beta ),z(\alpha )\}`$ is in $`E`$. It follows that $`\{\beta <\alpha :\{g(\beta ),g(\alpha )\}E\}=A_\alpha `$. Therefore by induction, $`g`$ is the desired increasing isomorphism into $`H`$. ∎
###### Lemma 9.5.
Suppose $`𝔾=(\kappa ,E)`$ is a $`\kappa `$-Rado graph with tree embedding $`\sigma `$ and $`S=\sigma [\kappa ]`$. For any $`<`$-increasing map $`g:\kappa \kappa `$ with $`𝔾(g[\kappa ],Eg[\kappa ])`$, the composition $`\sigma g\sigma ^1:SS`$ is a pnp map.
###### Proof.
Let $`f:=\sigma g\sigma ^1`$ for some $`<`$-increasing isomorphism of $`𝔾`$ into itself. Suppose $`s,tS`$ and $`\beta :=\mathrm{lg}(s)<\mathrm{lg}(t)=\alpha `$. Then $`\sigma ^1(s)=\beta `$ and $`\sigma ^1(t)=\alpha `$. Since $`g`$ is $`<`$-increasing, $`g(\beta )<g(\alpha )`$. Hence $`\mathrm{lg}(f(s))=g(\beta )<g(\alpha )=\mathrm{lg}(f(t))`$. Moreover, $`t(\mathrm{lg}(s))=t(\beta )=1`$ if and only if $`\{\beta ,\alpha \}E`$. Since $`g`$ is an isomorphism, $`t(\mathrm{lg}(s))=1`$ if and only if $`\{g(\beta ),g(\alpha )\}E`$. By definition of tree embedding, it follows that $`t(\mathrm{lg}(s))=1`$ if and only if $`f(t)(\mathrm{lg}(s))=1`$. Thus $`f`$ is a pnp map from $`S`$ into $`S`$. ∎
By much the same reasoning, one can show the converse.
###### Theorem 9.6.
\[Translation Theorem\] Suppose $`𝔾=(\kappa ,E)`$ is a $`\kappa `$-Rado graph with tree embedding $`\sigma `$ and $`S=\sigma [\kappa ]`$. For any pnp map $`f:SS`$, the composition $`g:=\sigma ^1f\sigma :\kappa \kappa `$ is an $`<`$-increasing map with $`𝔾(g[\kappa ],Eg[\kappa ])`$.
###### Proof.
Let $`g:=\sigma ^1f\sigma `$ for some pnp map $`f:SS`$. Suppose $`\beta <\alpha <\kappa `$. Then $`\mathrm{lg}(\sigma (\beta ))=\beta <\alpha =\mathrm{lg}(\sigma (\alpha ))`$. Since $`f`$ is a pnp map, $`\sigma ^1f\sigma (\beta )=\mathrm{lg}(f(\sigma (\beta )))<\mathrm{lg}(f(\sigma (\alpha )))=\sigma ^1f\sigma (\alpha )`$, so $`g`$ is a $`<`$-increasing map.
By the definition of the tree embedding, $`\{\beta ,\alpha \}`$ is an edge of $`𝔾`$ if and only if $`\sigma (\alpha )(\mathrm{lg}(\sigma (beta)))=1`$. Since $`f`$ is a pnp map, it follows that $`\{\beta ,\alpha \}`$ is an edge of $`𝔾`$ if and only if $`f(\sigma (\alpha ))(\mathrm{lg}(f(\sigma (beta))))=1`$. Apply the definition of tree embedding to $`g(\alpha )=\sigma ^1(f(\sigma (\alpha )))`$ and $`g(\beta )=\sigma ^1(f(\sigma (\beta )))`$, to see that $`\{\beta ,\alpha \}`$ is an edge of $`𝔾`$ if and only if $`\{g(\beta ),g(\alpha )\}`$ is an edge.
Thus $`g`$ is a $`<`$-increasing isomorphism of $`𝔾`$ into itself. ∎
The next definition identifies sufficient conditions for a map to carry a strongly diagonal set to one of the same $`m`$-type.
###### Definition 9.7.
Call a map $`f:{}_{}{}^{\kappa >}2{}_{}{}^{\kappa >}2`$ *polite* if it satisfies the following conditions for all $`x,y,u,v`$:
1. (preservation of lexicographic order) if $`x`$ and $`y`$ are incomparable and $`x<_{\text{lex}}y`$, then $`f(x)`$ and $`f(y)`$ are incomparable and $`f(x)<_{\text{lex}}(y)`$;
2. (meet regularity) if $`\{x,u,v\}`$ is diagonal and $`xu=xv`$, then $`f(x)f(u)=f(x)f(v)`$;
3. (preservation of meet length order) if $`\mathrm{lg}(xy)<\mathrm{lg}(uv)`$, then $`\mathrm{lg}(f(x)f(y))<\mathrm{lg}(f(u)f(v))`$.
Call it *polite to strongly diagonal sets* if it is a pnp map which satisfies the above conditions for all $`x,y,u,v`$ with $`\{x,y,u,v\}`$ a strongly diagonal set.
The next lemma follows immediately from the above definition.
###### Lemma 9.8.
Strong embeddings are polite and polite to strongly diagonal sets. The collection of polite embeddings is closed under composition as is the collection of embeddings polite to strongly diagonal sets.
###### Lemma 9.9.
Suppose $`\varphi :{}_{}{}^{\kappa >}2{}_{}{}^{\kappa >}2`$ is a map which is polite to strongly diagonal sets and whose image is a strongly diagonal set. Then for any strongly diagonal set $`A`$, $`\mathrm{clp}(A)=\mathrm{clp}(\varphi [A])`$ and there is a pnp map $`\overline{\varphi }:A^{}(\varphi [A])^{}`$ such that for all $`x,y`$ in $`A`$, $`\overline{\varphi }(xy)=\varphi (x)\varphi (y)`$.
###### Proof.
Fix a strongly diagonal set $`A`$. Let $`\overline{\varphi }(a)=\varphi (a)`$ for $`aA`$ and let $`\overline{\varphi }(ab)=\varphi (a)\varphi (b)`$ for $`a,bA`$. The proof of the lemma for $`A`$ proceeds by a series of claims.
###### Claim 9.9.a.
The map $`\overline{\varphi }`$ is well-defined.
###### Proof.
We must show that for all $`x,y,u,vA`$, if $`xy=uv`$, then $`\varphi (x)\varphi (y)=\varphi (u)\varphi (v)`$. If $`x=y`$ or $`u=v`$, then $`x=y=u=v`$ since $`A`$ is diagonal. In this case the claim follows immediately, so assume $`xy`$ and $`uv`$. If $`\{x,y\}\{u,v\}`$ is non-empty, then the claim follows from the assumption of meet regularity. So assume $`\{x,y\}\{u,v\}`$ is empty. Since $`\{x,y,u\}`$ is a three element diagonal set and $`xy`$ is an initial segment of all three elements, either $`xy=xu`$ or $`xy=yu`$. Hence by meet regularity, either $`\varphi (x)\varphi (y)=\varphi (x)\varphi (u)=\varphi (u)\varphi (v)`$ or $`\varphi (x)\varphi (y)=\varphi (y)\varphi (u)=\varphi (u)\varphi (v)`$, and the claim follows. ∎
###### Claim 9.9.b.
The map $`\overline{\varphi }`$ preserves length order.
###### Proof.
Assume $`s,tA^{}`$ satisfy $`\mathrm{lg}(s)<\mathrm{lg}(t)`$. Let $`x,yA`$ be such that $`s=xy`$ and $`u,vA`$ be such that $`t=uv`$. By preservation of meet length order, since $`\mathrm{lg}(xy)<\mathrm{lg}(uv)`$, it follows that $`\mathrm{lg}(\overline{\varphi }(s))=\mathrm{lg}(\varphi (x)\varphi (y))<\mathrm{lg}(\varphi (u)\varphi (v))=\mathrm{lg}(\overline{\varphi }(t))`$. ∎
###### Claim 9.9.c.
The map $`\overline{\varphi }`$ is a pnp map such that for all $`x,y`$ in $`A`$, $`\overline{\varphi }(st)=\varphi (s)\varphi (t)`$.
###### Proof.
By the definition of $`\overline{\varphi }`$ and the previous claims, it is enough to show that $`\overline{\varphi }`$ preserves passing numbers. Suppose $`s`$ and $`t`$ are in $`A^{}`$ and $`\mathrm{lg}(s)<\mathrm{lg}(t)`$. Since any element $`t^{}`$ of the closure of $`A`$ is either in $`A`$ or has a proper extension $`t`$ in $`A`$ with $`\overline{\varphi }(t^{})\overline{\varphi }(t)=\varphi (t)`$, we may assume without loss of generality that $`t`$ is in $`A`$. If $`sA`$, then the conclusion follows since $`\varphi `$ is a pnp map.
So suppose $`s=xy`$ for $`x`$ and $`y`$ distinct elements of $`A`$. Consider $`st`$. Either it has the same length as $`s`$ or it is shorter.
If $`\mathrm{lg}(st)<\mathrm{lg}(s)`$, then $`\mathrm{lg}(\overline{\varphi }(st))<\mathrm{lg}(\overline{\varphi }(s))`$ by the previous claim. In this case, $`\overline{\varphi }(t)(\mathrm{lg}(\overline{\varphi }(s)))=0=t(\mathrm{lg}(s))`$, since $`A`$ and its image under $`\varphi `$ are both strongly diagonal.
If $`\mathrm{lg}(st)=\mathrm{lg}(s)`$, then $`st`$. Let $`wA`$ be such that $`s=tw`$. Then the value of $`t(\mathrm{lg}(s))`$ and $`\overline{\varphi }(t)(\mathrm{lg}(\overline{\varphi }(s)))`$ are determined by the lexicographic order of the pairs $`t,w`$ and $`\varphi (t),\varphi (w)`$. Since $`\varphi `$ preserves lexicographic order, $`\overline{\varphi }(t)(\mathrm{lg}(\overline{\varphi }(s)))=t(\mathrm{lg}(s))`$, as required. ∎
###### Claim 9.9.d.
$`\mathrm{clp}(A)=\mathrm{clp}(\varphi [A])`$.
###### Proof.
Enumerate $`A^{}`$ in increasing order of length as $`a_\alpha :\alpha <\mu `$ for some $`\mu <\kappa `$. Let $`B=\varphi [A]`$. Then $`B`$ is a strongly diagonal set since it is a subset of a strongly diagonal set. Enumerate $`B^{}`$ in increasing order of length as $`b_\beta :\beta <\nu `$ for some $`\nu <\kappa `$.
Since $`\overline{\varphi }`$ is a pnp map that carries $`A^{}`$ onto $`B^{}`$, it is a bijection from $`A^{}`$ to $`B^{}`$. Since the order type of $`\{aA^{}:\mathrm{lg}(a)<\mathrm{lg}(a_\alpha )\}`$ is $`\alpha `$ and the order type of $`\{bB^{}:\mathrm{lg}(b)<\mathrm{lg}(b_\beta )\}`$ is $`\beta `$, it follows that $`\overline{\varphi }(a_\alpha )=b_\beta `$ and $`\mu =\nu `$.
For $`\alpha <\mu `$, let $`A_\alpha :=\{a\mathrm{lg}(a_\alpha ):aA\}`$ and $`B_\alpha :=\{b\mathrm{lg}(b_\alpha ):bB\}`$. Let $`A_\mu =A`$ and $`B_\mu =B`$. Use induction to prove that for all positive $`\alpha \mu `$, $`\mathrm{clp}(A_\alpha )=\mathrm{clp}(B_\alpha )`$. To start the induction, observe that $`|A_0|=|B_0|=1`$, so $`\mathrm{clp}(A_0)=\left\{\mathrm{}\right\}=\mathrm{clp}(B_0)`$. For the limit case, assume $`\alpha `$ is a limit ordinal and for all $`\beta <\alpha `$, $`\mathrm{clp}(A_\beta )=\mathrm{clp}(B_\beta )`$. In this case, $`\mathrm{clp}(A_\alpha )=\mathrm{clp}(B_\alpha )`$, since $`\mathrm{clp}(A_\alpha )=\{\mathrm{clp}(A_\beta ):\beta <\alpha \}`$ and $`\mathrm{clp}(B_\alpha )=\{\mathrm{clp}(B_\beta ):\beta <\alpha \}`$. For the successor case, assume $`\alpha =\beta +1`$ and $`\mathrm{clp}(A_\beta )=\mathrm{clp}(B_\beta )`$. Let $`\sigma =\sigma _\alpha `$ be the increasing enumeration of $`\{\mathrm{lg}(a_\gamma ):\gamma <\alpha \}`$ and let $`\tau =\tau _\alpha `$ be the increasing enumeration of $`\{\mathrm{lg}(b_\gamma ):\gamma <\alpha \}`$. Then $`\mathrm{clp}(A_\alpha )`$ is the similarity tree whose set of leaves is $`\{a\sigma :aA_\alpha \}`$ and $`\mathrm{clp}(B_\alpha )`$ is the similarity tree whose set of leaves is $`\{b\sigma :bB_\alpha \}`$. Moreover $`\mathrm{clp}(A_\beta )`$ is the similarity tree whose set of leaves is $`\{a(\sigma \beta ):aA_\beta \}`$ and $`\mathrm{clp}(B_\beta )`$ is the similarity tree whose set of leaves is $`\{b(\sigma \beta ):bB_\beta \}`$. For $`aA_\alpha A_\beta `$, $`a\sigma (\beta )`$ is the passing number of $`a`$ at $`\mathrm{lg}(a_\beta )`$. Since $`\overline{\varphi }`$ preserves passing numbers and carries elements of $`A`$ to elements of $`B`$ and elements of $`A^{}A`$ to elements of $`B^{}B`$, it follows that $`\mathrm{clp}(A_\alpha )=\mathrm{clp}(B_\alpha )`$ in this case as well. Therefore by induction, $`\mathrm{clp}(A^{})=\mathrm{clp}(B^{})`$, and since $`\mathrm{clp}(A)=\mathrm{clp}(A^{})`$ and $`\mathrm{clp}(B)=\mathrm{clp}(B^{})`$, the claim follows. ∎
Since $`A`$ was an arbitrary strongly diagonal set, by the last two claims, the lemma follows. ∎
Our next goal is to prove the existence of a *pnp diagonalization*.
###### Definition 9.10.
Suppose $`w{}_{}{}^{\kappa >}2`$ and $`S{}_{}{}^{\kappa >}2`$. Call $`f`$ a *pnp diagonalization into $`S\mathrm{Cone}(w)`$* if $`f`$ is a polite injective $`<_\text{Q}`$-preserving pnp map whose range is a strongly diagonal subset $`D`$ with $`D^{}S\mathrm{Cone}(w)`$. Call $`f`$ a *pnp diagonalization* if it is a pnp diagonalization into $`{}_{}{}^{\kappa >}2\mathrm{Cone}(\mathrm{})`$.
An extra quality we desire for our diagonalization is *level harmony*, which will be used in the section on lower bounds.
###### Definition 9.11.
Suppose $`f:{}_{}{}^{\kappa >}2{}_{}{}^{\kappa >}2`$ is an injective map. Define $`\widehat{f}:{}_{}{}^{\kappa >}2{}_{}{}^{\kappa >}2`$ by $`\widehat{f}(s)=f(s{}_{}{}^{}0)f(s{}_{}{}^{}1)`$. The function *$`f`$ has level harmony* if $`\widehat{f}`$ is an extension and $`<_{\text{lex}}`$-order preserving map such that for all $`s,t{}_{}{}^{\kappa >}2`$, the following conditions hold:
1. $`\widehat{f}(s)f(s)`$;
2. $`\mathrm{lg}(s)<\mathrm{lg}(t)`$ implies $`\mathrm{lg}(f(s))<\mathrm{lg}(\widehat{f}(t))`$;
3. $`\mathrm{lg}(s)=\mathrm{lg}(t)`$ implies $`\mathrm{lg}(\widehat{f}(s))<\mathrm{lg}(f(t))`$.
###### Lemma 9.12.
\[Second diagonalization lemma\] Suppose $`w{}_{}{}^{\kappa >}2`$ and that $`S{}_{}{}^{\kappa >}2`$ is cofinal and transverse. Then there is a pnp diagonalization into $`S\mathrm{Cone}(w)`$ which has level harmony.
###### Proof.
Our plan is to approach the problem in pieces by using recursion to define three functions, $`\phi _0,\phi _1,\phi :{}_{}{}^{\kappa >}2S`$ so that $`\phi `$ is the desired diagonalization, $`\widehat{\phi }=\phi _0`$, $`\phi _1(t)`$ is the minimal extension in $`S`$ of $`\phi _0{}_{}{}^{}1`$, $`\phi (t)\phi (t{}_{}{}^{}1)=\phi _1(t)`$, and $`\phi (t)<_{\text{lex}}\phi (t{}_{}{}^{}1)`$.
For notational convenience, if $`\phi `$ has been defined on $`{}_{}{}^{\alpha >}2`$, then we let $`\mathrm{}_0(\alpha )`$ be the least $`\theta `$ such that $`\mathrm{lg}(\phi (t))<\theta `$ for all $`t{}_{}{}^{\alpha >}2`$. Also, if $`\phi _1`$ has been defined on $`{}_{}{}^{\alpha }2`$, then we let $`\mathrm{}_1(\alpha )`$ be the least $`\theta `$ such that $`\mathrm{lg}(\phi _1(t))<\theta `$ for all $`t{}_{}{}^{\alpha }2`$.
Let $``$ be a well-ordering of the levels of $`{}_{}{}^{\kappa >}2`$. We use recursion on $`\alpha <\kappa `$ to define the restrictions to $`{}_{}{}^{\alpha }2`$ of $`\phi _0`$, $`\phi _1`$ and $`\phi `$ so that the following properties hold:
1. extension and lexicographic order:
1. the restriction of $`\phi _0`$ to $`{}_{}{}^{\alpha }2`$ is extension and $`<_{\text{lex}}`$-order preserving;
2. for all $`s{}_{}{}^{\alpha }2`$, $`\phi _1(s)`$ is the minimal extension in $`S\mathrm{Cone}(w)`$ of $`\phi _0(s){}_{}{}^{}1`$;
3. for all $`s{}_{}{}^{\alpha }2`$, $`\phi (s)`$ is an extension in $`S\mathrm{Cone}(w)`$ of $`\phi _1(s){}_{}{}^{}0`$;
4. for all $`s{}_{}{}^{\alpha >}2`$, $`\phi _0(s{}_{}{}^{}0)`$ is an extension of $`\phi _0(s){}_{}{}^{}0`$ and
$`\phi _0(s{}_{}{}^{}1)`$ is an extension of $`\phi _1(s){}_{}{}^{}1`$;
2. length order:
1. for all $`t{}_{}{}^{\alpha }2`$ and $`s{}_{}{}^{\alpha }2`$, if $`st`$, then
$`\mathrm{}_0(\mathrm{lg}(s))\mathrm{lg}(\phi _0(s))<\mathrm{lg}(\phi _0(t))`$ and
$`\mathrm{}_1(\mathrm{lg}(s))\mathrm{lg}(\phi (s))<\mathrm{lg}(\phi (t))`$;
3. passing number:
1. for all $`t{}_{}{}^{\alpha }2`$ and $`s{}_{}{}^{\alpha }2`$, if $`st`$ and $`st`$, then
$`\phi _0(t)(\mathrm{lg}(\phi _0(s))=0`$ and $`\phi _0(t)(\mathrm{lg}(\phi _1(s))=0`$;
2. for all $`t{}_{}{}^{\alpha }2`$ and $`s{}_{}{}^{\alpha }2`$, if $`st`$ and $`st`$, then
$`\phi (t)(\mathrm{lg}(\phi _0(s))=0`$ and $`\phi (t)(\mathrm{lg}(\phi _1(s))=0`$;
3. for all $`t{}_{}{}^{\alpha }2`$ and $`s{}_{}{}^{\alpha >}2`$, $`\phi _0(t)(\mathrm{lg}(\phi (s))=t(\mathrm{lg}(s))`$;
4. for all $`s,t{}_{}{}^{\alpha }2`$, if $`st`$, then $`\phi (t)(\mathrm{lg}(\phi (s))=0`$.
Suppose $`\alpha <\kappa `$ is arbitrary and for all $`\beta <\alpha `$, the restrictions to $`{}_{}{}^{\beta }2`$ of $`\phi _0`$, $`\phi _1`$ and $`\phi `$ have been defined. To maintain length order, we first define $`\phi _0`$ and $`\phi _1`$ by recursion on $``$ restricted to level $`{}_{}{}^{\alpha }2`$. So suppose $`\mathrm{lg}(t)=\alpha `$ and for all $`st`$, $`\phi _0`$ and $`\phi _1(s)`$ have been defined.
Use extension and $`<_{\text{lex}}`$-order properties to identify an element $`\phi _0^{}(t)`$ of which $`\phi _0(t)`$ is to be an extension. If $`\alpha =0`$, set $`\phi _0^{}(t)=\mathrm{}`$. If $`\alpha `$ is a limit ordinal, let $`\phi _0^{}(t)`$ be $`\{\phi _0(t\beta ):\beta <\alpha \}`$. If $`\alpha `$ is a successor ordinal and $`t=t^{}{}_{}{}^{}\delta `$, let $`\phi _0^{}(t)=\phi _\delta (t^{}){}_{}{}^{}\delta `$.
Next determine an ordinal $`\gamma _0(t)`$ sufficiently large that if $`\phi _0(t)`$ is at least that length, it will satisfy the length order property. If $`t`$ is the $``$-least element of length $`\alpha `$, let $`\gamma _0(t)=\mathrm{}_0(\alpha )`$. If $`t`$ has a $``$-immediate predecessor $`t^{}`$ of length $`\alpha `$, let $`\gamma _0(t)=\mathrm{lg}(\phi _1(t^{}))+1`$. If $`t`$ is a $``$-limit of elements of length $`\alpha `$, then let $`\gamma _0(t)`$ be the supremeum of $`\mathrm{lg}(\phi _1(s))+1`$ for $`s`$ of length $`\alpha `$ with $`st`$.
Next define an extension $`\phi _0^+(t)`$ of $`\phi _0^{}(t)`$ of length $`\gamma _0(t)`$ so that the passing number properties are satisfied by $`\phi _0^+(t)`$. If $`\alpha =0`$, then $`t=\mathrm{}`$, $`\phi _0^{}(t)=\mathrm{}`$, $`\gamma _0(0)=0`$ and $`\phi _0^+(t)=\mathrm{}`$. If $`\alpha >0`$ is a limit ordinal, then by induction on $`\beta <\alpha `$, $`\mathrm{}_0(\beta )`$ is an increasing sequence. Moreover, the limit of this sequence is the length of $`\phi _0^{}(t)`$. It follows that $`\phi _0^{}(t)`$ satisfies the passing number properties for $`s{}_{}{}^{\alpha >}2`$. Let $`\phi _0^+(t)`$ be the sequence extending $`\phi _0^{}(t)`$ by zeros, as needed, to a length of $`\gamma _0(t)`$. If $`\alpha `$ is a successor ordinal and $`t=t^{}{}_{}{}^{}\delta `$, then let $`\phi _0^+(t)`$ be the extension of $`\phi _0^{}(t)`$ of length $`\gamma _0(t)`$ such that for all $`\eta `$ with $`\mathrm{lg}(\phi _0^{}(t))\eta <\gamma _0(t)`$, $`\phi _0^+(\eta )=\delta `$ if $`\eta =\mathrm{lg}(\phi (s))`$ for some $`s`$ with $`\mathrm{lg}(s)+1=\alpha `$, and $`\phi _0^+(\eta )=0`$ otherwise.
Next let $`\phi _0(t)`$ be an extension in $`S\mathrm{Cone}(w)`$ of $`\phi ^+(t)`$ and let $`\phi _1(t)`$ be an extension in $`S\mathrm{Cone}(w)`$ of $`\phi _0(t){}_{}{}^{}1`$ as required by the extension and lexicographic order properties. The careful reader may now check that the various properties hold for the restrictions of $`\phi _0`$ and $`\phi _1`$ to $`{}_{}{}^{\alpha }2`$.
Use a similar process to define the restriction of $`\phi `$ to $`{}_{}{}^{\alpha }2`$ by recursion on $``$ restricted to $`{}_{}{}^{\alpha }2`$. Suppose that $`\mathrm{lg}(t)=\alpha `$ and for all $`st`$, $`\phi (s)`$ has been defined. Let $`\phi ^{}(t)=\phi _1(t){}_{}{}^{}0`$.
If $`t`$ is the $``$-least element of length $`\alpha `$, let $`\gamma _1(t)=\mathrm{}_1(\alpha )`$. If $`t`$ has a $``$-immediate predecessor $`t^{}`$ of length $`\alpha `$, let $`\gamma _1(t)=\mathrm{lg}(\phi (t^{}))+1`$. If $`t`$ is a $``$-limit of elements of length $`\alpha `$, then let $`\gamma _1(t)`$ be the supremum of $`\mathrm{lg}(\phi (s))+1`$ for $`s`$ of length $`\alpha `$ with $`st`$.
Next define an extension $`\phi ^+(t)`$ of $`\phi ^{}(t)`$ of length $`\gamma _1(t)`$ so that the passing number properties are satisfied by $`\phi ^+(t)`$. If $`\alpha =0`$, there are no passing number properties that need be checked, and we set $`\phi ^+(t)=\phi ^{}(t)`$. If $`\alpha >0`$, then let $`\phi ^+(t)`$ be the extension by zeros of $`\phi ^{}(t)`$ of length $`\gamma _1(t)`$. Since $`\phi _0(t)`$ and $`\phi _1(t)`$ satisfy the passing numbers properties, it follows that $`\phi ^+(t)`$ does as well, since all passing numbers longer than $`\mathrm{lg}(\phi ^{}(t))`$ will be zero.
Finally let $`\phi (t)`$ be an extension in $`S\mathrm{Cone}(w)`$ of $`\phi ^+(t)`$. The careful reader may now check that the various properties hold for the restriction of $`\phi `$ to $`{}_{}{}^{\alpha }2`$.
This completes the recursive construction of $`\phi _0`$, $`\phi _1`$ and $`\phi `$. By induction, the various properties hold for all $`\alpha <\kappa `$.
Thus $`\phi _0=\widehat{\phi }`$ is extension and $`<_{\text{lex}}`$-order preserving, and by the length order property, different elements of the union of the ranges of $`\phi _0`$, $`\phi _1`$ and $`\phi `$ have different lengths. It also follows that these three maps are injective. Moreover the union of their ranges is a subset of $`S\mathrm{Cone}(w)`$. By the extension and lexicographic order properties and the length order property, $`\phi `$ has level harmony.
By the passing number properties, $`\phi `$ is a pnp map. By the extension and lexicographic order properties, $`\phi `$ preserves $`<_\text{Q}`$-order.
By the extension and lexicographic order properties, $`\phi `$ carries incomparable elements into incomparable elements and preserves $`<_{\text{lex}}`$-order. By the length order property, $`\phi `$ preserves meet length order. Since $`\widehat{\phi }=\phi _0`$ preserves extension, $`\phi `$ satisfies meet regularity. Thus $`\phi `$ is polite.
From the extension and lexicographic order properties, it follows that the meet closure of $`D:=\mathrm{ran}(\phi )`$ is the union of the ranges of $`\phi `$, $`\phi _0`$ and $`\phi _1`$ and all elements of the range of $`\phi `$ are incomparable. Hence $`D`$ is an antichain and $`D^{}S\mathrm{Cone}(w)`$ is transverse, so $`D`$ is diagonal. Note that passing numbers of $`1`$ were introduced only to keep $`\phi _0`$ extension and $`<_{\text{lex}}`$-order preserving, to ensure $`\phi (t)<_{\text{lex}}\phi _1(t)`$ so that $`<_\text{Q}`$-order is preserved, and to ensure $`\phi `$ is a pnp map. It follows that $`D^{}`$ is strongly diagonal.
Therefore, $`\phi `$ is the required pnp diagonalization into $`S\mathrm{Cone}(w)`$ with level harmony. ∎
###### Lemma 9.13.
Suppose $`S{}_{}{}^{\kappa >}2`$ is cofinal and transverse. Then there are a diagonal set $`D`$, maps $`\phi _t:t{}_{}{}^{\kappa >}2`$ and a pre-$`S`$-vip order $``$ such that for all $`t{}_{}{}^{\kappa >}2`$, the following conditions hold:
1. $`\phi _t`$ is a pnp diagonalization with level harmony into $`S\mathrm{Cone}(t)`$;
2. the meet closure of the set $`D_t:=\mathrm{ran}(\phi _t)`$ is a subset of $`S`$ disjoint from $`D_s^{}`$ for all $`st`$; and
3. $``$ is a $`D_t`$-vip order.
###### Proof.
Use Lemma 3.5 to find $`^{}`$, a pre-$`S`$-vip order on $`{}_{}{}^{\kappa >}2`$. By Lemma 3.11, $`S`$ is dense.
Apply the Second Diagonalization Lemma 9.12 to each $`t{}_{}{}^{\kappa >}2`$ to obtain $`\phi _t^{}`$, a pnp diagonalization into $`S\mathrm{Cone}(t)`$ which has level harmony. Use recursion on $``$ to define $`\pi :{}_{}{}^{\kappa >}2{}_{}{}^{\kappa >}2`$, $`\phi _t:t{}_{}{}^{\kappa >}2`$ and $`D_t:t{}_{}{}^{\kappa >}2`$ such that for all $`t{}_{}{}^{\kappa >}2`$, $`D_t:=\mathrm{ran}(\phi _t)`$, $`\pi (t)`$ is an extension of $`t`$ with $`\mathrm{Cone}(\pi (t))`$ disjoint from the union over all $`st`$ of $`D_s^{}`$. Since the order type of $`\{s{}_{}{}^{\kappa >}2:st\}`$ is less than $`\kappa `$ and each $`D_s`$ is a strongly diagonal set whose meet closure is a of $`S`$, it is always possible to continue the recursion.
Use induction on the recursive construction to show that the meet closures of the sets $`D_t`$ are disjoint.
Let $`D=\{D_t^{}:t{}_{}{}^{\kappa >}2\}`$. Then $`D`$ is transverse since it is a subset of $`S`$ and $`S`$ is transverse. Let $``$ agree with $`^{}`$ on all pairs from different levels, and use recursion on $`\alpha <\kappa `$ to define $``$ from $`^{}`$ as follows. If there is no element of $`D`$ in $`{}_{}{}^{\alpha }2`$ then $``$ and $`^{}`$ agree on $`{}_{}{}^{\alpha }2`$.
So suppose $`dD_t^{}`$ and $`\mathrm{lg}(d)=\alpha `$. For each $`\beta <\alpha `$, let $`C(\beta )`$ be the set of all $`x{}_{}{}^{\alpha }2`$ such that $`x\beta =d\beta `$ and $`x(\beta )d(\beta )`$. Since $`^{}`$ is a pre-$`S`$-vip order, if $`\beta <\gamma <\alpha `$, then $`C(\beta )^{}C(\gamma )`$ in the sense that for every element $`x`$ of $`C(\beta )`$ and $`y`$ of $`C(\gamma )`$, one has $`x^{}y`$. Use the fact that $`D_t`$ is a diagonal set to partition $`{}_{}{}^{\alpha }2=\{d\}A_\alpha (0)A_\alpha (1)A_\alpha (2)`$ into disjoint pieces where for $`\delta <2`$, $`A_\alpha (\delta ):=\{u\alpha :uD_tu(\alpha )=\delta \}`$. Let the restriction of $``$ to $`{}_{}{}^{\alpha }2`$ be such that for each $`\beta <\alpha `$,
$$C(\beta )A_\alpha (0)C(\beta )A_\alpha (1)C(\beta )A_\alpha (2)$$
and otherwise $``$ agrees with $`^{}`$. Then the restriction of $``$ to $`{}_{}{}^{\alpha }2`$ is a well-order, since the restriction of $`^{}`$ is and because $`^{}`$ is a pre-$`S`$-vip order.
Since the restriction of $``$ to each level is a well-order, it follows that $``$ is a well-ordering of the levels of $`{}_{}{}^{\kappa >}2`$. For each $`t{}_{}{}^{\kappa >}2`$, since $`^{}`$ is a pre-$`S`$-vip order, it follows that $``$ is a pre-$`D_{t}^{}{}_{}{}^{}`$-vip order, so by construction, $``$ is a $`D_t`$-vip order. ∎
###### Theorem 9.14.
Let $`m2`$ and suppose that $`\kappa `$ is a cardinal which is measurable in the generic extension obtained by adding $`\lambda `$ Cohen subsets of $`\kappa `$, where $`\lambda (\kappa )_{2^\kappa }^{2m}`$. Then for $`r_m^+`$ equal to the number of vip $`m`$-types, any $`\kappa `$-Rado graph $`𝔾=(\kappa ,E)`$ satisfies
$$𝔾(𝔾)_{<\kappa ,r_m^+}^m.$$
###### Proof.
Let $`\sigma :\kappa {}_{}{}^{\kappa >}2`$ be the tree embedding and set $`S=\mathrm{ran}\sigma `$. Then by Lemma 9.2, $`S`$ is cofinal and transverse. Apply Lemma 9.13 to obtain a pre-$`S`$-vip order $``$ and a sequence $`\phi _t:t{}_{}{}^{\kappa >}2`$ such that for all $`t{}_{}{}^{\kappa >}2`$, the three listed properties of the lemma hold.
Fix a coloring $`c:[\kappa ]^m\mu `$ where $`\mu <\kappa `$. Define $`d:[{}_{}{}^{\kappa >}2]^m\mu `$ by $`d(z)=c(\sigma ^1[\phi _0[z]])`$.
Apply Shelah’s Theorem 2.5 to the restriction of $`d`$ to $`m`$-element antichains to obtain a strong embedding $`e`$ and a node $`w`$ such that $`e`$ preserves $``$ on $`\mathrm{Cone}(w)`$ and $`d(e[a])=d(e[b])`$ for all $``$-similar $`m`$-element antichains $`a`$ and $`b`$ of $`\mathrm{Cone}(w)`$.
Now $`D_w:=\mathrm{ran}(\phi _w)`$ is a strongly diagonal set and $``$ is $`D_w`$-vip. Let $`D=e[\phi _w[S]]`$. Since $`\phi _w[S]`$ is a subset of $`D_w`$, $`\phi _w[S]`$ is a strongly diagonal set since $`D_w`$ is. Also, $``$ is $`\phi _w[S]`$-vip level order on the downwards closure under initial segments of $`\phi _w[S]`$, since $``$ is a $`D_w`$-vip level order. Since $`e`$ is a strong embedding, $`D`$ is a strongly diagonal set. Since $`e`$ preserves $``$ on $`\mathrm{Cone}(w)`$ and $`\phi _w[S]D_w\mathrm{Cone}(w)`$, it follows that the restriction of $``$ to the downward closure of $`D`$ is a $`D`$-vip order. Hence for all $`x[D]^m`$, the ordered similarity type $`(\mathrm{clp}(x),_x)`$ is a vip $`m`$-type.
Finally let $`K:=\sigma ^1[\phi _0[D]]=\sigma ^1\phi _0e\phi _w\sigma [\kappa ]`$. Note that $`D=\phi _0^1[\sigma [K]]`$, so for all $`m`$-element subsets $`u`$ of $`K`$, the image, $`x=\phi _0^1[\sigma [K]][u]D`$, is a strongly diagonal set, and $`(\mathrm{clp}(x),_x)`$ is a vip $`m`$-type.
Since $`\phi _0`$, $`e`$ and $`\phi _w`$ are all pnp maps, so is their composition. By the First Translation Theorem 9.6, this mapping is an isomorphism of $`𝔾`$ into itself.
We claim that $`c`$ is constant on $`m`$-element subsets of $`K`$ whose images under $`\phi _0^1\sigma `$ are $``$-similar. Consider two such $`m`$-element subsets $`u,v`$ of $`K`$. Let $`a^{}`$ and $`b^{}`$ be the subsets of $`\kappa `$ for which $`\sigma ^1\phi _0e\phi _w\sigma [a^{}]=u`$ and $`\sigma ^1\phi _0e\phi _w\sigma [b^{}]=v`$. Let $`a=\phi _w[\sigma [a^{}]]`$ and $`b=\phi _w[\sigma [b^{}]]`$. Then $`a`$ and $`b`$ are $`m`$-element strongly diagonal subsets of $`\mathrm{Cone}(w)`$. Notice that $`u=\sigma ^1[\phi _0[e[a]]]`$ and $`v=\sigma ^1[\phi _0[e[b]]]`$. Thus $`e[a]`$ and $`e[b]`$ are $``$-similar. Since $`e`$ is a strong embedding which preserves $``$ on $`\mathrm{Cone}(w)`$, it follows that $`a`$ and $`b`$ are $``$-similar. By the application of Shelah’s Theorem above, $`d(e[a])=d(e[b])`$. From the definition of $`d`$ it follows that $`c(u)=c(\sigma ^1[\phi _0[e[a]]]=d(e[a])`$ and $`c(v)=c(\sigma ^1[\phi _0[e[b]]]=d(e[b])`$. Consequently, $`c(u)=c(v)`$.
Since the image under $`\phi _0^{}1\sigma `$ of any $`m`$-element subset of $`K`$ is similar to a vip $`m`$-type and any two $``$-similar subsets receive the same color from $`c`$, the $`m`$-element subsets of $`K`$ are colored with at most $`t_m^+`$ colors, so the theorem follows. ∎
## 10 Lower bounds for Rado graphs
The computation of lower bounds for Rado graphs is a bit more complicated than the computation for $`\kappa `$-dense linear orders. We reduce the problem by showing for suitable $`\kappa `$ that if $`D{}_{}{}^{\kappa >}2`$ is the range of a pnp diagonalization with level harmony and $``$ is a $`D`$-vip level order, then every vip $`m`$-type is realized as $`(\mathrm{clp}(x),_x)`$ for some $`xD`$. This theorem is the companion to Theorem 5.7. Its proof uses a pnp diagonalization with level harmony in place of a semi-strong embedding.
###### Theorem 10.1.
Suppose that $`\kappa `$ is a cardinal which is measurable in the generic extension obtained by adding $`\lambda `$ Cohen subsets of $`\kappa `$, where $`\lambda (\kappa )_{2^\kappa }^6`$. Further suppose $`f:{}_{}{}^{\kappa >}2{}_{}{}^{\kappa >}2`$ is a pnp diagonalization with level harmony, $`D:=f[{}_{}{}^{\kappa >}2]`$, and $``$ is a $`D`$-vip order of the levels of $`{}_{}{}^{\kappa >}2`$. Then every vip $`m`$-type $`(\tau ,)`$ is realized as $`(\mathrm{clp}(x),_x)`$ for some $`xD`$.
###### Proof.
For $`t{}_{}{}^{\alpha }2`$, $`i=0,1`$ and $`\delta =0,1`$, define well-orderings $`_t^{i,\delta }`$ on $`{}_{}{}^{\alpha }2`$ as follows. Let $`\beta _t^0=\mathrm{lg}(\widehat{f}(t))`$ and set $`\beta _t^1=\mathrm{lg}(f(t))`$. Then $`s_t^{i,\delta }s^{}`$ if and only if $`f(s{}_{}{}^{}\delta )\beta _t^if(s^{}{}_{}{}^{}\delta )\beta _t^i`$.
Let $`^{}`$ be any small well-ordering of the levels of $`{}_{}{}^{\kappa >}2`$. Call a triple $`\{s,s^{},t\}`$ *local* if $`\mathrm{lg}(s)=\mathrm{lg}(s^{})=\mathrm{lg}(t)`$, $`s<_{\text{lex}}s^{}`$, $`t^{}s`$, and $`t^{}s^{}`$, and $`ss^{}t`$.
Let $`d`$ be a coloring of the triples of $`{}_{}{}^{\kappa >}2`$ defined as follows: if $`\{s,s^{},t\}`$ is not local, let $`d^{i,\delta }(\{s,s^{},t\}):=2`$ and otherwise set $`d^{i,\delta }(\{s,s^{},t\}):=s^{}s^{}s_t^{i,\delta }s^{}`$. For $`b=\{s,s^{},t\}[{}_{}{}^{\alpha }2]^3`$, define $`d(b):=(d^{\mathrm{\hspace{0.17em}0},0}(b),d^{\mathrm{\hspace{0.17em}0},1}(b),d^{\mathrm{\hspace{0.17em}1},0}(b),d^{\mathrm{\hspace{0.17em}1},1}(b))`$.
Apply Shelah’s Theorem 2.5 to $`d`$ and $`^{}`$ to obtain a strong embedding $`e:{}_{}{}^{\kappa >}2{}_{}{}^{\kappa >}2`$ and an element $`w`$ so that for triples from $`T:=e[\mathrm{Cone}(w)]`$, the coloring depends only on the $`^{}`$-ordered similarity type of the triple.
Then two local triples $`\{s,s^{},t\}`$ and $`\{u,u^{},v\}`$ of $`T`$ are colored the same if and only if for all $`i=0,1`$ and $`\delta =0,1`$,
$$s_t^{i,\delta }s^{}u_v^{i,\delta }u^{}.$$
Thus for $`tT`$, the orderings $`_t^{i,\delta }`$ must always agree with one of $`^{}`$ and its converse on $`T`$ on pairs $`\{s,s^{}\}T`$ with $`\{s,s^{},t\}`$ local and $`s^{}s^{}`$. Similarly, they must always agree with one of $`^{}`$ and its converse on $`T`$ on pairs $`\{s,s^{}\}T`$ with $`\{s,s^{},t\}`$ local and $`s^{}^{}s`$. Since $``$ is a well-order, all of the orderings $`_t^{i,\delta }`$ are also well-orders. Thus they always agree with $`^{}`$.
Let $`(\tau ,)`$ be an arbitrary vip $`m`$-type. Let $`L`$ be the set of leaves of $`\tau `$. Then $`\tau `$ is a subtree of $`{}_{}{}^{2m2}2`$ and every level of $`{}_{}{}^{2m2}2`$ has exactly one element of $`L^{}`$. Extend $``$ defined on $`\tau `$ to $`^{}`$ defined on all of $`{}_{}{}^{2n}2`$ in such a way that the extension is still a $`L^{}`$-vip order.
Apply Lemma 5.3 to get an order preserving strong embedding $`j`$ of $`({}_{}{}^{2m2}2,^{})`$ into $`(\mathrm{Cone}(w),^{})`$.
Let $`t_{\mathrm{}}:\mathrm{}2m2`$ enumerate the elements of $`L^{}`$ in increasing order of length. Note that $`\mathrm{lg}(t_{\mathrm{}})=\mathrm{}`$. For $`\mathrm{}2m2`$, define
$$\beta _{\mathrm{}}:=\{\begin{array}{cc}\mathrm{lg}(\widehat{f}(e(j(t_{\mathrm{}}))))\hfill & \text{if }t_{\mathrm{}}L\text{,}\hfill \\ \mathrm{lg}(f(e(j(t_{\mathrm{}}))))\hfill & \text{if }t_{\mathrm{}}L\text{.}\hfill \end{array}$$
Finally define $`\rho :\tau {}_{}{}^{\kappa >}2`$ by recursion on $`\mathrm{}2m2`$. For $`\mathrm{}=0`$, let $`\rho (\mathrm{})=\widehat{f}(e(j(\mathrm{})))`$. For $`\mathrm{}>0`$, consider three cases for elements of $`\tau {}_{}{}^{\mathrm{}}2`$. If $`t_{\mathrm{}}L`$, let $`\rho (t_{\mathrm{}})=f(e(j(t_{\mathrm{}})))`$. If $`t_{\mathrm{}}L`$, let $`\rho (t_{\mathrm{}})=\widehat{f}(e(j(t_{\mathrm{}})))`$. Note that in both these cases, $`\beta _{\mathrm{}}=\mathrm{lg}(\rho (t_{\mathrm{}}))`$. If $`s\tau L^{}`$ has length $`\mathrm{}`$, then there is a unique immediate successor in $`\tau `$, $`s{}_{}{}^{}\delta `$. In this case, let $`\rho (s)=f(e(j(s)){}_{}{}^{}\delta )\beta _{\mathrm{}}`$.
Since $`j`$ sends $`^{}`$-increasing pairs to $`^{}`$-increasing pairs and $`e`$ is a $`^{}`$-order preserving strong embedding, their composition sends sends $`^{}`$-increasing pairs to $`^{}`$-increasing pairs. Since for $`v_{\mathrm{}}=e(j(t_{\mathrm{}}))T`$, the order $`^{}`$ agrees with $`_v_{\mathrm{}}^i`$ on $`T{}_{}{}^{\gamma }2`$ where $`\gamma =\mathrm{lg}(v_{\mathrm{}})`$, it follows that $`\rho `$ sends $`^{}`$-increasing pairs to $``$-increasing pairs.
Since $`\widehat{f}`$ preserves extension and lexicographic order and $`\widehat{f}(s)f(s)`$, $`\rho `$ preserves extension and lexicographic order. By construction $`\rho `$ sends levels to levels, meets to meets and leaves to leaves. Let $`x=\rho [L]`$ be the image under $`\rho `$ of the leaves of $`\tau `$. By construction, $`x\mathrm{ran}(f)`$. Also $`(\mathrm{clp}(x),<_x)=(\tau ,)`$, as required.
Since $`(\tau ,)`$ was arbitrary, the theorem follows. ∎
We would like to define a coloring of the $`m`$-tuples of $`\kappa `$ using $`t_m^+`$ colors such that for any $`H\kappa `$ with $`(H,EH)`$ isomorphic to our Rado graph $`𝔾`$, every color is the color of some $`m`$-tuple from $`H`$.
By Lemma 9.4, there is an $`<`$-increasing map $`h:\kappa H`$ such that the graph $`(h[\kappa ],Eh[\kappa ])`$ is isomorphic to $`𝔾`$. By Lemma 9.5, $`g=\sigma h\sigma ^1`$ is a pnp map. By the previous theorem, every vip $`m`$-type can be realized in the range of a pnp diagonalization with level harmony. By the Translation Theorem, a pnp map gives rise to an induced subgraph of the Rado graph which is isomorphic to the whole graph.
Our plan is to define a coloring $`c`$ using $`t_m^+`$ as follows. Apply Lemma 9.13 to obtain a pre-$`S`$-vip order $``$ and a sequence $`\phi _t:t{}_{}{}^{\kappa >}2`$ such that for all $`t{}_{}{}^{\kappa >}2`$, the three listed properties of the lemma hold. In particular, $`\phi _0`$ is a pnp diagonalization into $`S`$ with level harmony, $`D_0:=\phi _0[{}_{}{}^{\kappa >}2]`$ is a strongly diagonal set and $``$ is a $`D`$-vip order. Enumerate the vip $`m`$-types as $`(\tau _j,_j):j<t_m^+`$; define $`d`$ on $`m`$-element subsets of $`{}_{}{}^{\kappa }2`$ by $`d(x)=j`$ if $`(\mathrm{clp}(\phi _0[x]),_{\mathrm{clp}(\phi _0[x])})=(\tau _j,_j)`$; and set $`c(b)=d(\sigma [b])`$.
Then for every isomorphic copy $`H`$ of the Rado graph, the associated pnp map $`\phi _0g`$ described above has the property that different colors of $`m`$-element subsets of $`K`$ are distinguished by the ordered similarity types of elements of the range of $`\phi _0g`$. Thus it is enough to show there is a pnp diagonalization $`f`$ with level harmony whose range is a subset of $`\phi _0g`$, and use Theorem 10.1 with $`f`$ to show that all the colors appear in every isomorphic copy.
We show that for any pnp map $`g`$ with range a strongly diagonal set there is a pnp diagonalization $`f`$ with level harmony with $`\mathrm{ran}(f)\mathrm{ran}(g)`$ in Theorem 10.7 below. We work by successive approximation, using the next lemma.
###### Lemma 10.2.
Let $`𝒟`$ be the collection of pnp maps whose domain is $`{}_{}{}^{\kappa }2`$ and whose range is a strongly diagonal subset of $`{}_{}{}^{\kappa }2`$. Then $`𝒟`$ is closed under composition as are the following subfamilies:
1. maps in $`𝒟`$ with meet regularity;
2. maps in $`𝒟`$ with meet regularity that preserve lexicographic order of incomparable pairs;
3. maps in $`𝒟`$ that are polite on strongly diagonal sets;
4. maps in $`𝒟`$ that are polite.
Moreover, if $`g`$ is in one of these families and $`e`$ is a strong embedding, then $`eg`$ is also in that family.
###### Proof.
Apply Lemma 9.8 and use the definitions of terms to see that the various families are closed under composition. Since strong embeddings preserve all the properties in question, and carry strongly diagonal sets to strongly diagonal sets, their composition with any function in one of the various families is also in that family. ∎
###### Lemma 10.3.
Suppose $`w{}_{}{}^{\kappa }2`$ and $`S`$ is cofinal and transverse. Further suppose $`f`$ is polite to strongly diagonal sets and has range a strongly diagonal set $`D`$ with $`D^{}S\mathrm{Cone}(w)`$ and $`g`$ is a pnp diagonalization which has level harmony. Then $`fg`$ is a pnp diagonalization into $`S\mathrm{Cone}(w)`$ which has level harmony.
###### Proof.
First notice that $`f`$ and $`g`$ are in $`𝒟`$, so by Lemma 10.2, their composition is a polite pnp map whose range is a strongly diagonal set. Since $`<_\text{Q}`$ agrees with the lexicographic order on incomparable pairs, $`g`$ preserves $`<_\text{Q}`$ and sends all pairs to incomparable pairs, the composition $`fg`$ preserves $`<_\text{Q}`$. Since $`g`$ is injective with range a strongly diagonal set and $`f`$ is a pnp map, the composition is injective. Thus, since the range of $`fg`$ is a subset of the range of $`f`$, the composition, $`fg`$, is a pnp diagonalization into $`S\mathrm{Cone}(w)`$.
For notational convenience, let $`h=fg`$. Then $`\widehat{h}(s):=h(s{}_{}{}^{}0)h(s{}_{}{}^{}1)=f(g(s{}_{}{}^{}0))f(g(s{}_{}{}^{}1))`$.
###### Claim 10.3.a.
For any $`s{}_{}{}^{\kappa }2`$, $`\widehat{h}(s)=h(s{}_{}{}^{}1\delta )h(s)h(s)`$, where $`\delta =g(s)(\widehat{g}(s))`$.
###### Proof.
Since $`g`$ has level harmony, $`\widehat{g}(s)g(s)`$. Set $`\delta :=g(s)(\mathrm{lg}(\widehat{g}(s))`$. Since $`s{}_{}{}^{}0<_{\text{lex}}s{}_{}{}^{}1`$ and $`g`$ preserves lexicographic order, it follows that $`g(s{}_{}{}^{}0)<_{\text{lex}}g(s{}_{}{}^{}1)`$, so $`g(s{}_{}{}^{}\delta )(\mathrm{lg}(\widehat{g}(s))=\delta `$.
Since $`f`$ is polite, by meet regularity, $`f(g(s{}_{}{}^{}1\delta ))f(g(s{}_{}{}^{}\delta ))=f(g(s{}_{}{}^{}1\delta ))f(g(s))`$. That is, $`\widehat{h}(s)=h(s{}_{}{}^{}1\delta )h(s)`$. ∎
###### Claim 10.3.b.
The function $`\widehat{h}`$ preserves lexicographic order.
###### Proof.
Suppose $`s<_{\text{lex}}t`$. Since $`g`$ has level harmony, $`\widehat{g}(s)<_{\text{lex}}\widehat{g}(t)`$, $`\widehat{g}(s)g(s)`$ and $`\widehat{g}(t)g(t)`$. Hence $`\widehat{g}(s)\widehat{g}(t)=g(s)g(t)`$. Thus $`\mathrm{lg}(g(s)g(t))<\mathrm{lg}(g(s{}_{}{}^{}0)g(s{}_{}{}^{}0))`$ and $`\mathrm{lg}(g(s)g(t))<\mathrm{lg}(g(t{}_{}{}^{}0)g(t{}_{}{}^{}0))`$. Since $`f`$ preserves meet length order, it follows that $`\mathrm{lg}(h(s)h(t))<\mathrm{lg}(\widehat{h}(s))`$ and $`\mathrm{lg}(h(s)h(t))<\mathrm{lg}(\widehat{h}(t))`$.
Since $`f`$ preserves lexicographic order, $`h(s)<_{\text{lex}}h(t)`$. By the Claim 10.3.a, $`\widehat{h}(s)h(s)`$ and $`\widehat{h}(t)h(t)`$. Thus $`\widehat{h}(s)<_{\text{lex}}\widehat{h}(t)`$. ∎
###### Claim 10.3.c.
For all $`s`$ and $`t`$, $`\mathrm{lg}(s)<\mathrm{lg}(t)`$ implies $`\mathrm{lg}(h(s))<\mathrm{lg}(\widehat{h}(t))`$.
###### Proof.
Suppose $`\mathrm{lg}(s)<\mathrm{lg}(t)`$. Then $`\mathrm{lg}(g(s))<\mathrm{lg}(\widehat{g}(t))=lg(g(s{}_{}{}^{}0)g(s{}_{}{}^{}1)`$. Since $`g(s)g(s)=g(s)`$, by preservation of meet length order by $`f`$, it follows that $`\mathrm{lg}(h(s)<\mathrm{lg}(\widehat{h}(t))`$. ∎
###### Claim 10.3.d.
For all $`s`$ and $`t`$, $`\mathrm{lg}(s)=\mathrm{lg}(t)`$ implies $`\mathrm{lg}(\widehat{h}(s))<\mathrm{lg}(h(t))`$.
###### Proof.
Suppose $`\mathrm{lg}(s)=\mathrm{lg}(t)`$. Then $`\mathrm{lg}(\widehat{g}(s))<\mathrm{lg}(g(t))`$. Argue as in the previous claim: by preservation of meet length order by $`f`$, it follows that $`\mathrm{lg}(\widehat{h}(s)<\mathrm{lg}(h(t))`$. ∎
###### Claim 10.3.e.
The function $`\widehat{h}`$ preserves extension.
###### Proof.
Suppose $`st`$. Then $`\widehat{g}(s)\widehat{g}(t)`$, since $`g`$ has level harmony. Since $`f`$ satisfies preservation of meet length order, $`\mathrm{lg}(\widehat{h}(s))<\mathrm{lg}(\widehat{g}(s))`$. By Claim 10.3.a, $`\widehat{h}(t)h(t)`$, so $`\mathrm{lg}(\widehat{h}(s))<\mathrm{lg}(h(t))`$. Thus to show $`\widehat{h}(s)\widehat{h}(t)`$, it is enough to show $`\widehat{h}(s)h(t)`$.
If $`h(t)(\mathrm{lg}(\widehat{h}(s)))=1`$, then $`\widehat{h}(s)h(t)`$, since the range of $`h`$ is strongly diagonal. If $`t`$ is one of $`s{}_{}{}^{}0`$ and $`s{}_{}{}^{}1`$, then $`\widehat{h}(s)h(t)`$ by definition of $`\widehat{h}(s)`$.
So suppose $`h(t)(\mathrm{lg}(\widehat{h}(s)))=0`$ and $`\mathrm{lg}(t)>\mathrm{lg}(s)+1`$. Since $`g`$ has level harmony, $`\widehat{g}(s)g(s)`$ and $`\widehat{g}(s)\widehat{g(t)}g(t)`$. By Claim 10.3.a, $`\widehat{h}(s)=h(t{}_{}{}^{}1\delta )h(s)`$ where $`\delta =g(\mathrm{lg}(\widehat{g}(s))`$. For notational convenience, let $`\alpha =\mathrm{lg}(\widehat{g}(s))`$. For the first subcase, suppose $`g(t)(\alpha )g(s)(\alpha )`$. Then $`g(t)(\alpha )=g(s{}_{}{}^{}1\delta )(\alpha )`$, so $`g(s)g(s{}_{}{}^{}1\delta )=g(s)g(t)`$. Since the range of $`g`$ is strongly diagonal and $`f`$ is polite, it follows that $`f(g(s))f(g(s{}_{}{}^{}1\delta )=f(g(s))f(g(t))`$, so $`\widehat{h}(s)=h(s)h(t)h(t)`$. For the second subcase, in which $`g(t)(\alpha )=g(s)(\alpha )`$, interchange the roles of $`g(s)`$ and $`g(s{}_{}{}^{}1\delta )`$. The parallel argument concludes with the inclusion $`\widehat{h}(s)=h(s{}_{}{}^{}1\delta )h(t)h(t)`$. ∎
Now the lemma follows from the claims. ∎
The next lemma gives an inequality, for pnp maps, which compares lengths of meets of images with lengths of images of meets.
###### Lemma 10.4.
Suppose $`g:{}_{}{}^{\kappa >}2{}_{}{}^{\kappa >}2`$ is a pnp map and $`\{x,u,v\}`$ is a three element strongly diagonal set with $`xu=xvuv`$. Then $`\mathrm{lg}(g(x)g(u))\mathrm{lg}(g(xu))`$ and $`\mathrm{lg}(g(x)g(v))\mathrm{lg}(g(xu))`$.
###### Proof.
Let $`\alpha :=\mathrm{lg}(ux)=\mathrm{lg}(uv)`$ and set $`\beta :=\mathrm{lg}(g(xu))=\mathrm{lg}(g(xv)`$. Since $`g`$ is a pnp map, $`\beta <\mathrm{lg}(g(x))`$ and $`g(x)(\beta )=x(\alpha )`$. Similarly, $`\beta <\mathrm{lg}(g(u))`$ and $`g(u)(\beta )=u(\alpha )`$. Also, $`\beta <\mathrm{lg}(g(v))`$ and $`g(v)(\beta )=v(\alpha )`$.
Since $`xu=xv`$, it follows that $`x(\alpha )u(\alpha )=v(\alpha )`$. Consequently, $`g(x)(\beta )g(u)(\beta )=g(v)(\beta )`$. Thus $`\mathrm{lg}(g(x)g(u))\beta `$ and $`\mathrm{lg}(g(x)g(v))\beta `$, so the lemma follows. ∎
Next we show how to use Shelah’s Theorem 2.5 to obtain a pnp map with meet regularity from a pnp map whose range is a strongly diagonal set.
###### Lemma 10.5.
Suppose that $`\kappa `$ is a cardinal which is measurable in the generic extension obtained by adding $`\lambda `$ Cohen subsets of $`\kappa `$, where $`\lambda (\kappa )_{2^\kappa }^6`$. Further suppose $`g:{}_{}{}^{\kappa >}2{}_{}{}^{\kappa >}2`$ is a pnp map whose range is a strongly diagonal set. Then there is a pnp map $`f`$ which satisfies meet regularity and whose range is a subset of the range of $`g`$.
###### Proof.
Apply Lemma 9.13 to obtain a pre-$`S`$-vip order $``$ and a sequence $`\phi _t:t{}_{}{}^{\kappa >}2`$ such that for all $`t{}_{}{}^{\kappa >}2`$, the three listed properties of the lemma hold. Say a three element subset $`\{b_0,b_1,b_2\}`$ listed in increasing lexicographic order *witnesses meet regularity for $`g`$* if it is diagonal and either ($`b_0b_1=b_0b_2`$ and $`g(b_0)g(b_1)=g(b_0)g(b_2)`$) or ($`b_2b_0=b_2b_1`$ and $`g(b_2)g(b_0)=g(b_2)g(b_1)`$). Say a three element subset $`\{b_0,b_1,b_2\}`$ listed in increasing lexicographic order *refutes meet regularity for $`g`$* if it is diagonal, $`b_0b_1=b_0b_2`$ implies $`g(b_0)g(b_1)g(b_0)g(b_2)`$ and $`b_2b_0=b_2b_1`$ implies $`g(b_2)g(b_0)g(b_2)g(b_1)`$.
Define a coloring $`d^{}`$ on three element subsets of $`{}_{}{}^{\kappa }2`$ by $`d^{}(x)=0`$ if $`x`$ is strongly diagonal and witnesses meet regularity for $`g`$, $`d^{}(x)=1`$ if $`x`$ is strongly diagonal and refutes meet regularity for $`g`$ and $`d^{}(x)=2`$ otherwise. Apply Shelah’s Theorem 2.5 to $`d^{}`$ and $``$ to obtain a strong embedding $`e`$ and $`w{}_{}{}^{\kappa }2`$ such that $`e`$ preserves $``$ on $`\mathrm{Cone}(w)`$ and for all three element subsets $`x`$ of $`T=e[\mathrm{Cone}(w)]`$, the value of $`d^{}(x)`$ depends only on the $``$-ordered similarity type of $`x`$.
Let $`\psi =\phi _w`$. Then $`\psi `$ is a pnp diagonalization into $`S\mathrm{Cone}(w)`$. We claim that $`f:=ge\psi `$ is a pnp map which satisfies meet regularity and whose range is a subset of the range of $`g`$.
Since $`e\psi `$ is a pnp map which witnesses meet regularity and whose range is a strongly diagonal set, it is enough to show that every strongly diagonal subset $`xT`$ witnesses meet regularity for $`g`$.
Let $`(\tau ,)`$ be an arbitrary vip $`m`$-type. Enumerate the leaves of $`\tau `$ in increasing order of length as $`a_0,a_1,a_2`$. We must show that $`(\tau ,)`$ witnesses meet regularity for $`g`$.
###### Claim 10.5.a.
If $`\mathrm{lg}(a_0)=1`$, then every $`xT`$ with ordered similarity type $`(\tau ,)`$ witnesses meet regularity for $`g`$.
###### Proof.
Notice that since that since $`\mathrm{lg}(a_0)=1`$ and $`\tau `$ has an elements of the meet closure of the leaves of lengths $`0,1,2,3,4`$, we must have $`\mathrm{lg}(a_0)=1<\mathrm{lg}(a_1a_2)`$.
Let $`\nu `$ be a cardinal larger than $`\mathrm{lg}(g(e(\psi (a_0))))`$. For $`\alpha <\nu `$, let $`b_\alpha `$ be the sequence of length $`2\alpha +3`$ if $`\alpha <\omega `$ and of length $`\gamma +2n+1`$ if $`\alpha =\gamma +n`$ for some limit ordinal $`\gamma \omega `$ and $`n<\omega `$ such that
$$b_\alpha (\eta )=\{\begin{array}{cc}a_1(0),\hfill & \text{if }\eta =0\text{,}\hfill \\ a_1(1),\hfill & \text{if }\eta =1\text{,}\hfill \\ a_2(2),\hfill & \text{if }0<\eta <\mathrm{lg}(b_\alpha )1\text{ even,}\hfill \\ a_1(2),\hfill & \text{if }\eta =\mathrm{lg}(b_\alpha )1\text{,}\hfill \\ a_2(3),\hfill & \text{otherwise.}\hfill \end{array}$$
Note that the length of $`b_\alpha `$ is always odd and at least $`3`$, and that for odd $`\eta 3`$, $`b_\alpha (\eta )=a_2(3)`$. For $`\beta <\alpha `$, by construction, $`b_\beta b_\alpha =b_\beta (\mathrm{lg}(b_\beta )1)`$. Using the above calculations, the careful reader may check that for $`\beta <\alpha `$, one has $`\mathrm{clp}(\{a_0,b_\beta ,b_\alpha \})=\tau `$. Notice that $`a_0b_\beta =a_0b_\alpha =\mathrm{}`$. For $`\beta <\nu `$, let $`s_\beta =e(\psi (b_\beta ))`$, set $`t_0=e(\psi (a_0))`$ and let $`x(\beta ,\alpha )=\{t_0,s_\beta ,s_\alpha \}`$.
By Lemma 9.9, since $`e\psi `$ is a polite pnp map, we have $`\mathrm{clp}(x(\beta ,\alpha ))=\mathrm{clp}(\{a_0,b_\beta ,b_\alpha \})=\tau `$. Since there is only one vip order on $`\tau `$, it follows that $`(\mathrm{clp}(x(\beta ,\alpha )),)=(\tau ,)`$.
Since each $`x(\beta ,\alpha )`$ is a strongly diagonal set, it either witnesses or refutes meet regularity for $`g`$. Since each $`x(\beta ,\alpha )`$ is a subset of $`T`$, either they all witness or all refute meet regularity for $`g`$.
Since $`e\psi `$ satisfies meet regularity, $`t_0s_\beta =t_0s_\alpha `$. Suppose each $`x(\beta ,\alpha )`$ refutes meet regularity. Then $`g(t_0)g(s_\beta )g(t_0)g(s_\alpha )`$ for $`\beta <\alpha <\nu `$. Since $`\nu `$ is a cardinal larger than $`\mathrm{lg}(g(e(\psi (a_0))))=\mathrm{lg}(g(t_0))`$, by the Pigeonhole Principle, there are $`\beta _0<\alpha _0`$ with $`g(t_0)g(s_\beta )=g(t_0)g(s_\alpha )`$. This contradiction shows that each $`x(\beta ,\alpha )`$ witnesses meet regularity, so by choice of $`d^{}`$, $`e`$ and $`w`$, the claim follows. ∎
###### Claim 10.5.b.
If $`\mathrm{lg}(a_0a_1)=1`$, then every $`xT`$ with ordered similarity type $`(\tau ,)`$ witnesses meet regularity for $`g`$.
###### Proof.
Notice that since that since $`\mathrm{lg}(a_0a_1)=1`$ and $`\tau `$ has an elements of the meet closure of the leaves of lengths $`0,1,2,3,4`$, we must have $`\mathrm{lg}(a_0a_1)=1<\mathrm{lg}(a_0)<\mathrm{lg}(a_1)<\mathrm{lg}(a_2)`$. As in the previous case, there is only one vip level order on $`\tau `$.
Let $`z=e(\psi (a_0))`$ and let $`\nu =\left(2^{|\mathrm{lg}(g(z))|}\right)^+`$. We proceed much as in the previous case, except we start by constructing $`b_\alpha :\alpha <\nu `$ and $`c_\alpha :\alpha <\nu `$ such that for all $`\beta <\alpha `$, $`c_\alpha b_\beta =\mathrm{}=c_\alpha b_\alpha `$. Then we set $`t_\alpha =e(\psi (c_\alpha ))`$, $`s_\alpha =e(\psi (b_\alpha ))`$, and $`x(\beta ,\alpha )=\{s_\beta ,s_\alpha ,t_\alpha \}`$. In the previous case we had only $`t_0`$, so in this case we will use the Pigeonhole Principle to select a collection of $`\alpha `$’s for which the behavior of $`g`$ on $`t_\alpha `$ is sufficiently uniform to reduce this case to one like the previous one.
For $`\alpha =\gamma +n`$, where $`\gamma =0`$ or $`\gamma `$ limit, let $`b_\alpha `$ be the sequence of length $`\gamma +2n+2`$ such that
$$b_\alpha (\eta )=\{\begin{array}{cc}a_1(0),\hfill & \text{if }\eta =0\text{,}\hfill \\ a_1(1),\hfill & \text{if }\eta <\mathrm{lg}(b_\alpha )1\text{ odd,}\hfill \\ a_1(2),\hfill & \text{if }0<\eta <\mathrm{lg}(b_\alpha )1\text{ even,}\hfill \\ a_0(1),\hfill & \text{if }\eta =\mathrm{lg}(b_\alpha )1\text{.}\hfill \end{array}$$
Note that the length of $`b_\alpha `$ is always even and its Cantor normal form has an even finite part that is at least $`2`$. Also, for positive even $`\eta `$, $`b_\alpha (\eta )=a_1(2)`$. For $`\beta <\alpha `$, by construction, $`b_\beta b_\alpha =b_\beta (\mathrm{lg}(b_\beta )1)`$. For $`\alpha <\eta `$, let $`c_\alpha `$ be the sequence of length $`\mathrm{lg}(b_\alpha )+1`$ such that $`c_\alpha (0)=a_2(0)`$, $`c_\alpha (1)=a_2(1)`$, $`c_\alpha (\mathrm{lg}(b_\alpha ))=a_2(3)`$, and for all $`\eta `$ with $`1<\eta <\mathrm{lg}(b_\alpha )`$, $`c_\alpha (\eta )=a_2(2)`$. Then for all $`\alpha ,\beta <\nu `$, $`b_\beta c_\alpha =\mathrm{}`$. Using these definitions and calculations, the careful reader may check that for $`\beta <\alpha `$, one has $`\mathrm{clp}(\{b_\beta ,b_\alpha ,c_\alpha \})=\tau `$.
Since $`e\psi `$ is a polite pnp map, by Lemma 9.9, we have $`\mathrm{clp}(x(\beta ,\alpha ))=\mathrm{clp}(\{b_\beta ,b_\alpha ,c_\alpha \})=\tau `$. Since there is only one vip order on $`\tau `$, it follows that $`(\mathrm{clp}(x(\beta ,\alpha )),)=(\tau ,)`$.
Since $`e\psi `$ satisfies meet regularity, $`t_\alpha s_\beta =t_\alpha s_\alpha `$ for all $`\beta <\alpha <\nu `$.
By Lemma 10.4, $`\mathrm{lg}(g(t_\alpha )g(s_\beta ))\mathrm{lg}(g(e(\psi (c_\alpha b_\alpha ))))=\mathrm{lg}(g(z))`$. Similarly, $`\mathrm{lg}(g(t_\alpha )g(s_\beta ))\mathrm{lg}(g(z))`$.
Set $`\gamma =\mathrm{lg}(g(z))`$. Apply the Pigeonhole Principle, to find a sequence $`p`$ of length $`\gamma `$ so that the collection $`A=A_p=\{\alpha :g(t_\alpha )\gamma =p\}`$ has cardinality $`\nu `$.
Since each $`x(\beta ,\alpha )`$ is a strongly diagonal set, it either witnesses or refutes meet regularity for $`g`$. Since each $`x(\beta ,\alpha )`$ is a subset of $`T`$, either they all witness or all refute meet regularity for $`g`$. Suppose each $`x(\beta ,\alpha )`$ refutes meet regularity. Then $`g(t_\alpha )g(s_\beta )g(t_\alpha )g(s_\alpha )`$ for $`\beta <\alpha <\nu `$.
In particular, for all $`\beta <\alpha `$ from $`A`$, one has
$$\begin{array}{cc}\hfill g(t_\beta )g(s_\beta )& =pg(s_\beta )\hfill \\ & =g(t_\alpha )g(s_\beta )\hfill \\ & g(t_\alpha )g(s_\alpha ).\hfill \end{array}$$
Now we have a contradiction that $`\{g(t_\alpha )g(s_\alpha ):\alpha A\}`$ is a set of $`\nu `$ distinct initial segments of $`p`$ and $`\mathrm{lg}(p)=\gamma <\nu `$. This contradiction shows that each $`x(\beta ,\alpha )`$ witnesses meet regularity, so by choice of $`d^{}`$, $`e`$ and $`w`$, the claim follows. ∎
###### Claim 10.5.c.
If $`\mathrm{lg}(a_0a_2)=1`$, then every $`xT`$ with ordered similarity type $`(\tau ,)`$ witnesses meet regularity for $`g`$.
###### Proof.
Notice that since that since $`\mathrm{lg}(a_0a_2)=1`$ and $`\tau `$ has an elements of the meet closure of the leaves of lengths $`0,1,2,3,4`$, we must have $`\mathrm{lg}(a_0a_2)=1<\mathrm{lg}(a_0)<\mathrm{lg}(a_1)<\mathrm{lg}(a_2)`$. As in the previous case, there is only one vip level order on $`\tau `$.
Let $`z=e(\psi (a_0))`$ and let $`\nu =\left(2^{|\mathrm{lg}(g(z))|}\right)^+`$. We proceed much as in the previous case, except we start by constructing $`b_\beta :\beta <\nu `$ and $`c_\beta :\alpha <\nu `$ such that for all $`\beta <\alpha `$, $`c_\beta b_\beta =\mathrm{}=c_\beta b_\alpha `$. Then we set $`t_\beta =e(\psi (c_\beta ))`$, $`s_\beta =e(\psi (b_\beta ))`$, and $`x(\beta ,\alpha )=\{s_\beta ,t_\beta ,s_\alpha \}`$.
For $`\beta =\gamma +n`$, where $`\gamma =0`$ or $`\gamma `$ limit, let $`b_\beta `$ be the sequence of length $`\gamma +4n+2`$ such that for $`\eta =\zeta +4k+\mathrm{}`$ where $`\zeta =0`$ or $`\zeta `$ limit, $`k<\omega `$ and $`\mathrm{}<4`$,
$$b_\beta (\eta )=\{\begin{array}{cc}a_2(\mathrm{}),\hfill & \text{if }\eta <\mathrm{lg}(b_\beta )1\text{,}\hfill \\ a_0(1),\hfill & \text{if }\eta =\mathrm{lg}(b_\beta )1\text{.}\hfill \end{array}$$
Note that the length of $`b_\beta `$ is always even and its Cantor normal form has an even finite part that is at least $`2`$. For $`\beta <\alpha `$, by construction, $`b_\beta b_\alpha =b_\beta (\mathrm{lg}(b_\beta )1)`$, and $`b_\alpha (\mathrm{lg}(b_\beta )1)=a_2(1)`$. For $`\beta <\nu `$, let $`c_\beta `$ be the sequence of length $`\mathrm{lg}(b_\beta )+1`$ such that $`c_\beta (0)=a_1(0)`$, $`c_\beta (\mathrm{lg}(b_\beta )1)=a_1(1)`$, $`c_\beta (\mathrm{lg}(b_\beta ))=a_1(2)`$, and for all $`\eta `$ with $`0<\eta <\mathrm{lg}(b_\beta )1`$, $`c_\beta (\eta )=0`$. Then for all $`\beta \alpha <\nu `$, $`b_\beta c_\alpha =\mathrm{}`$. Using these definitions and calculations, the careful reader may check that for $`\beta <\alpha `$, one has $`\mathrm{clp}(\{b_\beta ,c_\beta ,b_\alpha \})=\tau `$.
The rest of the proof in this case parallels that of the previous one and the details are left to the reader. ∎
Since $`S`$ is cofinal and transverse, its intersection with $`\mathrm{Cone}(p_1)`$ is cofinal above $`p_1`$ and transverse. Thus by Lemma 3.11, $`(S\mathrm{Cone}(w),<_\text{Q})`$ is $`\kappa `$-dense. Then $`U:=T(S\mathrm{Cone}(w))`$ is an almost perfect tree by Lemma 4.16.
###### Claim 10.5.d.
If $`\mathrm{lg}(a_1a_2)=1`$ and $`p_0,p_1`$ are incomparable elements of $`U`$ such that $`p_0p_1`$ a densely splitting point of $`S\mathrm{Cone}(w)`$ and $`p_0<_{\text{lex}}p_1`$ if and only if $`a_0<_{\text{lex}}a_1`$, then there are $`cU\mathrm{Cone}(p_0)`$ and $`b,b^{}U\mathrm{Cone}(p_1)`$ with $`\mathrm{max}(\mathrm{lg}(p_0),\mathrm{lg}(p_1))<\mathrm{lg}(bb^{})<\mathrm{lg}(b)<\mathrm{lg}(b^{})`$ such that $`(\{c,b,b^{}\},)`$ has the same ordered similarity type as $`(\tau ,)`$.
###### Proof.
By Lemma 4.14, $`U`$ is almost perfect. Use that fact to find a densely splitting node $`q_1`$ of $`U`$ extending $`p_1`$ of length longer than $`\mathrm{max}(\mathrm{lg}(p_0),\mathrm{lg}(p_1))`$.
Let $`C(S\mathrm{Cone}(w))`$ be the set of all limit ordinals $`\alpha >0`$ such that every $`tU{}_{}{}^{\alpha >}2`$ has proper extensions in both $`S{}_{}{}^{\alpha >}2`$ and $`𝒲(S){}_{}{}^{\alpha >}2`$ By Lemma4.16, $`C(S\mathrm{Cone}(w))`$ is closed and unbounded in $`\kappa `$.
Let $`\lambda C(S\mathrm{Cone}(w))`$ be a regular cardinal greater than $`\mathrm{lg}(q_1)`$. Use the fact that $`U`$ is almost perfect to find an extension $`c`$ of $`p_0`$ in $`S`$ of length greater than $`\lambda `$.
Define $`h:{}_{}{}^{\lambda }2U\mathrm{Cone}(q_1)`$ by recursion as follows and prove by induction that for all $`s{}_{}{}^{\lambda }2`$, $`h(s)`$ is a densely splitting point of $`U\mathrm{Cone}(q_1){}_{}{}^{\lambda }2`$. To start the recursion, let $`h(\mathrm{})=q_1`$. If $`h(s)`$ has been defined and $`\delta <2`$, let $`h_0(s{}_{}{}^{}\delta )`$ be an extension of $`h(s){}_{}{}^{}\delta `$ of length less than $`\lambda `$ which is a densely splitting node of $`S\mathrm{Cone}(w)`$ and let $`h(s{}_{}{}^{}\delta )`$ be an extension of $`h_0(s){}_{}{}^{}1\delta `$ of length less than $`\lambda `$ which is a densely splitting node of $`S\mathrm{Cone}(w)`$. If $`\gamma <\lambda `$ is a limit ordinal and $`h(s\beta )`$ has been defined for all $`\beta <\gamma `$, let $`h(s)`$ be an extension of $`\{h(s)\beta :\beta <\gamma \}`$ of length less than $`\lambda `$ which is a densely splitting node of $`S\mathrm{Cone}(w)`$. If $`h(s\beta )`$ has been defined for all $`\beta <\lambda `$, let $`h(s)`$ be $`\{h(s)\beta :\beta <\lambda \}`$. Since $`U`$ is an almost perfect tree and for each $`s`$ of limit length $`\gamma `$, the node $`\{h(s)\beta :\beta <\gamma \}`$ is an even-handed limit of densely splitting points, the recursion is well-defined.
Set $`A:=h[{}_{}{}^{\lambda }2]`$. Since $`\lambda `$ is regular, it follows that $`A{}_{}{}^{\lambda }2`$. Also, $`|A|=2^\lambda `$.
Let $`\xi =\mathrm{lg}(c)`$ and let $`h^{}:AU{}_{}{}^{\xi }2`$ be an injection with $`sh^{}(s)`$ for all $`sA`$. Such an injection exists since $`U`$ is almost perfect. Define $`^{}`$ on $`A`$ by $`s^{}t`$ if and only if $`h^{}(s)h^{}(t)`$. Notice that $`^{}`$ is a well-order, since $``$ is a well order.
Since $`h`$ preserves lexicographic order, we can find $`r_0A`$ such that both $`\{sA:s<_{\text{lex}}r_0\}`$ and $`\{tA:r_0<_{\text{lex}}t\}`$ have cardinality $`2^\lambda `$. Thus there are $`s_0`$ and $`t_0`$ with $`s_0<_{\text{lex}}r_0<_{\text{lex}}t_0`$ and $`r_0^{}s_0`$, $`r_0^{}t_0`$. That is there are pairs from $`A`$ where the lexicographic order and the $`^{}`$-order agree and where they disagree.
Choose $`r_1,r_2A`$ such that $`r_1<_{\text{lex}}r_2`$ if and only if $`a_1<_{\text{lex}}a_2`$ and $`r_1^{}r_2`$ if and only if $`a_12a_22`$.
Use the fact that $`U`$ is almost perfect to find be an element $`b`$ of $`S`$ extending $`h^{}(a_1)`$ of length greater than $`\xi =\mathrm{lg}(c)`$ and to find an element $`b^{}`$ of $`S`$ extending $`h^{}(a_2)`$ of length greater than $`\mathrm{lg}(b)`$. Then $`(\{c,b,b^{}\},)`$ has the same ordered similarity type as $`(\tau ,)`$. ∎
###### Claim 10.5.e.
If $`\mathrm{lg}(a_1a_2)=1`$, then every $`xT`$ with ordered similarity type $`(\tau ,)`$ witnesses meet regularity for $`g`$.
###### Proof.
Notice that since that since $`\mathrm{lg}(a_1a_2)=1`$ and $`\tau `$ has an elements of the meet closure of the leaves of lengths $`0,1,2,3,4`$, we must have $`\mathrm{lg}(a_1a_2)=1<\mathrm{lg}(a_0)<\mathrm{lg}(a_1)<\mathrm{lg}(a_2)`$.
The proof of this claim shares similarities with the proofs of Claims 10.5.a, 10.5.c and 10.5.b, once the construction of the family of sets $`x(\beta ,\alpha )`$ has been completed.
Let $`p_0`$ and $`p_1`$ be incomparable elements of $`U`$ such that $`p_0p_1`$ is a densely splitting node of $`S\mathrm{Cone}(w)`$ and $`p_0<_{\text{lex}}p_1`$ if and only if $`a_0<_{\text{lex}}a_1`$.
Set $`z=e(p_0p_1)`$ and let $`\nu =\left(2^{|\mathrm{lg}(g(z))|}\right)^+`$.
Define by recursion sequences $`p_{\delta ,\alpha }:\alpha <\nu `$ for $`\delta <2`$, $`c_\alpha :\alpha <\nu `$ $`b_\alpha :\alpha <\nu `$ and $`b_\alpha ^{}:\alpha <\nu `$ as follows. To start the recursion, let $`p_{0,0}=p_0`$ and $`p_{1,0}=p_1`$. If $`p_{0,\alpha }`$ and $`p_{1,\alpha }`$ have been defined with $`p_0p_{0,\alpha }`$ and $`p_1p_{1,\alpha }`$, then apply Claim 10.5.d to $`p_{0,\alpha }`$ and $`p_{1,\alpha }`$ to obtain $`c_\alpha S\mathrm{Cone}(p_{0,\alpha })`$ and $`b_\alpha ,b_\alpha ^{}S\mathrm{Cone}(p_{1,\alpha })`$ such that $`(\{c_\alpha ,b_\alpha ,b_\alpha ^{}\},)`$ has the same ordered similarity type as $`(\tau ,)`$ and the following inequalities hold:
$$\mathrm{max}(\mathrm{lg}(p_{0,\alpha }),\mathrm{lg}(p_{1,\alpha }))<\mathrm{lg}(b_\alpha b_\alpha ^{})<\mathrm{lg}(b_\alpha )<\mathrm{lg}(b_\alpha ^{}).$$
Note that $`c_\alpha b_\alpha =p_0p_1`$.
If $`\alpha >0`$ and $`p_{0,\beta }`$, $`p_{1,\beta }`$, $`c_\beta `$, $`b_\beta `$ and $`b_\beta ^{}`$ have all been defined for $`\beta <\alpha `$, then set $`p_{0,\alpha }:=\{c_\beta :\beta <\alpha \}`$ and $`p_{1,\alpha }:=\{b_\beta ^{}:\beta <\alpha \}`$.
By construction, the sequence $`𝒮=b_\alpha :\alpha <\nu `$ is increasing in length. Fix attention on a specific pair $`\beta <\alpha <\nu `$. Since $`s_\alpha \mathrm{Cone}(p_{1,\alpha })`$, it follows from the definition of $`p_{1,\alpha }`$ that $`b_\beta ^{}s_\alpha `$. Thus $`(\{c_\beta ,b_\beta ,b_b^{}eta\},)`$ and $`(\{c_\beta ,b_\beta ,b_\alpha \},)`$ have the same $``$-ordered similarity type, namely $`(\tau ,)`$. Furthermore, $`c_\beta b_\beta =p_0p_1`$ and $`c_\beta b_\alpha =p_0p_1`$.
For $`\alpha <\nu `$, let $`t_\alpha =e(c_\alpha )`$ and let $`s_\alpha =e(b_\alpha )`$. For $`\beta <\alpha <\nu `$, let $`x(\beta ,\alpha )=\{t_\beta ,s_\beta ,s_\alpha \}`$. Since $`e`$ preserves meets, it follows that $`e(c_\beta )e(b_\beta )=e(p_0p_1)=z`$ and $`e(c_\beta )e(b_\alpha )=e(p_0p_1)=z`$. Since $`e`$ is a strong embedding which preserves $``$ on $`\mathrm{Cone}(w)`$, and $`x(\beta ,\alpha )\mathrm{Cone}(w)`$, it follows that $`(x(\beta ,\alpha ),)`$ and $`(\{c_\beta ,b_\beta ,b_\alpha \},)`$ have the same $``$-ordered similarity type, namely $`(\tau ,)`$.
The rest of the proof in this case parallels those of Claims 10.5.c and 10.5.b and the details are left to the reader. ∎
By Claims 10.5.a, 10.5.c, 10.5.b and 10.5.e, we have shown that $`(\tau ,)`$ witnesses meet regularity for $`g`$. Since $`(\tau ,)`$ was an arbitrary vip $`m`$-type, every such type witness meet regularity for $`g`$. Thus $`f=ge\psi `$ is a pnp map with meet regularity whose range is a subset of the range of $`g`$. ∎
###### Lemma 10.6.
Suppose that $`\kappa `$ is a cardinal which is measurable in the generic extension obtained by adding $`\lambda `$ Cohen subsets of $`\kappa `$, where $`\lambda (\kappa )_{2^\kappa }^6`$. Further suppose $`g:{}_{}{}^{\kappa >}2{}_{}{}^{\kappa >}2`$ is a pnp map which satisfies meet regularity and whose range is a strongly diagonal set. Then there is a pnp map $`f`$ which satisfies meet regularity and preserves lexicographic order and whose range is a subset of the range of $`g`$.
###### Proof.
Apply Lemma 9.13 to obtain a pre-$`S`$-vip order $``$ and a sequence $`\phi _t:t{}_{}{}^{\kappa >}2`$ such that for all $`t{}_{}{}^{\kappa >}2`$, the three listed properties of the lemma hold. Let $`\tau _0`$ be the $`2`$-type whose leaves are $`0`$ and $`1,0`$. Let $`\tau _0`$ be the $`2`$-type whose leaves are $`1`$ and $`0,0`$. Let $`\tau _2`$ be the $`2`$-type whose leaves are $`0`$ and $`1,1`$. Let $`\tau _3`$ be the $`2`$-type whose leaves are $`1`$ and $`0,1`$. Then for all $`x[{}_{}{}^{\kappa }2]^2`$, the set $`\mathrm{clp}(x)`$ is one of $`\tau _0`$, $`\tau _1`$, $`\tau _2`$ and $`\tau _3`$.
Define a coloring $`d^{\prime \prime }`$ on pairs from $`{}_{}{}^{\kappa }2`$ by $`d^{\prime \prime }(x)=j`$ where $`\mathrm{clp}(g[x])=\tau _j`$.
Apply Shelah’s Theorem 2.5 to $`d^{\prime \prime }`$ and $``$ to obtain a strong embedding $`e`$ and $`w{}_{}{}^{\kappa }2`$ such that $`e`$ preserves $``$ on $`\mathrm{Cone}(w)`$ and for all pairs $`x`$ of $`T=e[\mathrm{Cone}(w)]`$, the value of $`d^{\prime \prime }(x)`$ depends only on the $``$-ordered similarity type of $`x`$.
Let $`\psi =\phi _w`$. Then $`\psi `$ is a pnp diagonalization into $`S\mathrm{Cone}(w)`$. We claim that $`f:=ge\psi ge\psi `$ is a pnp map which satisfies meet regularity and preserves lexicographic order. Clearly the range of $`f`$ is a subset of the range of $`g`$.
Since $`e\psi `$ is a pnp map which witnesses meet regularity and whose range is a strongly diagonal set, by Lemma 10.5, the functions $`ge\psi `$ and $`f`$ satisfy meet regularity.
Since $`g`$ is a pnp map, for any pair $`x`$ from $`e[\mathrm{Cone}(w)]`$ with $`\mathrm{clp}(x)=\tau _0`$ or $`\mathrm{clp}(x)=\tau _1`$, we must have $`\mathrm{clp}(g(x))=\tau _0`$ or $`\mathrm{clp}(g(x))=\tau _1`$. Similarly, for any pair $`x`$ from $`e[cone(w)]`$ with $`\mathrm{clp}(x)=\tau _1`$ or $`\mathrm{clp}(x)=\tau _2`$, we must have $`\mathrm{clp}(g(x))=\tau _1`$ or $`\mathrm{clp}(g(x))=\tau _2`$.
Say $`g`$ *sends $`i`$ to $`j`$* if for all incomparable pairs $`x`$ from $`e[\mathrm{Cone}(w)]`$ with $`\mathrm{clp}(x)=\tau _i`$ one has $`\mathrm{clp}(g[x])=\tau _j`$.
###### Claim 10.6.a.
Either (a) $`g`$ sends $`0`$ to $`1`$ and $`1`$ to $`0`$, or (b) $`g`$ sends $`0`$ to $`0`$ and $`1`$ to $`1`$.
###### Proof.
Consider $`a=0,0,0,0`$, $`b=0,1`$ and $`c=1,0,0`$. Then $`\{a,b,c\}`$ is diagonal, and $`ca=\mathrm{}=cb`$.
Now $`\mathrm{clp}(\{a,c\})=\tau _1`$ and $`\mathrm{clp}(\{b,c\})=\tau _0`$. Let $`s=e(\psi (a))`$, $`t=e(\psi (b))`$ and $`u=e(\psi (c))`$. Set $`x:=\{s,u\}`$ and $`y=\{t,u\}`$. Since $`e\psi `$ is a polite pnp map, by Lemma 9.9, we have $`\mathrm{clp}(x)=\tau _1`$ and $`\mathrm{clp}(y)=\tau _0`$.
Since $`ge\psi `$ satisfies meet regularity, $`g(u)g(s)=g(u)g(t)`$. That is, $`g(s)`$ and $`g(t)`$ are on the same side of $`g(u)`$. Let $`\gamma =\mathrm{lg}(g(u)g(s))`$. Then $`g(s)(\gamma )=g(t)(\gamma )=1g(u)(\gamma )`$. If $`g(s)(\gamma )=0`$ then $`\mathrm{clp}(g[x])=\tau _1`$ and $`\mathrm{clp}(g[y])=\tau _0`$, so $`g`$ sends $`1`$ to $`1`$ and $`0`$ to $`0`$. Otherwise $`g(s)(\gamma )=1`$. In this case $`\mathrm{clp}(g[x])=\tau _0`$ and $`\mathrm{clp}(g[y])=\tau _1`$, so $`g`$ sends $`1`$ to $`0`$ and $`0`$ to $`1`$. ∎
###### Claim 10.6.b.
Either (a) $`g`$ sends $`2`$ to $`3`$ and $`3`$ to $`2`$, or (b) $`g`$ sends $`2`$ to $`2`$ and $`3`$ to $`3`$.
###### Proof.
The proof parallels that of the previous claim using $`a=1,1,1,1`$, $`b=1,0`$ and $`c=0,0,1`$. The details are left to the reader. ∎
Notice that $`e\psi `$ sends pairs from $`{}_{}{}^{\kappa }2`$ to incomparable pairs from $`\mathrm{Cone}(w)`$. For incomparable pairs, $`e\psi `$ preserves the similarity type by Lemma 9.9. Hence by the above claims, for any incomparable pair $`z`$ from $`{}_{}{}^{\kappa }2`$ with $`\mathrm{clp}(z)=\tau _i`$, we have $`\mathrm{clp}(f[z])=\tau _i`$. Therefore $`f`$ is a pnp map which satisfies meet regularity and preserves lexicographic order and has range a subset of the range of $`g`$. ∎
###### Theorem 10.7.
Suppose that $`\kappa `$ is a cardinal which is measurable in the generic extension obtained by adding $`\lambda `$ Cohen subsets of $`\kappa `$, where $`\lambda (\kappa )_{2^\kappa }^6`$. If $`g:{}_{}{}^{\kappa >}2{}_{}{}^{\kappa >}2`$ is a pnp map whose range is a strongly diagonal set whose meet closure is a subset of $`S=\sigma [\kappa ]`$, then there is a pnp diagonalization $`f:{}_{}{}^{\kappa >}2\mathrm{ran}(g)`$ into $`S`$ with level harmony.
###### Proof.
Apply Lemma 9.13 to obtain a pre-$`S`$-vip order $``$ and a sequence $`\phi _t:t{}_{}{}^{\kappa >}2`$ such that for all $`t{}_{}{}^{\kappa >}2`$, the three listed properties of the lemma hold. Apply Lemma 10.5 to $`g`$ to obtain a pnp map $`g_0`$ which satisfies meet regularity and whose range is strongly diagonal subset of the range of $`g`$. Apply Lemma 10.6 to $`g_0`$ to obtain a pnp map $`h`$ which satisfies meet regularity and preserves lexicographic order and whose range is a strongly diagonal subset of $`\mathrm{ran}(g_0)\mathrm{ran}(g)`$.
Define a coloring $`d^{}`$ on three element antichains $`b=\{b_0,b_1,b_2\}`$ listed in increasing order of length as follows: $`d^{}(b)=0`$ if $`\mathrm{lg}(b_0)<\mathrm{lg}(b_1b_2)`$ and $`\mathrm{lg}(h(b_0))<\mathrm{lg}(h(b_1)h(b_2))`$; $`d^{}(b)=1`$ if $`\mathrm{lg}(b_0)<\mathrm{lg}(b_1b_2)`$ and $`\mathrm{lg}(h(b_0))\mathrm{lg}(h(b_1)h(b_2))`$; and $`d^{}(b)=2`$ otherwise.
Apply Shelah’s Theorem 2.5 to $`d^{}`$ and $``$ to obtain a strong embedding $`e`$ and $`w{}_{}{}^{\kappa }2`$ such that $`e`$ preserves $``$ on $`\mathrm{Cone}(w)`$ and for all pairs $`x`$ of $`T=e[\mathrm{Cone}(w)]`$, the value of $`d^{}(x)`$ depends only on the $``$-ordered similarity type of $`x`$.
Let $`\psi =\phi _w`$. Then $`\psi `$ is a pnp diagonalization into $`S\mathrm{Cone}(w)`$ with level harmony. We claim that $`f^{}:=he\psi `$ is a pnp map which satisfies meet regularity and preserves lexicographic order. Clearly the range of $`f^{}`$ is a subset of the range of $`g`$.
Since $`e\psi `$ is a pnp map which witnesses meet regularity and preserves lexicographic order, and whose range is a strongly diagonal set, by Lemma 10.6, the function $`f^{}`$ satisfies meet regularity and preserves lexicographic order.
Our next goal is to prove $`f^{}`$ also preserves meet length order. The claim below is a preliminary step toward that goal.
###### Claim 10.7.a.
If $`b=\{b_0,b_1,b_2\}e[\mathrm{Cone}(w)]`$ is a strongly diagonal set listed in increasing order of length and $`\mathrm{lg}(b_0)<\mathrm{lg}(b_1b_2)`$, then $`d^{}(b)=0`$ and $`\mathrm{lg}(h(b_0))<\mathrm{lg}(h(b_1)\mathrm{lg}(h(b_2))`$.
###### Proof.
Assume toward a contradiction that $`b=\{b_0,b_1,b_2\}e[\mathrm{Cone}(w)]`$ is a strongly diagonal set listed in increasing order of length with $`\mathrm{lg}(b_0)<\mathrm{lg}(b_1b_2)`$ and $`d^{}(b)0`$. Then $`d^{}(b)=1`$. Let $`\tau =\mathrm{clp}(b)`$ and list the elements of $`a`$ in increasing order of length as $`\tau =\{a_0,a_1,a_2\}`$. Then $`\mathrm{lg}(a_0)`$.
Let $`z=e(\psi (a_0))`$ and let $`\nu `$ be an uncountable cardinal greater than $`2^{|\mathrm{lg}(h(z))|}`$. Using the proof of Claim 10.5.a as a guide, construct a sequence $`b_\alpha :\alpha <\nu `$ such that for all $`\alpha <\nu `$, $`a_0b_\alpha =\mathrm{}`$ and for all $`\beta <\alpha <\nu `$, $`\mathrm{clp}(\{a_0,b_\beta ,b_\alpha \})=\tau `$ and $`\mathrm{lg}(b_\beta )<\mathrm{lg}(b_\alpha )`$. Set $`s_\alpha :=e(\psi (b_\alpha ))`$ for all $`\alpha <\nu `$. For $`\beta <\alpha <\nu `$, let $`x(\beta ,\alpha ):=\{z,s_\beta ,s_\alpha \}`$. Thus for all $`\beta <\alpha <\nu `$, $`\mathrm{clp}(x(\beta ,\alpha )=\tau `$.
There is only one vip order $``$ on $`\tau `$. Since $`\psi `$ is a pnp diagonalization into $`S\mathrm{Cone}(w)`$, $``$ is a pre-$`S`$-vip order, and $`e`$ preserves $``$ on $`\mathrm{Cone}(w)`$, it follows that $`(\mathrm{clp}(x(\beta ,\alpha ),_{x(\beta ,\alpha )})=(\tau ,)`$, since both are vip $`3`$-types and $`\mathrm{clp}(x(\beta ,\alpha )=\tau `$.
Since $`d^{}`$ takes the same value on all triples from $`e[\mathrm{Cone}(w)]`$ of the same $``$-ordered similarity type, it follows that $`\mathrm{lg}(h(z))\mathrm{lg}(h(s_\beta )h(s_\alpha ))`$ for all $`\beta <\alpha <\nu `$. Since the range of $`h`$ is a strongly diagonal set, the inequalities must be strict.
Since $`\nu `$ is an uncountable cardinal greater than $`2^{|\mathrm{lg}(h(z))|}`$, by the Pigeonhole Principle, there are $`\beta <\gamma `$ with $`h(t_\beta )\mathrm{lg}(h(z))=h(t_\gamma )\mathrm{lg}(h(z))`$. Thus we have reached the contradiction that $`h(t_\beta )h(t_\gamma )`$ must have length short and greater than or equal to $`\mathrm{lg}(h(z))`$. Thus the claim follows. ∎
###### Claim 10.7.b.
If $`\mathrm{lg}(xy)<\mathrm{lg}(uv)`$ and $`|\{x,y,u,v\}|=3`$, then $`\mathrm{lg}(f^{}(x)f^{}(y))<\mathrm{lg}(f^{}(u)f^{}(v))`$.
###### Proof.
If $`x=y`$, then the claim follows from Claim 10.7.a. So assume $`xy`$. Then one of $`x`$ and $`y`$ is either $`u`$ or $`v`$, so $`xyuv`$. Let $`z`$ be the unique one of $`x`$ and $`y`$ which is not in $`\{u,v\}`$. Then $`zu=zv=xy`$. By meet regularity, $`f^{}(z)f^{}(u)=f^{}(z)f^{}(v)=f^{}(x)f^{}(y)`$. Let $`\beta =\mathrm{lg}(f^{}(x)f^{}(y)`$. Since $`f^{}`$ preserves lexicographic order, $`f^{}(z)(\beta )f^{}(u)(\beta )=f^{}(v)(\beta )`$. It follows that $`\mathrm{lg}(f^{}(u)f^{}(v))>\beta =\mathrm{lg}(f^{}(x)f^{}(y))`$. ∎
###### Claim 10.7.c.
If $`xyuv`$ and $`|\{x,y,u,v\}|=4`$, then $`\mathrm{lg}(f^{}(x)f^{}(y))<\mathrm{lg}(f^{}(u)f^{}(v))`$.
###### Proof.
If $`xy=x`$ or $`xy=y`$, then the claim follows from Claim 10.7.a. So assume $`xxyy`$. Since $`xy`$ is a proper initial segment of $`uv`$, either $`(xy){}_{}{}^{}0uv`$ or $`(xy){}_{}{}^{}1uv`$. Consequently, either $`xy=xu=xv`$ or $`xy=yu=yv`$. In the first case, the claim follows from Claim 10.7.b applied to $`\{x,u,v\}`$ and in the second case, it follows from Claim 10.7.b applied to $`\{y,u,v\}`$. ∎
###### Claim 10.7.d.
If $`\mathrm{lg}(xy)<\mathrm{lg}(uv)`$, $`xvuv`$ and $`|\{x,y,u,v\}|=4`$, then $`\mathrm{lg}(f^{}(x)f^{}(y))<\mathrm{lg}(f^{}(u)f^{}(v))`$.
###### Proof.
Since $`\mathrm{lg}(xy)<\mathrm{lg}(uv)`$ and $`xvuv`$, it follows that $`xy`$ and $`uv`$ are incomparable. By Lemma 10.4, $`\mathrm{lg}(f^{}(x)f^{}(y))lg(f^{}(xy))`$.
If one of $`u`$ and $`v`$ is an initial segment of the other. Since $`e\psi `$ is a pnp map, $`e(\psi (xy))`$ is shorter than both $`e(\psi (u))`$ and $`e(\psi (v))`$. If neither is an initial segment of the other, the $`e(\psi (xy))`$ is shorter than than both $`e(\psi (u))`$ and $`e(\psi (v))`$ since $`e\psi `$ preserves meet length order. Thus by Claim 10.7.a applied to $`\{e(\psi (xy)),e(\psi (u)),e(\psi (v))\}`$, $`\mathrm{lg}(f^{}(xy))<\mathrm{lg}f^{}(u)\mathrm{lg}(f^{}(v))`$ and the claim follows. ∎
###### Claim 10.7.e.
The function $`f^{}`$ preserves meet length order.
###### Proof.
Let $`x,y,u,v`$ be arbitrary with $`\mathrm{lg}(xy)<\mathrm{lg}(uv)`$. Consider the set $`\{x,y,u,v\}`$. It must have at least two elements and at most four. If it has three or four elements, then $`\mathrm{lg}(f^{}(x)f^{}(y))<\mathrm{lg}(f^{}(u)f^{}(v))`$ by one of Claims 10.7.b, 10.7.c and 10.7.d. If it has two elements, say $`z`$ and $`w`$ with $`\mathrm{lg}(z)<\mathrm{lg}(w)`$, then there are only three possible meets, listed here in increasing order of length: $`zw`$, $`zz`$ and $`ww`$. Since $`f^{}`$ is a pnp map, $`\mathrm{lg}(f^{}(z))<\mathrm{lg}(f^{}(w))`$. Since the range of $`f^{}`$ is strongly diagonal set, $`f^{}(z)`$ and $`f^{}(w)`$ are incomparable, so $`f^{}(z)f^{}(w)`$ is a proper initial segment of both $`f^{}(z)`$ and $`f^{}(w)`$. Thus $`f^{}`$ preserves meet length order. ∎
Since $`f^{}`$ satisfies meet regularity and preserves both lexicographic order and meet length order, it is polite. We saw above that $`f^{}`$ is a pnp map whose range is a subset of the range of $`g`$. Recall that $`\psi `$ is a pnp diagonalization with level harmony into $`S\mathrm{Cone}(w)`$. Thus by Lemma 10.3, the function $`f=gf^{}`$ is a pnp diagonalization into $`S`$ with $`\mathrm{ran}(f)\mathrm{ran}(g)S`$. ∎
Note that for $`m=2`$, there are four $`m`$-types and each of them admits a single vip order. Any copy of an uncountable Rado $`𝔾`$ has an induced subgraph which is a countable Rado graph. Since Laflamme, Sauer and Vuksanovic have shown that these four types must appear in translations of every induced subgraph of the countable Rado graph which is itself isomorphic to the countable Rado graph, it follows from their work that $`𝔾_\kappa (𝔾_\kappa )_{<\omega ,r_2^+1}^2`$. For larger values of $`m`$, we use Shelah’s Theorem.
###### Question 10.8.
Suppose $`\kappa `$ is an uncountable cardinal with $`\kappa ^{<\kappa }=\kappa `$, $`𝔾_\kappa `$ is a $`\kappa `$-Rado graph and $`2<m<\omega `$. Does $`𝔾(𝔾)_{<\omega ,r_m^+1}^m`$? That is, does the lower bound hold even when $`\kappa `$ does not satisfy the hypothesis of Shelah’s Theorem?
###### Theorem 10.9.
Let $`m3`$ be a natural number and suppose that $`\kappa `$ is a cardinal which is measurable in the generic extension obtained by adding $`\lambda `$ Cohen subsets of $`\kappa `$, where $`\lambda (\kappa )_{2^\kappa }^6`$. Further suppose $`𝔾_\kappa `$ is a $`\kappa `$-Rado graph. Then for $`r_m^+`$ equal to the number of vip $`m`$-types, $`𝔾_\kappa `$ satisfies
$$𝔾_\kappa (𝔾_\kappa )_{<\omega ,r_m^+1}^m.$$
###### Proof.
Apply Lemma 9.13 to obtain a pre-$`S`$-vip order $``$ and a sequence $`\phi _t:t{}_{}{}^{\kappa >}2`$ such that for all $`t{}_{}{}^{\kappa >}2`$, the three listed properties of the lemma hold. Enumerate the vip $`m`$-types: $`(\tau _0,_0)`$, $`(\tau _1,_1)`$, …, $`(\tau _{r1},_{r1})`$.
Define a coloring $`d:[{}_{}{}^{\kappa >}2]^mr`$ by $`d(x)=i`$ if $`(\mathrm{clp}(\phi _0[x]),_{\phi _0[x]})=(\tau _i,_i)`$. Then $`c:[\kappa ]^mr`$ defined by $`c(b)=d(\sigma [b])`$ is a coloring of the $`m`$-element subsets of $`\kappa `$ with $`r`$ colors.
To prove the theorem, we must show that every isomorphic copy of $`𝔾_\kappa `$ contains sets of every color. Toward that end, let $`K\kappa `$ be an arbitrary set such that $`(K,EK)(\kappa ,E)`$ and let $`\rho :\kappa K`$ be the isomorphism and let $`j<r`$ be arbitrary. By Lemmas 9.4 and 9.6, we may assume that $`\sigma \rho \sigma ^1`$ is a pnp map with domain $`S`$. Hence $`g:=\phi _0\sigma \rho \sigma ^1\phi _0`$ is a pnp map with domain $`{}_{}{}^{\kappa >}2`$ whose range is a strongly diagonal subset of $`S`$ and $`\mathrm{ran}(g)^{}S`$.
Apply Theorem 10.7 to $`g`$ obtain a pnp diagonalization $`f:{}_{}{}^{\kappa >}2\mathrm{ran}(g)`$ into $`S`$ with level harmony. Since $``$ is a pre-$`S`$-vip order, it is also a $`D`$-vip order for $`D=\mathrm{ran}(f)`$.
By Theorem 10.1 applied to $`f`$ and $`D`$, there is some $`yD`$ such that $`(\tau _j,_j)=(\mathrm{clp}(y),^y)`$. Recall that $`\mathrm{ran}(f)\mathrm{ran}(g)`$. Let $`a`$ be such that $`g[a]=x`$. Then $`b:=\rho [\sigma ^1[\phi _0[a]]]`$ is well-defined and by definition of $`\sigma `$ and $`\rho `$, we can see that $`bK`$. Let $`x=\sigma [b]`$. Then $`\phi _0[x]=g[a]=y`$ by definition of $`g`$ and choice of $`a`$. So $`d(x)=j`$, by definition of $`d`$. Thus $`c(b)=d(\sigma [b])=j`$. Since $`K`$ and $`j`$ were arbitrary, every isomorphic copy of $`𝔾_\kappa `$ contains sets of every color. ∎
Figure 3 summarizes the calculation from of values of $`r_m^+`$ for $`m5`$. A comparison with $`r_m`$, the number of $`m`$-types, is also included, where $`r_m`$ is the critical value for finite colorings of $`m`$-tuples of the countable Rado graph.
|
warning/0506/hep-ex0506019.html
|
ar5iv
|
text
|
# BEAUTY PRODUCTION WITH THE ALICE DETECTOR
## 1 Why studying heavy quarks in high energy nuclear collisions?
ALICE (A Large Ion Collider Experiment) is a detector dedicated to the study of nucleus-nucleus interactions at the LHC. It will investigate the physics of strongly interacting matter at extreme energy densities, where the formation of a new phase of matter, the QGP (Quark-Gluon Plasma) is expected. Heavy flavour physics is an important part of the ALICE experimental programme in many respects. As an intrinsically perturbative phenomenon heavy quark production is a key benchmark for testing QCD and parton model concepts $`^{\mathrm{?},\mathrm{?},\mathrm{?}}`$ in a novel kinematic region of large $`Q^2`$ and very low Bjorken-$`x`$ (as low as about $`10^5`$) becoming accessible at the LHC. In the context of heavy ion collisions, open heavy quark production is sensitive to gluon shadowing effects in nuclei $`^{\mathrm{?},\mathrm{?}}`$. Moreover, due to their long lifetime heavy quarks live through the thermalization phase of a QGP and therefore carry information about the deconfined medium $`^\mathrm{?}`$. Measuring the beauty hadron production down to the very low $`p_\mathrm{t}`$ – region where the ALICE detector has been optimized for tracking and particle identification (cf. section 3) – is essential to minimize the extrapolation uncertainty on the total cross section. In addition, the low $`p_\mathrm{t}`$ region will be influenced by non-perturbative effects which can differ from p-p to A-A collisions. For heavy quarkonium physics, the measurement of the beauty quark yield in the same kinematic region as for the $`\mathrm{{\rm Y}}`$ measurement will provide a valuable reference against which to compare $`\mathrm{{\rm Y}}`$ production in order to observe a possible suppression in A-A collisions. In the following, we will present the expected performance of the ALICE detector – presently under construction – for beauty production measurements in semileptonic decay channels.
## 2 Production of $`𝒃`$-quarks at LHC
Heavy quarks are produced in hadron-hadron collisions at LHC energies by strong interaction processes described within the QCD theory. Production cross section can be computed in the framework of the factorization theorem of the QCD parton model as the convolution of parton-parton scattering cross sections (calculable in pQCD) with parton distribution functions. QCD higher order corrections to the Born reactions have to be considered at LHC energies. Fixed-order NLO calculations of heavy quark production cross sections are available $`^\mathrm{?}`$ and have been used to define the ALICE baseline $`^\mathrm{?}`$. In the analysis presented hereafter, the Pythia 6.214 event generator $`^\mathrm{?}`$ has been used to produce heavy quarks after tuning to reproduce the relevant NLO differential distributions (namely single quark $`p_\mathrm{t}`$, rapidity, heavy quark pair invariant mass and azimuthal correlation). Rates in heavy ion collisions are then obtained from binary scaling in accordance with the Glauber multiple scattering model $`^\mathrm{?}`$. “Known” initial state effects are included (namely primordial transverse momentum and nuclear shadowing). The expected production rates are presented in Table 1.
## 3 Overview of the ALICE detector for lepton detection
Both electrons and muons are measured in ALICE, electrons in the central barrel and muons in a dedicated forward spectrometer. Central detectors, covering mid-rapidity ($`|\eta |0.9`$) over the full azimuth, are embedded inside the L3 solenoidal magnet providing a magnetic field of $`0.5\mathrm{T}`$. Electrons are identified in ALICE combining the TRD (Transition Radiation Detector) and TPC (Time Projection Chamber) particle identification capabilities. TRD provides an $`e`$/$`\pi `$ rejection power of 100 supplemented by the specific electron energy loss in the TPC which allows to reach a combined pion rejection factor of $`10,000`$. The ALICE forward muon spectrometer has been designed to detect muons emitted in the pseudo-rapidity region $`4\eta 2.5`$. The forward muon spectrometer is made of a passive front absorber of 10 interaction lengths to absorb hadrons and photons from the interaction vertex, a high-granularity tracking system of 10 cathod pad chambers, a large dipole magnet creating a field of 0.7 T (field integral of $`3\mathrm{T}\mathrm{m}`$), and a trigger system performing the selection of heavy flavour decay muons by a transverse momentum cut-off made of four resistive plate chambers. Muons penetrating the whole spectrometer length are finally identified with a momentum resolution of about 1 % and 90 % efficiency (for $`p_\mathrm{t}>3\mathrm{GeV}/\mathrm{c}`$).
## 4 Beauty production measurements from semileptonic decays
The copious beauty production at the LHC will be measured with the ALICE detector using $`b`$-hadron semileptonic decays characterized by large branching ratios. Leptons can indeed be produced both from direct decay $`b\mathrm{}^{}`$ ($`\mathrm{BR}10\%`$) and cascade $`bc\mathrm{}^+`$ ($`\mathrm{BR}8\%`$). After extracting the beauty signal in lepton data sets, $`b`$-hadron production cross sections are assessed by unfolding $`b\mathrm{}`$ decay.
### Single inclusive $`b`$-quark production via muon detection
Beauty production in Pb-Pb collisions will be measured in ALICE using semileptonic decay muons in the pseudo-rapidity region $`4\eta ^\mu 2.5`$. Both inclusive muon and opposite sign dimuon productions are considered, the dimuon sample being divided into two topologically distinct contributions: $`b`$-chain decays (named $`\mathrm{BD}_{\mathrm{same}}`$ hereafter) of low mass $`M_{\mu \mu }<5\mathrm{GeV}/\mathrm{c}^2`$ and high transverse momentum (cf. Fig. 1(a)) and muon pairs where the two muons originate from different quarks ($`\mathrm{BB}_{\mathrm{diff}}`$) emitted at large angles resulting in large invariant masses $`M_{\mu \mu }>5\mathrm{GeV}/\mathrm{c}^2`$ (cf. Fig. 1(b)). Beauty signal is enhanced with respect to other sources (charm and $`\pi /K`$ decay-in-flight) applying a low $`p_\mathrm{t}`$ cut-off of $`1.5\mathrm{GeV}/\mathrm{c}`$ $`^\mathrm{?}`$. Using Monte Carlo predicted line shapes for $`b\overline{b}`$, $`c\overline{c}`$, and decay background, fits are carried out to find the $`b\overline{b}`$ fraction in the different data sets as presented in Fig. 1(a), (b), and (c). Finally muon level cross sections are converted into inclusive $`b`$-hadron cross section following the method initially developped for the UA1 experiment $`^\mathrm{?}`$. The $`b`$-hadron cross section measurement is plotted in Fig. 1(d), it shows no systematic bias with respect to the initial distribution used in the simulation to produce beauty signal even if a detailed study of systematic uncertainties has still to be done. $`b`$-decay muon statistics is large over the whole $`p_\mathrm{t}`$ range allowing a tight mapping of the production cross section up to $`p_\mathrm{t}=30\mathrm{GeV}/\mathrm{c}`$.
### Single inclusive $`b`$-quark production via electron detection
Electrons from semileptonic decay of $`b`$-quarks are characterized by a hard $`p_\mathrm{t}`$ spectrum and a large average impact parameter<sup>b</sup><sup>b</sup>bDefined as the distance of closest approach of the electron track to the primary vertex in the transverse plane. $`d_0`$ is measured by the ALICE Inner Tracking System (ITS) with a resolution $`70\mu \mathrm{m}`$ at $`p_\mathrm{t}=1\mathrm{GeV}/\mathrm{c}`$. $`d_0`$ ($`300\mu \mathrm{m}`$) with respect to other electron sources (pion misidentification, charm, light mesons, and photon conversions). The beauty signal purity and statistics as a function of the impact parameter cut-off for different values of the $`p_\mathrm{t}`$ threshold are shown in Fig. 2 a) and b) respectively. $`p_\mathrm{t}>2\mathrm{GeV}/\mathrm{c}`$ and $`200<|d_0|<600\mu \mathrm{m}`$ provide an electron sample of $`8\times 10^4`$ with a 90 % purity $`^\mathrm{?}`$. An upper limit on $`d_0`$ is needed to reduce long lived strange particles and tracks suffering from large angle scatterings in detector materials.
## 5 Conclusion
The ALICE detector has good capabilities for heavy flavour measurements $`^\mathrm{?}`$ with the unique ability to address the $`b`$-quark sector in p-p, p-A and A-A collisions with relatively low transverse momentum thresholds. The measurement of $`b`$ production provides an important test of the theory of QCD in heavy ion collisions where new effects are expected as compared to nucleon-nucleon interactions. A precise measurement of the inclusive $`\mathrm{d}\sigma ^B/\mathrm{d}\mathrm{p}_\mathrm{t}`$ cross section will allow to probe in-medium quenching effects. Channels discussed in these proceedings will be supplemented by measuring dilepton correlations, multi-muon topologies, and $`b`$-tagged jets. Quantitative studies are currently underway.
## Acknowledgments
The author would like to thank F. Antinori and A. Dainese, members of the ALICE Collaboration, for providing me the relevant material to prepare this talk and valuable discussions.
## References
|
warning/0506/math0506517.html
|
ar5iv
|
text
|
# Self-adjointness of Cauchy singular integral operator
## Abstract.
We extend Krupnik’s criterion of self-adjointness of the Cauchy singular integral operator to the case of finitely connected domains. The main aim of the paper is to present a new approach for proof of the criterion.
###### Key words and phrases:
Cauchy singular integral operator, Smirnov spaces
Let $`G_+`$ be a finitely connected domain bounded by the rectifiable curve $`C=G_+`$, $`G_{}=\mathrm{clos}G_+`$ and $`\mathrm{}G_{}`$. Suppose also that $`w(z),zC`$ is a nonnegative weight such that $`w(z)0`$ on each connected component of the curve $`C`$. For any $`fL^2(C,|dz|)`$, we denote by $`f_\pm (z),zC`$ the angular boundary values of the Cauchy transform
$$𝒦(f,\lambda ):=\frac{1}{2\pi i}\underset{C}{}\frac{f(z)}{z\lambda }𝑑z,\lambda C$$
from the domains $`G_\pm `$ respectively. The well-known David’s theorem says that the mappings $`P_\pm :f\pm f_\pm `$ are bounded linear operators in $`L^2(C,|dz|)`$ if and only if the curve $`C`$ is a Carleson curve. Moreover, the operators $`P_\pm `$ are bounded in $`L^2(C,w(z)|dz|)`$ if and only if the weight $`w`$ is a Mackenhoupt weight (see, e.g. ), where the vector space $`L^2(C,w(z)|dz|)`$ is endowed with the inner product
$$(f,g)_{L^2(C,w)}=\frac{1}{2\pi }\underset{C}{}f(z)\overline{g(z)}w(z)|dz|,$$
and $`|dz|`$ is the arc-length measure. In the sequel, we always assume that $`C`$ is a Carleson curve.
In the paper we are interested in finding necessary and sufficient conditions for self-adjointness of the projections $`P_\pm `$ and therefore (note that $`P_++P_{}=I`$) for self-adjointness of the corresponding Cauchy singular integral operator
$$𝒦_S(f,\lambda ):=\underset{\epsilon 0}{lim}\frac{1}{2\pi i}\underset{C(\lambda ,\epsilon )}{}\frac{f(z)}{z\lambda }𝑑z=\frac{1}{2}((P_+f)(\lambda )(P_{}f)(\lambda )),$$
where $`\lambda C`$ and $`C(\lambda ,\epsilon )=\{zC:|z\lambda |<\epsilon \}`$. For simple connected domains (i.e., when C is a simple closed curve) N.Krupnik has established the criterion: the bounded operator $`𝒦_S`$ is self-adjoint if and only if $`C`$ is a circle and $`w(z)const`$ (see also for previous results and for applications). Our extension of Krupnik’s criterion to the case of multiply connected domains is a consequence of the following theorem, which is a slightly more strong assertion than Krupnik’s result.
###### Theorem.
Let $`wL^1(C,|dz|)`$. The following conditions are equivalent:
1) $`\lambda G_+\mu G_{}(\frac{1}{z\lambda },\frac{1}{z\mu })_{L^2(C,w)}=0`$;
2) the curve $`C`$ is a circle and $`w(z)const`$.
###### Proof.
The implication $`2)\mathrm{\hspace{0.17em}1})`$ is obvious.
$`1)\mathrm{\hspace{0.17em}2})`$ . Let $`w(z)|dz|=k(z)dz`$. Obviously, $`kL^1(C,|dz|)`$ . Then $`\lambda G_+`$
$$0=\underset{\mu \mathrm{}}{lim}\underset{C}{}\frac{1}{z\lambda }\frac{\overline{\mu }}{\overline{z}\overline{\mu }}k(z)𝑑z=\underset{C}{}\frac{k(z)}{z\lambda }𝑑z.$$
By Smirnov’s theorem , we get $`k(z)E^1(G_{})`$. For the same reason, we have $`\lambda G_+\mu G_{}n`$
$$0=\underset{C}{}\frac{1}{z\lambda }\frac{k(z)}{(\overline{z}\overline{\mu })^n}𝑑z\frac{k(z)}{(\overline{z}\overline{\mu })^n}E^1(G_{}).$$
Let $`f`$ be defined by $`f(z)=\overline{z},zC`$. Evidently, we have
$$f(z)=\overline{\mu }+\frac{k(z)}{g(z)},zC,\text{where}g(z)=\frac{k(z)}{\overline{z}\overline{\mu }}E^1(G_{})$$
and therefore the function $`f(z)`$ admits meromorphic continuation into the domain $`G_{}`$. Since $`n`$ $`\frac{k(z)}{(f(z)\overline{\mu })^n}E^1(G_{})`$, we get $`f:G_{}\overline{\mathrm{clos}G_+}`$ and $`f(z)H^{\mathrm{}}(G_{})`$.
The curve $`C`$ can be represented in the form $`C=_{k=0}^nC_k`$, where $`C_k`$ are simple closed contours. We have $`G_+=_{k=0}^nG_{k+}`$, $`G_{}=_{k=0}^nG_k`$, and $`\mathrm{}G_0`$, where $`G_{0+}=\mathrm{Int}C_0,G_0=\mathrm{Ext}C_0`$, $`G_{k+}=\mathrm{Ext}C_k,G_k=\mathrm{Int}C_k,k=\overline{1,n}`$. Evidently, $`f:G_0\overline{\mathrm{clos}G_{0+}}`$ and $`f(z)H^{\mathrm{}}(G_0)`$. Besides, $`f(z)=\overline{z}`$ is an one-to-one correspondence between $`C_0`$ and $`\overline{C_0}=\{\overline{z}:zC_0\}`$. Hence the function $`f`$ is a conformal mapping of $`G_0`$ onto $`\overline{G_{0+}}`$.
Let $`z=\phi (\zeta )`$ be a conformal mapping of the unit disk $`𝔻`$ onto $`G_{0+}`$ and $`\zeta =\psi (z)`$ be its inverse. Consider the function $`\psi _{\mathrm{}}(z)=\overline{\psi (\overline{f(z)})}`$ and its inverse $`z=\phi _{\mathrm{}}(\zeta )`$. It is clear that $`\phi _{\mathrm{}}`$ is a conformal mapping of $`𝔻`$ onto $`G_0`$. Further, we have
$$\psi _{\mathrm{}}(z)=\overline{\psi (\overline{\overline{z}})}=\overline{\psi (z)},zC_0;\psi _{\mathrm{}}(\phi (\zeta ))=\overline{\psi (\phi (\zeta ))}=\overline{\zeta },|\zeta |=1$$
and $`\phi (\zeta )=\phi _{\mathrm{}}(\overline{\zeta })`$. Therefore the function
$$\mathrm{\Phi }(\zeta )=\{\begin{array}{cc}\phi (\zeta ),\hfill & |\zeta |1,\hfill \\ \phi _{\mathrm{}}(1/\zeta ),\hfill & |\zeta |1\hfill \end{array}$$
is a conformal mapping of the whole complex plane $``$ onto itself and we obtain that $`\mathrm{\Phi }(\zeta )=\frac{a\zeta +b}{c\zeta +d}`$, and $`C_0`$ is a circle.
The curves $`C_k,k=\overline{1,n}`$ are circles too. This claim can be reduced to the case of $`C_0`$: it follows easily from the observation that the operator
$$\begin{array}{c}C_\phi :L^2(\phi (C),w(\stackrel{~}{z})|d\stackrel{~}{z}|)L^2(C,w(\phi (z))|dz|),\hfill \\ (C_\phi f())(z):=\sqrt{\phi ^{}(z)}f(\phi (z)),zC,fL^2(\phi (C),w(\stackrel{~}{z})|d\stackrel{~}{z}|)\hfill \end{array}$$
is an unitary operator and from the straightforward computation
$$(C_\phi f())(z)=\frac{c\lambda +d}{\sqrt{adbc}}\frac{1}{z\lambda },\stackrel{~}{z}=\phi (z)=\frac{az+b}{cz+d},f(\stackrel{~}{z})=\frac{1}{\stackrel{~}{z}\phi (\lambda )}.$$
Without loss of generality we can assume that $`C_0`$ is the unit circle. Other curves $`C_k`$ are also circles (with centers $`a_k`$ and radii $`r_k`$). By the same argument as above, $`\mu G_{}\lambda G_+`$
$$0=\underset{C}{}\frac{1}{z\mu }\frac{k(z)}{\overline{z}\overline{\lambda }}𝑑zh(z)=\frac{k(z)}{\overline{z}\overline{\lambda }}E^1(G_+).$$
Since $`k(z)E^1(G_{})`$, the function $`h`$ admits the meromorphic continuation $`h(z)={\displaystyle \frac{k(z)}{\frac{1}{z}\overline{\lambda }}}`$ into the domain $`G_0=\{z:|z|>1\}`$ and $`h(z)={\displaystyle \frac{k(z)}{\overline{a}_k+\frac{r_k^2}{za_k}\overline{\lambda }}}`$ into the domains $`G_k=\{z:|za_k|<r_k\}`$. This meromorphic continuation has only simple poles at the points $`b_0=\frac{1}{\overline{\lambda }}G_0`$ and $`b_k=a_k+\frac{r_k^2}{\overline{\lambda }\overline{a}_k}G_k`$. Therefore, we have
$$h(z)\frac{c_0}{zb_0}\frac{c_1}{zb_1}\mathrm{}\frac{c_n}{zb_n}E^1(G_+)E^1(G_{})=\{0\},$$
where $`c_k=c_k(\lambda )=\mathrm{res}(h(z),b_k),k=\overline{0,n}`$. Hence,
$$k(z)=c_0\frac{\overline{z}\overline{\lambda }}{zb_0}+c_1\frac{\overline{z}\overline{\lambda }}{zb_1}+\mathrm{}+c_n\frac{\overline{z}\overline{\lambda }}{zb_n},zC.$$
In particular,
$$k(z)=\frac{\overline{\lambda }}{z}\left(c_0+c_1\frac{zb_0}{zb_1}+\mathrm{}+c_n\frac{zb_0}{zb_n}\right),zC_0.$$
Since coefficients $`c_k`$ depend analytically on $`\overline{\lambda }`$ and the functions $`b_k=b_k(\lambda )`$ are not constants, we obtain $`c_k=0,k=\overline{1,n}`$. If $`n0`$, in the same way, we have
$$k(z)=\frac{\overline{a}_1\overline{\lambda }}{za_1}\left(c_0\frac{zb_1}{zb_0}+c_1+c_2\frac{zb_1}{zb_2}+\mathrm{}+c_n\frac{zb_1}{zb_n}\right),zC_1$$
and $`c_k=0,k1`$. Hence, $`c_k=0,k=\overline{0,n}`$ and $`k(z)0,zC`$. This contradicts our assumption $`w(z)0`$. Thus, $`n=0`$, $`k(z)=\frac{C}{z},|z|=1`$ and therefore $`w(z)const`$. ∎
In the context of the Smirnov spaces $`E^2(G_\pm )`$, the self-adjointness of the projections $`P_\pm `$ is equivalent to the orthogonality $`E^2(G_+)E^2(G_{})`$ (for the definition of $`E^2(G_\pm )`$, see ). Recall that $`\mathrm{Ran}P_\pm =E^2(G_\pm )`$ and $`\mathrm{Ker}P_\pm =E^2(G_{})`$.
###### Corollary 1.
If $`E^2(G_+)E^2(G_{})`$ with respect to the inner product $`(,)_{L^2(C,|dz|)}`$, then the curve $`C`$ is a circle.
Recall that the operator $`𝒦_S(,\lambda )`$ are bounded in $`L^2(C,w(z)|dz|)`$ if and only if $`(C,w^{1/2})A_2`$ (i.e., $`w^{1/2}`$ is a Mackenhoupt weight). Under this condition we evidently have $`wL^1(C,|dz|)`$) and therefore we can establish a desired extension of criterion of self-adjointness.
###### Corollary 2.
The bounded Cauchy singular integral operator $`𝒦_S(,\lambda )`$ is self-adjoint in the Hilbert space $`L^2(C,w(z)|dz|)`$ if and only if the curve $`C`$ is a circle and $`w(z)const`$.
Note that this extension to the case of multiply connected domains can be obtained from the corresponding result for simple connected domains. However, we prefer to regard it as consequence of our main theorem mainly with the aim to present a new proof of Krupnik’s result.
###### Remark.
In B.Sz.-Nagy and C.Foiaş employed the decomposition $`L^2=H^2H_{}^2`$ to the construction of their functional model for contractions, where the spaces $`H^2=E^2(𝔻),H_{}^2=E^2(𝔻_{})`$ are Hardy’s spaces. We emphasize that along with orthogonality there is an analyticity in both the domains $`𝔻`$ and $`𝔻_{}`$, respectively. However, if we intend to extend the Sz.-Nagy-C.Foiaş functional model to arbitrary domains, we cannot keep simultaneously both these analyticities and orthogonality because of nonorthogonality of the decomposition $`L^2(C)=E^2(G_+)+E^2(G_{})`$.
The combination “analyticity only in $`G_+`$ plus orthogonality” is a mainstream of development in the multiply connected case (see, e.g., or ). In particular, the Riemann surface (=double of the planar domain $`G_+`$) is used therein and the authors have to deal with the “finite-rank defect” decomposition $`L^2(C)=H_+^2H_{}^2𝔐`$, where $`0dim𝔐<\mathrm{}`$ and the subspace $`H_{}^2`$ corresponds to the duplicate of $`G_+`$ (and therefore we lose entirely geometrical information concerning the domain $`G_{}`$). Another drawback of this approach is the use of uniformization technique or analytic vector bundles (with fairly large amount of algebraic geometry).
On the other hand, the nonorthogonal theory keeping both analyticities and free of the above drawbacks was developed recently in . Note that the duality with respect to the Cauchy pairing
$$<f,g>_C:=\frac{1}{2\pi i}\underset{C}{}f(z)\overline{g(\overline{z})}dz,fL^2(C),gL^2(\overline{C})$$
is a substitute for orthogonality in our approach. For instance, $`E^2(G_\pm )^{<>}=E^2(\overline{G_\pm })`$ (see also ). Note also that linear similarity (instead of unitary equivalence) is a natural kind of equivalence in our theory.
|
warning/0506/hep-th0506047.html
|
ar5iv
|
text
|
# DYONS OF ONE HALF MONOPOLE CHARGE
## 1 Introduction
The SU(2) Yang-Mills-Higgs (YMH) field theory in $`3+1`$ dimensions, with the Higgs field in the adjoint representation possess both the magnetic monopole and multimonopole solutions -. The ’t Hooft-Polyakov monopole solution with non zero Higgs mass and self-interaction is the first monopole solution that possess finite energy. This numerical, spherically symmetric monopole solution of unit magnetic charge is invariant under a U(1) subgroup of the local SU(2) gauge group .
Configurations of the YMH field theory with a unit magnetic charge are in general spherically symmetric - although exceptions do exit . All other monopole configurations with magnetic charges greater than unity possess at most axial symmetry and it has been shown that these solutions cannot possess spherical symmetry .
Until now, exact monopole and multimonopoles solutions exist only in the Bogomol’nyi-Prasad-Sommerfield (BPS) limit -. Outside of this limit, where the Higgs field potential is non-vanishing only numerical solutions exist. Asymmetric multimonopole solutions with no rotational symmetry are all numerical solutions even in the BPS limit .
Recently, we have also shown that the extended ansatz of Ref. possesses more exact multimonopole-antimonopole configurations. The anti-configurations of all these multimonopole-antimonopole solutions with all the magnetic monopole charges reversing their sign were also constructed . Hence monopoles becomes antimonopoles and vice versa. Solutions with vortex rings in the presence of an antimonopole-monopole-antimonopole (A-M-A) chain were also discussed . We have also shown the existence of half-integer topological monopole charge solutions. This configuration possesses axial symmetry and a Dirac-like string singularity along the negative z-axis . The existence of smooth Yang-Mills potentials which correspond to monopoles and vortices of one-half winding number has been demonstrated in Ref. but no exact or numerical solutions of the YMH equations have been given.
In this paper, we would like to reexamine the axially symmetric one half monopole solutions of the Ref. once again in more detail by introducing electric charge to the system, hence creating what is called a dyon solution with a one half monopole charge. The procedure of introducing an electric charge to a monopole solution is standard and was first shown by Julia and Zee in 1975 .
The one half monopole that was presented in Ref. is a particular case when the solution parameter $`m=1/2`$. In this paper, we would like to show that the one half monopole solution actually possesses a parameter $`m`$, where $`1/2m<0`$. The Higgs field of this one half monopole solution does not possess any point zeros and hence there are no other monopoles in the configuration except the one half monopole which is located at $`r=0`$, where the Higgs unit vector, $`\widehat{\mathrm{\Phi }}^a`$, becomes indeterminate and the Higgs field is singular.
By applying Gauss’ Law to the electric field of this dyon solution at large $`r`$, we find that the electric flux seems to come from a net electric charge of decreasing magnitude as $`m`$ approaches zero. When $`m=0`$, the dyon solution of one half monopole charge becomes the dyon of the Wu-Yang type monopole with unit magnetic charge and zero net electric charge.
We briefly review the SU(2) Yang-Mills-Higgs field theory in the next section and introduce the modified axially symmetric magnetic ansatz in section 3. We also discuss the dyon solutions of one half monopole charge and their electromagnetic properties in section 3 and end with some comments in section 4.
## 2 The SU(2) Yang-Mills-Higgs Theory
The SU(2) YMH Lagrangian in 3+1 dimensions is given by
$$=\frac{1}{4}F_{\mu \nu }^aF^{a\mu \nu }+\frac{1}{2}D^\mu \mathrm{\Phi }^aD_\mu \mathrm{\Phi }^a\frac{1}{4}\lambda (\mathrm{\Phi }^a\mathrm{\Phi }^a\frac{\mu ^2}{\lambda })^2,$$
(1)
where $`\mu `$ is the Higgs field mass, $`\lambda `$ is the strength of the Higgs potential and $`\mu /\sqrt{\lambda }`$ is the vacuum expectation value of the Higgs field. The Lagrangian (1) is gauge invariant under the set of independent local SU(2) transformations at each space-time point. The covariant derivative of the Higgs field and the gauge field strength tensor are given respectively by
$`D_\mu \mathrm{\Phi }^a`$ $`=`$ $`_\mu \mathrm{\Phi }^a+ϵ^{abc}A_\mu ^b\mathrm{\Phi }^c,`$ (2)
$`F_{\mu \nu }^a`$ $`=`$ $`_\mu A_\nu ^a_\nu A_\mu ^a+ϵ^{abc}A_\mu ^bA_\nu ^c.`$ (3)
Since the gauge field coupling constant, can be scaled away, we set it to one without any loss of generality. The metric used is $`g_{\mu \nu }=(+++)`$. The SU(2) internal group indices $`a,b,c`$ run from 1 to 3 and the spatial indices are $`\mu ,\nu ,\alpha =0,1,2`$, and $`3`$ in Minkowski space.
The equations of motion that follow from the Lagrangian (1) are
$`D^\mu F_{\mu \nu }^a`$ $`=`$ $`^\mu F_{\mu \nu }^a+ϵ^{abc}A^{b\mu }F_{\mu \nu }^c=ϵ^{abc}\mathrm{\Phi }^bD_\nu \mathrm{\Phi }^c,`$
$`D^\mu D_\mu \mathrm{\Phi }^a`$ $`=`$ $`\lambda \mathrm{\Phi }^a(\mathrm{\Phi }^b\mathrm{\Phi }^b{\displaystyle \frac{\mu ^2}{\lambda }}).`$ (4)
Non-BPS solutions to the YMH theory are obtained by solving the second order differential equations of motion (4), whereas BPS solutions can be more easily obtained by solving the Bogomol’nyi equations,
$$B_i^a\pm D_i\mathrm{\Phi }^a=0,$$
(5)
which is of first order.
The Abelian electromagnetic field tensor as proposed by ’t Hooft is
$`F_{\mu \nu }`$ $`=`$ $`\widehat{\mathrm{\Phi }}^aF_{\mu \nu }^aϵ^{abc}\widehat{\mathrm{\Phi }}^aD_\mu \widehat{\mathrm{\Phi }}^bD_\nu \widehat{\mathrm{\Phi }}^c`$ (6)
$`=`$ $`_\mu A_\nu _\nu A_\mu ϵ^{abc}\widehat{\mathrm{\Phi }}^a_\mu \widehat{\mathrm{\Phi }}^b_\nu \widehat{\mathrm{\Phi }}^c,`$
where $`A_\mu =\widehat{\mathrm{\Phi }}^aA_\mu ^a,\widehat{\mathrm{\Phi }}^a=\mathrm{\Phi }^a/|\mathrm{\Phi }|,|\mathrm{\Phi }|=\sqrt{\mathrm{\Phi }^a\mathrm{\Phi }^a}`$. Hence the ’t Hooft electric field is $`E_i=F_{0i}`$, and the ’t Hooft magnetic field is $`B_i=\frac{1}{2}ϵ_{ijk}F_{jk}`$, where the indices, $`i,j,k=1,2,3`$. The topological magnetic current, which is also the topological current density of the system is
$$k_\mu =\frac{1}{8\pi }ϵ_{\mu \nu \rho \sigma }ϵ_{abc}^\nu \widehat{\mathrm{\Phi }}^a^\rho \widehat{\mathrm{\Phi }}^b^\sigma \widehat{\mathrm{\Phi }}^c.$$
(7)
Therefore the corresponding conserved topological magnetic charge is
$`M={\displaystyle d^3xk_0}={\displaystyle \frac{1}{4\pi }}{\displaystyle d^2\sigma _iB_i}.`$ (8)
In the BPS limit when the Higgs potential vanishes, the energy can be written in the form
$`E`$ $`=`$ $`{\displaystyle _i(B_i^a\mathrm{\Phi }^a)d^3x}+{\displaystyle \frac{1}{2}(B_i^a\pm D_i\mathrm{\Phi }^a)^2d^3x}`$ (9)
$`=`$ $`{\displaystyle _i(B_i^a\mathrm{\Phi }^a)d^3x}=4\pi M{\displaystyle \frac{\mu }{\sqrt{\lambda }}},`$
where $`M`$ is the “topological charge” when the vacuum expectation value of the Higgs field, $`\frac{\mu }{\sqrt{\lambda }}`$, is non zero coupled with some non-trivial topological structure of the fields at large $`r`$.
## 3 The Dyons
### 3.1 The Ansatz
The static gauge fields and Higgs field that will lead to the dyon solutions of one half monopole charge are given respectively by, ,
$`A_0^a`$ $`=`$ $`\mathrm{sinh}\gamma (\mathrm{\Phi }_1\widehat{r}^a+\mathrm{\Phi }_2\widehat{\theta }^a),`$
$`A_i^a`$ $`=`$ $`{\displaystyle \frac{1}{r}}\psi (r)\left(\widehat{\theta }^a\widehat{\varphi }_i\widehat{\varphi }^a\widehat{\theta }_i\right)+{\displaystyle \frac{1}{r}}R(\theta )\left(\widehat{\varphi }^a\widehat{r}_i\widehat{r}^a\widehat{\varphi }_i\right),`$
$`\mathrm{\Phi }^a`$ $`=`$ $`\mathrm{cosh}\gamma (\mathrm{\Phi }_1\widehat{r}^a+\mathrm{\Phi }_2\widehat{\theta }^a),`$ (10)
where $`\mathrm{\Phi }_1=\frac{1}{r}\psi (r),\mathrm{\Phi }_2=\frac{1}{r}R(\theta )`$ and $`\gamma `$ is an arbitrary constant. The spherical coordinate orthonormal unit vectors, $`\widehat{r}^a`$, $`\widehat{\theta }^a`$, and $`\widehat{\varphi }^a`$ are defined by
$`\widehat{r}^a`$ $`=`$ $`\mathrm{sin}\theta \mathrm{cos}\varphi \delta _1^a+\mathrm{sin}\theta \mathrm{sin}\varphi \delta _2^a+\mathrm{cos}\theta \delta _3^a,`$
$`\widehat{\theta }^a`$ $`=`$ $`\mathrm{cos}\theta \mathrm{cos}\varphi \delta _1^a+\mathrm{cos}\theta \mathrm{sin}\varphi \delta _2^a\mathrm{sin}\theta \delta _3^a,`$
$`\widehat{\varphi }^a`$ $`=`$ $`\mathrm{sin}\varphi \delta _1^a+\mathrm{cos}\varphi \delta _2^a,`$ (11)
where $`r=\sqrt{x^ix_i}`$, $`\theta =\mathrm{cos}^1(x_3/r)`$, and $`\varphi =\mathrm{tan}^1(x_2/x_1)`$. Ansatz (10) satisfies the radiation or Coulomb gauge, $`^iA_i^a=A_0^a=0`$.
### 3.2 The Solutions
Upon substituting ansatz (10) into the equations of motion (4), the resulting equations are simplified to two first order differential equations,
$`r\psi ^{}+\psi \psi ^2=m(m+1),`$
$`\dot{R}+R\mathrm{cot}\theta R^2=m(m+1),`$ (12)
where $`\psi ^{}`$ means $`\frac{\psi }{r}`$ and $`\dot{R}`$ means $`\frac{R}{\theta }`$. The solutions for $`\psi `$ and $`R`$ are then given respectively by
$`\psi (r)`$ $`=`$ $`{\displaystyle \frac{(m+1)m(br)^{2m+1}}{1+(br)^{2m+1}}},`$
$`R(\theta )`$ $`=`$ $`(m+1)\mathrm{csc}\theta \left\{\mathrm{cos}\theta {\displaystyle \frac{(P_{m+1}(\mathrm{cos}\theta )+aQ_{m+1}(\mathrm{cos}\theta ))}{(P_m(\mathrm{cos}\theta )+aQ_m(\mathrm{cos}\theta ))}}\right\},`$ (13)
where $`P_m`$ and $`Q_m`$ are the Legendre polynomials of the first and second kind of degree $`m`$, respectively, and $`1/2m<0`$. For solutions regular along the positive $`z`$-axis, we required, $`R(0)=0`$, and the integration constant $`a`$ is set to zero. The integration constant $`b`$ is just a scaling factor and without any loss in generality, we set $`b=1`$ and the boundary conditions for $`\psi (r)`$ are $`\psi (0)=m+1`$ and $`\psi (\mathrm{})=m`$. In these dyon solutions there are no zeros of the Higgs field and the Higgs unit vector is only indeterminate at the origin where a monopole of half unit charge is located.
From the ansatz (10), the ’t Hooft gauge potential becomes,
$$A_\mu =\widehat{\mathrm{\Phi }}^aA_\mu ^a=\frac{\mathrm{sinh}\gamma }{r}\sqrt{\psi (r)^2+R(\theta )^2}\delta _\mu ^0.$$
(14)
Hence the ’t Hooft electric field is non zero when $`\gamma 0`$ and the monopole becomes a dyon.
### 3.3 The ’t Hooft Magnetic Field
The ’t Hooft magnetic field is however independent of the gauge fields $`A_\mu ^a`$. To calculate for the ’t Hooft magnetic field $`B_i`$, we rewrite the Higgs field from the spherical to the Cartesian coordinate system,
$`\mathrm{\Phi }^a`$ $`=`$ $`\mathrm{\Phi }_1\widehat{r}^a+\mathrm{\Phi }_2\widehat{\theta }^a+\mathrm{\Phi }_3\widehat{\varphi }^a`$ (15)
$`=`$ $`\stackrel{~}{\mathrm{\Phi }}_1\delta ^{a1}+\stackrel{~}{\mathrm{\Phi }}_2\delta ^{a2}+\stackrel{~}{\mathrm{\Phi }}_3\delta ^{a3}`$
$`\text{where}\stackrel{~}{\mathrm{\Phi }}_1`$ $`=`$ $`\mathrm{sin}\theta \mathrm{cos}\varphi \mathrm{\Phi }_1+\mathrm{cos}\theta \mathrm{cos}\varphi \mathrm{\Phi }_2\mathrm{sin}\varphi \mathrm{\Phi }_3=|\mathrm{\Phi }|\mathrm{cos}\alpha \mathrm{sin}\beta `$
$`\stackrel{~}{\mathrm{\Phi }}_2`$ $`=`$ $`\mathrm{sin}\theta \mathrm{sin}\varphi \mathrm{\Phi }_1+\mathrm{cos}\theta \mathrm{sin}\varphi \mathrm{\Phi }_2+\mathrm{cos}\varphi \mathrm{\Phi }_3=|\mathrm{\Phi }|\mathrm{cos}\alpha \mathrm{cos}\beta `$
$`\stackrel{~}{\mathrm{\Phi }}_3`$ $`=`$ $`\mathrm{cos}\theta \mathrm{\Phi }_1\mathrm{sin}\theta \mathrm{\Phi }_2=|\mathrm{\Phi }|\mathrm{sin}\alpha .`$ (16)
The Higgs unit vector can be simplified to
$`\widehat{\mathrm{\Phi }}^a`$ $`=`$ $`\mathrm{cos}\alpha \mathrm{sin}\beta \delta ^{a1}+\mathrm{cos}\alpha \mathrm{cos}\beta \delta ^{a2}+\mathrm{sin}\alpha \delta ^{a3},`$ (17)
$`\text{where},\mathrm{sin}\alpha `$ $`=`$ $`{\displaystyle \frac{\psi \mathrm{cos}\theta R\mathrm{sin}\theta }{\sqrt{\psi ^2+R^2}}},\beta ={\displaystyle \frac{\pi }{2}}\varphi ,`$
and the ’t Hooft magnetic field is reduce to only the $`\widehat{r}_i`$ and $`\widehat{\theta }_i`$ components,
$`B_i={\displaystyle \frac{1}{r^2\mathrm{sin}\theta }}\left\{{\displaystyle \frac{\mathrm{sin}\alpha }{\theta }}\right\}\widehat{r}_i+{\displaystyle \frac{1}{r\mathrm{sin}\theta }}\left\{{\displaystyle \frac{\mathrm{sin}\alpha }{r}}\right\}\widehat{\theta }_i.`$ (18)
The function, $`\mathrm{sin}\alpha `$, is a regular function over all space and the ’t Hooft magnetic field is also regular everywhere except at the location of the monopole at the origin and along the negative $`z`$-axis where there is a semi infinite string singularity. When $`m=1/2`$ and $`m=0`$, $`\mathrm{sin}\alpha `$ is independent of $`r`$ and the magnetic field is purely radial in direction. We also notice that, $`B_i`$, can be written as
$`B_i=ϵ_{ijk}^j(\mathrm{sin}\alpha )^k\beta =ϵ_{ijk}^j(\mathrm{sin}\alpha ^k\beta ),`$ (19)
and that a suitable Maxwell four-vector gauge potential for this magnetic field is
$`𝒜_0={\displaystyle \frac{\mathrm{sinh}\gamma }{r}}\sqrt{\psi ^2+R^2},𝒜_i=(\mathrm{sin}\alpha 1)_i\beta ={\displaystyle \frac{(\mathrm{sin}\alpha 1)}{r\mathrm{sin}\theta }}\widehat{\varphi }_i.`$ (20)
When $`m=0`$, the gauge potential, $`𝒜_i`$, is just the usual Dirac string potential and it is singular along the negative z-axis. However when $`1/2m<0`$, the gauge potential $`𝒜_i`$ still possesses a similar string singularity along the negative $`z`$-axis, but it is no longer the Dirac string.
From Eq.(19), it is obvious that the magnetic field is always perpendicular to the gradients of $`\mathrm{sin}\alpha `$ and $`\beta `$. Hence the magnetic field lines lie on the line $`\mathrm{sin}\alpha =k`$, where $`1<k<1`$, and $`\varphi =`$ constant. By plotting $`\mathrm{sin}\alpha =k`$ on a vertical plane through the origin; we manage to draw the magnetic field lines for the configurations when $`m=0.01`$, Fig.(1); $`m=0.05`$, Fig.(2); and $`m=0.2`$, Fig.(3). We have notice that for every field line through the origin, one end of the line goes to infinity, whereas the other end will make a loop and return back to $`r=0`$. Therefore the magnetic flux going to infinity is only half of that of a monopole of unit charge.
### 3.4 The Monopole Charge
From Eq.(8), the net magnetic charge enclosed by the sphere at infinity is calculated to be,
$`M_{\mathrm{}}={\displaystyle \frac{1}{2}}\mathrm{sin}\alpha |_{0,r\mathrm{}}^\pi `$ $`=`$ $`{\displaystyle \frac{1}{2}},\text{when}{\displaystyle \frac{1}{2}}m<0,`$ (21)
$`=`$ $`1,\text{when}m=0.`$
Hence the magnetic charge of the system is always one half over all space when $`m`$ is a non integer less than zero and it is concentrated at only one point in space at $`r=0`$. When $`m=0`$, the half monopole becomes a unit monopole.
### 3.5 The ’t Hooft Electric Field
The ’t Hooft electric field of the dyons is given by
$`E_i`$ $`=`$ $`_iA_0,`$ (22)
$`=`$ $`\mathrm{sinh}\gamma \left\{\sqrt{\psi ^2+R^2}{\displaystyle \frac{\psi (\psi +m)(\psi m1)}{\sqrt{\psi ^2+R^2}}}\right\}\left({\displaystyle \frac{\widehat{r}_i}{r^2}}\right)`$
$``$ $`\mathrm{sinh}\gamma \left\{{\displaystyle \frac{R(R^2R\mathrm{cot}\theta +m(m+1))}{\sqrt{\psi ^2+R^2}}}\right\}\left({\displaystyle \frac{\widehat{\theta }_i}{r^2}}\right).`$
Unlike the magnetic field, the electric field depends on the constant $`\gamma `$. Hence the electric field can be switched off by setting $`\gamma =0`$. As the parameter $`m`$ decreases from $`1/2`$ to zero, the $`\widehat{\theta }_i`$ component of the electric field diminishes to zero while the radial component approaches
$`E_i=\left\{{\displaystyle \frac{2}{(1+r)}}{\displaystyle \frac{1}{(1+r)^2}}\right\}{\displaystyle \frac{\widehat{r_i}}{r^2}},`$ (23)
and the solution becomes spherically symmetric. The 3D field plot of the electric field for different values of $`m`$ in the range $`1/2m<0`$, shows the presence of a positive point charge at $`r=0`$ and a positive line charge along the negative $`z`$-axis. Fig.(4) shows the 3D field plot for the case of $`m=1/2`$. The electric field which is pointing radially outward diminishes as $`1/r^2`$ along the negative $`z`$-axis.
### 3.6 The Electric Charge
The electric charge densities, $`q=^iE_i`$, of these dyon solutions consist of a negative cloud charge distribution concentrated in regions around the origin and along the negative $`z`$-axis diminishing along the axis as $`\frac{1}{r^3}`$. The positive charge densities are delta functions distributions at $`r=0`$ and along the negative $`z`$-axis. Fig.(5) and (6) show the electric charge density distribution for the $`m=0.05`$ and $`m=0.5`$ dyons respectively. The negative electric cloud charge density distribution is in blue and the positive charges are indicated in red. The charge distribution along the negative $`z`$-axis is greatest when $`m=1/2`$ and continuously diminishes to zero as $`m`$ approaches zero when the dyon’s monopole charge changes from half to one. The total electric charge of the dyons can be obtained by Gauss’ law,
$`Q`$ $`=`$ $`{\displaystyle _r\mathrm{}}E_i\widehat{r}_ir^2\mathrm{sin}\theta d\theta d\varphi ,`$ (24)
$`=`$ $`2\pi \mathrm{sinh}\gamma {\displaystyle \sqrt{m^2+R(\theta )^2}\mathrm{sin}\theta d\theta }.`$
The charge concentrated at the origin is calculated to be
$`Q_0=2\pi \mathrm{sinh}\gamma {\displaystyle \sqrt{(m+1)^2+R(\theta )^2}\mathrm{sin}\theta d\theta }.`$ (25)
$`Q`$ and $`Q_0`$ can be numerically integrated using Maple 9.5. The charges of the dyons $`Q`$ and and $`Q_0`$ depend on the parameter $`m`$ and a point plot of $`Q`$ and $`Q_0`$ versus $`m`$ is given in Fig.(7). The graph branches out at $`m=1/2`$ indicating that there is totally no screening of the positive charge at $`r=0`$ of the $`m=1/2`$ dyon. However as $`m`$ increases from $`1/2`$ to zero, screening of the delta function point source at $`r=0`$ takes place and when $`m=0`$, total screening occurs and the net electric charge of the Wu-Yang type dyon is zero.
## 4 Comments
The ansatz (10) used to construct the dyon solutions is introduced in the same standard way as in the construction of the Julia-Zee dyons years ago . Hence when the parameter $`\gamma `$ is zero, the electric field is switched off and only the magnetic field remains and the dyon becomes monopole.
The energy of these dyon solutions are infinite as they possess delta functions electric charge sources at the origin and along the negative $`z`$-axis. Infinite energy dyons solutions that are complex in the gauge potentials have also been discussed in the literature .
These dyon solutions of one half monopole charge are axially symmetric about the $`z`$-axis. Their gauge potentials and electromagnetic fields possess a string singularity along the negative $`z`$-axis. Configurations of one half monopole charge that possess axial symmetry and a semi infinite string singularity both in the gauge potentials and the electromagnetic fields have also been discussed in the literature .
Unlike the monopole solutions of the ansatz (10), where the parameter $`m`$ can take only discrete positive integer values , the dyons of one half monopole charge possess a continuous parameter, $`1/2m<0`$, and as $`m`$ varies, the properties of the solutions like the electric charges $`Q`$ and $`Q_0`$, the electric charge distribution, and the electric and magnetic fields change. However there is a small range of values of $`0.005m<0`$, for which the monopole charges become indeterminate by Maple 9.5 although we believe that the monopole charge should be one half as long as $`m<0`$.
Calculations show that the electric charges that give rise to the $`\widehat{\theta }_i`$ component of the electric field are composed of a negative cloud charge distribution around the negative $`z`$-axis and a positive delta function charge distribution along the negative $`z`$-axis which exactly cancelled out the charges of each other at $`r`$ infinity.
The electric charges that give rise to the radial component of the electric field are all concentrated at $`r=0`$ when $`m=1/2`$. Hence $`Q=Q_0`$ for the $`m=1/2`$ dyon. However when, $`1/2<m0`$, there is a negative cloud charge distribution around the origin and this negative cloud charge increases in intensity until it reaches its’ maximum value at $`m=0`$ where it totally screened off the positive delta function source at $`r=0`$ giving a zero net electric charge for the Wu-Yang type dyon when $`m=0`$. Hence for the Wu-Yang type dyon, $`Q=0`$, and $`Q_0=4\pi \mathrm{sinh}\gamma `$.
## 5 Acknowlegements
The authors would like to thank Universiti Sains Malaysia and the Academy of Sciences Malaysia for the Scientific Advancement Grant Allocation, SAGA, (Account No.: 304/pfizik/653004/A118). The author, Rosy Teh, would also like to thank Prof. Izumi Tsutsui for reading through the manuscript and the Institute of Particle and Nuclear Studies, High Energy Accelerator Research Organization (KEK) Tsukuba, Ibaraki 305-0801, Japan for their hospitality.
## FIGURE CAPTIONS
|
warning/0506/nucl-th0506059.html
|
ar5iv
|
text
|
# Shell-model test of the rotational-model relation between static quadrupole moments 𝑄(2⁺₁), 𝐵(𝐸2)’s, and orbital 𝑀1 transitions
## I Introduction
In this work we will make a comparison of the shell model and the collective model for several quantities that are sensitive to nuclear deformation. These include $`B(E2)`$’s, static quadrupole moments, and orbital magnetic dipole transitions. This will be a theory versus theory work. Some experimental results are quoted and serve as anchors for our results, but we will not be inhibited by the lack of experimental data in doing these calculations. We plan in the near future to make a more extensive theory–experiment comparison. But there are holes in the experimental data which must be filled.
The main thrust of our work will be to understand the relationship of orbital magnetic dipole transitions to quadrupole deformations in the nucleus. For example, after the experimental discovery in heavy deformed nuclei of the relation between the orbital magnetic dipole strength and nuclear deformation zrrs90 , there have been many works which relate the orbital $`M1`$ (scissors mode) strength to electric quadrupole transition rates ($`B(E2)_{0_12_1}`$), often assuming that they are proportional to each other hcrw92 ; zz92 ; ir93 ; ngbr95 ; eknrr99 .
First, though, we shall do survey calculations of $`B(E2)`$’s and static quadrupole moments in the $`fp`$ shell to see how well the shell model relates to the simple rotational model of Bohr and Mottelson bm75 . In this rotational model, the formulae for $`B(E2)`$’s and static quadrupole moments involve a single parameter—the intrinsic quadrupole moment. These formulae are, respectively,
$$B(E2)=\frac{5}{16\pi }Q_0^2(B)\left|I_1K20|I_2K\right|^2$$
(1a)
$$Q(I)=\frac{3K^2I(I+1)}{(I+1)(2I+3)}Q_0(S),$$
(1b)
where $`B`$ and $`S`$ stand for $`B(E2)`$ and “Static”, respectively. Here $`Q_0`$ is the intrinsic quadrupole moment—what we would see in the rotational frame. On the other hand, $`Q(I)`$ is what we measure in the laboratory. In the simple rotational model, $`Q_0(B)`$ is equal to $`Q_0(S)`$.
For the case $`I_1=0`$, $`I_2=2`$, the Clebsch-Gordan coefficient above is 1. For a simple $`K=0`$ band in an even–even nucleus, we obtain
$$B(E2)_{02}=\frac{5}{16\pi }Q_0^2(B)$$
(2a)
$$Q(2^+)=\frac{2}{7}Q_0(S).$$
(2b)
Note that the laboratory quadrupole moment has the opposite sign of the intrinsic quadrupole moment—a well known result. It can be understood physically by imagining rotating a cigar (which has a positive quadrupole moment, i.e., prolate) about an axis perpendicular to the line of the cigar. This will trace out a flat pancake shape which is oblate.
We then find that the ratio
$$\frac{Q_0(S)}{Q_0(B)}=\frac{7}{2}\sqrt{\frac{5}{16\pi }}\frac{Q(2^+)}{\sqrt{B(E2)}}=1.1038705\frac{Q(2^+)}{\sqrt{B(E2)}}.$$
(3)
## II Quadrupole properties in the $`sd`$ shell
Although we will be performing calculations in the $`fp`$ shell, we shall here briefly look over the experimental situation in the $`sd`$ shell. In Table 1 we show experimental values of $`Q(2_1^+)`$ s01 , $`B(E2)`$ rnt01 , and the ratio $`|Q_0(S)/Q_0(B)|`$ as given by Eq. (3). We also show the experimental values of $`E(4_1^+)/E(2_1^+)`$ as a measure of how close we are to the rotational limit of $`10/3`$, or the vibrational limit of $`2/1`$.
Note that the static quadrupole moments of <sup>20</sup>Ne, <sup>22</sup>Ne, <sup>24</sup>Mg, and <sup>32</sup>S are negative, while those of <sup>28</sup>Si and <sup>36</sup>Ar are positive. If we limit ourselves to axial symmetry, this indicates that the first group has prolate ground state bands and the second group has oblate ones. Skyrme II Hartree-Fock results by Jaqaman and Zamick jz84 correctly give the signs of all the static quadrupole moments. The small static quadrupole moment of <sup>40</sup>Ar is consistent with magnetic moment results of the $`2_1^+`$ state by Stefanova et al. setal05 .
The ratio $`|Q_0(S)/Q_0(B)|`$ for <sup>20</sup>Ne is larger than the rotational limit, 1.377 versus 1; likewise <sup>22</sup>Ne. In the case of <sup>24</sup>Mg, $`|Q_0(S)/Q_0(B)|`$ is smaller than for <sup>20</sup>Ne or <sup>22</sup>Ne, despite the fact that the spectrum is closer to rotational for <sup>24</sup>Mg. Also surprisingly for <sup>32</sup>S, the ratio $`E(4)/E(2)`$ is 2.000, the vibrational limit, for which one might expect a near zero static quadrupole moment. But the ratio $`|Q_0(S)/Q_0(B)|`$ is 0.950, close to the simple rotational prediction of unity.
In general, it is difficult to correlate $`|Q_0(S)/Q_0(B)|`$ with $`E(4)/E(2)`$ assuming a simple axially symmetric rotor.
We should mention that an analysis of the relationship of $`Q_0(S)`$ and $`Q_0(B)`$ has already been performed by Bender, Flocard, and Heenen bfh03 and Bender et al. bbdh04 , albeit not for the $`fp`$-shell nuclei considered here and using a different method. They perform angular momentum projections on BCS–Hartree-Fock states obtained with the Skyrme interaction SLy6 for the particle–hole channel and a density-dependent contact force in the pairing channel bfh03 . Their calculations are mainly in the $`sd`$ shell bfh03 and neutron-deficient lead region bbdh04 . For one nucleus in common, <sup>40</sup>Ca, their results for $`0p0h`$, $`2p2h`$, $`4p4h`$, $`6p6h`$, $`8p8h`$, and $`12p12h`$ do not differ so much from previous calculations of Zheng, Berdichevsky, and Zamick zbz88 as far as the intrinsic properties are concerned, but their calculation has the added feature of providing an energy spectrum and expectation values in the laboratory frame.
In Ref. jz84 , the authors predict that <sup>36</sup>Ar is oblate. This is confirmed by the fact that the static quadrupole moment of the $`2_1^+`$ state is positive: $`+11`$ e fm<sup>2</sup> s01 . The experimental $`B(E2)`$ is 340 e<sup>2</sup> fm<sup>4</sup> and $`|\beta _2|=0.273`$ rnt01 . Using Eq. (3), we find
$$\frac{Q_0(S)}{Q_0(B)}=0.6505236.$$
(4)
The energy ratio is
$$\frac{E(4_1^+)}{E(2_1^+)}=\frac{4414.36}{1970.35}=2.240.$$
(5)
These results are consistent with a nucleus not being too rotational.
The corresponding numbers in the calculation of Bender et al. bfh03 are
$$Q(2_1^+)_{\text{lab}}=13\text{e fm}\text{2},B(E2)=220\text{e}\text{2}\text{ fm}\text{4},\beta =0.21.$$
(6)
The calculated ratios are
$$\frac{Q_0(S)}{Q_0(B)}=0.9675,\frac{E(4_1^+)}{E(2_1^+)}=2.6545.$$
(7)
These calculations bfh03 give a more rotational picture than experiment. There is a consistency, however, in that a larger ratio $`E(4)/E(2)`$ yields a larger ratio $`Q_0(S)/Q_0(B)`$.
## III Shell model calculations of $`B(E2)`$ and $`Q(2_1^+)`$, and how they relate to the simple rotational model
We will put the above relation (3) to the test in a shell model approach for the following nuclei: <sup>44</sup>Ti, <sup>46</sup>Ti, <sup>48</sup>Ti, <sup>48</sup>Cr, and <sup>50</sup>Cr. We use the OXBASH program eetal85 and the FPD6 interaction rmjb91 .
The nuclei that we have chosen are far from being perfect rotors. Their description falls somewhere between vibrational and rotational. The ratios $`E(4)/E(2)`$, which would all be $`10/3`$ in the simple rotational case, are as follows: 1.922, 2.010, 2.118, 2.459, 2.342, for <sup>44</sup>Ti, <sup>46</sup>Ti, <sup>48</sup>Ti, <sup>48</sup>Cr, and <sup>50</sup>Cr, respectively.
We perform shell model calculations in a complete $`fp`$ space using the FPD6 interaction. We assign effective charges of 1.5 for the protons and 0.5 for the neutrons. We calculate $`B(E2)_{0_12_1}`$ and $`Q(2^+)`$ (the laboratory $`Q`$, of course) and put them into Eq. (3) in order to get operational values of $`Q_0(S)/Q_0(B)`$. The results are given in Table 2.
Except for <sup>48</sup>Ti, the FPD6 results for the ratios are all greater than $`0.9`$, reading a maximum of $`0.9892`$ for <sup>48</sup>Cr. It is somewhat surprising that these ratios are so close to 1, given that the ratios $`E(4)/E(2)`$ are much further away from the rotational limit $`10/3`$.
We can also obtain some of the above ratios from experiment. We refer to the compilation of nuclear moments of Stone s01 and of $`B(E2)`$’s by Raman et al. rnt01 . Taking these experiments at face value, we see that the ratio $`Q_0(S)/Q_0(B)`$ reduces to about 0.75 for <sup>46</sup>Ti and <sup>48</sup>Ti, but is bigger than 1 for <sup>50</sup>Cr.
It should be noted that in the simplest version of the vibrational mode, $`Q_0(S)`$ is zero. We can imagine a nucleus vibrating between a prolate shape and an oblate shape, and causing the quadrupole moment to average to zero. On the other hand, the $`B(E2)_{0_12_1}`$ is quite large in this vibrational limit, causing the ratio $`Q_0(S)/Q_0(B)`$ to be zero or, in more sophisticated vibrational models, quite small.
## IV Results with an alternate interaction T0FPD6
For systems of identical particles, e.g., the tin isotopes, which in the simple shell model involve only valence neutrons, one does not get rotational behaviour. One does go closer to the rotational limit when one has many open-shell neutrons and protons. Whereas two identical nucleons must have isospin 1, a neutron and a proton can have both isospin 0 and 1. This suggests that the $`T=0`$ part of the nucleon–nucleon interaction plays an important role in enhancing nuclear rotational collectivity.
In this section, we will use an interaction TOFPD6 that is the same as the FPD6 interaction for $`T=1`$ states, but vanishes for $`T=0`$ states. We thus expect that the rotational collectivity will be reduced. This interaction serves as a counterpoint of the full interaction in the previous section. It has been discussed before by Robinson and Zamick rz01 .
We present the results for T0FPD6 in Table 3.
We see that both $`Q(2_1^+)`$ and $`B(E2)`$ decrease in magnitude, consistent with the above discussion. However, $`Q_0(S)`$ decreases more rapidly than $`\sqrt{B(E2)}`$, so the ratio $`Q_0(S)/Q_0(B)`$ is less for this case than when the full interaction is present. For <sup>44</sup>Ti this ratio is 0.0502, close to the vibrational limit of zero. For <sup>48</sup>Cr the ratio decreases from 0.9892 to 0.8299; nevertheless, it is still substantial, indicating that the $`T=1`$ interaction, acting alone, can lead us to some extent in the direction of the rotational limit.
## V Random interaction studies
Nuclei in the region we are considering have undergone Random Interaction studies. We refer to the works of Velázquez et al. vhfz03 and Zelevinsky and Volya zv04 . These works were stimulated by that of Johnson, Bertsch, and Dean jbd98 .
In particular, in the work of Zelevinsky, a quantity is considered which is proportional to the square of $`Q_0(S)/Q_0(B)`$ and which is normalized to $`(2/7)^2`$ if $`Q_0(S)/Q_0(B)`$ equals 1. He calls this the Alaga ratio. He selects cases for which the random interaction yields a $`J=0,J=2`$ sequence of lowest energies. For these he finds two peaks, one corresponding to $`Q_0(S)/Q_0(B)=1`$ and the other to $`Q_0(S)/Q_0(B)=0`$. In our terminology, these would correspond to the rotational limit in the former case and either the simple vibrational limit or the spherical limit in the latter case. He cites early work of R. Rockmore r61 as affording an explanation of this surprising behaviour.
## VI Orbital magnetic dipole transitions in <sup>44</sup>Ti, <sup>46</sup>Ti, and <sup>48</sup>Ti
The orbital magnetic isovector dipole transitions, i.e., scissors mode excitations, also display collective behaviour zrrs90 . There are systematics which suggest that $`B(M1)_{\text{orbital}}`$ is roughly proportional to $`B(E2)`$. There are more detailed sophisticated relations as well. If one uses a simple quadrupole–quadrupole interaction, the energy weighted $`B(M1)_{\text{orbital}}`$ is proportional to the difference $`\left(B(E2)_{\text{isoscalar}}B(E2)_{\text{isovector}}\right)`$ zz92 .
The bare orbital $`M1`$ operator is
$$\sqrt{\frac{3}{4\pi }}l(i)g_l(i),$$
(8)
where $`g_l`$ is 1 for a proton and 0 for a neutron. This is the operator that we use in the calculations.
How to extract the scissors mode strength is not completely unambiguous. The mode is associated with low-lying $`1^+`$ excitations at around 3 MeV. But the strength can be fragmented even at this lowest energy. Besides this, there is orbital strength at higher energies, a somewhat grassy behaviour where individual states are very weakly excited but, because there are so many of them, the total orbital strength can be significant.
Therefore, we will give three sets of values (see Table 4). First, we give the strength to the lowest state, then to the lowest 10 states, and finally to the lowest 1000 states (except for <sup>48</sup>Ca, where we include only 300 states). The 10-states strength should encompass what we usually call the scissors mode, while the 1000-states strength is close to the total strength including the grassy, non-collective part. It would appear that the highest excitation energies reached in the experiments zrrs90 ; gdrjvw90 are not sufficient to reach the $`T+1`$ part of the spectrum.
We first discuss the nuclei <sup>46</sup>Ti and <sup>48</sup>Ti, for which there are some data on $`B(M1)`$. We see consistently that the orbital $`B(M1)`$ strength is larger in <sup>46</sup>Ti than in <sup>48</sup>Ti. This is consistent with the fact that <sup>46</sup>Ti has a greater $`B(E2)`$ and static $`2^+`$ quadrupole moment than <sup>48</sup>Ti.
We next consider the $`N=Z`$ nucleus <sup>44</sup>Ti, for which there is no data because this nucleus is unstable. The isoscalar orbital $`B(M1)`$ strength is very weak. This is also true for the spin $`B(M1)`$, but for a different reason. The isoscalar spin coupling is much smaller than the isovector one. For the orbital case, the couplings are equal because the operator is $`_{\text{ protons}}\stackrel{}{\mathrm{}}`$. For the orbital case, the $`B(M1)`$ isoscalar is very weak because the correlations due to the nuclear interaction move the ground state towards the $`SU(4)`$ limit, in which $`LS`$ coupling holds and for which the ground state is a pure $`L=0`$ state. For this extreme case, the $`B(M1)`$ orbital isoscalar will vanish.
The transitions of interest for <sup>44</sup>Ti are, therefore, the isovector orbital dipole ones. The ($`TT+1`$) $`B(M1)_{\text{orbital}}`$ summed strength is larger in <sup>44</sup>Ti than the ($`TT`$) and \[$`(TT)+(TT+1)`$\] strengths in <sup>46</sup>Ti and <sup>48</sup>Ti, which are 0.5615 (1.4810) and 0.3099 (1.0288) $`\mu _\text{N}^2`$, respectively. On the other hand, <sup>44</sup>Ti is not more deformed than <sup>46</sup>Ti. According to Raman et al. rnt01 , the values of the quadrupole deformation parameters $`\beta `$ for <sup>44,46,48</sup>Ti and <sup>48,50</sup>Cr are, respectively, 0.27, 0.317, 0.269, 0.335, and 0.293. Thus, we have here in $`fp`$-shell nuclei a counter-example to the experimentally established proportionality between the orbital $`B(M1)`$ and the $`B(E2)`$ in heavy deformed nuclei. Perhaps there are correlations which cause an enhancement for $`N=2`$ nuclei.
An analysis by Retamosa et al. rupm90 in the $`sd`$ shell comparing <sup>20</sup>Ne, <sup>22</sup>Ne, and <sup>24</sup>Mg was performed (somewhat analogous to <sup>44</sup>Ti and <sup>46</sup>Ti for the first two cases), but no anomaly was reported there. In the $`SU(3)`$ model, they found consistency in the relation of $`B(M1)_{\text{orbital}}`$ to deformation. In this limit, the summed $`M1`$ strengths (all orbital) for <sup>20</sup>Ne, <sup>22</sup>Ne, and <sup>24</sup>Mg were 1.1, 1.17, and 1.6 $`\mu _\text{N}^2`$, respectively. Looking at the Raman tables rnt01 for these nuclei, there is some complication—the deformation parameters $`\beta _2`$ are not in one-to-one correspondence with the $`B(E2)`$’s. The values of $`(B(E2),\beta )`$ for these three nuclei from the Raman tables rnt01 are, respectively, $`(0.034,0.728)`$, $`(0.0236,0.562)`$, and $`(0.0432,0.606)`$, where the units for $`B(E2)`$ are b<sup>2</sup>. The authors also do calculations with a more realistic interaction, but no enough strengths are listed in order to make a comparison for the point we are trying to make. Retamosa et al. rupm90 also give strengths to the first $`10^+`$ states in <sup>44</sup>Ti; our numbers are consistent with theirs.
Earlier works on the shell model for light nuclei include L. Zamick z85 and A. Poves prm89 .
## VII Closing remarks
In this work we have examined what predictions the shell model make for collective properties which are after dealt with in the rotational model. Although the nuclei are far from perfect rotors, the calculated ratio $`Q_0(S)/Q_0(B)`$ is fairly close to 1 in many cases. When the FPD6 interaction is used, the orbital magnetic dipole transitions for <sup>46,48</sup>Ti also fit into this picture, although there is the added complication of separating the collective from the non-collective part in this case. Also there is a substantial enhancement for the $`N=Z`$ nucleus <sup>44</sup>Ti, which cannot be explained as purely a deformation effect. We hope our work will stimulate more experimental investigations. There is information of $`B(M1)`$ rates in <sup>46</sup>Ti and <sup>48</sup>Ti, but thus far the orbital $`B(M1)`$ has only been extracted in <sup>48</sup>Ti. However, in a short time, we will be able to make a more extensive theory–experiment study of these magnetic dipole transitions.
###### Acknowledgements.
This work has been supported by the DFG under contracts SFB 634 and 445 SUA-113/6/0-1, and by the NRF, South Africa. AE acknowledges support from the Secretaría de Estado de Educación y Universidades (Spain) and the European Social Fund. We thank Y. Y. Sharon for his help.
|
warning/0506/astro-ph0506685.html
|
ar5iv
|
text
|
# Stellar populations and Ly𝛼 emission in two lensed 𝑧≳6 galaxies
## 1 Introduction
Little is known about the stellar properties, extinction, and the expected intrinsic Ly$`\alpha `$ emission of distant, high redshift galaxies. Indeed, although it has in the recent past become possible through various techniques to detect already sizeable numbers of galaxies at $`z5`$ (see e.g. the reviews of Taniguchi et al. 2003 and Spinrad 2004) the information available on these objects remains generally scant. For example, in many cases the galaxies are just detected in two photometric bands and Ly$`\alpha `$ line emission, when present, serves to determine the spectroscopic redshift (e.g. Bremer et al. 2004, Dickinson et al. 2004, Bunker et al. 2004). Then the photometry is basically used to estimate the star formation rate (SFR) assuming standard conversion factors between the UV restframe light and the SFR, and nothing is known about the extinction, and the properties of the stellar population (such as age, detailed star formation histories etc.)
At higher redshift ($`z6`$) even less information is generally available (but see a recent study of Eyles et al. 2005 on two $`z6`$ galaxies observed with HST and Spitzer). Many objects are found by Ly$`\alpha `$ emission, but remain weak or sometimes even undetected in the continuum (e.g. Rhoads & Malhotra 2001, Kodaira et al. 2003, Cuby et al. 2003, Ajiki et al. 2003, Taniguchi et al. 2004). In these cases the Ly$`\alpha `$ luminosity can be determined and used to estimate a SFR using again standard conversion factors. Also the Ly$`\alpha `$ equivalent width is estimated, providing some possible clue on the nature of these sources. However, this has lead to puzzling results e.g. for the sources from the LALA survey which seem to show unusually large Ly$`\alpha `$ equivalent widths that are difficult to understand without invoking exceptional conditions (PopIII stars?; Malhotra & Rhoads 2002, Rhoads et al. 2003). Given the few data available for the LALA sources it is fair to say that the nature of these objects, their stellar populations, extinction etc. remain currently largely unknown (cf. Dawson et al. 2004). When possible, a simple comparison between the UV and Ly$`\alpha `$ SFR is undertaken providing possibly information on the Ly$`\alpha `$ transmission, i.e. the partial absorption of Ly$`\alpha `$ photons on their sight line through the intergalactic medium (e.g. Haiman 2002, Santos 2004) and/or on partial Ly$`\alpha `$ “destruction” processes close to the source (e.g. due to dust or ISM geometry; Charlot & Fall 1993, Valls-Gabaud 1993, Tenorio-Tagle et al. 1999, Mas-Hesse et al. 2003).
Notable exceptions of $`z4`$ samples for which some estimate of extinction is available from multi-band photometry include work on the Subaru Deep Survey (Ouchi et al. 2004) and GOODS data (e.g. Papovich et al. 2004). Lehnert & Bremer (2004) also discuss some preliminary information on little extinction in their $`z>5`$ sources. Interestingly, in their study of a $`z=5.34`$ galaxy discovered by Dey et al. (1998), Armus et al. (1998) find indications for significant reddening ($`A_V>0.5`$ mag) from analysis of the observed SED and from the presence of Ly$`\alpha `$ emission.
In a similar manner we will here present a consistent study of the stellar population properties, extinction, and Ly$`\alpha `$ emission for two galaxies at redshift $`z6`$. For this aim we use two distant ($`z6`$) gravitationally lensed galaxies for which multi-band photometry is available (detection in at least 3–4 bands). Through a quantitative analysis of their SED, using a vast library of empirical and theoretical template spectra, we aim to constrain properties of the stellar populations, such as age and star formation (hereafter SF) history (burst or constant SF?) and their extinction. Furthermore by comparing the Ly$`\alpha `$ emission expected from the stellar population constraint with the observed Ly$`\alpha `$ flux we estimate consistently the Ly$`\alpha `$ “transmission” for the individual sources.
The Ly$`\alpha `$ transmission and SF properties derived here can in principle be used to infer the ionisation fraction of hydrogen in the IGM at a given redshift (cf. Haiman 2002, Santos 2004), a key quantity of interest for the study of the reionisation history of the Universe (cf. review from Barkana & Loeb 2001). Obviously the present “exploratory” work will have to be extended to larger galaxy samples, and sophisticated tools will probably be needed to interpret such results in terms of IGM properties (cf. Gnedin & Prada 2004). However, this approach should be complementary to other methods probing the reionisation history by measuring the Gunn-Peterson optical depth observed in quasar spectra as a function of redshift (e.g. Becker et al. 2001, Fan et al. 2003), or by comparing Ly$`\alpha `$ luminosity functions at different redshifts (e.g. Malhotra & Rhoads 2004).
The remainder of the paper is structured as follows. In Sect. 2 we summarise the adopted observational constraints from the literature. Our modeling technique is described in Sect. 3. The detailed results for each galaxy are presented in Sects. 5 and 4. Our main conclusion are summarised in Sect. 6.
## 2 Observational constraints
The two galaxies studied here are: 1) The probable $`z7`$ galaxy recently discovered by Kneib et al. (2004, hereafter KESR), which presently lacks of a spectroscopic redshift but for which rather accurate multi-band HST observations are available, allowing us in particular also to derive a fairly reliable photometric redshift. 2) the $`z=6.56`$ Ly$`\alpha `$ emitter HCM 6A behind the lensing cluster Abell 370. We now summarise the observational data, taken from the literature. The adopted redshift and gravitational magnification factors are listed in Table 1.
Before proceeding let us mention for clarity that these two objects are generally considered to be star forming galaxies (starbursts), not AGN (narrow line - type II - or others), as no contradicting information is available so far. However, one must bear in mind that some of the interpretations presented below (and in the literature) may need to be revised, should this assumption be incorrect.
Triple arc in Abell 2218: The observational data for this object, named Abell 2218 KESR hereafter, is taken from Kneib et al. (2004, hereafter KESR) and from Egami et al. (2005). The photometry from KESR includes observations with HST (WFPC2, ACS, NICMOS) in V<sub>606W</sub>(undetected), I<sub>814W</sub>, z<sub>850LP</sub>, and H<sub>160W</sub>, and with NIRC/Keck in $`J`$. Subsequently, additional photometry was obtained with NICMOS/HST in the $`J`$ band (F110W), and with IRAC/Spitzer at 3.6 and 4.5 $`\mu `$m (see Egami et al.) For our computations (see below) we have used the appropriate filter transmission curves. In particular, updated transmission curves were used for the ACS and NICMOS filters (M. Sirianni 2003, private communication; Sirianni et al. 2004; R. Thompson 2003, private communication).
Few brief comments concerning the photometry are needed here. First, KESR present photometry for two multiple images (a and b). Apparently sources a and b differ in the z<sub>850LP</sub> flux (with quoted errors of $`\pm `$ 0.05 mag) by 3.2 $`\sigma `$, whereas the fluxes in the other filters agree well within 1 $`\sigma `$. Differential lensing across the images together with sampling effects could be responsible for this small discrepancy. As we are interested in a global representative SED for this source, we have chosen to use the averaged photometric SED, the magnification factors being the same for the two images. Finally, we have also noted some apparent discrepancies between the measurements reported in KESR and Egami et al., the most important one being the H<sub>160W</sub> flux, which is $``$ 15–20 % (3–4 $`\sigma `$) higher in the latter publication. These differences are mostly due to the use of different apertures on different repixeled/rescaled images (J. Richard, 2004, private communication). Again, this illustrates the difficulty in deriving reliable colors for extended arcs. To account for these small discrepancies and for the possible error underestimate we therefore adopt a minimum photometric error of 0.15 mag in all filters. It is worth noting that photometric errors translate into absolute flux calibration errors for fitting purposes. As we will see below, adopting the latter minimum errorbars significantly improves the SED fits.
In addition to the photometry, the non-detection of the source with Keck LRIS spectroscopy provides an upper limit on the continuum flux between 9000 and 9300 Å (KESR). This upper limit will be used as an additional constraint in our SED modeling. KESR also indicate a possible drop of the continuum below $``$ 9800 Å from their Keck II NIRSPEC spectrum. For various reasons the reality of this spectral break is questionable. First a true neutral hydrogen break (“Ly$`\alpha `$ break”) at $``$ 9800 Å, far in the red wing of the z<sub>850LP</sub> filter, seems incompatible with the relatively strong flux measured in this filter. Furthermore, test computations show that such a break is difficult if not impossible to reconcile with our spectral modeling. In any case the significance of this finding appears questionable as the detected continuum is extremely faint and noisy. The reality of this spectral feature is now also questioned by Egami et al. (2005). For these reasons this information is discarded from our spectral fitting.
In practice we have retained the following two variants to describe the observed SED of this source: SED1) The average fluxes (I<sub>814W</sub>, z<sub>850LP</sub>, H<sub>110W</sub>, H<sub>160W</sub>) of images a and b from KESR plus the IRAC/Spitzer data of image b from Egami et al. SED2) All fluxes from image b from Egami et al. These are treated as SEDs from two different objects. Furthermore, for each of these “objects” we have computed two cases in our SED fitting described below: i) The observed SEDs in I<sub>814W</sub>, z<sub>850LP</sub>, H<sub>110W</sub>, H<sub>160W</sub>, 3.6, and 4.5 $`\mu `$m. ii) Same as (i) plus the flux limits from the V<sub>606W</sub> and Keck LRIS non-detections.
No emission line has so far been detected for Abell 2218 KESR. Its spectroscopic redshift remains therefore presently unknown but the well-constrained mass model for the cluster strongly suggests a redshift $`z`$ 6 for this source (KESR, Egami et al. 2005). The magnification factors of both images a and b is $`\mu =25\pm 3`$, according to KESR.
Abell 370 HCM6A: The observational data of this $`z=6.56`$ galaxy is taken from Hu et al. (2002). The photometry includes $`VRIZJHK^{}`$ from Keck I and II (LRIS and Echellette Spectrograph and Imager) and from Subaru (CISCO/OHS). The gravitational magnification of the source is $`\mu =4.5`$ according to Hu et al. (2002).
Photometric fluxes and errors were adopted from their Fig. 3. Where possible the appropriate filter transmission curves were used. The “$`Z`$” band filter transmission is somewhat uncertain, as these observations were undertaken using an RG850 filter, which together with the LRIS optics and the CCD response, yields a transmission similar to a $`Z`$ band filter (Hu et al. 1999). Our approximate filter curve shows a blueward shift of $`\lambda _{\mathrm{eff}}`$ by $``$ 200 Å compared to the information given by Hu et al. (1999). However, since the redshift is known for this source and since we adjust the observed flux (not magnitude) in this band, this should not affect our conclusions.
## 3 SED modeling
### 3.1 Main restframe UV-optical SED features of high-z galaxies and their “information content”
Before proceeding to the fits of the individual SEDs a brief comment on the available SED features seems appropriate.
For obvious reasons the available SED (basically from broad-band photometry) of high-z ($`z6`$) galaxies is primarily limited to the rest-frame UV (when observed from the ground) or optical spectrum (when available e.g. with Spitzer and future satellite missions). The main information “encoded” in this SED is therefore: 1) the neutral HI break shortward of Ly$`\alpha `$ (hereafter the “Ly$`\alpha `$” break) due to the strong or complete Gunn-Peterson trough, 2) the slope of the UV spectrum, and 3) possibly a 4000 Å break (hereafter denoted Balmer break), if present and covered by the observations. In addition the presence of the Ly$`\alpha `$ line, mostly used to determine spectroscopically the redshift, provides clear evidence for ongoing massive star formation (hereafter SF).
The position of the Ly$`\alpha `$ break depends essentially on redshift. The UV slope depends on the intrinsic spectrum – in turn depending mostly on age and SF history – and on the extinction, i.e. the extinction law and the amount of reddening. The Balmer break becomes visible (in absorption) in the continuum of stellar populations after $``$ 10–30 Myr. Ly$`\alpha `$ emission, if due to stellar photoionisation and not AGN activity, indicates the presence of young ($``$ 10 Myr) massive ionizing stars.
Concerning the UV slope, it is useful to recall that this quantity <sup>1</sup><sup>1</sup>1Various definitions of the UV slope exist. The most commonly used ones, generally denoted $`\beta `$, are defined as the power-law index of the SED in $`F_\lambda `$ versus $`\lambda `$ over a certain wavelength interval. does not lend itself to determine the metallicity of a star forming galaxy from a theoretical point of view and in terms of individual objects. The reasons are that intrinsically the UV slope shows only small variations with metallicity, and that the slope depends strongly on the exact SF history (see e.g. Leitherer & Heckman 1995, Meurer et al. 1995). This is illustrated in Fig. 1 where $`\beta `$ (measured over the interval 1300-1800 Å) is plotted as a function of age for populations of metallicities between solar and zero (PopIII) and for the limiting cases of bursts and SFR=const. Furthermore, as pointed out in Schaerer (2002, 2003) and also shown in this figure, for very low metallicities ($`Z1/50Z_{}`$) nebular continuous emission becomes dominant even down to UV wavelengths (longward of Ly$`\alpha `$), such that the observed integrated (stellar+nebular) spectrum has even a flatter UV slope than high metallicity starbursts. In other words, even for bursts, there are strong intrinsic degeneracies of $`\beta `$ between age and metallicity, to which the additional effect of reddening must be added, including the uncertainties on the a priori unknown extinction law. It is therefore evident that on an individual object basis there are in general degeneracies between age, metallicity, SF history, and extinction. However, this does not preclude the possible existence of statistical correlations between quantities such as e.g. $`\beta `$ and metallicity in large samples of galaxies, as known to hold e.g. for local UV selected starbursts (cf. Heckman et al. 1998). Also, as we will see below, there are cases where the UV slope and the mere fact of the presence of an emission line allow us nevertheless to lift some degeneracies and therefore to determine interesting constraints on the stellar population and on extinction.
The behaviour of the 4000 Å break and its use as an age indicator has extensively been discussed in the literature (e.g. Bruzual 1983, and recently Kauffmann et al. 2003). As in simple stellar populations its amplitude is basically a monotonically increasing function of age, an estimate of the break, e.g. obtained from 3.8-4.5 $`\mu `$m photometry with IRAC/Spitzer and JHK photometry in $`z6`$ galaxies, provides information on the age of the light emitting stellar population. Since the exact SF history cannot be determined in these cases (in contrast to studies at low $`z`$, cf. Kauffmann et al.) the amplitude of the break provides a range of ages, the minimum age being given by instantaneous bursts, the maximum age from models with constant SF. For obvious reasons this maximum “luminosity weighted” age derived from a measure of the Balmer break is also unaffected by considerations of possible multiple stellar populations. The same is not true for the minimum age, which is, however, of less cosmological interest.
### 3.2 Spectral fitting
For the spectral fitting we use a slightly adapted version of the photometric redshift code Hyperz of Bolzonella et al. (2000). Hyperz does standard SED fitting using a number of modeling parameters. The free parameters for the SED modeling are:
* the spectral template,
* extinction and the reddening law,
* a parameter $`f_{\mathrm{Lyf}}`$ describing possible deviations from the average Lyman forest attenuation from Madau (1995).
For Abell 2218 KESR the source redshift is also a free parameter.
For the spectral templates we use a large compilation of empirical and theoretical SED, including starbursts, QSO, and galaxies of all Hubble types, and covering various star formation histories (bursts, exponentially decreasing, constant SF) and various metallicities. For most applications we group the templates in the following way:
* Starbursts and QSOs (hereafter SB+QSO): this group includes the starburst templates with $`E(BV)`$ from $`<0.1`$ to 0.7 from the Calzetti et al. (1994) and Kinney et al. (1996) atlas, the HST QSO template of Zheng et al. (1997), as well as UV-optical spectrum of the metal-poor galaxy SBS 0335-052 with numerous strong optical emission lines (and an extinction of $`E(BV)0.09`$, Izotov & Thuan 1998) kindly communicated to us by Yuri Izotov (2002, private communication)
* BCCWW+: Bruzual & Charlot (1998, private communication; cf. Bruzual & Charlot 1993) evolving synthesis models assuming bursts, constant star formation, and exponentially decaying star formation histories reproducing present day spectra of galaxies of various types (E, S0, Sa, Sb, Sc, Sd, and Im) plus the empirical E, Sbc, Scd, and Im templates from Coleman et al. (1980), as included in the public Hyperz version.
* S03+: Theoretical templates of starburst galaxies from Schaerer (2003) covering metallicities of $`Z=0.02`$ (solar), 0.008, 0.004, 0.001, 1/50 Z, $`Z=10^5`$, $`10^7`$, and zero metallicity (PopIII). For low metallicities ($`Z10^5`$) these templates have been computed for 3 different assumptions on the IMF. The spectral library includes burst models and models with a constant star formation rate (SFR). For more details see Schaerer (2003). For the present work these computations were extended to cover ages of up to 1 Gyr. These SEDs are available on request from the first author and on the Web<sup>2</sup><sup>2</sup>2http://obswww.unige.ch/sfr.
The standard extinction law adopted here is the one from Calzetti et al. (2000) determined empirically from nearby starbursts. We also explore the possible implications of other laws, such as the Galactic law of Seaton (1979) including the 2200 Å bump, and the SMC law from Prévot et al. (1984) and Bouchet et al. (1985) showing no UV bump, but a steeper increase of the extinction in the UV compared to Calzetti et al..
For the Lyman forest attenuation, Hyperz follows Madau (1995). However, we allow for possible deviations from the mean attenuation by varying the Lyman forest optical depths $`\tau _{\mathrm{eff}}^{\alpha ,\beta }`$ by a multiplicative factor taking the values of ($`f_{\mathrm{Lyf}}`$, 1., and $`1/f_{\mathrm{Lyf}}`$). Typically we adopted $`f_{\mathrm{Lyf}}=`$ 2 or 3. Here $`\tau _{\mathrm{eff}}^{\alpha ,\beta }`$ stands for the optical depths corresponding to the absorption between Ly$`\alpha `$ and Ly$`\beta `$, and between Ly$`\beta `$ and the Lyman limit respectively.
The following other minor changes have been made in our version (1.3ds) of Hyperz. The calculation of the synthetic photometry deals correctly with templates including strong spectral lines (emission or absorption). Furthermore we make sure to use the proper filter transmission curves usually given in photon units. Earlier versions of Hyperz and other codes (e.g. evolutionary synthesis codes) assume sometimes (for “historical” reasons) that transmission curves be given in flux units. In case of wide filters, e.g. such as some ACS/HST filters, this may lead to small differences. Other modifications concern essentially features related to the user interface (additional outputs etc.).
For given choices of the above parameters, theHyperz code performs a standard minimisation fit to the observed SEDs and determines, for each point in the parameter space, the corresponding $`\chi ^2`$ value. Using these $`\chi ^2`$ values, it is possible to quantify the probabilities for the main free parameters, namely extinction, age of the spectral template, SF history, etc. When the SED fitting is based on theoretical templates, the SFR value is easily obtained and allows us to compare the expected values for the Ly$`\alpha `$ flux to the actual ones.
To convert the observed/adjusted quantities to absolute values we adopt the following cosmological parameters: $`\mathrm{\Omega }_m=0.3`$, $`\mathrm{\Omega }_\mathrm{\Lambda }=0.7`$, and $`H_0=70`$ km s<sup>-1</sup> Mpc<sup>-1</sup>.
## 4 Results for Abell 2218 KESR
### 4.1 Photometric redshift estimate
As a spectroscopic redshift has not been obtained (yet) for this galaxy we here examine its photometric redshift estimate. In Fig. 2 we show the photometric redshift probability distributions $`P(z)`$ for the two SEDs (SED1, SED2) of Abell 2218 KESR described above using the three spectral template groups and adopting a minimum photometric error of 0.15 mag. For each redshift, $`P(z)`$ quantifies the quality of the best fit model obtained varying all other parameters (i.e. extinction, $`f_{\mathrm{Lyf}}`$, spectral template among template group). Given the excellent HST (WFPC2, ACS and NICMOS) photometry, $`P(z)`$ is quite well defined: the photometric redshift ranges typically between $`z_{\mathrm{phot}}`$ 5.5 and 7.3. Outside of the plotted redshift range $`P(z)`$ is essentially zero. If we assume the (smaller) quoted formal photometric errors (but note the discrepancies discussed in Sect. 2) $`P(z)`$ becomes more peaked, i.e. the photometric redshift better defined. This is driven by the error on the z<sub>850LP</sub> flux, which determines the red side of the “Ly$`\alpha `$” break. However, the resulting best fit value $`z_{\mathrm{phot}}`$ does not change much. Furthermore, the fit quality is considerably decreased. This demonstrates the interest of such high accuracy measurements and the need for reliable error estimates.
The predicted redshift distribution is found to be quite insensitive to the exact template (as shown in Fig. 2), to the exact value of $`f_{\mathrm{Lyf}}`$, and to the adopted extinction law (variations of the latter two are not shown). However, we note that for this object the fits (and $`P(z)`$) are improved when allowing for deviations from the average Madau (1995) attenuation law. The curves shown here have been computed for $`f_{\mathrm{Lyf}}=2`$.
More important in determining $`P(z)`$ is the exact SED. As seen from Fig. 2 the use of SED1 or SED2 lead to somewhat different $`P(z)`$ distributions. SED2 (cf. Sect. 2) yields a somewhat larger redshifts, albeit with a somewhat yreduced fit quality. These differences illustrate how uncertainties and difficulties in the photometric measurements of such faint sources, whose origin are briefly discussed in Sect. 2, propagate to the photometric redshift estimate.
All our best-fit solutions have redshift $`z_{\mathrm{phot}}`$ 6.25–6.63, lower than the redshift range estimated by KESR, but compatible with the more recent quantitative analysis of Egami et al. (2005). Given the various free parameters, uncertainties on the intrinsic SED, etc. we conclude that the redshift of Abell 2218 KESR is likely $`z`$ 6.0 – 7.2. taking into account both our photometric determination and the lensing considerations of KESR.
### 4.2 SED fits and inferences on the stellar population and on Ly$`\alpha `$.
A large number of models have been computed using the different variants of the SEDs describing this object (SED1-2), the different filter combinations (non-detections + spectroscopic constraint) discussed in Sect. 2, and varying the various model parameters. We first discuss briefly the main salient results with the help of some illustrations. A more general discussion of the results and their dependence on various assumptions follows.
#### 4.2.1 Age, star formation history, and extinction
Figure 3 shows the best fit models to the SED2 including the upper limits from the I<sub>814W</sub> and LRIS spectroscopy for the S03+ and SB+QSO template groups. The best fit redshifts are $`z_{\mathrm{phot}}`$ $`=6.63`$ and 6.54 respectively. These fits show in particular that the spectroscopic constraint can be accommodated simultaneously with the observed z<sub>850LP</sub> flux; the resulting fits are within the 1 $`\sigma `$ errors in all bands. The best fit from the S03+ group corresponds to a burst with an age of 15 Myr at solar metallicity and no extinction (solid line). Similarly good fits are also obtained for lower metallicity. The best fit with empirical starburst and QSO templates (SB+QSO) is obtained with the spectrum of the metal-poor H ii galaxy SBS 0335-052 (dotted line in Fig. 3). In this case the apparent Balmer break observed between the NICMOS/HST and IRAC/Spitzer domain is simply explained by the presence of strong emission lines in the 3.6 and 4.5 $`\mu `$m filters <sup>3</sup><sup>3</sup>3The main lines are between H$`\gamma `$, H$`\beta `$ and \[O iii\] $`\lambda \lambda `$4959,5007 in the 3.6 $`\mu `$m filter and He i $`\lambda `$5876 in the 4.5 $`\mu `$m filter. E.g. for the emission lines between H$`\gamma `$ and \[O iii\] $`\lambda \lambda `$4959,5007 the total observed equivalent width (boosted by the $`(1+z)`$ factor) is $``$ 9130 Å, as estimated from the data of Izotov & Thuan (1998), compared to a filter surface of $``$ 6600 Å. and some additional extinction to reduce the restframe UV flux. The extinction needed is $`A_V=0.6`$ for the Calzetti et al. law, or $`A_V=0.2`$ for the Prévot et al. extinction law. In terms of age the restframe UV to optical spectrum (continuum and lines) of SBS 0335-052 corresponds to a young population of $``$ 3–5 Myr according to the analysis of Papaderos et al. (1998) and Vanzi et al. (2000). Of course, the presence of an older population in addition to the starburst cannot be excluded on the present grounds. In short, the observed SED of Abell 2218 KESR can be explained by a young population withour or with emission lines. Spectroscopy in the 3–4 $`\mu `$m range would be needed to distinguish the latter solution from others.
Alternatively, good SED fits are also obtained with relatively “old” populations. The oldest ages are obtained when invoking the longest SF timescale, i.e. constant SF. In this case the UV restframe flux remains high (due to the continuous formation of massive stars) and older ages need to be attained to build up a sufficient population of evolved stars with strong Balmer breaks, in order to reproduce the observed break. This case is shown in Fig. 4 with fits with the S03+ templates to the observed SED1 and SED2. The best fits solutions obtained here correspond to ages of 500 and 400 Myr, no extinction, and redshifts $`z_{\mathrm{phot}}`$ $``$ 6.40 and 6.57 respectively. Similar, but somewhat older ages are obtained for metallicities $`Z<0.008`$, below the ones shown here.
A quantitative examination of the “maximum age” allowed by the observations (here SED2) is presented in Fig. 5, showing $`\chi ^2`$ contours in the extinction–age plane for a given set of spectral templates (S03+ group with constant SFR), the Calzetti et al. extinction law, and a fixed redshift of $`z=6.55`$. For these conditions the best fit is corresponds to 400 Myr, zero extinction, and $`z_{\mathrm{phot}}=6.57`$ (see Fig. 4). This Figure shows that a maximum age of $``$ (250–650) Myr (1 $`\sigma `$ interval) is obtained, in good agreement with the modeling of Egami et al. (2005). If true, this would correspond to a formation redshift of $`z_{\mathrm{form}}`$ 8.7–20 for our adopted cosmological parameters. As clear from Fig. 5, even with constant SF models, younger populations with some extinction can also fit, although less well, the present observations. However, solutions with low or zero extinction are generally preferred.
When varying the star formation history between these extreme cases (burst or SFR$`=`$const), i.e. considering e.g. exponentially declining SF histories, any intermediate age can be found for obvious reasons. Such cases are e.g. obtained when fitting templates from the Bruzual & Charlot models (not shown here) with exponentially decreasing SF histories and can be found in Egami et al. As discussed in Sect. 3.1, considering multiple stellar populations (cf. Eyles et al. 2005) does not alter the above estimate of the maximum age determined from constant SFR models. In any case, the data available here does not allow us to constrain the SF history and age further.
#### 4.2.2 Ly$`\alpha `$ emission
The observations obtained so far have not revealed any emission line from this object (KESR). In particular Ly$`\alpha `$ emission is lacking, which could be puzzling for a source with intense star formation. As we have just seen a variety of star formation histories and ages are possible for Abell 2218 KESR. Therefore one may or may not expect intrinsic Ly$`\alpha `$ emission.
A simple explanation for the apparent absence of Ly$`\alpha `$ emission could be to invoke an advanced age (in the post starburst phase). However, even with a young age it is not necessary that Ly$`\alpha `$ emission be observed. E.g. the spectrum of the metal-poor H ii galaxy SBS 03335-052 which provides an excellent fit to the observed SED and shows strong emission lines (cf. Fig. 3) does not show Ly$`\alpha `$ emission (Thuan et al. 1997, Kunth et al. 2003). Alternatively, if intrinsically present, the Ly$`\alpha `$ non-detection could be due a variety of factors: a redshift $`z6.4`$ placing Ly$`\alpha `$ below the spectral range discussed in detail by KESR<sup>4</sup><sup>4</sup>4This is probably excluded as, according to J.-P. Kneib (2005, private communication), no emission line was found in the blue part of the spectrum taken with LRIS., a flux below their strongly varying detection threshold, or other factors depressing the Ly$`\alpha `$ emission within the host galaxy (dust, ISM+HI geometry) and in the intervening IGM.
In conclusion, from the available data the apparent lack of Ly$`\alpha `$ emission from this source is not puzzling. However, it is not completely excluded that the galaxy truely shows Ly$`\alpha `$ emission, which has so far eluded detection.
#### 4.2.3 General comments on fits and discussion
After these main findings we shall now quickly mention more “technical” results about the influence of various fit parameters.
Quite generally, the results on the age, SF history, extinction etc. depend little on the different variants of the observed SEDs (SED1-2), on the inclusion or not of the non-detections in the fits, and on the use of the published formal errors or a minimum error of 0.15 mag (cf. Sect. 2). The results discussed above are therefore quite robust with respect to these assumptions. Small differences in the best fit values can, however, be obtained. E.g. the best fit photometric redshift can vary by up to $``$ 0.2 depending on adopting SED1 or SED2. In all cases we note that SED1 allows better fits (smaller $`\chi ^2`$) than SED2. Adopting $`\sigma _{\mathrm{min}}=0.15`$ mag also significantly increases the fit quality. Finally, considering variations around the mean Lyman forest attenuation improves the fits (especially as the HST photometry determining the “Ly$`\alpha `$ break” is quite accurate). In practice all best fits are found with an increased Lyman-forest opacity ($`f_{\mathrm{Lyf}}=2`$).
To summarise, given the absence of a spectroscopic redshift, a fair number of good fits is found to the observations of Abell 2218 KESR when considering all the free parameters. The main conclusions from these “best fits” are:
* Generally the determined extinction is negligible or zero quite independently of the adopted extinction law. The best fit with the empirical starburst spectrum of SBS 0335-052 represents an exception to this case, requiring an additional $`A_V`$ 0.2–0.6 mag, depending on the adopted extinction law.
* Although generally burst models fit somewhat better than those with constant star formation among the theoretical templates (BC, S03+), the data does not strongly constrain the star formation history.
* Typical ages between $``$ 15 and 400 Myr are obtained. A reasonable 1-$`\sigma `$ upper bound on the age of $``$ 650 Myr can be obtained assuming constant star formation. However, the data can also be well fit with a very young ($``$ 3–5 Myr) stellar population with strong emission lines (using e.g. the spectrum of the metal-poor galaxy SBS 0335-052). In this case the apparent Balmer break observed between the HST and Spitzer broad-band photometry is simply due to the presence of strong emission lines affecting the red 3.6 and 4.5 $`\mu `$m filters.
* Given degeneracies of the restframe UV spectra between age and metallicity (cf. above) no clear indication on the galaxian metallicity can be derived, in contrast to the claim of KESR. Good fits to the available data can even be found with solar metallicity starburst templates.
* Depending on the star formation history and age one may or may not expect intrinsic Ly$`\alpha `$ emission, i.e. an important H ii region around the object. The apparent absence of observed Ly$`\alpha `$ emission does therefore not provide much insight.
A more complete error analysis beyond the level presented here is difficult to achieve for a variety of reasons and clearly beyond the scope of this publication.
#### 4.2.4 SFR, stellar mass and luminosity
The theoretical templates can also be used to estimate the stellar mass involved in the starburst or the star formation rate when constant star formation is assumed. For this aim we use all the best fits to the three SEDs (SED1-3) with the S03+ templates, we assume a typical redshift of $`z=6.6`$, and the magnification $`\mu =25`$ determined by KESR. For the adopted cosmology the distance luminosity is then $`d_L=`$64457.8 Mpc.
When constant SF is assumed one obtains the following star formation rate: $`SFR(0.91.1)`$ M yr<sup>-1</sup> (for a Salpeter IMF from 1 to 100 M). For the best fit ages of $``$ 400–570 Myr the total mass of stars formed would then correspond to $`(3.66.3)\times 10^8`$ M. The mass estimated from best fit burst models (of ages $``$ 6–20 Myr) is slightly smaller, $`M_{}(0.31)\times 10^8`$ M. If we assume a Salpeter IMF with M<sub>low</sub> $`=0.1`$ M the mass and SFR estimates would be higher by a factor 2.55, and in good agreement with the values derived by KESR and Egami et al. (2005). In all the above cases the total luminosity (unlensed) is typically $`L_{\mathrm{bol}}2\times 10^{10}`$ L.
## 5 Results for Abell 370 HCM 6A
### 5.1 SED fits and inferences on the stellar population
Overall the published SED of HCM 6A (see Fig. 6) is “reddish”, showing an increase of the flux from $`Z`$ to $`H`$ and even to $`K^{}`$<sup>5</sup><sup>5</sup>5The significance of a change of the SED slope between $`ZJH`$ and $`HK^{}`$ seems weak, and difficult to understand.. From this simple fact and the above explanations it is already clear qualitatively that one is driven towards solutions with a) “advanced” age and little extinction or b) constant or young star formation plus extinction. However, a) can be excluded as no Ly$`\alpha `$ emission would be expected in this case.
Quantitatively, the best solutions obtained for each spectral template group is shown in the left panel of Figure 6. Indeed, the solutions shown correspond to bursts of ages $``$ 50–130 Myr (BCCWW+, S03 templates) and little or no extinction. However, as just mentioned, solutions lacking young ($``$ 10 Myr) massive stars can be excluded since Ly$`\alpha `$ emission is observed. The best fit empirical SB+QSO template shown corresponds to the spectrum of the H ii galaxy SBS 0335-052 with an additional extinction of $`A_V=1.`$ To reconcile the observed SED with Ly$`\alpha `$, a young population, e.g. such as SBS 0335-052, or constant SF is required. In any of these cases fitting the “reddish” SED requires a non negligible amount of reddening.
To illustrate the typical range of possible results we show in Fig. 7 $`\chi ^2`$ contour maps and corresponding confidence intervals for solar metallicity models (S03+ template group) and reddened with the Calzetti law. The left panel (burst models) illustrates in particular the need for progressively higher extinctions the younger the bursts. All ages $``$ 10 Myr are excluded for the absence of Ly$`\alpha `$ emission. From the constant SF models (right panel) we see that for a given age $`A_V`$ is typically $``$ 0.5–1.8 mag at the 68 % confidence level. For obvious reasons, no constraint can be set on the age since the onset of (constant) SF.
Hence, from the photometry of HCM 6A and from the presence of Ly$`\alpha `$ we are led to conclude that this object must suffer from reddening with typical values of $`A_V1.`$ for a Calzetti attenuation law. A somewhat smaller extinction ($`A_V0.4`$) can be obtained if the steeper SMC extinction law of Prévot et al. (1984) is adopted. From the present data it is not possible to distinguish the different extinction/attenuation laws. Also, it is not possible to draw any constraints on the metallicity of HCM 6A from the available data (cf. Sect. 3.1).
What if we are dealing with composite stellar populations? Indeed it is conceivable that the Ly$`\alpha `$ emission originates from a population of young stars and the “reddish” restframe UV flux be due to another, older population. Assuming constant SF, no loss of Ly$`\alpha `$ and standard SFR conversion factors, the observed Ly$`\alpha `$ emission implies a maximum UV flux of the order of 0.1 $`\mu `$Jy (and approximately constant in $`F_\nu `$ over the observed wavelength range: $`\lambda _{\mathrm{rest}}3000`$ Å) for an unreddened population. The bulk of the observed flux could then be from an older population. In this case the rising spectrum from the z<sub>850LP</sub> over the JH (and presumably K) bands could even be due to an unreddened population; a strongly increasing flux and probably a significant “Balmer” break is then expected, similar to the aged burst shown in the right panel of Fig. 6. This explanation should in principle be testable with Spitzer observations as discussed below.
How does our possible indication for a high extinction fit in with other studies? At redshift $`z4`$ the extinction of Lyman break galaxies (LBGs) has been estimated by various authors (e.g. Sawicky & Yee 1998, Meurer et al. 1999, Adelberger & Steidel 2000, Shapley et al. 2001, Ouchi et al. 2004). Given their mean/median values (typically $`<E(BV)>`$ 0.15–0.2) and the $`E(BV)`$, our finding of a “high” extinction is not exceptional. Furthermore, Armus et al. (1998) find indications for $`A_V>0.5`$ mag from their analysis of a $`z=5.34`$ galaxy. However, this implies that dust extinction is likely present in starburst galaxies with redshifts above 6. Large amounts of dust have already been observed in QSOs up to similar redshift (Bertoldi et al. 2003, Walter et al. 2003).
### 5.2 Properties of HCM6A: SFR, mass, Ly$`\alpha `$ transmission
To estimate properties such as the mass, SFR, and the intrinsic Ly$`\alpha `$ emission from HCM 6A we simply examine the predictions from the best fit models, scale them appropriately to the luminosity distance<sup>6</sup><sup>6</sup>6For the adopted cosmology and $`z=6.56`$ one has $`d_L=`$ 64005.7 Mpc., and correct for the gravitational magnification (here $`\mu =4.5`$). The derived quantities are summarised in Table 1.
First we consider single (non-composite) stellar populations. From the best fit constant SF models (with variable ages) we deduce an extinction corrected star formation rate of the order of SFR(UV) $``$ 11 – 41 M yr<sup>-1</sup> for a Salpeter IMF from 1 to 100 M. For a commonly adopted, although unjustified, Salpeter IMF down to 0.1 M this would increase by a factor 2.55. Actually this estimate is not very different than the one obtained from standard SFR calibrations provided the same assumptions on the IMF. Indeed the observed restframe UV luminosity, e.g. derived from the average J, H, and K flux of $`F_{\mathrm{UV}}=(2.6\pm 0.7)\times 10^{30}`$ erg s<sup>-1</sup> cm<sup>-2</sup> Hz<sup>-1</sup> is $`L_{\mathrm{restUV}}=4\pi d_L^2F_{\mathrm{UV}}/(1+z)/\mu 4.\times 10^{28}`$ erg s<sup>-1</sup>Hz<sup>-1</sup> (Hu et al. 2002), translates to $`SFR_{\mathrm{UV}}=cL_{\mathrm{restUV}}\mathrm{\hspace{0.17em}10}^{0.4A_{\mathrm{UV}}}`$, where $`c`$ is the usual SFR conversion coefficient, and $`A_{\mathrm{UV}}`$ the UV extinction. For the standard value $`c=1.4\times 10^{28}`$ from Kennicutt (1998), assuming a Salpeter IMF down to 0.1 M, and $`A_{\mathrm{UV}}`$ 2.–3., one has $`SFR`$ 35–88 M yr<sup>-1</sup>; for the IMF used in this work (Salpeter from 1-100 M) this becomes $`SFR`$ 14–34 M yr<sup>-1</sup>. This assumption, and the absence of extinction correction also explains the difference with $`SFR`$ estimate of Hu et al. (2002) <sup>7</sup><sup>7</sup>7Actually Hu et al. (2002) derive without further explanation $`SFR=9`$ M yr<sup>-1</sup> from $`L_{\mathrm{restUV}}=4.\times 10^{28}`$ erg s<sup>-1</sup>, whereas the classical Kennicutt (1998) calibration would yield $`SFR=5.6`$ M yr<sup>-1</sup> without extinction correction..
For continuous SF over timescales $`t_{\mathrm{SF}}`$ longer than $``$ 10 Myr, the total (bolometric) luminosity output is typically $`10^{10}`$ L per unit SFR (in M yr<sup>-1</sup>) for a Salpeter IMF from 1-100 M, quite independently of metallicity. The total luminosity associated with the observed SF is therefore $`L(14)\times 10^{11}L_{}`$, close to or just above the limit to possibly qualify as a luminous infrared galaxy ($`L_{\mathrm{IR}}>10^{11}L_{}`$; cf. Sanders & Mirabel 1996) if a significant fraction of its bolometric flux emerges in the (restframe) IR. For $`t_{\mathrm{SF}}`$ 10 Myr the estimated stellar mass is $`M_{}t_{\mathrm{SF}}\times SFR(14)\times 10^8`$ M.
From the data given by Hu et al. (2002), the observed Ly$`\alpha `$ flux is $`F(\mathrm{Ly}\alpha )=\mu \times 3.8\times 10^{18}`$ erg s<sup>-1</sup> cm<sup>-2</sup>, with the magnification factor $`\mu `$. The Ly$`\alpha `$ luminosity per unit SFR from the same S03 models used above is $`L(\mathrm{Ly}\alpha )=(2.44.4)\times 10^{42}`$ erg s<sup>-1</sup> (M yr<sup>-1</sup>)<sup>-1</sup> for metallicities between solar and 1/50 Z. The SFR deduced from Ly$`\alpha `$ would then be SFR$`(\mathrm{Ly}\alpha )`$ 0.4–0.8 M yr<sup>-1</sup> for HCM 6A. Taking an extinction of $`A_V=1.`$ (for the Calzetti law) into account implies a reddening corrected SFR$`(\mathrm{Ly}\alpha )`$ 7–12 M yr<sup>-1</sup>. The ratio between SFR$`(\mathrm{Ly}\alpha )`$/SFR(UV) presumably reflects the incomplete Ly$`\alpha `$ transmission $`t_{\mathrm{Ly}\alpha }`$, which can be estimated in various ways. The most consistent estimate is obtained from the comparison of the predicted Ly$`\alpha `$ luminosity of each best fit model (obtained from fitting the broad-band SED) to the observed Ly$`\alpha `$ luminosity. From the best fit models with $`A_V1`$ and the Calzetti law we obtain $`t_{\mathrm{Ly}\alpha }`$ 23–54 %; in the case of the fit with the Prévot et al. extinction law we find a higher transmission $`t_{\mathrm{Ly}\alpha }`$ 90 %. For comparison Haiman (2002) assumed a Ly$`\alpha `$ transmission of 20 % from the data of Hu et al. Per definition this Ly$`\alpha `$ “transmission” corresponds to the ratio of the expected/intrinsic Ly$`\alpha `$ emission from the starburst over the observed one. The physical causes for a partial (i.e. $`<`$ 100 %) transmission are of course open to various interpretations (e.g. physical processes destroying Ly$`\alpha `$ photons in the host galaxy, absorption in the intervening IGM etc.). In fact the relatively high Ly$`\alpha `$ transmission estimated here could be somewhat surprising, given the Gunn-Peterson trough observations in $`z6`$ quasars (cf. Becker et al. 2001, Fan et al. 2003) and the possible presence of dust (this work).
Consider now the case of composite stellar populations. In this case we retain as a rough estimate in Table 1 the $`SFR(\mathrm{Ly}\alpha )`$ (from the young population) as a lower limit, and the mass and total luminosity is derived from the best fit burst model of age $``$ 100 Myr assuming that this “older” population dominates the observed continuum flux. Formally we then have no handle on the Ly$`\alpha `$ transmission, except that it cannot be very low (say $``$ 40 % $`0.1/0.26=F_{\mathrm{young}}/<F_{\mathrm{obs}}>`$) since otherwise the associated UV flux from the young population $`F_{\mathrm{young}}`$ would dominate the observed continuum flux $`<F_{\mathrm{obs}}>`$.
### 5.3 Spitzer Observatory predictions
It is interesting to examine the SEDs predicted by the various models at longer wavelengths, including the rest-frame optical domain, which is potentially observable with the sensitive IRAC camera onboard the Spitzer Observatory and other future missions. In the right panel of Fig. 6 we plot the 3 best fits to the observed data for the BCCWW+ and S03+ template groups (“burst” solutions with no extinction) and the SBS 03352-052 template (with additional $`A_V=1`$) showing strong optical emission lines. We see that these solutions have fluxes comparable to or above the detection limit of IRAC/Spitzer <sup>8</sup><sup>8</sup>8The IRAC detection limits plotted here correspond to the values given by the Spitzer Science Center on http://ssc.spitzer.caltech.edu/irac/sens.html as 1 $`\sigma `$ point-source sensitivity for low and medium backgrounds for frame times of 200s and described by Fazio et al. (2004). These values do not include “confusion noise”.. On the other hand the strongly reddened constant SF or young burst solutions do not exhibit a Balmer break and are hence expected to show fluxes just below the IRAC sensitivity at 3.6 $`\mu `$m and significantly lower at longer wavelengths. As Ly$`\alpha `$ emission is expected only for the reddened SEDs the latter solutions (low 3.6-4.5 $`\mu `$m flux) are predicted to apply to HCM 6A, except if composite stellar populations are considered. Indeed, a high 3.6 and 4.5 $`\mu `$m flux could be a good indication for a composite stellar population, as discussed above. In addition it is important to secure higher accuracy photometry especially in the near-IR (JHK) to assess the accuracy of the redward increasing shape of the spectrum (in $`F_\nu `$), which drives one towards solutions with non-negligible extinction.
## 6 Conclusion
Using SED fitting techniques considering a large number of parameters (mainly a vast library of empirical and theoretical template spectra, variable extinction and extinction laws) we have attempted to constrain the properties of the stellar populations and Ly$`\alpha `$ emission of two strongly lensed galaxies with redshifts $`z`$ 6 from their observed SED including various ground-based observations, HST, and Spitzer observations.
The following main results have been obtained for these objects (see Sects. 5, 4), and summary in Table 1):
* Triple arc in Abell 2218 discovered by Kneib et al. (2004, KESR). The most likely redshift of this source is $`z`$ 6.0–7.2 taking into account both our photometric determination and lensing considerations.
SED fits indicate generally a low extinction ($`E(BV)0.05`$) but do not strongly constrain the SF history. Best fits have typical ages of $``$ 3 to 400 Myr. A reasonable maximum age of (250–650) Myr (1 $`\sigma `$ interval) can be estimated. However, the apparent 4000 Å break observed recently from combination of IRAC/Spitzer and HST observations, can also equally well be reproduced with the template of a young ($``$ 3–5 Myr) burst where strong restframe optical emission lines enhance the 3.6 and 4.5 $`\mu `$m fluxes.
The estimated SFR is typically $``$ 1 M yr<sup>-1</sup> for a Salpeter IMF from 1-100 M, in agreement with previous estimates.
Given the poor constraint on age and SF history, we conclude that intrinsic Ly$`\alpha `$ emission may or may not be present in this galaxy. The apparent non-detection of Ly$`\alpha `$ by KESR can therefore even be understood without invoking Ly$`\alpha `$ destruction.
* Abell 370 HCM 6A discovered by Hu et al. (2002). The relatively red SED and the presence of Ly$`\alpha `$ emission indicate basically two possible solutions: 1) a young burst or ongoing constant SF with non-negligible extinction ($`A_V`$ 0.5–1.8 at a 1 $`\sigma `$ level) or 2) a composite young + “old” stellar population.
For the first case, best fits are obtained for constant SF with $`E(BV)0.25`$. In consequence previous SFR estimates for this source must likely be revised upward. If correct, the bolometric luminosity of this galaxy is estimated to be $`L(14)\times 10^{11}L_{}`$, comparable to the luminosity of infrared luminous galaxies. Furthermore a Ly$`\alpha `$ transmission of $``$ 23–90 % is estimated from our best fit models.
Alternatively the observed 0.9-2.2 $`\mu `$m SED could also be fit without extinction by a composite “young” and “old” stellar population, where the former would be responsible for the Ly$`\alpha `$ emission and a fraction of the restframe UV flux. The SFR, stellar mass, and total luminosity are then lower than in case 1. The two scenarios may be distinguishable with IRAC/Spitzer observations at 3.6 and 4.5 $`\mu `$m.
Given the limited observed spectral range, the present data does not allow to draw any firm constraints on the maximum age of the stellar population.
In general it should also be noted that broad-band SED fits or measurements of the UV slope do not allow one to determine the metallicity of a star forming galaxy from a theoretical point of view and in terms of individual objects given important degeneracies (cf. Sect. 3.1).
The estimates of the Ly$`\alpha `$ transmissions presented here can in principle be used to constrain the intervening IGM properties, and therefore probe the reionisation of the Universe.
Although the results obtained here from this exploratory study of just two lensed galaxies, the highest known redshift galaxies with photometric detections in at least 3–4 filters, cannot provide a general view on the SF and IGM properties at $`z6`$, there is good hope that the sample of such objects will considerably increase in the near future with the availability of large ground-based telescopes and sensitive space borne observatories such as Spitzer and even more so with the planned JWST.
## Acknowledgements
We thank an anonymous referee for critical comments which helped to improve the paper. We thank Eichi Egami and Johan Richard for comments on the HST and Spitzer photometry of the arc in Abell 2218 KESR, Jean-Paul Kneib for information on Keck spectroscopy of this object, and Yuri Izotov for communicating spectra of metal-poor galaxies. Part of this work was supported by the Swiss National Science Foundation and the CNRS.
|
warning/0506/cond-mat0506745.html
|
ar5iv
|
text
|
# Phase-sensitive tests of the pairing state symmetry in Sr2RuO4
## Abstract
Exotic superconducting properties of Sr<sub>2</sub>RuO<sub>4</sub> have provided strong support for an unconventional pairing symmetry. However, the extensive efforts over the past decade have not yet unambiguously resolved the controversy about the pairing symmetry in this material. While recent phase-sensitive experiments using flux modulation in Josephson junctions consisting of Sr<sub>2</sub>RuO<sub>4</sub> and a conventional superconductor have been interpreted as conclusive evidence for a chiral spin-triplet pairing, we propose here an alternative interpretation. We show that an overlooked chiral spin-singlet pairing is also compatible with the observed phase shifts in Josephson junctions and propose further experiments which would distinguish it from its spin-triplet counterpart.
Unambiguous determination of the pairing state symmetry is one of the key steps towards understanding the pairing mechanism in a continously growing class of unconventional superconductors unconv . Phase-sensitive experiments, capable of identifying angular dependence of the superconducting order parameter, have provided a crucial evidence for a dominant $`d`$-wave orbital symmetry in cuprate superconductors Wollman1993:PRL ; Wei1998:PRL ; Klemm . However, much less is known for other unconventional superconductors such as heavy fermions, charge transfer salts, and cobaltates. In particular, there is compelling evidence for an unconventional pairing in Sr<sub>2</sub>RuO<sub>4</sub> Maeno1994:N ; Mackenzie2003:RMP , with the strong possibility of spin-triplet superconductivity which would be a solid-state analog of superfluid He<sup>3</sup> Agterberg1997:PRL .
In superconductors with inversion symmetry an order parameter (gap matrix) can be expressed as $`\widehat{\mathrm{\Delta }}(𝐤)=\mathrm{\Delta }_0(𝐤)i\widehat{\sigma }_y`$ for spin singlet and $`\widehat{\mathrm{\Delta }}(𝐤)=\widehat{\sigma }𝐝(𝐤)i\widehat{\sigma }_y`$ for spin-triplet pairing. Here $`\widehat{\sigma }`$ are the Pauli spin matrices and scalar (vector) $`\mathrm{\Delta }_0`$ $`(𝐝)`$ has even (odd) parity in the wavevector k. Often the symmetry of both the orbital and the spin part of $`\widehat{\mathrm{\Delta }}(𝐤)`$ remains to be identified and the lack of related understanding comes from the difficulty in performing phase-sensitive experiments.
While numerous previous experiments probed the pairing symmetry of Sr<sub>2</sub>RuO<sub>4</sub> Mackenzie2003:RMP , in this context, recent phase-sensitive experiments Nelson2004:S that provide angle-resolved information are particularly important. The corresponding measurements were performed in a superconducting quantum interference device (SQUID) geometry, consisting of a pair of Au<sub>0.5</sub>In<sub>0.5</sub>/Sr<sub>2</sub>RuO<sub>4</sub> Josephson junctions. Since Au<sub>0.5</sub>In<sub>0.5</sub> is a conventional $`s`$-wave superconductor, the observed modulation of critical current in an applied magnetic field was interpreted as conclusive support for the phase shifts from an odd-parity spin-triplet pairing in Sr<sub>2</sub>RuO<sub>4</sub> Nelson2004:S ; Rice2004:S .
A similar SQUID geometry was initially proposed Geshkenbein1987:PRB to study possible $`p`$-wave pairing in heavy fermions and later also used for identifying $`d`$-wave pairing in cuprates Wollman1993:PRL . The critical current $`I_c`$ is modulated in the applied magnetic field as typo
$$I_c\mathrm{cos}(\mathrm{\Phi }/\mathrm{\Phi }_0+\delta _{12}/2),$$
(1)
where $`\mathrm{\Phi }`$ is the flux threading the SQUID, $`\mathrm{\Phi }_0`$ is the flux quantum, and $`\delta _{12}`$ is the intrinsic phase shift of the order parameter between the two tunneling directions. For a conventional $`s`$-wave SQUID $`\delta _{12}=0`$ and $`I_c`$ has a maximum at $`\mathrm{\Phi }=0`$. In contrast, a phase shift $`\delta _{12}=\pi `$, characteristic of unconventional pairing pi , yields a minimum $`I_c`$ at $`\mathrm{\Phi }=0`$. The modulation of external flux together with the fabrication of junctions with varying tunneling directions in SQUID geometry therefore provides an angle-resolved phase-sensitive information about the superconducting pairing symmetry.
The suggested chiral $`p`$-wave (CpW) state with the triplet order parameter Ishida1998:N ,
$$𝐝(𝐤)(k_x+ik_y)\widehat{𝐳},$$
(2)
in which the spins of the Cooper pairs lie in the RuO<sub>2</sub> plane ($`𝐝`$), is indeed compatible with the experiment Nelson2004:S . However, we show here that it is not the only candidate. There exists another pairing state, allowed by the tetragonal symmetry of Sr<sub>2</sub>RuO<sub>4</sub>, the singlet chiral $`d`$-wave (CdW) state $`{}_{}{}^{1}E_{g}^{}(c)`$ with $`\mathrm{\Delta }_0(𝐤)(k_x+ik_y)k_z`$, or, more accurately Annet1990:AP
$$\mathrm{\Delta }_0(𝐤)(k_x+ik_y)\mathrm{sin}k_zc,$$
(3)
which is equally consistent T-note with the phase shifts observed in Ref. Nelson2004:S . We use our findings to propose an experimental test which would discriminate between CpW and CdW pairing symmetries.
Could experimental and theoretical reasons be used to rule out CdW and favor only the CpW state? The two main arguments in favor of the CpW come from muon spin resonance and Knight shift experiments Mackenzie2003:RMP ; Luke1998:N . The former indicate a time-reversal symmetry breaking below the transition temperature $`T_c`$, incompatible with the $`d_{x^2y^2}`$-wave state in cuprates, but fully compatible with either CpW or CdW symmetry. The Knight shift ($`K`$) was initially interpreted as firm evidence for a triplet state with in-plane spins (like CpW), since no change of the in-plane spin susceptibility below $`T_c`$ was found.
Even in singlet superconductors ($`e.g.`$, vanadium), $`K`$ could remain invariant below $`T_c`$. Such behavior is usually attributed to (a) spin-orbit induced spin-flip scattering which suppresses the effect of the superconductivity on $`K`$ and (b) an accidental cancellation of the spin, dipole, and orbital contributions of the Fermi-level electrons to $`K`$ which leaves only superconductivity-insensitive contributions such as the Van Vleck susceptibility. However, a quantitative analysis Pavarini2005:P shows that the spin-orbit coupling in Sr<sub>2</sub>RuO<sub>4</sub> is too weak for scenario (a) while the accidential cancellation, required for scenario (b), does not occur Knight . Thus, neither of the two explanations of a constant $`K`$ arising from singlet pairing is applicable.
This would have made the Knight shift argument for CpW very convincing, if not for the recent experiment showing the same result in a field perpendicular to the plane Murukawa2004:PRL . It was proposed Murukawa2004:PRL that the testing field of 0.02 T may be enough to induce a phase transition from the CpW in Eq. (2) to a state with d$`\widehat{𝐱}`$. However, this is highly unlikely: (i) the d$`\widehat{𝐱}`$ state would have an additional horizontal line node, as compared to the d$`(k_x+ik_y)\widehat{𝐳}`$ state and therefore lose a large part of the pairing energy ( $``$ $`\mathrm{\Delta }`$ per electron, $`\mathrm{\Delta }1.4`$ K $`\mu _B\times `$0.02 T); (ii) although in the d$`\widehat{𝐱}`$ state the spins of the pairs lie in the $`yz`$ plane, there is no $`yz`$ symmetry (as opposed to the $`xy`$ plane) and it is not $`a`$ $`priori`$ clear whether the magnetic susceptibility of the Cooper pair will be the same as for the normal electrons. Since d$`\widehat{𝐱}`$ is not allowed for a tetragonal symmetry, it may only appear as a result of a second phase transition below $`T_c`$, which has never been observed in Sr<sub>2</sub>RuO<sub>4</sub>; (iii) the spin-orbit part of the pairing interaction, which keeps the spins in the $`xy`$ plane, despite $`z`$ being the easy magnetization axis Maeno1997:JPSJ , would have to be weaker than $`\mu _B\times `$0.02 T=1.1 $`\mu `$eV=0.013 K, an energy scale much too small for the spin-orbit coupling in Sr<sub>2</sub>RuO<sub>4</sub>. So, neither the old theories for the lack of a Knight shift reduction below $`T_c`$, nor the new explanation in terms of the magnetic-field rotated order parameter withstand quantitative scrutiny; the Knight shift in Sr<sub>2</sub>RuO<sub>4</sub> remains a challenge for theorists. Until this puzzle is resolved, we cannot use the Knight shift argument for the pairing symmetry determination.
We now turn to the experiments of Ref. Nelson2004:S and compare Josephson tunneling between an $`s`$-wave superconductor and either (a) an even parity (spin singlet) superconductor or (b) an odd parity (triplet) superconductor. In the first case, the Josephson current between a conventional $`s`$-wave superconductor and an unconventional spin-singlet superconductor, represented by the order parameters $`\mathrm{\Delta }_{s\mathrm{wave}}`$ and $`\mathrm{\Delta }_0(𝐤)`$, respectively, can be expressed as Geshkenbein1986:PZETF
$$JT_𝐤\mathrm{Im}[\mathrm{\Delta }_{s\mathrm{wave}}^{}\mathrm{\Delta }_0(𝐤)]_{FS},$$
(4)
which depends on the relative phase between the superconducting order parameters. The averaging is over all states at the Fermi surface (FS) where the Fermi velocity, $`𝐯_F`$, has a positive projection on the tunneling direction represented by the unit normal n (perpendicular to the interface plane, see Fig. 1) and $`T_𝐤`$ is the tunneling probability. For a thick rectangular barrier of width $`w`$ and height $`U`$ Mazin2001:EPL we can obtain
$$T_𝐤=\frac{16m^2\kappa ^2v_Lv_R\mathrm{exp}[2\kappa wk_{}^2w/\kappa ]}{(\kappa ^2+m^2v_L^2)(\kappa ^2+m^2v_R^2)},$$
(5)
where $`\kappa =\sqrt{2m(U\mu )}`$ such that $`w\kappa 1`$ (in the thick barrier limit), $`m`$ is the free-electron mass, $`\mu `$ is the Fermi energy, and we set $`\mathrm{}=1`$. We use $`v_{L,R}`$ to denote normal components of the Fermi velocities in the two superconductors and $`k_{}`$ is the component parallel to the interface. From Eq. (5) we see that $`T_𝐤`$ is sharply peaked when $`𝐯_F𝐧`$.
In the second case, the Josephson current between a singlet and a triplet superconductor becomes Geshkenbein1986:PZETF ; Millis1988:PRB
$$J\stackrel{~}{T}_𝐤\mathrm{Im}[\mathrm{\Delta }_0^{}(𝐤)𝐝(𝐤)(𝐧\times 𝐤)]_{FS},$$
(6)
where we use $`\stackrel{~}{T}_𝐤`$ to denote that, unlike $`T_𝐤`$, it contains matrix elements corresponding to spin-flip tunneling, for example, due to magnetic interfaces or spin-orbit coupling. For nonmagnetic barriers and in the absence of spin-orbit, there is no spin-flip scattering, therefore $`\stackrel{~}{T}_𝐤=0`$ and the Josephson current vanishes identically Geshkenbein1986:PZETF ; Millis1988:PRB ; Pals1977:PRB .
From Eqs. (4) and (6) we can directly infer that for c-axis tunneling, $`𝐧𝐜`$, $`J=0`$ for both CdW and CpW states \[$`𝑑k_x𝑑k_y(k_x+ik_y)=0`$\]. For tunneling precisely in the $`ab`$ plane, the current is also zero for CdW while for CpW it only vanishes at $`𝐧𝐤`$.
We consider a model of a quasi-two-dimensional (2D) layered superconductor which has a nearly cylindrical FS with a small $`c`$-axis dispersion originating from the weak inter-layer hopping. In Fig. 1(a) we show the sample geometry used Ref. Nelson2004:S and in Fig. 1(b) represent a warping of the Fermi surface. In the first approximation for Sr<sub>2</sub>RuO<sub>4</sub> such a warping can be expressed as Bergemann2003:AP
$$\mu =\frac{k_F^2(z)}{2m}[1+\epsilon \mathrm{cos}k_zc],$$
(7)
where $`|\epsilon |1`$ is the warping parameter ($`\epsilon 7\times 10^4`$ Bergemann2003:AP ), $`c`$ is the lattice constant along the crystallographic $`z`$-direction, and $`𝐤_F(z)`$ is the $`z`$-dependent projection of the Fermi wavevector in the $`xy`$ plane \[see Fig. 1(c)\]. It is convenient to resolve the Fermi wavevector in cylindrical coordinates ($`k_F(z)`$, $`\phi `$, $`k_z`$) with
$$k_F(z)=k_{F0}/[1+\epsilon \mathrm{cos}k_zc]^{1/2},$$
(8)
where $`k_{F0}=(2m\mu )^{1/2}`$ and for $`\phi =0`$ \[see Fig. 1(c)\] $`k_F(z)k_{Fx}`$.
Schematic representation of the SQUID geometry in Fig. 1(a) (adapted from Ref. Nelson2004:S ) is an oversimplification. While efforts were made to fabricate edges either precisely parallel or perpendicular to the $`c`$-axis, in the actual samples the direction of interface planes or their corresponding normals changes gradually from $`a`$ to $`c`$-direction. In several samples Nelson2004:S an interface nearly parallel to the $`ab`$ plane, at Au<sub>0.5</sub>In<sub>0.5</sub>/Sr<sub>2</sub>RuO<sub>4</sub> junction, was covered by an insulating oxide \[see Fig. 1(a)\]. It is then plausible to expect that the normal to such interface could deviate from the $`ab`$ plane. In Fig. 1(b) we depict a generalized situation in which an interface normal, $`𝐧=(n_\rho ,n_\phi ,n_z)`$ with $`|n_z|1`$, need not lie exactly in the crystallographic $`ab`$ plane of Sr<sub>2</sub>RuO<sub>4</sub>. We show below that the analysis of phase-sensitive measurements in terms of the two small but *finite* parameters, $`\epsilon `$ and $`n_z`$, can provide a qualitatively different interpretation from those which a priori assume $`\epsilon =n_z0`$.
For a conventional superconductor with the FS larger than the one of Sr<sub>2</sub>RuO<sub>4</sub>, the Josephson tunneling across a thick rectangular barrier can be obtained from Eqs. (4) and (5) as
$`J`$ $`{\displaystyle _{𝐯_F𝐧>0}}𝑑𝐤\delta (ϵ_𝐤\mu )𝐯_F𝐧\mathrm{exp}(k_{}^2w/2\kappa )`$
$`\times \mathrm{Im}(k_x+ik_y)\mathrm{sin}k_zc,`$ (9)
where $`k_{}^2=𝐤_F^2(𝐤_F𝐧)^2`$, $`\kappa ^2mv_{L,R}^2`$, and the projection of the Fermi velocity in Sr<sub>2</sub>RuO<sub>4</sub> along $`𝐧`$ is
$$𝐯_F𝐧=\frac{k_{F0}[1+\epsilon \mathrm{cos}k_zc]^{1/2}}{m}n_\rho \frac{k_{F0}^2\epsilon c\mathrm{sin}k_zc}{2m[1+\epsilon \mathrm{cos}k_zc]}n_z.$$
(10)
For a thick barrier, the integration can be simplified by noting that the dominant contribution comes from $`k_{}=0`$. The right hand side of Eq. (9), in the leading order in $`\epsilon `$ and $`n_z`$, can be then reduced to $`\sqrt{\pi \kappa /w}ck_{F0}^2n_z(1\epsilon )`$, such that
$$J_{}=An_z(1\epsilon ),$$
(11)
where $`A`$ characterizes the normal state barrier transparency. Thus, with a tilted interface ($`n_z0`$) there is a finite Josephson current even in the absence of any Fermi surface warping ($`\epsilon =0`$).
To verify that our findings of finite Josephson current in CdW state are not limited to the specific assumption of a thick rectangular barrier, we also consider the rather different case of a strong $`\delta `$-function barrier. The corresponding transmission probability is Zutic1999:PRB ; Mazin2001:EPL
$$T_𝐤=\frac{4v_Lv_R}{(v_L+v_R)^2+4U^2},$$
(12)
where $`v_{L,R}`$ are the normal components of the Fermi velocities in the two superconductors and $`U`$ ($`v_{L,R}^2`$) is the scattering strength. From Eqs. (4), (12) we obtain
$$J_{𝐯_F𝐧>0}𝑑𝐤\delta (ϵ_𝐤\mu )𝐯_F𝐧\mathrm{Im}(k_x+ik_y)\mathrm{sin}k_zc,$$
(13)
where, unlike in the case of a thick barrier, we perform $`\phi `$ and $`k_z`$ integration. In the leading order, the right hand side of Eq. (13) is $`\pi k_{F0}^2\epsilon n_z,`$ and yields
$$J_\delta =A\epsilon n_z,$$
(14)
where again $`A`$ characterizes the normal state transparency. In contrast to the thick-barrier result, the current now vanishes in the absence of FS warping. From Eqs. (11), (14) one can conjecture that for a general case $`JA(s\epsilon )`$, where $`0s1`$.
The presence of small parameters $`\epsilon `$ and $`n_z`$ in Eqs. (11), (14) shows that the Josephson current in the CdW state would be reduced as compared to the conventional SQUID with $`s`$-wave electrodes. However, the alternative picture, based on the CpW state, also contains small parameters which should be kept in mind when interpreting the experiment of Ref. Nelson2004:S . In addition to the small relative strength of the spin-orbit coupling (quantified by the admixture of $`S_{}`$ into a nonrelativistic $`S_{}`$ state, or the spin-orbit induced band shift relative to the band width small ), there can also be another small factor — a ratio of the lattice constant and the superconducting coherence length Fenton1985:SSC , approximately $`6\times 10^3`$ Mackenzie2003:RMP .
Results from Eq. (11), (14) confirm that the CdW state could be compatible with the phase shifts observed in Ref. Nelson2004:S . Furthermore, the azimuthal dependence of an order parameter coincides for both CdW and CpW states. While the proposed symmetry arguments Nelson2004:S ; Rice2004:S exclude most of superconducting states allowed in the tetragonal symmetry Asano2005:P , these arguments alone are not sufficient to unambiguously identify the odd-pairing of CpW state. Instead, to confirm that a CpW state has indeed been observed, one would need to accurately calculate the expected magnitudes of the Josephson current for both chiral states. In particular we propose a modification of the experimental configuration Nelson2004:S such that the interface plane would be slanted at $`45^o`$ with the $`c`$-axis. If the corresponding ratio of the Josephson current to the normal state conductivity becomes substantially larger ($`n_z`$ is no longer small) than in Ref. Nelson2004:S , it would be strong support for the chiral singlet state in Eq. (3).
Another important distinction between the two chiral states is the presence of nodes in the superconducting gap. In contrast to the CpW state, CdW requires by symmetry a horizontal line node \[see Eqs. (2), (3)\]. The idea of a horizontal line node Mackenzie2003:RMP has been entertained by experimentalists Tanatar2001:PRL and theorists Zhitomirsky2001:PRL for a while, although recently it has fallen out of favor. Still, some researchers insist on the existence of a horizontal line node Contreras2004:PRB . Moreover, in the Josephson experiments of Ref. Nelson2004:S evidence was found for a substantial $`k_z`$ dependence, albeit not necessarily for horizontal nodes, of the order parameter in Sr<sub>2</sub>RuO<sub>4</sub> Liu2005:PC . How could such a material with nearly 2D electronic structure develop a highly 3D superconducting state? To answer this question we point out the following facts: (a) practically no ferromagnetic spin fluctuations, favorable for a $`p`$-wave pairing, have been experimentally found in Sr<sub>2</sub>RuO<sub>4</sub>; (b) antiferromagnetic spin fluctuations at q$`=(2/3,2/3,q_z)`$ have negligible $`z`$ dispersion; (c) the crystal structure of Sr<sub>2</sub>RuO<sub>4</sub>, as opposed to its electronic structure, is fairly 3D, so one can expect the electron-phonon coupling to be quite 3D as well (d) there is a sizeable O isotope effect in Sr<sub>2</sub>RuO<sub>4</sub>, which strongly changes with introduction of pair-breaking symmetries Mao2001:PRB . While electron-phonon coupling $`per`$ $`se`$ can only induce an $`s`$-wave pairing, such a pairing would be prevented by the strong antiferromagnetic spin fluctuations. However, for the proposed CdW state, any 2D interaction cancels out, including the magnetic interaction. Should the electron-phonon coupling have a maximum say, at $`(1/2,1/2,1/2)`$, as opposed to $`(1/2,1/2,0)`$, the CdW state would have been immediately stabilized providing a plausible scenario for spin-singlet superconductivity in Sr<sub>2</sub>RuO<sub>4</sub>.
In conclusion, we have revealed that a completely overlooked chiral $`d`$-wave pairing state in Sr<sub>2</sub>RuO<sub>4</sub> is equally compatible with the existing body of experimental data as the generally accepted chiral $`p`$-wave state. We have proposed phase-sensitive experiments in a SQUID geometry with a variable tilting angle capable of unambiguously distinguishing between the two chiral states.
We thank R. A. Klemm, Y. Liu, and D. J. Van Harlingen for useful discussions. This work was supported by the US ONR. I. Ž. acknowledges financial support from the National Research Council.
|
warning/0506/gr-qc0506118.html
|
ar5iv
|
text
|
# The Bianchi identity and weak gravitational lensing
## I Introduction
The Bianchi identity in general relativity is usually expressed as
$$_{[a}R_{bc]d}^{}{}_{}{}^{e}=0$$
(1)
and is considered to be a consequence of the geometrical properties of the covariant derivative operator. Most students of relativity come across the Bianchi identity in a contracted form that is quite useful in discussions of various properties of the Einstein tensor.
The decomposition of the Riemann tensor into the Ricci ($`R_{ab}`$) and Weyl ($`C_{abcd}`$) tensors,
$$R_{abcd}=C_{abcd}+(g_{a[c}R_{d]b}g_{b[c}R_{d]a})\frac{1}{3}Rg_{a[c}g_{d]b},$$
(2)
presents a powerful use of the Bianchi identity. As Hawking & Ellis (1973) point out, the Ricci tensor is determined by the Einstein field equation, while the Weyl tensor is the part of the curvature not determined locally by the matter distribution. However, the application of the Bianchi identity to Eq. 2 yields a constraint on the Weyl tensor once the Ricci tensor is found from the Einstein field equations. By treating the Bianchi identity as a field equation for the Weyl tensor given a known Ricci tensor, Newman & Penrose (1962) (hereafter NP) developed a consistent methodology for the study of gravitational radiation.
In this paper, we consider the Bianchi identity in the reverse manner. Specifically, we assume that the Weyl tensor is known and treat the Bianchi identity as a partial differential equation for the Ricci tensor.
Our motivation for this approach is the study of weak gravitational lensing, where measurements are made of the distortion of images due the presence of gravitational fields. From the observed distortion of images, a matter distribution can be inferred, so weak gravitational lensing is now a very active area of observational research with many papers published each year. Pivotal early work relating the observed distortion and inferred mass distribution was done by Miralda-Escude (1991) and Kaiser & Squires (1993) amongst others. Substantial reviews of the subject are provided by Mellier (1999) and Schneider & Bartelmann (2001). For this paper, we are most interested in the presentation given in Miralda-Escude (1996).
In this paper, we present four results. First, we show that the observed distortion and inferred matter distributions of weak lensing are equal to projected components of the Ricci and Weyl tensors as expressed in the formalism introduced by Newman & Penrose (1962). Next, we demonstrate that the integral equation used in observational weak lensing to determine the projected mass distribution from observed image distortions is actually an integral relation between the $`\mathrm{\Psi }_0`$ and $`\mathrm{\Phi }_{00}`$ components of the Weyl and Ricci tensors.
Our most important result is that we prove that a particular component of the Bianchi identity yields a partial differential equation for the $`\mathrm{\Phi }_{00}`$ component of the Ricci tensor given a known $`\mathrm{\Psi }_0`$ Weyl tensor. Using a Green’s function, we derive the standard equation used in observational weak lensing as formulated in Miralda-Escude (1996) from the Bianchi identity. Thus, we have developed a PDE approach to weak lensing that may provide a new calculational approach to the field that allows for more accurate determinations of mass mappings.
Finally, we consider the important case of axially symmetric lenses and show that in this case the Bianchi identity is easily integrated. Two explicit examples are shown: a point lens and a singular isothermal sphere.
Our results on weak lensing extend a new way of thinking about gravitational lensing based on a space-time perspective. Strong lensing and general image distortion have been studied by several researchers including Frittelli & Newman (1999) and two connected papers by Frittelli et al. (2001, 2001) with a substantial review provided by Perlick (2004). This paper is the first paper that seeks a first principles equation for observational weak lensing.
## II Computation of Ricci and Weyl tensors
In this section, we compute the complex Ricci and Weyl tensor components and the spin coefficients introduced in Newman & Penrose (1962). Our goal is to determine the spin coefficient form of the Ricci and Weyl tensor components to first order in a static metric perturbation off flat space.
While modern evidence strongly indicates that the universe is accelerating and dominated by dark energy, it is reasonable to chose to perturb off flat space for several reasons. First, we wish to directly compare our derivation of the equations of weak lensing with the standard derivations using the thin-lens approximation. Because all the “lensing activity” in the thin-lens approximation happens when light rays passes the lens, the choice of cosmology does not directly enter the equations. In fact, both in the literature and in actual application, the only place the cosmological model enters weak lensing calculations is in the conversion between redshift and angular diameter distance. Hence, one of the strengths of lensing studies is that the basic equations of the theory are relatively independent of cosmological model.
Furthermore, since the basic cosmological models are conformally flat, our basic results can be translated to them. Beginning with a more “realistic” metric does not add to the calculation and introduces a number of conceptual issues best handled in a different paper.
The physical situation that we wish to study is the appearance and distortion of a small patch of the sky containing extended galaxies that has been weakly lensed by some matter distribution. Both the astronomer in question and the sky of galaxies are assumed to be situated in reasonably isolated regions of space-time where the space-time is flat.
The observed galaxies are connected to our astronomer by light rays along her past light cone. In fact, we may think of pencils of light rays that travel from each extended galaxy to the telescope. The path of these light rays is to be determined by integrating the null geodesic equations of the space-time metric, so that we may find a tangent vector, $`\mathrm{}^a`$, to the light rays.
The matter distribution that lenses the light rays is encoded into a space-time metric. For our purposes, we may assume a weakly perturbed Minkowski metric,
$$ds^2=(1+2\phi )dt^2(12\phi )(dx^2+dy^2+dz^2),$$
(3)
where $`\phi `$ is a static gravitational potential satisfying
$$^2\phi =4\pi \rho (x,y,z).$$
(4)
As the astronomer and observed galaxies are to be isolated, we assume that the matter distribution, $`\rho (x,y,z)`$, is contained in some region of space far from the astronomer and galaxies. In practice, this is always the case.
Following the program of NP, we chose a tetrad of null vectors,
$$\lambda _i^a=(\lambda _1^a,\lambda _2^a,\lambda _3^a,\lambda _4^a)=(\mathrm{}^a,n^a,m^a,\overline{m}^a),$$
(5)
associated with the pencil of light rays connecting our astronomer and each individual observed galaxy, where $`\mathrm{}^a`$ is the real vector tangent to the pencil. By convention, the other vectors are chosen such that
$$\mathrm{}^an_a=1\&m^a\overline{m}_a=1,$$
(6)
with all other products zero. $`m^a`$ and $`\overline{m}^a`$ are spatial, complex null vectors parallel propagated along the pencil.
In principle, the space-time metric can be written in terms of a general null tetrad as $`g^{ab}=\eta ^{ij}\lambda _i^a\lambda _j^b`$, although this is not central to our presentation. Further, we will make physical restrictions on the tetrad vectors tied to our goal of finding the Weyl and Ricci tensor components to first order in the perturbation.
We orient our spatial coordinates, $`(x,y,z)`$, such that the astronomer is located at $`x=y=0`$ and $`z=z_a`$ with the “center” of the matter distribution at the spatial origin. With this placement, our astronomer’s telescope points straight down the $`\widehat{z}`$ axis.
In principle, each null tetrad associated with an individual pencil of light connecting the astronomer to a lensed galaxy varies along the pencil. However, in our calculations, we are justified in considering each tetrad to be a constant null tetrad given by
$$\mathrm{}^a=\frac{1}{\sqrt{2}}(1,0,0,1),n^a=\frac{1}{\sqrt{2}}(1,0,0,1)\&m^a=\frac{1}{\sqrt{2}}(0,1,i,0),$$
(7)
for two reasons. First, the wide field telescopes used today see a very small portion of the sky. For example, the proposed $`8.4`$ m Dark Matter Telescope would have a field of view of only $`260`$ square milli degrees Tyson et al. (2000). Hence, with our orientation, all the pencils of light are parallel to the $`\widehat{z}`$ axis. Second, even though the individual pencils will be deflected by the lens, the deflection of the light ray will be proportional to the 2-dimensional $`(x,y)`$ gradient of $`\phi `$. Since we will be contracting the tetrad vectors with the Ricci and Weyl tensors, to work to first order in $`\phi `$, we must neglect any variation in the null tetrad.
The basis of the NP formalism is twelve complex spin coefficients that are analogous to the twenty-four real Ricci rotation coefficients defined by
$$\gamma _{jk}^i=\lambda _j^a\lambda _k^b_b\lambda _a^i.$$
(8)
Because our choice of tetrad involves vectors whose components are all constant, all the NP spin coefficients are zero for our physical situation (to zeroth order in the metric perturbation).
In the NP formalism, tetrad components of the Ricci and Weyl tensors are computed by contracting the coordinate components with the tetrad vectors. This yields five complex Weyl tensor and ten complex Ricci tensor components (see appendix A).
Using the null tetrad above, the non-zero NP formalism Ricci curvature components are
$$\mathrm{\Phi }_{00}=\mathrm{\Phi }_{22}=2\mathrm{\Phi }_{11}=\frac{1}{2}^2\phi .$$
(9)
The first order Weyl tensor components are
$`\mathrm{\Psi }_0=\mathrm{\Psi }_4`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left(\phi _{xx}\phi _{yy}+2i\phi _{xy}\right),`$
$`\mathrm{\Psi }_1=\overline{\mathrm{\Psi }}_3`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left(\phi _{xz}+i\phi _{yz}\right),`$
$`\mathrm{\Psi }_2`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left(\phi _{zz}{\displaystyle \frac{1}{3}}^2\phi \right).`$ (10)
We note that up to numerical factors, $`\mathrm{\Phi }_{00}`$ is equal to the matter density $`\rho (x,y,z)`$ that determines the gravitational perturbation $`\phi `$.
The four directional derivatives associated with the null tetrad play an important role in the NP formalism and are given special names:
$$D=\mathrm{}^a_a,\mathrm{\Delta }=n^a_a,\delta =m^a_a,\&\overline{\delta }=\overline{m}^a_a.$$
(11)
Using our choice of tetrad, and the fact that the metric perturbation is static, these derivative operators will act as
$`D={\displaystyle \frac{1}{\sqrt{2}}}{\displaystyle \frac{}{z}},\mathrm{\Delta }={\displaystyle \frac{1}{\sqrt{2}}}{\displaystyle \frac{}{z}},`$
$`\delta ={\displaystyle \frac{1}{\sqrt{2}}}\left({\displaystyle \frac{}{x}}+i{\displaystyle \frac{}{y}}\right),\overline{\delta }={\displaystyle \frac{1}{\sqrt{2}}}\left({\displaystyle \frac{}{x}}i{\displaystyle \frac{}{y}}\right).`$ (12)
## III Observational weak lensing in practice
In this section, we briefly outline the usual presentation of the equations of weak lensing by a thin lens so that we can draw parallels to our presentation that uses the Weyl and Ricci tensor. The derivation presented here relies heavily on that of Miralda-Escude (1996).
The practicing astrophysicist assumes a thin (or two dimensional) lens that divides a background space-time into an observer and a source side. The lens lies in a two (spatial) dimensional “lens plane” that is perpendicular to the line of sight of the telescope. The observed background galaxy is said to lie in a two dimensional “source plane,” also perpendicular to the line of sight.
Using dimensionless cartesian coordinates $`\stackrel{}{r}_s`$ in the source plane and $`\stackrel{}{r}_l=(x,y)`$ for the lens plane, the starting point for thin-lens lensing is a mapping from the lens plane to the source plane given by
$$\stackrel{}{r}_s=\stackrel{}{r}_l\stackrel{}{}_2\psi (\stackrel{}{r}_l).$$
(13)
The two-dimensional gravitational potential $`\psi `$ is determined from an ordinary three-dimensional potential $`\phi `$ by projection,
$$\psi (\stackrel{}{r}_l)=_{z_a}^{z_s}𝑑z\phi (x,y,z),$$
(14)
where $`z_a`$ and $`z_s`$ are the $`z`$ coordinates of the observing astronomer and distant source, respectively. The two dimensional potential $`\psi `$ will satisfy
$$\frac{1}{2}\mathrm{\Delta }^2\psi =\frac{1}{2}\left(_x^2+_y^2\right)\psi =\frac{\mathrm{\Sigma }(\stackrel{}{r}_l)}{\mathrm{\Sigma }_{crit}},$$
(15)
where $`\mathrm{\Sigma }(\stackrel{}{r}_l)`$ is a projected mass density and $`\mathrm{\Sigma }_{crit}`$ is a critical density for strong lensing, or the appearance of multiple images.
To consider weak lensing, one forms a Jacobian matrix
$$𝒥=\frac{\stackrel{}{r}_s}{\stackrel{}{r}_l}=\left(\begin{array}{cc}1\psi _{xx}& \psi _{xy}\\ \psi _{xy}& 1\psi _{yy}\end{array}\right)=\left(\begin{array}{cc}1\kappa \lambda & \mu \\ \mu & 1\kappa +\lambda \end{array}\right),$$
(16)
where
$$\kappa =\frac{1}{2}\left(\psi _{xx}+\psi _{yy}\right)=\frac{1}{2}\mathrm{\Delta }^2\psi =\frac{\mathrm{\Sigma }}{\mathrm{\Sigma }_{crit}}$$
(17)
is referred to as the “convergence,” and the two quantities
$$\lambda =\frac{1}{2}\left(\psi _{xx}\psi _{yy}\right)\&\mu =\psi _{xy}$$
(18)
are called the “shears” in the thin-lens literature. Frittelli et al. (2001) shows how these quantities are related to the convergence ($`\rho `$) and shear ($`\sigma `$) in general relativity. For this paper, it is critical to note the similarities that $`\kappa `$ has with $`\mathrm{\Phi }_{00}`$ and that $`\mathrm{\Psi }_0`$ has with $`\lambda `$ and $`\mu `$.
If the outer surface of a circular, extended source is parameterized by $`\stackrel{}{t}_s=R(\mathrm{cos}\delta ,\mathrm{sin}\delta )`$, then under the thin lens mapping the inverse of the Jacobian matrix in Eq. 16 maps $`\stackrel{}{t}_s\stackrel{}{t}_l`$, a new elliptical curve in the lens plane. The orientation and ellipticity of the resulting ellipse is determined by $`\lambda `$ and $`\mu `$ – see Schneider et al. (1992) for the details.
Since the orientation and ellipticity is observable, the goal in observational weak lensing is to measure the shears $`(\lambda ,\mu )`$ and infer the projected mass density, $`\kappa `$. To do this, one inverts Eq. 17 using the two dimensional Green’s function and applies the differentiations in Eq. 18 to the result. This yields
$`\lambda `$ $`=`$ $`{\displaystyle 𝑑\stackrel{}{r}^{}\frac{\kappa (\stackrel{}{r}^{})}{\pi }\frac{(\mathrm{cos}2\eta )}{|\stackrel{}{r}\stackrel{}{r}^{}|^2}},`$
$`\mu `$ $`=`$ $`{\displaystyle 𝑑\stackrel{}{r}^{}\frac{\kappa (\stackrel{}{r}^{})}{\pi }\frac{(\mathrm{sin}2\eta )}{|\stackrel{}{r}\stackrel{}{r}^{}|^2}},`$ (19)
where $`\eta `$ is the angle in the lens plane between $`\stackrel{}{r}\stackrel{}{r}^{}`$ and the $`\widehat{x}`$ axis.
By employing a Fourier transform technique, Miralda-Escude (1996) shows that one can invert Eq. 19 to obtain
$$\kappa (\stackrel{}{r})=𝑑\stackrel{}{r}^{}\frac{[\lambda (\stackrel{}{r}^{}),\mu (\stackrel{}{r}^{})]}{\pi }\frac{[\mathrm{cos}2\eta ,\mathrm{sin}2\eta ]}{|\stackrel{}{r}\stackrel{}{r}^{}|^2}.$$
(20)
Equation 20 is the primary equation for observational weak lensing. It is an integral relation that allows one to determine the projected mass density at every point if one has measured the “shears.”
## IV Weak lensing with curvature tensors
We may exploit the similarities between the definitions of the thin-lens convergence and shears and $`\mathrm{\Phi }_{00}`$ and $`\mathrm{\Psi }_0`$ to recast Eq. 20 in different language. The first step is to project our gravitational potential and curvature tensors into a “thin-lens” form. Equation 14 projects our metric perturbation $`\phi (x,y,z)`$ to a two dimensional $`\psi (x,y)`$. Then we define
$${}_{L}{}^{}\mathrm{\Phi }_{00}^{}_{z_a}^{z_s}𝑑z\mathrm{\Phi }_{00}=\frac{1}{2}\left(\psi _{xx}+\psi _{yy}\right),$$
(21)
where we use Eq. 9 with Eq. 14 and the property that $`\phi _z`$ is zero far from the lens. Likewise, we can define
$${}_{L}{}^{}\mathrm{\Psi }_{0}^{}_{z_a}^{z_s}𝑑z\mathrm{\Psi }_0=\frac{1}{2}\left(\psi _{xx}\psi _{yy}+2i\psi _{xy}\right).$$
(22)
We note that $`{}_{L}{}^{}\mathrm{\Phi }_{00}^{}=\kappa `$ and $`{}_{L}{}^{}\mathrm{\Psi }_{0}^{}=\lambda +i\mu `$, so that the fundamental observable in weak gravitational lensing is a projected component of the Weyl tensor, while the fundamental inferred quantity of observational interest (the projected mass density) is a projected component of the Ricci tensor.
Then exactly as in section III, one can use a two-dimensional Green’s function to invert Eq. 21 for $`\psi `$ and plug this relation into Eq. 22 to obtain an integral equation specifying $`{}_{L}{}^{}\mathrm{\Psi }_{0}^{}`$ given a known $`{}_{L}{}^{}\mathrm{\Phi }_{00}^{}`$. Taking the Fourier transform yields a relation between the transforms of $`{}_{L}{}^{}\mathrm{\Psi }_{0}^{}`$ and $`{}_{L}{}^{}\mathrm{\Phi }_{00}^{}`$ which is easily rearranged such that the inverse Fourier transform produces the desired integral equation for $`{}_{L}{}^{}\mathrm{\Phi }_{00}^{}`$,
$${}_{L}{}^{}\mathrm{\Phi }_{00}^{}(\stackrel{}{r})=𝑑\stackrel{}{r}^{}\frac{{}_{L}{}^{}\mathrm{\Psi }_{0}^{}(\stackrel{}{r}^{})}{\pi }\frac{e^{2i\eta }}{|\stackrel{}{r}\stackrel{}{r}^{}|^2},$$
(23)
where $`\eta `$ again is the angle between $`\stackrel{}{r}\stackrel{}{r}^{}`$ and the $`\widehat{x}`$ axis in the lens plane.
Equation 23 demonstrates that the fundamental equation of weak lensing is an integral relation between components of the Weyl and Ricci curvature tensors. Specifically, a projected version of $`\mathrm{\Psi }_0`$ is known through observation and used to determine a projected mass density $`{}_{L}{}^{}\mathrm{\Phi }_{00}^{}`$.
## V The Bianchi identity
As discussed in section I, applying the Bianchi identity, Eq. 1, to the decomposition of the Riemann tensor into the Ricci and Weyl curvature tensors produces a differential constraint on the Weyl tensor. To obtain the Bianchi identity in the NP spin coefficient formalism, one contracts the Bianchi identity with all possible combinations of the tetrad vectors and writes the resulting equations out using the definitions of the spin coefficients and Ricci and Weyl tensor components. In the non-vacuum case, this results in twelve equations, which are listed in a number of references including Newman & Tod (1980).
The first full Bianchi identity equation is
$`\overline{\delta }\mathrm{\Psi }_0D\mathrm{\Psi }_1+D\mathrm{\Phi }_{01}\delta \mathrm{\Phi }_{00}`$ $`=`$ $`(4\alpha \pi )\mathrm{\Psi }_02(2\rho +ϵ)\mathrm{\Psi }_1+3\kappa \mathrm{\Psi }_2+(\overline{\pi }2\overline{\alpha }2\beta )\mathrm{\Phi }_{00}`$ (24)
$`+2(ϵ+\overline{\rho })\mathrm{\Phi }_{01}+2\sigma \mathrm{\Phi }_{10}2\kappa \mathrm{\Phi }_{11}\overline{\kappa }\mathrm{\Phi }_{02}.`$
Due to our choice of tetrad, all the spin coefficients are zero, so the entire right hand side of Eq. 24 vanishes. Also, $`\mathrm{\Phi }_{01}`$ is zero, so to first order in the gravitational perturbation, we have
$$\overline{\delta }\mathrm{\Psi }_0D\mathrm{\Psi }_1\delta \mathrm{\Phi }_{00}=0.$$
(25)
Equation 25 is a partial differential equation that holds at every point along the pencil of rays connecting our astronomer with the distant observed galaxy. To compare this relation with Eq. 23, we project into a lens plane by integrating out the $`z`$ coordinate. Integrating $`D\mathrm{\Psi }_1`$ in $`z`$ from $`z_a`$ to $`z_s`$ effectively yields zero since $`D_z`$ and $`\mathrm{\Psi }_1`$ evaluated far from the lens is assumed to be zero. Passing the $`z`$ integration through the $`\delta `$ derivative operators, we have
$$\delta _L\mathrm{\Phi }_{00}=\overline{\delta }_L\mathrm{\Psi }_0.$$
(26)
We consider Eq. 26 as a field equation for $`{}_{L}{}^{}\mathrm{\Phi }_{00}^{}`$ given a known $`{}_{L}{}^{}\mathrm{\Psi }_{0}^{}`$. Equation 26 relates the fundamental quantities of weak gravitational lensing and is derived from first principles in general relativity.
## VI Green’s function
In this section, we show that the relation between the Weyl and Ricci tensor derived from the Bianchi identity, Eq. 26, is the differential version of the integral relation used in standard weak lensing studies, Eq. 23. To do this, we employ a set of Green’s functions first developed by Porter (1981) and expanded in Ivancovich et al. (1989), which we refer to as the Porter Green’s functions.
The Porter Green’s functions are designed to work with powers of the $`ð`$ and $`\overline{ð}`$ differential operators. The $`ð`$ derivative operator acts on functions on the sphere of different spin weight in different ways and serves as a spin-weight raising and lowering operator Newman & Tod (1980). Functions on the sphere of a spin weight $`s`$ are expandable in a series of spin-weighted spherical harmonics $`{}_{s}{}^{}Y_{lm}^{}`$. Because different powers of $`ð`$ and $`\overline{ð}`$ annihilate the $`{}_{s}{}^{}Y_{lm}^{}`$ with $`|s|=l`$, a set of Green’s functions can be developed.
Our main result, Eq. 26, is a partial differential equation that holds in an $`(x,y)`$ plane that we consider to be the lens plane. To make contact with the Porter Green’s functions, we introduce the pair of complex coordinates $`(\zeta ,\overline{\zeta })`$ defined by
$$\zeta =\frac{1}{\sqrt{2}}(xiy).$$
(27)
In these coordinates, $`m^a`$ points in the $`\zeta `$ direction and $`\delta =_\zeta `$.
By stereographic projection of the sphere into a equatorial plane from the pole, the $`(\zeta ,\overline{\zeta })`$ coordinates are complex coordinates for the sphere. With the exception of the point at infinity, where the projected curvature tensors $`{}_{L}{}^{}\mathrm{\Psi }_{0}^{}`$ and $`{}_{L}{}^{}\mathrm{\Phi }_{00}^{}`$ are assumed to be zero, introduction of the complex stereographic coordinates turns Eq. 26 into a partial differential equation on the sphere.
Since $`{}_{L}{}^{}\mathrm{\Phi }_{00}^{}`$ is a spin-weight zero function, the application of $`ð`$ to it will take the form
$$ð_L\mathrm{\Phi }_{00}=(1+\zeta \overline{\zeta })\delta _L\mathrm{\Phi }_{00}.$$
(28)
For this reason, we multiply both sides of Eq. 26 by $`(1+\zeta \overline{\zeta })`$ and our Bianchi identity takes the form
$$ð_L\mathrm{\Phi }_{00}=(1+\zeta \overline{\zeta })\overline{\delta }_L\mathrm{\Psi }_0=A_1(\zeta ,\overline{\zeta }),$$
(29)
where $`A_1(\zeta ,\overline{\zeta })`$ is a function of spin-weight 1.
The Porter Green’s function for an equation of the form in Eq. 29 is
$${}_{L}{}^{}\mathrm{\Phi }_{00}^{}(\zeta ,\overline{\zeta })=_{S^2}K_{0,1}(\zeta ,\overline{\zeta };\eta ,\overline{\eta })(1+\eta \overline{\eta })\overline{\delta }_\eta {}_{L}{}^{}\mathrm{\Psi }_{0}^{}𝑑\mu _\eta ,$$
(30)
where the integral is taken over the sphere parameterized by complex coordinates $`(\eta ,\overline{\eta })`$ with the area element
$$d\mu _\eta =\frac{2}{i}\frac{d\eta d\overline{\eta }}{(1+\eta \overline{\eta })^2}.$$
The kernel of the Green’s function for an equation of the type in Eq. 29 is
$$K_{0,1}(\zeta ,\overline{\zeta };\eta ,\overline{\eta })=\frac{1}{4\pi }\frac{1+\eta \overline{\eta }}{\overline{\zeta }\overline{\eta }}.$$
(31)
Putting all this together, we have
$${}_{L}{}^{}\mathrm{\Phi }_{00}^{}(\zeta )=\frac{1}{2i\pi }_{S^2}\frac{\overline{\delta }_\eta {}_{L}{}^{}\mathrm{\Psi }_{0}^{}(\eta )}{\overline{\zeta }\overline{\eta }}𝑑\eta d\overline{\eta }.$$
(32)
It is convenient to multiply by $`1`$ in the form $`1=(\zeta \eta )/(\zeta \eta )`$ to obtain
$${}_{L}{}^{}\mathrm{\Phi }_{00}^{}(\zeta )=\frac{1}{2i\pi }_{S^2}\frac{\zeta \eta }{(\overline{\zeta }\overline{\eta })(\zeta \eta )}\overline{\delta }_\eta {}_{L}{}^{}\mathrm{\Psi }_{0}^{}(\eta )𝑑\eta d\overline{\eta }.$$
(33)
Because we want to show that Eq. 33 is equivalent to Eq. 23, we reintroduce cartesian coordinates
$$\zeta =\frac{1}{\sqrt{2}}(xiy)\eta =\frac{1}{\sqrt{2}}(x^{}iy^{}).$$
(34)
In these coordinates, Eq. 33 becomes
$${}_{L}{}^{}\mathrm{\Phi }_{00}^{}(x,y)=\frac{1}{2\pi }\frac{[(xx^{})i(yy^{})]}{(xx^{})^2+(yy^{})^2}\left(_x^{}i_y^{}\right){}_{L}{}^{}\mathrm{\Psi }_{0}^{}𝑑x^{}𝑑y^{}.$$
(35)
Comparing Eq. 35 with Eq. 23, we see that we need to flip the $`\overline{\delta }_\eta `$ derivative operator that is inside the integral by integrating by parts. Using the definition of the angle $`\eta `$ as the angle the vector $`\stackrel{}{r}\stackrel{}{r}^{}`$ makes with the $`+\widehat{x}`$ axis and integrating Eq. 35 by parts yields
$${}_{L}{}^{}\mathrm{\Phi }_{00}^{}(x,y)=𝑑\stackrel{}{r}^{}\frac{{}_{L}{}^{}\mathrm{\Psi }_{0}^{}}{\pi }\frac{e^{2i\eta }}{|\stackrel{}{r}\stackrel{}{r}^{}|^2}$$
(36)
plus the surface term
$$\frac{1}{2\pi }𝑑x^{}𝑑y^{}\left(_x^{}i_y^{}\right)\left\{\frac{[(xx^{})i(yy^{})]}{|\stackrel{}{r}\stackrel{}{r}^{}|^2}{}_{L}{}^{}\mathrm{\Psi }_{0}^{}\right\}.$$
(37)
One can see that the surface term will vanish by breaking the integral into two parts for the $`_x^{}`$ and $`_y^{}`$ derivatives. Each part vanishes as the integrand is evaluated at either $`x^{}=\mathrm{}`$ or $`y^{}=\mathrm{}`$.
Therefore, we have shown that the application of the Porter Green’s function to the Bianchi identity, Eq. 26, results in the integral relation used in weak gravitational lensing studies.
## VII Axial Symmetry
In this section, we consider those mass distributions which are axially symmetric around the line of sight. In practice, this is a very important simplifying assumption used in gravitational lensing studies. Under this assumption, Eq. 26 is easily integrated directly, without using the Porter Green’s function.
First, we change coordinates in the lens plane from the cartesian $`(x,y)`$ coordinates to standard axial coordinates $`(r,\varphi )`$. In the axial coordinate system, the $`\delta `$ directional derivative operator in Eq. 12 is
$$\delta =\frac{e^{i\varphi }}{\sqrt{2}}\left(_r+\frac{i}{r}_\varphi \right).$$
(38)
As an ansatz for an axially symmetric lens, we assume that the Ricci tensor is a function of $`r`$ only,
$${}_{L}{}^{}\mathrm{\Phi }_{00}^{}={}_{L}{}^{}\mathrm{\Phi }_{00}^{}(r),$$
(39)
and that the Weyl tensor can be written as
$${}_{L}{}^{}\mathrm{\Psi }_{0}^{}=\mathrm{{\rm Y}}(r)e^{2i\varphi }.$$
(40)
This second assumption is motivated by the concept of spin weight in the NP formalism Newman & Tod (1980).
Applying Eq. 38 to our ansatz for the solution to Eq. 26, the $`\varphi `$ dependence drops out, and one is left with
$$_r{}_{L}{}^{}\mathrm{\Phi }_{00}^{}(r)=_r\mathrm{{\rm Y}}(r)+\frac{2}{r}\mathrm{{\rm Y}}(r),$$
(41)
which is now a one dimensional equation.
Recall that we have shown that $`{}_{L}{}^{}\mathrm{\Psi }_{0}^{}`$ or $`\mathrm{{\rm Y}}(r)`$ is a known, measurable quantity and that $`{}_{L}{}^{}\mathrm{\Phi }_{00}^{}`$ is essentially the projected matter density. This means that after integrating over $`r`$, Eq. 41 gives us the functional form of the projected matter distribution:
$${}_{L}{}^{}\mathrm{\Phi }_{00}^{}(r)=\mathrm{{\rm Y}}(r)+\frac{2}{r}\mathrm{{\rm Y}}𝑑r+C.$$
(42)
The constant of integration represents the mass sheet degeneracy present in gravitational lensing and can be taken as zero.
## VIII Two axially symmetric examples
In this section, we examine two axially symmetric mass distributions in the context of weak lensing as expressed by Eq. 42. The models we consider include a singular, Schwarzschild-like point lens and a singular isothermal sphere (SIS) model. For each model, we will determine the gravitational potential, $`\psi (r)`$, then compute the projected Weyl and Ricci tensor components and show that they obey Eq. 42.
### VIII.1 Point lens
As our first example, we consider a Schwarzschild point lens of mass $`M`$ at the origin. For this mass distribution, the gravitational potential $`\psi `$ is given by
$$\psi =2M\mathrm{ln}(r)=2M\mathrm{ln}\left(\sqrt{x^2+y^2}\right).$$
(43)
One can show by direct computation that
$${}_{L}{}^{}\mathrm{\Psi }_{0}^{}=\frac{2M}{r^2}e^{2i\varphi },$$
(44)
which matches Eq. 40 yielding
$$\mathrm{{\rm Y}}=\frac{2M}{r^2}.$$
(45)
Inserting this functional form for $`\mathrm{{\rm Y}}`$ into the right hand side of Eq. 42 gives
$$\frac{2M}{r^2}+𝑑r\frac{4M}{r^3}=0.$$
(46)
Equation 46 indicates, through the Bianchi identity Eq. 42, that the projected Ricci tensor is zero at all radii $`r>0`$, which is the expected result for a Schwarzschild, point-like lens.
### VIII.2 SIS model
The SIS model is a one parameter model given by
$$\mathrm{\Sigma }(r)=\frac{\sigma _v^2}{2r},$$
(47)
where $`\sigma _v`$ is the velocity dispersion. Although unphysical due to the singularity at the origin and infinite total mass, SIS models do explain the flat rotation curves of galaxies and are widely used as models in gravitational lensing studies.
For our purposes, the simplest way to integrate the projected gravitational potential for the SIS model is to follow the argument of Schneider et al. (1992) in section 8.1, where it is shown that one can write the projected gravitational potential as
$$\psi (r)=8\pi _0^rr^{}𝑑r^{}\mathrm{\Sigma }(r^{})\mathrm{ln}\left(\frac{r}{r^{}}\right).$$
(48)
It is then relatively simple to show that
$$\psi =4\pi \sigma _v^2r,$$
(49)
so that the projected Weyl tensor is
$${}_{L}{}^{}\mathrm{\Psi }_{0}^{}=\frac{2\pi \sigma _v^2}{r}e^{2i\varphi }.$$
(50)
From Eq. 40, we have $`\mathrm{{\rm Y}}=2\pi \sigma _v^2/r`$ and from the Bianchi identity
$$\mathrm{{\rm Y}}+𝑑r\frac{2\mathrm{{\rm Y}}}{r}=\frac{2\pi \sigma _v^2}{r}.$$
(51)
By direct computation,
$${}_{L}{}^{}\mathrm{\Phi }_{00}^{}=\frac{1}{2}\left(\psi _{xx}+\psi _{yy}\right)=\frac{2\pi \sigma _v^2}{r},$$
(52)
so that the Bianchi identity, Eq. 42 is preserved.
## IX Discussion
The main results of this paper are that the fundamental quantities of weak gravitational lensing are projected components of the Ricci and Weyl tensors and that the Bianchi identity provides a field equation for the projected matter density derived from first principles. We explicitly show that the common integral equation used in weak gravitational is an integral version of the Bianchi identity. To our knowledge, these results are known in the lensing or relativity communities.
From this perspective, this paper extends the work presented in a series of recent papers that has helped reunite applied gravitational lensing with its roots in general relativity. The underlying perspective of these papers has been that any observed lensing phenomena must be encoded into the past light cone of the observer.
A primary purpose of these papers has been to identify the equation from general relativity that is approximated in the thin lens treatment used by practicing astrophysicists. Frittelli & Newman (1999) first pointed out that the lens should be coded into a space-time metric, and then solving the null geodesic equations would simultaneously give the “time of flight” equation and lens mapping. In a two-paper set, Frittelli et al. (2001) related general relativity’s optical scalars to the “shears” and “convergences” cited in the thin-lens literature. These papers also wrote out integral relations that showed how image distortion grew continuously along the pencil of rays connected a source and observer. A generalization of the Fermat principle of least time was present in Frittelli et al. (2002).
Because Eq. 23 is kinematically derived, it does not indicate a true connection between observational weak lensing and relativistic first principles. The derivation of Eq. 23 relies only on the definitions of $`\mathrm{\Phi }_{00}`$ and $`\mathrm{\Psi }_0`$.
However, we show in this paper that the projected Bianchi identity, Eq. 26 is the differential version of Eq. 23. Thus, we have a first principles derivation of the basic equation of weak gravitational lensing for the first time and a potentially useful new PDE approach to the subject.
Most directly, Eq. 26 provides a new PDE approach for determining the projected mass density in weak lensing studies. While numerical approaches to solving PDEs with real data tend to less stable than integral approaches, they also tend to provide higher accuracy, so it is not unreasonable to pursue approaches to data analysis based directly on these results.
Another potential application of this work might be in the cumulative analysis of thick lenses or repeated lensing by multiple clusters along the line of sight. In this case, one would not apply a thin-lens assumption, but would use Eq. 25 as a field equation along the line of sight for a full metric that encodes a more realistic cosmological model.
###### Acknowledgements.
The authors wish to acknowledge Simonetta Frittelli, Ted Newman and Al Janis for helpful suggestions and comments. BK would like to thank the Adrian Tinsley Program for Undergraduate Research of BSC for making his involvement possible.
## Appendix A Weak field Ricci and Weyl tensors
In this appendix, we give the usual component form of the Ricci and Weyl tensors considered in the text, along with their definitions in the Newman & Penrose (1962) formalism. In Eqs. 53-55, the usual Einstein summation conventions are not employed.
For the weak field metric in Eq. 3, to first order in the perturbation, $`\phi `$, the non-zero Christofel symbols are
$`\mathrm{\Gamma }_{}^{0}{}_{0i}{}^{}=\phi _i`$ $`\mathrm{\Gamma }_{}^{i}{}_{ik}{}^{}=\phi _k`$
$`\mathrm{\Gamma }_{}^{i}{}_{00}{}^{}=\phi _i`$ $`\mathrm{\Gamma }_{}^{i}{}_{kk}{}^{}=\phi _i(ik)`$ (53)
where $`\phi _i=_i\phi `$, $`0`$ denotes time and $`i,j,k`$ denote spatial components. The non-zero, first order Ricci tensor components are
$$R_{00}=^2\phi R_{ii}=^2\phi ,$$
(54)
with $`^2=_x^2+_y^2+_z^2`$, while the non-zero, first order Weyl tensor components are
$`C_{0i0i}={\displaystyle \frac{1}{3}}\left(3\phi _{ii}+^2\phi \right)`$ $`C_{0i0j}=\phi _{ij}ij`$
$`C_{ijij}={\displaystyle \frac{1}{3}}\left(3\phi _{kk}^2\phi \right)ijk`$ $`C_{ijik}=\phi _{jk}ijk.`$ (55)
In the NP formalism, the components of the Ricci and Weyl tensors are contracted with the null tetrad to give named components. The Weyl tensor components are defined by
$$\mathrm{\Psi }_0=C_{abcd}\mathrm{}^am^b\mathrm{}^cm^d,\mathrm{\Psi }_1=C_{abcd}\mathrm{}^an^b\mathrm{}^cm^d,$$
$$\mathrm{\Psi }_2=\frac{1}{2}\left(C_{abcd}\mathrm{}^an^b\mathrm{}^cn^dC_{abcd}\mathrm{}^an^bm^c\overline{m}^d\right),$$
$$\mathrm{\Psi }_3=C_{abcd}\mathrm{}^an^bn^c\overline{m}^d,\mathrm{\Psi }_4=C_{abcd}n^a\overline{m}^bn^c\overline{m}^d,$$
(56)
and the Ricci tensor components are
$`\mathrm{\Phi }_{00}={\displaystyle \frac{1}{2}}R_{ab}\mathrm{}^a\mathrm{}^b,`$ $`\mathrm{\Phi }_{10}={\displaystyle \frac{1}{2}}R_{ab}\mathrm{}^a\overline{m}^b,`$ $`\mathrm{\Phi }_{20}={\displaystyle \frac{1}{2}}R_{ab}\overline{m}^a\overline{m}^b,`$
$`\mathrm{\Phi }_{01}={\displaystyle \frac{1}{2}}R_{ab}\mathrm{}^am^b,`$ $`\mathrm{\Phi }_{11}={\displaystyle \frac{1}{4}}\left(R_{ab}\mathrm{}^an^b+R_{ab}m^a\overline{m}^b\right),`$ $`\mathrm{\Phi }_{21}={\displaystyle \frac{1}{2}}R_{ab}n^a\overline{m}^b,`$
$`\mathrm{\Phi }_{02}={\displaystyle \frac{1}{2}}R_{ab}m^am^b,`$ $`\mathrm{\Phi }_{12}={\displaystyle \frac{1}{2}}R_{ab}n^am^b,`$ $`\mathrm{\Phi }_{22}={\displaystyle \frac{1}{2}}R_{ab}n^an^b,`$ (57)
$`\mathrm{\Lambda }={\displaystyle \frac{1}{24}}R.`$
|
warning/0506/nucl-th0506027.html
|
ar5iv
|
text
|
# The Origin of the Difference between Multiplicities in 𝑒⁺𝑒⁻ Annihilation and Heavy Ion Collisions.
## Abstract
Multiplicities in $`e^+e^{}`$ annihilation and relativistic heavy ion collisions show remarkable similarities at high energies. A thermal-statistical model is proposed to explain the differences which occur mainly at low beam energies. Two different calculations are performed, one using an approximate thermodynamic relationship, the other using a full thermal model code. The results are in qualitative agreement, suggesting that the interplay of baryon density and temperature tends to systematically suppress the total multiplicity at lower beam energies.
Recently, the PHOBOS experiment at RHIC has shown results on the total charged particle multiplicity produced in heavy ion collisions over a wide range of collision energies as well as the collision centrality, characterized by the number of participating nucleons . Three surprising connections emerge from comparison of this data with particle multiplicities measured in elementary collisions ($`pp`$, $`\overline{p}p`$, and $`e^+e^{}`$):
1. The total number of primary charged particles produced in A+A reactions ($`N_{ch}^{A+A}`$) scales linearly with the number of participants ($`N_{part}`$).
2. The multiplicity per participant pair in A+A ($`N_{ch}^{A+A}/N_{part}/2`$) agrees with that measured in $`e^+e^{}`$ ($`N_{ch}^{e^+e^{}}`$) within 10% over a large range in $`\sqrt{s}`$ and $`\sqrt{s_{NN}}`$. In general, the $`e^+e^{}`$ data has not fully removed the contribution from weak decays, but this is generally less than a 10% correction.
3. This agreement is not simply in the total multiplicity, but extends over the full rapidity range (relative to the thrust axis in the $`e^+e^{}`$ case).
This agreement in total multiplicity suggests a certain universality in particle production, a result consistent with the thermal-statistical approach started by Fermi, Landau and Hagedorn . However, the same data set shows a systematic deviation between the multiplicities below $`\sqrt{s_{NN}}=30`$ GeV, and increasing as the energy gets lower. It is the purpose of this letter to explore a simple physics hypothesis which can explain this deviation in a semi-quantitative fashion. To compare multiplicities between $`e^+e^{}`$ and heavy ion collisions we introduce the quantity
$$\mathrm{\Delta }N_{ch}N_{e^+e^{}}\frac{N_{AA}}{N_{part}/2}$$
(1)
where, in an obvious notation, $`N_{part}`$ denotes the number of participants in A+A collisions, while $`N_{e^+e^{}}`$ is the total multiplicity in an $`e^+e^{}`$ collision, and $`N_{AA}`$ is the total multiplicity in an A+A collision. It is proposed that this difference is directly related to the thermodynamic variables determined from the total multiplicities measured in A+A collisions, namely the freeze-out values of the temperature, $`T`$, and baryon chemical potential $`\mu _B`$. For the multiplicity of charged particles this relation reduces to
$$\mathrm{\Delta }N_{ch}=\frac{\mu _B}{3T}$$
(2)
Various improvements on this relation will be discussed below.
Examination of the yields of different particle species over a wide range in $`\sqrt{s_{NN}}`$ shows a large variation of the baryon to pion ratio ($`p/\pi `$). Conversely, $`p+p`$ collisions show a somewhat more rapid dependence of this ratio with beam energy. This can be understood heuristically by saying that relative to proton-proton collisions, collisions involving nuclei (p+A, A+A) are distinguished by the larger “stopping power” of the nuclear targets. While this phenomenon is not well-understood theoretically, it has been characterized phenomenologically in several different ways. In proton-nucleus collisions, the concept of “rapidity loss” is usually used to measure how much energy the proton projectile loses in the multiple collisions in the nuclear target . While the data are not trivial to interpret, given the state of understanding of longitudinal dynamics, one generally finds a substantial net-baryon density near mid-rapidity in A+A collisions, higher than p+p and $`e^+e^{}`$ reactions.
In Au+Au collisions, it is often postulated that the large particle multiplicities create a thermally and chemically equilibrated system. This suggests using statistical models to characterize the relative population of hadronic states . In these models, the main parameters are the freeze-out temperature ($`T`$), which is associated with the energy density of the system, and the baryochemical potential ($`\mu _B`$), which is directly related to the density of the net baryon number distributed in the freeze-out volume ($`N(p)N(\overline{p})e^{\mu _B/T}e^{\mu _B/T}`$). It is less clear why similar fits should work for elementary collisions (e.g. $`pp`$ and $`e^+e^{}`$). However, the work of Becattini has shown that statistical models prove to be an equally useful tool in describing the relative yields of hadrons in collisions with relatively small multiplicities , although additional care must be taken to guarantee appropriate conservation of quantum numbers (e.g. strangeness and baryon number).
Thermal fits made by a number of authors show that increasing the $`\sqrt{s_{NN}}`$ in A+A collisions leads to an increase in $`T`$ and a correlated decrease in $`\mu _B`$, shown in Fig. 1. This has been interpreted by Cleymans and Redlich by postulating a fixed relationship of the freezeout parameters, such that $`E/N1`$ GeV . Whatever the physical scenario implied by this condition, it provides a useful way to determine these parameters as a function of beam energy, and to interpolate between available data points. However, it turns out that this criterion (called “Thermal I”) does not perfectly describe the existing data. A somewhat better description, although purely phenomenological, can be made by a sixth-order polynomial fit in $`\mu _B`$ to the same data in the ($`T`$,$`\mu _B`$) plane (“Thermal II”) :
$$T(\mu _B)=0.164460.11196\mu _B^20.139139\mu _B^4+0.0684637\mu _B^6$$
In this work, we will show both parametrizations where possible.
Also in this work, we use a parametrization of $`\mu _B`$ as a function of $`\sqrt{s}`$ made by the authors in Ref.
$$\mu _B(\sqrt{s})=\frac{1.2735}{(1+0.2576\sqrt{s})}$$
(3)
To apply this information to the heavy ion and $`e^+e^{}`$ data, we will invoke a simple thermodynamic condition. When dealing with blackbody radiation, one typically sets the Gibbs potential $`G=ETS+pV=_i\mu _iN_i\mu _BN_B`$, since the other chemical potentials (e.g. strangeness, charge, isospin) are usually smaller than the baryochemical potential. In this formula, $`E`$ is the internal energy, $`T`$ is the temperature, $`S`$ the entropy, $`p`$ the pressure, $`\mu _B`$ the baryochemical potential and $`N_B`$ the baryon number which must be conserved in the interaction. This expression can be rearranged to show how the entropy is related to the other variables:
$$S=\frac{(E+pV)\mu _BN_B}{T}=S_0S_B$$
(4)
where
$$S_0=\frac{E+pV}{T}$$
(5)
is the entropy due to the internal energy and the pressure of the system, while
$$S_B=\frac{\mu _BN_B}{T}$$
(6)
is interpreted as the entropy bound up in the conserved baryons, suppressing the total entropy.
The $`(E+pV)/T`$ term can be understood as the one that controls particle production in the absence of conserved baryon charges (i.e. $`\mu _B=0`$). It is assumed that this is universal for all strongly interacting collision systems with the same expansion features, most importantly the dominance of 1D expansion in the early stages. The second term is thus a correction which will only be important when $`\mu _BN_B/T`$ is non-negligible, i.e. at large $`\mu _B`$ or small $`T`$ or both.
This correction to the total entropy can be estimated in a crude way as follows:
* A factor of $`\alpha =4`$ to normalize entropy to the number of particles (as is relevant for a massless Boltzmann gas).
* A factor of $`N_{part}/2`$ to give the total change in multiplicity per participant baryon pair. This cancels the $`N_B`$ in the numerator since it is precisely the number of participants which determines the conserved baryon number.
* A factor of $`\beta =3/2`$ which accounts for unmeasured neutral pions. This is based on the assumption that we are calculating the entropy of the lighter pions that would have been produced except for the non-zero $`\mu _B`$ enforcing the presence of heavy baryons.
In other words, this scenario postulates $`S/N_{ch}=6`$. Dividing by all these factors gives:
$`\mathrm{\Delta }N_{ch}`$ $`=`$ $`{\displaystyle \frac{2}{\alpha \beta N_{part}}}{\displaystyle \frac{\mu _BN_B}{T}}`$
$`=`$ $`{\displaystyle \frac{2}{\alpha \beta }}\times {\displaystyle \frac{N_B}{N_{part}}}\times {\displaystyle \frac{\mu _B}{T}}`$
$`=`$ $`{\displaystyle \frac{\mu _B}{3T}}`$
This is the multiplicity that must be added to the low-energy results at a given $`\sqrt{s}`$ to account for the entropy that would have been available except for the need to conserve baryon charge.
It turns out that direct calculations with thermal models , give the result that the entropy divided by the multiplicity of final-state charged particles (after strong decays) is $`S/N_{ch}=7.2`$. This number should be compared to $`\alpha \beta =6`$ from the considerations above. For the subsequent calculations, the more theoretically relevant number will be used instead of the simpler estimate. The difference between them should be seen as contributing to an overall theoretical uncertainty.
When this is done, we get the results shown in Fig. 2, where the multiplicities have been divided by the Landau-Fermi expression ($`N_{ch}=2.2s^{1/4}`$. The $`e^+e^{}`$ results are shown as open squares, the original A+A results are shown as open circles, and the “$`\mu _B`$”-corrected A+A results are shown as closed circles. It is surprising that this simple model works as well as it does, since it is nothing more than correcting for the fact that the initial baryons must be present in the final state, and thus take up energy that would have normally gone to normal thermal particle production (mainly pions).
However, it can be argued that the $`4\pi `$ multiplicity collisions has an extra component that would not be found in $`e^+e^{}`$ collisions, namely the participant baryons themselves. To correct for these, we make two additional transformations on the A+A data.
* $`m_P`$”: Subtract $`2m_P`$ from $`\sqrt{s}`$ to correct for the difference in the mass of the beam particle
* $`n_B`$”: Subtract 1, assuming that the net baryons will be either $`p`$ or $`n`$ equally.
These corrections are shown to have relatively little effect on the basic result, as seen by the thick dotted line in Fig. 3. Thus, it is not possible to distinguish their relevance by comparison with the $`e^+e^{}`$ data. It should also be noted that just applying the $`m_P`$ and $`n_B`$ corrections, without the $`\mu _B`$ correction, has little effect on the initial result as well, as shown in Fig. 3.
A more quantitative check of the physical picture discussed here can be made by calculating the entropy density vs. $`\mu _B`$ in a full statistical-thermal model calculation . The freezeout contour is the same as in the calculation above. To relate this curve to experimental data, one considers the translation of the experimental variable $`N_{ch}^{A+A}/(N_{ch}^{e^+e^{}}N_{part}/2)`$ to parameters accessible in thermal models. From the fits performed on $`p+p`$ and $`e^+e^{}`$ data in Ref. , we find that the $`T`$ parameter is essentially constant over a wide range of energy $`T=T_0170`$ MeV. In this case, $`s(T,\mu _B)=s(T_0,0)`$ and thus should be constant with $`\sqrt{s}`$. It also appears to be the asymptotic value reached in ultra-high energy $`A+A`$ collisions, if current trends are to be believed. In this case, we assume that
$$N_{ch}^{e^+e^{}}=C^{e^+e^{}}V^{e^+e^{}}s_0$$
(8)
where $`V^{e^+e^{}}`$ is the freezeout volume and $`C_{e^+e^{}}`$ is a constant relating the number of detected particles to the total entropy (again, approximately 4). In this picture, the energy dependence of the total multiplicity is determined dominantly by the freezeout volume.
In A+A collisions, the multiplicity has been found to scale linearly with the number of participating nucleons $`N_{part}`$. Since the volume of the initial nuclei scales with $`A`$, and the freezeout entropy scales with $`V`$, this suggests that $`VN_{part}/2=V^{A+A}N_{part}/2`$. We then assume that
$$N_{ch}^{A+A}=\frac{N_{part}}{2}C^{A+A}V^{A+A}s(T,\mu _B)$$
(9)
where $`V_{A+A}`$ is the effective volume per participant pair.
With these assumptions, the ratio shown in Figure 3 is
$$\frac{2}{N_{part}}\frac{N_{ch}^{A+A}}{N_{ch}^{e^+e^{}}}=\frac{C^{A+A}}{C^{e^+e^{}}}\frac{V^{A+A}}{V^{e^+e^{}}}\frac{s(T,\mu _B)}{s_0}.$$
(10)
The right-hand side of this equation has two sets of constants (C and V) and one ratio that depends on beam energy. The constants C control the proportionality between the total entropy and the total charged particle multiplicity. Landau and Belenkij argued that this constant does not depend on the system size, so it makes sense to set this ratio to unity. This presumption is supported by the overall similarity in the particle production (although strangeness is clearly suppressed in the smaller systems). They also did not think there would be a change in the relation of the multiplicity to the entropy as a function of beam energy or initial baryon density. The ratio of the volumes might not be expected to be the same. However, the agreement (to the 10% level) of the total multiplicity per participant pair in A+A and the total multiplicity in $`e^+e^{}`$ suggests that these volumes are similar, even as a function of energy.
In Fig. 2 we compare the ratio $`s(T,\mu _B)/s_0`$, shown for the two parametrizations of $`T(\mu _B)`$ (Thermal I and Thermal II), with the ratios discussed previously. The agreement between the thermal model curve and the A+A data is in reasonable qualitative agreement. This suggests that the previously-made physics assumption
$$\frac{C^{A+A}}{C^{e^+e^{}}}\frac{V^{A+A}}{V^{e^+e^{}}}=1$$
(11)
is a reasonable one.
It should be noted that the concept of “pion suppression” was previously discussed by Gazdzicki et al by reference to data from a similar set of experiments as discussed in . However, the phenomenon discussed in this work is concerned with the suppression of the total entropy rather than just the pions, and is a feature which arises naturally in the context of thermal models. The models here also include all available meson and baryon resonances, as opposed to just delta resonances.
To make the relative energy dependence of mesons and baryons clearer, we show their respective contributions to the entropy density as a function of beam energy in Fig. 4, again for two parametrizations. The individual and total entropy densities are both divided by $`T^3`$ to remove the expected temperature dependence. They are then multiplied by a factor if $`\pi ^2/4`$, which transforms the quantity $`s/T^3`$ into the effective number of degrees of freedom $`n(T)`$ of a massless Boltzmann gas. The main result is that the baryon contribution completely dominates at low energies, but the mesons are equal at $`\sqrt{s_{NN}}10`$ GeV and their contribution exceeds that of the baryons by a factor of $`2`$ and saturates. However, as was also noted in Ref. , it is observed that the quantity $`s/T^3`$ is constant over a large range in center of mass energies (down to $`\sqrt{s_{NN}}=5`$ GeV) and diverges only at very low energies (presumably due to the associated rapid decrease in $`T`$). This result shows that not only are the number of degrees of freedom similar at freezeout for $`A+A`$ and $`e^+e^{}`$, but they are similar for heavy ion collisions over a large range of collision energies. From this result, that freezeout occurs for $`s/T^3=const.`$, one might also explain the suppression of the entropy density $`s=const.\times T^3`$, as due to the lower temperatures associated with larger $`\mu _B`$. In either case, the suppression of the total multiplicity results from the non-trivial interplay between $`\mu _B`$ and $`T`$.
In conclusion, the difference between the charged particle multiplicity per participant pair in A+A and the multiplicity in $`e^+e^{}`$ can be explained by the suppression of entropy due to the presence of a conserved quantum number, manifest as the net-baryon density. A semi-quantitative understanding of the existing data has been achieved both by simple thermodynamic arguments as well as more detailed thermal model comparisons.
This work was partially supported by the US DOE grant DE-AC02-98CH10886. P.S. acknowledges fruitful discussions with Mark Baker, Francesco Becattini, Nigel George, Dan Magestro and Gunther Roland.
|
warning/0506/astro-ph0506633.html
|
ar5iv
|
text
|
# A Catalogue of Field Horizontal Branch Stars Aligned with High Velocity Clouds
## 1. Introduction
Discovered by Muller et al. (1963) over 40 years ago, High-Velocity Clouds (HVCs) are H I clouds with velocities that do not conform to simple models of Galactic rotation (e.g. Wakker & van Woerden, 1997, and references therein). Since their discovery, HVC distances (and origins) have been the source of much debate, with various scenarios suggested. Shapiro & Field (1976) proposed the existence of a “Galactic Fountain” to explain the observed soft X-ray background and O VI absorption lines. This process, in which hot gas is driven up into the halo by supernovae and stellar winds, which then condenses and falls back onto the disk, was linked by Bregman (1980) to the HVCs (see also de Avillez, 2000). Others have suggested they may be the result of tidal disruption of satellite galaxies, e.g. the Magellanic Stream (Putman et al., 2003). Perhaps they are the infalling source of star formation fuel long required by chemical evolution models (e.g. Fenner & Gibson, 2003)? Oort (1966) first suggested that some subset of HVCs may be at much larger distances, being the remnants of the galaxy formation process (a suggestion revived by Blitz et al., 1999). Other authors have proposed that the smaller, isolated Compact HVCs (CHVCs) may be explained under this scenario (e.g. de Heij et al., 2002a), suggesting that such objects may also be observable around M31 with sensitive H I observations. Subsequent observations toward M31 have successfully detected HVC-analogs (Thilker et al., 2004), whilst others have failed to find them in searches of Local Group analogues (Pisano et al., 2004).
It is clear that to establish a definitive theory describing HVCs the largely unknown distances are a key element. Many physical properties scale with the distance; physical size ($`d`$), mass ($`d^2`$), density ($`d^1`$) and pressure ($`d^1`$) for example. Since HVCs do not conform to models of Galactic rotation, we cannot use those models to derive distances from their observed velocities. Hence another method is required. A distance limit may be set by high-resolution absorption spectroscopy (Danly et al., 1993; Schwarz et al., 1995; van Woerden et al., 1999). If a star is located behind the HVC, the stellar spectrum should contain the imprint of the HVC at the (known) velocity of the cloud. A sufficiently high resolution spectrum should be able to resolve such features. Detection of these absorption lines indicates the star is behind the cloud, and an upper distance limit is set, namely the distance to the star.
The interpretation of non-detections is more challenging and is succinctly summarised by Wakker & van Woerden (1997) (see their Fig. 4). Several possibilities must be taken into account before one can be convinced that the lack of an optical HVC absorption feature indicates that the star is in front of the HVC, and sets a significant lower distance limit. HVCs have been shown to have large column density variations over small (arcmin) angular scales (e.g. Wakker & Schwarz, 1991). Since HVCs are usually identified in H I surveys with large beam sizes, high-resolution (interferometric) radio imaging is necessary, to ensure the line-of-sight does not pass through a “sucker hole” in the cloud, missing the expected high column density gas. If the alignment is favorable, we also require sufficiently high metallicity intervening gas (a typical value for HVCs is $`0.10.3`$ solar; see Wakker (2001)); clearly if this is not the case, there will be no metal ions to create an absorption line. Further, knowledge of the gas metallicity comes from other lines-of-sight, so we must assume that there are no abundance variations within the cloud.
Optical absorption line studies are often conducted using the Ca II H and K lines ($`\lambda \lambda 3968.469`$ Å and 3933.663 Å respectively). These two resonance lines have high oscillator strengths and thus should be easily detectable. Prata & Wallerstein (1967) were the first to specifically target the HVCs (unsuccessfully) using absorption line studies, with Kepner (1968) producing the first catalogue of early-type stars aligned with HVCs. Despite almost 40 years of efforts by various groups, only a handful of stellar absorption line studies have successfully detected HVC absorption features<sup>1</sup><sup>1</sup>1There is also a large body of ultraviolet absorption line detections against stellar and extra-galactic sources. These ionized high-velocity clouds in some cases align with H I HVCs, in some cases not. See e.g. Sembach et al. (2003) for details., providing firm upper limits. van Woerden et al. (1999) detected HVC “Chain A” in Ca II absorption in the optical spectrum of the RR Lyrae star AD UMa, giving a maximum distance of 10 kpc. A few years earlier, Danly et al. (1993) used the IUE satellite to detect the HVC Complex M in the UV resonance lines O I, C II and Si II toward the early-type star $`\text{BD}+38\mathrm{°}2182`$, limiting the HVC to $`z<4.4\text{kpc}`$.
Field Horizontal Branch (FHB) stars are valuable probes for distance determination work, for two primary reasons: (i) Their intrinsic luminosity means that sufficiently high signal-to-noise optical spectra can be obtained for distant halo FHB stars ($`d20`$ kpc), meaning they can act as useful and interesting discriminators of the various origin scenarios, and (ii) Being hot stars, their spectra are relatively free of intrinsic absorption features, making the detection and interpretation of the HVC absorption features relatively straightforward. Previous authors have also successfully employed RR Lyrae stars (van Woerden et al., 1999) or bright B-stars in the ultraviolet (Danly et al., 1993) and optical (Smoker et al., 2004).
Here we present a table of 430 FHB stars which align with H I gas at high velocities, and the associated H I spectra and moment maps. These stars are ideal for high-resolution optical spectroscopy and distance determinations. Section 2 gives the details about the selection of HVC distance probes from the HES FHB catalogue of Christlieb et al. (2005, hereafter C05). The catalogue of HVC distance probes is presented in Section 3. A brief conclusion is given in Section 4. The full catalogue plus H I spectra and summed intensity moment maps towards each star in the catalogue are presented in the online edition.
## 2. Stellar selection and Matching
We have taken as our input sample, the C05 catalogue of 8321 FHB stars selected from the Hamburg/ESO Survey, which has been shown to have contamination from high-gravity main-sequence stars of $`<16\%`$. For each stellar line-of-sight, an H I spectrum was extracted from the H I Parkes<sup>2</sup><sup>2</sup>2The Parkes telescope is part of the Australia Telescope which is funded by the Commonwealth of Australia for operation as a National Facility managed by CSIRO. All-Sky Survey (HIPASS; Barnes et al., 2001; Putman et al., 2002, hereafter P02) and any emission above a threshold of $`0.15\text{Jy/Beam}`$ at high velocities ($`|v_{LSR}|90\text{km\hspace{0.17em}s}\text{-1}`$) identified. This threshold corresponds to a brightness temperature threshold of $`0.12\text{K}`$ for the Parkes Multibeam used for the HIPASS survey (e.g. Putman et al., 2003) and is set by the level at which Ca II absorption lines are expected to be detectable in optical spectra (see Sec 2.6). All lines-of-sight meeting these criteria were then screened by eye to reject obvious problems. In all cases, these were due to the wings of emission features well below the $`|v_{LSR}|90\text{km\hspace{0.17em}s}\text{-1}`$ limit extending outside this range above the level of $`0.12\text{K}`$. H I moment maps were created from the HIPASS data cubes around each line of sight, and the positions were matched against several HVC catalogues.
### 2.1. The Hamburg/ESO Survey
We briefly review the salient points of the HES, while a fuller description of the survey and its stellar products may be found elsewhere (e.g. Wisotzki et al., 1996, C05). Originally a survey for bright quasars, the HES is an objective prism survey which covers a large fraction of the southern extra-galactic sky ($`\delta <+2.5\mathrm{°};|b|30\mathrm{°}`$) to a limiting magnitude of $`B18.0`$. As well as finding quasars, it has also become a valuable resource for identifying many different classes of interesting stars; e.g. Carbon stars (Christlieb et al., 2001a), White Dwarfs (Christlieb et al., 2001b), metal-poor halo stars (e.g. Christlieb, 2003) and FHB stars (C05). Selection of the FHB stars was made first on the basis of a color cut. The second step involved automated selection criteria based on the summed Balmer line equivalent widths and the Strömgren $`c_1`$ index. This process resulted in 8321 stars, which have been shown to have a contamination from, e.g. higher surface gravity A-stars, of $`<16\%`$. See C05 for further details.
### 2.2. Spectrum Extraction
To identify stars aligned with HVCs, HIPASS H I spectra for all stellar lines-of-sight were first extracted from the HIPASS HVC cubes. These cubes were produced from the HIPASS data by P02 using the MINMED5 bandpass calibration, which recovers much of the extended emission lost in the original HIPASS bandpass calibration; since the survey was originally designed for point sources, this loss was not a critical issue. Nevertheless, when emission extends spatially to fill more than $`6\mathrm{°}`$ of a full HIPASS $`8\mathrm{°}`$ declination scan, the calibration will not be accurate e.g. for data close to the Galactic plane (see Putman et al., 2003, for details).
The miriad task mbspect was used to extract the spectra, averaging spatially over a 3x3 pixel box (approximately the HIPASS beam size of 15.5 arcmin; HIPASS pixels are 4 arcmin). mbspect uses the nearest pixel to the specified coordinate. All spectra containing emission above the level of $`0.12\text{K}`$ at velocities $`|v_{LSR}|90\text{km\hspace{0.17em}s}\text{-1}`$ were automatically flagged. An example of such a spectrum is shown<sup>3</sup><sup>3</sup>3Similar figures are provided for all probes in Table 3 in Figs. 5 – 148 of the online version. in Fig 1. We plot only the velocity range $`500+500\text{km\hspace{0.17em}s}\text{-1}`$ since this is the range occupied by nearly all Galactic HVCs<sup>4</sup><sup>4</sup>4Note that Thilker et al. (2004) have found HVCs associated with M31 at velocities more negative than the $`500\text{km\hspace{0.17em}s}\text{-1}`$ limit.. The associated moment map is shown in Fig 2 (see Section 2.3 for details).
To create the final FHB/HVC catalogue, the spectra for these 430 lines-of-sight were then visually inspected to reject obvious noise artefacts. In some cases, the peak flux was at an absolute velocity less than the enforced $`90\text{km\hspace{0.17em}s}\text{-1}`$ limit, while significant emission was still present outside this range. In these cases, stars were retained if the channel(s) outside the $`90\text{km\hspace{0.17em}s}\text{-1}`$ boundary had a flux comparable to the peak value. Fig 3 shows an example of such a case. In these cases, the peak emission with $`|v_{LSR}|90\text{km\hspace{0.17em}s}\text{-1}`$ was recorded.
During screening, it was noted that some spectra contained multiple high-velocity emission components. In cases where these were not immediately identifiable as the same cloud, a separate entry in Table 3 was recorded. These multiple components were flagged in the catalogue and separate moment maps produced. Obviously, separate H I spectra are not required, although the spectra are repeated for completeness. In all but a few cases, these doubles are flagged as associated with the Magellanic Stream (see Section 2.4). HE 1008$``$0438 also aligns by chance with the galaxy HIPASS J1010-04. The star HE 0053$``$3740 was found to align with a Sculptor Group galaxy and was excluded. HE 0056$``$4043 shows a noise spike at $`410\text{km\hspace{0.17em}s}\text{-1}`$ which is evident in the HIPASS data cubes.
### 2.3. Moment Maps
In order to visualize the spatial locations of the stars with respect to the gas, summed-intensity moment maps of a region approximately $`3\mathrm{°}\times 3\mathrm{°}`$ centered around each probe were created. Moment maps were generated by summing the velocity channel containing the peak flux outside the $`|v_{LSR}|90\text{km\hspace{0.17em}s}\text{-1}`$ boundary and the two adjacent channels. This summed width equates to 39.6 km s<sup>-1</sup>, which is approximately the 36 km s<sup>-1</sup> median FWHM of the HIPASS HVCs (P02) (note, though, that this median is almost certainly an overestimate of the true value and a product of the coarse velocity resolution of the HIPASS data). Maps were converted to column densities using the relation $`N(HI)=1.823\times 10^{18}T_b𝑑v`$ (Dickey & Lockman, 1990). Onto these moment maps, contours were drawn at H I column densities of $`2.0\times 10^{18}`$, $`5.0\times 10^{18}`$, $`1.0\times 10^{19}`$, $`2.5\times 10^{19}`$, $`5.0\times 10^{19}`$ and $`1.0\times 10^{20}`$. The positions of the FHB stars are marked by a “+” in the centers of these images, one example of which is shown<sup>5</sup><sup>5</sup>5Similar figures are provided for all probes in Table 3 in Figs. 5 – 148 of the online version. in Figure 2.
### 2.4. Cloud Matching and the Magellanic Stream
One of the most prominent features in the southern radio sky at 21 cm is the Magellanic Stream (MS). We wanted to differentiate the entries that probe gas possibly associated with the Stream, from those probing other HVCs. In this vein we follow the spatial and kinematic regions defined in Figs. 5 and 7 respectively of Putman et al. (2003). Table 1 shows the regions we adopt as being associated with the Stream. Probes within these regions are flagged in the catalogue. Note that exact definitions of which clouds clouds are associated with the Stream are difficult and somewhat subjective – the definitions here should be considered as a guide only. Further, there is some overlap between the Stream as defined here and the Sculptor group of galaxies. Despite this attempted flagging, some of the CHVCs and semi-compact HVCs catalogued by P02 and de Heij et al. (2002b, hereafter D02) are also flagged as Stream-related. Since their origin has been proposed to be quite different from that of the large complexes, and given the imperfect nature of the MS flagging, these probes may prove quite interesting and are not excluded from Table 3.
Using the HVC catalogues published by PO2, D02 and Wakker & van Woerden (1991, hereafter WvW91) we have attempted to identify which clouds are associated with each detection. From P02, we use both the tables of HVCs, plus their table of XHVCs – clouds which have emission at high velocities, but which clearly link to emission at lower deviation velocities<sup>6</sup><sup>6</sup>6Deviation velocity is the degree to which a velocity deviates from the maximum allowed by Galactic rotation in that direction (Wakker, 1991).. Since clouds have arbitrary shapes, this is a difficult process.
In general, the optimal solution to the problem of associating stellar lines-of-sight with individual clouds would be to obtain component lists for all clouds in the desired catalogue (i.e. which pixels make up a particular cloud). Unfortunately, such lists are not available for the catalogues we searched, so another strategy was required.
From the tables of HVCs and XHVCs reported in P02, we took a position and angular size. Assuming a circular cloud, we found all clouds which encompass the stellar position. We then required that the velocity measured in the HIPASS spectrum and the velocity of the cloud differ by no more than the FWHM of the cloud. For the few cases where multiple clouds met these criteria, the data cubes were examined and it was immediately apparent which was the correct entry.
The D02 catalogue is very similar in structure; indeed, the same algorithm was used by P02 to create the HIPASS catalogue. The one exception is that D02 do not publish the size of individual clouds, based on the number of associated pixels, as do P02. They do, however, provide the major and minor axes of an ellipse which is fit to half the peak flux level. We used the major axis length to search a circular area around the reported cloud position. To avoid associating some clouds with large fractions of the sky (i.e. some clouds have major axes up to $`165\mathrm{°}`$), we limited the search radius to 10 deg. The same FWHM velocity criteria as above were also applied.
For the clouds in the catalogue of WvW91 the same process was repeated. Since those authors do not tabulate the FWHM for individual clouds, we assumed a FWHM of 36 km s<sup>-1</sup>, based on the global properties of the HIPASS HVCs listed in Table 2 of P02.
### 2.5. Stellar Velocities
It will be difficult or impossible to detect HVC absorption features in high resolution optical spectra if the velocity of the gas and stellar radial velocity (RV) are too close. In this case, stellar absorption will dominate the spectrum and the HVC feature will be lost. As part of a separate program, we have obtained medium-resolution spectra of 18 stars in our catalogue using the Double-Beam Spectrograph on the SSO 2.3m. An example of such a spectrum is shown in Fig. 4. Note the general lack of metal absorption lines, with the exception of the stellar Ca II K line at 3933.663 Å. From these spectra we obtained radial velocities by fitting Sersic profiles to the $`H\beta `$, $`H\gamma `$ and $`H\delta `$ Balmer lines. Table 2 lists the average stellar RV corrected to the Local Standard of Rest (LSR) frame, along with the H I velocity and the absolute difference of the two. Clearly, the greater the difference between the stellar and HVC features, the easier a detection will be and we suggest that a velocity difference of greater than $`50\text{km\hspace{0.17em}s}\text{-1}`$ will be needed. Such observations must be of sufficient resolution (i.e. R = 40000 – 60000, corresponding to $`7.55\text{km\hspace{0.17em}s}\text{-1}`$ at Ca II K) to resolve the separation between the HVC, stellar and disk absorption features.
We have performed a simple simulation to estimate, to first-order, the percentage of halo stars that are expected to have velocities within 50 km s<sup>-1</sup> of an HVC. Stars in the halo are drawn from a population with a total dispersion of $`\sigma 200\text{km\hspace{0.17em}s}\text{-1}`$. Assuming an isotropic distribution of velocities, a single projection (e.g. the line-of-sight component) will then follow a Gaussian distribution with $`\sigma _{LOS}200/\sqrt{3}120\text{km\hspace{0.17em}s}\text{-1}`$. We randomly generated stellar velocities according to such a distribution. For each of these velocities, we also generated a random HVC velocity using the HIPASS HVC velocity distribution (e.g. Fig 10 of Putman et al., 2002). Comparing these two velocities, we find that they are within 50 km s<sup>-1</sup> of each other $`13\%`$ of the time. It should be noted, however, that this is a global average; clearly HVCs with lower velocities will have, on average, a chance velocity alignment more often, while very high velocity HVCs will suffer such problems less than the average.
### 2.6. Optical Detections
In order to detect an absorption line in the optical, there must be a sufficient number of absorbing atoms along the line-of-sight. Assuming an optically thin gas, Savage & Sembach (1996) give a relation between the column density $`N`$ and the equivalent width of the absorption line $`W_\lambda `$. Since an H I column density can be measured from the HIPASS spectra by integrating over the HVC emission line, if the ratio of $`\text{Ca}^+`$ ions to H I atoms, N(Ca<sup>+</sup>)/N(H I) is known, we can predict the equivalent width of the Ca II absorption features. It should be noted that an accurate determination of N(Ca<sup>+</sup>)/N(H I) is important, since this quantity can vary by up to 2 orders of magnitude from cloud to cloud (Wakker, 2001)<sup>7</sup><sup>7</sup>7However Wakker & Mathis (2000) show a relation between the Ca II abundance and column density with much less variation at a given column density.. For the lowest column density features in Table 3, the expected equivalent width is $`20`$ mÅ, which should be detectable using large telescopes in a reasonable amount of time.
## 3. The Catalogue
After these selection criteria were applied, we had a total of 430 stars probing sufficiently dense HVC gas. These stars are listed in Table 3. The table contains the following columns: Column (1) gives the running number. Column (2) indicates the HES name of the star. Multiple components in a single line-of-sight have separate entries and are numbered. Columns (3) and (4) list the right ascension and declination (J2000) of the stellar probe, with Columns (5) and (6) showing the corresponding Galactic coordinates of the star. The large gap between RA $``$ 13:30 – 21:30 is due to the HES avoidance of the Galactic plane. Column (7) gives the HES B magnitude, which is accurate to $`0.2\text{mag}`$ (Wisotzki et al., 2000). The distance to the star (in kpc) is shown in column (8) and is taken directly from C05. Briefly, the distance was derived by those authors by estimating the absolute magnitude $`M_\mathrm{v}`$ from the B V color. A metallicity correction was applied based on an assumed average halo metallicity and the distance moduli computed. For any higher gravity A-star contaminants, this distance will be incorrect. The peak brightness temperature (in K) is shown in column (9); P02 quote a 5$`\sigma `$ sensitivity of 0.04 K. Column (10) shows the H I column density along the line-of-sight, which was measured from the moment maps. The LSR velocity ( km s<sup>-1</sup>) of the peak H I emission outside the $`|v_{LSR}|90\text{km\hspace{0.17em}s}\text{-1}`$ zone is given in column (11). This is the velocity of the HIPASS channel containing the peak flux – no attempt at profile fitting was made. The HIPASS velocity resolution is 26.4 km s<sup>-1</sup> after hanning smoothing. An asterisk in column (12) indicates an association with the Magellanic Stream, (see Sec. 2.4). The matches to the P02 HVCs and XHVCs, the WvW91 HVCs and the D02 HVCs are given in columns (13), (14), (15) and (16) respectively. For the P02 and D02 catalogues, the original nomenclature classifying the cloud (e.g. HVC, CHVC, HVC: etc) has been retained.
## 4. Conclusion
We have presented a catalogue of FHB stars, selected from the Hamburg/ESO Survey, that fortuitously align with high-velocity H I gas. These stars should make ideal candidates for echelle spectroscopy. Such observations offer the possibility of detecting the doppler-shifted absorption features in the spectrum, placing an upper limit on the distance to the HVC. In a similar fashion, non-detections place lower limits on the distance. Distance limits will help to resolve the question of the origins of HVCs and their role in the formation and evolution of Galaxies.
The financial support of the Australian Research Council, through its Discovery Project and Linkage International Award schemes, is gratefully acknowledged. We thank the scientific editor, W. Butler Burton, for his extensive and helpful suggestions.
|
warning/0506/hep-ph0506125.html
|
ar5iv
|
text
|
# 1 Distributions for 𝐺₀→𝑞̄𝑞𝑔 in arbitrary units. In figure (1a) the dot-dashed line is 𝑑𝑁/𝑑𝐸_𝑔 and the dashed line is 𝑑𝑁/𝑑𝐸_𝑞. In figure (1b) the dot-dashed line is 𝑑𝑁/cos𝜃_{𝑞𝑔} and the dashed line is 𝑑𝑁/𝑑cos𝜃_{𝑞̄𝑞}.
1. Introduction. — The existence of gluonic states is a quintessential prediction of Quantum Chromodynamics (QCD). The key difference between Quantum Electrodynamics (QED) and QCD is that gluons carry color charge while photons are electrically neutral. Gluon pairs can then form color singlet hadronic bound states, “glueballs,” like mesons and baryons, which are color singlet bound states of valence quarks. Because of formidable experimental and theoretical difficulties, it is frustrating, though not surprising, that this simple, dramatic prediction has resisted experimental verification for more than two decades. Quenched lattice simulations predict that the mass of the lightest glueball, $`G_0`$, a scalar, is near $`1.65`$ GeV, but the prediction is complicated by mixing with $`\overline{q}q`$ mesons that require more powerful computations. Experimentally the outstanding difficulty is that glueballs are not easily distinguished from ordinary $`\overline{q}q`$ mesons, themselves imperfectly understood. This difficulty is also exacerbated if mixing is appreciable.
The most robust identification criterion, necessary but not sufficient, is that glueballs are extra states, beyond those of the $`\overline{q}q`$ meson spectrum. This is difficult to apply in practice, though ultimately essential. In addition, glueballs are expected to be copiously produced in gluon rich channels such as radiative $`J/\psi `$ decay, and to have small two photon decay widths. These two expectations are encapsulated in the quantitative measure “stickiness,” which characterizes the relative strength of gluonic versus photonic couplings.
Another popular criterion is based on the fact that glueballs are $`SU(3)_{\mathrm{Flavor}}`$ singlets which should then couple equally to different flavors of quarks. However we show here that the amplitude for the decay of the ground state scalar glueball to quark-antiquark is proportional to the quark mass, $`(G_0\overline{q}q)m_q`$, so that decays to $`\overline{s}s`$ pairs are greatly enhanced over $`\overline{u}u+\overline{d}d`$, and mixing with $`\overline{s}s`$ mesons is enhanced relative to $`\overline{u}u+\overline{d}d`$. We exhibit the result at leading order and show that it holds to all orders in standard QCD perturbation theory.
The result has a simple nonperturbative physical explanation, similar, though different in detail, to the well known enhancement of $`\pi \mu \nu `$ relative to $`\pi e\nu `$. For $`m_q=0`$ chiral symmetry requires the final $`q`$ and $`\overline{q}`$ to have equal chirality, hence unequal helicity, so that in the $`G_0`$ rest frame with $`z`$ axis in the quark direction of motion, the total $`z`$ component of spin is nonvanishing, $`|S_Z|=1`$. Because the ground state $`G_0`$ $`gg`$ wave function is isotropic ($`L=S=0`$), the $`\overline{q}q`$ final state is pure s-wave,<sup>2</sup><sup>2</sup>2Integrating over the gluon direction to project out the s-wave $`gg`$ wave function is equivalent to integrating over the final quark direction with the initial gluon direction fixed. $`L=0`$. The total angular momentum is zero, and since there is no way to cancel the nonvanishing spin contribution, the amplitude vanishes. With one power of $`m_q0`$, the $`q`$ and $`\overline{q}`$ have unequal chirality and the amplitude is allowed.
The enhancement is substantial, since $`m_s`$ is an order of magnitude larger than $`m_u`$ and $`m_d`$. But for scalar glueballs of mass $`1.52`$ GeV, $`\mathrm{\Gamma }(G_0\overline{s}s)`$, is suppressed of order $`(m_s/m_{G_0})^2`$, so that it may be smaller than the nominally higher order $`G_0\overline{q}qg`$ process, which is $`SU(3)_{\mathrm{Flavor}}`$ symmetric. We find that the soft and collinear quark-gluon singularities of $`G_0\overline{q}qg`$ vanish for $`m_q=0`$, as they must if $`G_0\overline{q}q`$ is to vanish at one loop order for $`m_q=0`$. Unsuppressed, flavor-symmetric $`G_0\overline{q}qg`$ decays are dominated by configurations in which the gluon is well separated from the quarks, which hadronize predominantly to multi-body final states. $`G_0`$ may also decay flavor symmetrically to multigluon final states ($`n3`$), via the three and four gluon components of $`G^2`$ in equation (1), by higher dimension operators that arise nonperturbatively, or by higher orders in perturbation theory. The IR singularities of $`G_0ggg`$ are cancelled by virtual corrections to the $`G_0`$ wave function, while configurations with three (or more) well separated gluons hadronize to multihadron final states. The enhancement of $`\overline{s}s`$ relative to $`\overline{u}u+\overline{d}d`$ is then most strongly reflected in two body decays: we expect $`K^+K^{}`$ to be enhanced relative to $`\pi ^+\pi ^{}`$, while multibody decays are more nearly flavor symmetric.
Glueball decay to light quarks and/or gluons cannot be computed reliably in any fixed order of perturbation theory. However, the predicted ratio, $`\mathrm{\Gamma }(G_0\overline{s}s)/\mathrm{\Gamma }(G_0\overline{u}u+\overline{d}d)1`$, is credible since it follows from an analysis to all orders in perturbation theory and, in addition, from a physical argument that does not depend on perturbation theory. The implication that $`\mathrm{\Gamma }(G_0K^+K^{})\mathrm{\Gamma }(G_0\pi ^+\pi ^{})`$ is less secure and is best studied on the lattice. Remarkably, it is supported by an early quenched study of $`G_0`$ decay to pseudoscalar meson pairs for two “relatively heavy” $`SU(3)_{\mathrm{Flavor}}`$ symmetric values of $`m_q`$, corresponding to $`m_{\mathrm{PS}}400`$ and $`630`$ MeV. Linear dependence on $`m_q`$ implies quadratic dependence on $`m_{\mathrm{PS}}`$, which is consistent at $`1\sigma `$ with the lattice computations. Chiral suppression could then be the physical basis for the unexpected and unexplained lattice result. With subsequent computational and theoretical advances in lattice QCD, it should be possible today to verify the earlier study and to extend it to smaller values of $`m_{PS}`$, nearer to the chiral limit and to the physical pion mass. If the explanation is indeed chiral suppression, then the couplings of higher spin glueballs should be approximately flavor symmetric and independent of $`m_{\mathrm{PS}}`$, a prediction which can also be tested on the lattice.
Enhanced strange quark decay changes the expected experimental signature and supports the hypothesis that $`f_0(1710)`$ is predominantly the ground state scalar glueball. This identification was advocated by Sexton, Vaccarino, and Weingarten, and is even more compelling today in view of recent results from $`J/\psi `$ decay obtained by BES — see for an overview of the experimental situation.
In section 2 we compute $`(G_0\overline{q}q)`$ at leading order for massive quarks, with the expected linear dependence on $`m_q`$. In section 3 we show that $`(G_0\overline{q}q)m_q`$ to any order in $`\alpha _S`$. In section 4 we describe the infrared singularities of $`(G_0\overline{q}qg)`$. We conclude with a brief discussion, including experimental implications.
2. $`G_0\overline{q}q`$ at leading order. — Consider the decay of a scalar glueball $`G_0`$ with mass $`M_G`$ to a $`\overline{q}q`$ pair<sup>3</sup><sup>3</sup>3 Elastic scattering, $`gggg`$, contributes to the glueball wave function, not to the decay amplitude. The dissociation process, $`gg\overline{q}q+\overline{q}q`$, in which each gluon makes a transistion to a color-octet $`\overline{q}q`$ pair, is kinematically forbidden for $`m_q0`$; an additional gluon exchange is required to allow it to proceed on-mass-shell, which is therefore of order $`g^4`$ in the amplitude. with quark mass $`m_q`$. The effective glueball-gluon-gluon coupling is parameterized by
$$_{eff}=f_0G_0G_{a\mu \nu }G_a^{\mu \nu }$$
$`(1)`$
where $`G_0`$ is an interpolating field for the glueball, $`G_{a\mu \nu }`$ is the gluon field strength tensor with color index $`a`$, and $`f_0`$ is an effective coupling constant with dimension $`1/M`$ that depends on the $`G_0`$ wave function. The $`gg\overline{q}q`$ scattering amplitude can be written as
$$(gg\overline{q}q)=ϵ_{1\mu }ϵ_{2\nu }^{\mu \nu }(gg\overline{q}q)$$
$`(2)`$
where $`ϵ_{i\mu }=ϵ_\mu (p_i,\lambda _i)`$ with $`i=1,2`$ are the polarization vectors for massless constituent gluons with four momentum $`p_i`$ and polarization $`\lambda _i`$. Using equations (1) and (2), summing over the polarizations $`\lambda _i`$, and averaging over the gluon direction in the $`G_0`$ rest frame to project out the s-wave, we obtain
$$(G_0\overline{q}q)=\frac{f_0}{4\pi }𝑑\mathrm{\Omega }X^{\mu \nu }_{\mu \nu }(gg\overline{q}q),$$
$`(3)`$
where $`X^{\mu \nu }=2p_2^\mu p_1^\nu M_G^2g^{\mu \nu }`$ projects out the $`|(++)+()>`$ helicity state that couples to $`G_{a\mu \nu }G_a^{\mu \nu }`$ in equation (1).
From the lowest order Feynman diagrams we obtain
$$X^{\mu \nu }_{\mu \nu }=\frac{32\pi \sqrt{2}\alpha _S}{3}\frac{m_q}{1\beta ^2\mathrm{cos}^2\theta }\overline{u}(p_3,h_3)v(p_4,h_4)\delta _{ij}$$
$`(4)`$
where $`u_3,v_4`$ are the $`q,\overline{q}`$ spinors for quark and antiquark with center of mass momenta $`p_3,p_4`$, helicities $`h_3,h_4`$, color indices $`i,j`$ and center of mass velocity $`\beta `$. Equation (4) includes a color factor from the color singlet $`gg`$ wave function,
$$C_{ij}=\frac{\delta _{a,b}}{\sqrt{8}}\frac{\lambda _{ik}^a}{2}\frac{\lambda _{kj}^b}{2}=\frac{\sqrt{2}}{3}\delta _{ij}$$
$`(5)`$
Performing the angular integration, the decay amplitude is
$$(G_0\overline{q}q)=f_0\alpha _S\frac{16\pi \sqrt{2}}{3}\frac{m_q}{\beta }\mathrm{log}\frac{1+\beta }{1\beta }\overline{u}_3v_4\delta _{ij}.$$
$`(6)`$
Squaring equation (6), summing over quark helicities and color indices, and performing the phase space integration, the decay width is
$$\mathrm{\Gamma }(G_0\overline{q}q)=\frac{16\pi }{3}\alpha _S^2f_0^2m_q^2M_G\beta \mathrm{log}^2\frac{1+\beta }{1\beta }.$$
$`(7)`$
Notice that an explicit factor $`m_q`$ appears in the $`gg\overline{q}q`$ amplitude, equation (4), which is not averaged over the initial gluon direction and which clearly has contributions from higher partial waves, $`J>0`$. It may then appear that chiral suppression applies not just to spin 0 glueballs but also to glueballs of higher spin. However when equation (4) is squared and the phase space integration is performed, a factor $`1/m_q^2`$ results from the $`t`$ and $`u`$ channel poles, which cancels the explicit factor $`m_q^2`$ in the numerator, so that the total cross section $`\sigma (gg\overline{q}q)`$ does not vanish in the chiral limit, because of the $`J>0`$ partial waves.
3. $`G_0\overline{q}q`$ to all orders. — We now show that $`(G_0\overline{q}q)`$ vanishes to all orders in perturbation theory for $`m_q=0`$. Consider the Lorentz invariant amplitude
$$_X(p_1,p_2,p_3,p_4)=X^{\mu \nu }_{\mu \nu }$$
$`(8)`$
where $`_{\mu \nu }`$ is defined in equation (2) and $`X^{\mu \nu }`$ below (3). The perturbative expansion for $`_X`$ is a sum of terms arising from Feynman diagrams with arbitrary numbers of loops. After evaluation of the loop integrals, regularized as necesary, $`_X`$ is a sum of terms,
$$_X=\underset{i}{}\overline{u}(p_3,\chi _3)\mathrm{\Gamma }_iu(p_4,\chi _4),$$
$`(9)`$
where $`u_3,u_4`$ are respectively massless fermion and antifermion spinors of chirality $`\chi _3,\chi _4`$. The $`\mathrm{\Gamma }_i`$ are $`4\times 4`$ matrices, each a product of $`n_i`$ momentum-contracted Dirac matrices,
$$\mathrm{\Gamma }_i=\mathit{}_{i1}\mathit{}_{i2}\mathrm{}\mathit{}_{in_i}$$
$`(10)`$
where each $`k_{ia}`$ is one of the external four-momenta, $`p_1,p_2,p_3,p_4`$.
Chiral invariance for $`m_q=0`$ implies that the number of factors, $`n_i`$, in equation (10) is always odd. Since all external momenta vanish and the spinors obey $`\mathit{}_3u_3=\mathit{}_4u_4=0`$, by suitably anticommuting the $`\mathit{}_{ia}`$, each term in equation (9) can be reduced to a sum of terms linear in $`\mathit{}_1`$ and $`\mathit{}_2`$, which we choose to be symmetric and antisymmetric,
$$\overline{u}(p_3,\chi _3)\mathrm{\Gamma }_iu(p_4,\chi _4)=\overline{u}(p_3,\chi _3)[S_i(s,t,u)(\mathit{}_1+\mathit{}_2)+A_i(s,t,u)(\mathit{}_1\mathit{}_2)]u(p_4,\chi _4).$$
$`(11)`$
The coefficients $`A_i,S_i`$ are Lorentz invariant functions of the Mandelstam variables $`s,t,u`$. Since $`p_1+p_2=p_3+p_4`$, the symmetric term vanishes and equation (9) reduces to
$$_X=A(s,t,u)\overline{u}(p_3,\chi _3)(\mathit{}_1\mathit{}_2)u(p_4,\chi _4).$$
$`(12)`$
where $`A(s,t,u)=_iA_i(s,t,u)`$.
Next consider the integration over the gluon direction, equation (3). In the $`G_0`$ rest frame with the z-axis chosen along the quark direction of motion, $`\widehat{z}=\widehat{p}_3`$, we integrate over $`d\mathrm{\Omega }=d^2\widehat{p}_1=d\varphi _1d\mathrm{cos}\theta _1`$. The Mandelstam variables are then $`s=M_G^2`$ and $`u,t=\frac{1}{2}M_G^2(1\pm \mathrm{cos}\theta _1)`$. Since the color and helicity components of the $`G_0`$ wave function are symmetric under interchange of the two gluons, Bose symmetry requires $`A(s,t,u)`$ to be odd under $`p_1p_2`$. In our chosen coordinate system $`A`$ is a function only of $`\mathrm{cos}\theta _1`$, and must therefore be odd, $`A(\mathrm{cos}\theta _1)=A(\mathrm{cos}\theta _1)`$. But evaluating $`\overline{u}_3(\mathit{}_1\mathit{}_2)u_4`$ explicitly we find
$$\overline{u}_3(\mathit{}_1\mathit{}_2)u_4=M_G^2e^{i\varphi _1}\mathrm{sin}\theta _1$$
$`(13)`$
which is even in $`\mathrm{cos}\theta _1`$, while the azimuthal factor, $`e^{i\varphi _1}`$, provides the required oddness under $`p_1p_2`$: $`e^{i\varphi _1}e^{i(\varphi _1+\pi )}=e^{i\varphi _1}`$. Consequently the integral $`𝑑\mathrm{cos}\theta _1𝒜`$ vanishes, and $`(G_0\overline{q}q)=0`$ to all orders in the chiral limit. In fact, because of our choice of axis, $`\widehat{z}=\widehat{p}_3`$, the integral over $`\varphi _1`$ also vanishes. For other choices of $`\widehat{z}`$ the azimuthal and polar integrals do not vanish separately, but the full angular integral, $`d^2\widehat{p}_1`$, vanishes in any case.
For nonvanishing quark mass, $`m_q0`$, chirality-flip amplitudes contribute. With one factor of $`m_q`$ from the fermion line connecting the external quark and antiquark, the $`\mathrm{\Gamma }_i`$ matrices in equation (9) include products of even numbers of Dirac matrices, i.e., $`n_i`$ in equation (10) may be even. Beginning in order $`m_q`$ there are then nonvanishing contributions to $`(G_0\overline{q}q)`$, like the leading order term shown explicitly in equation (6).
The vanishing azimuthal integration for $`\widehat{z}=\widehat{p}_3`$ reflects the physical argument given in the introduction. The factor $`e^{i\varphi _1}`$ corresponds to $`S_Z=1`$ from the aligned spins of the $`q`$ and $`\overline{q}`$, while the absence of a compensating factor in $`𝒜`$ is due to the projection of the orbital s-wave by the $`d^2\widehat{p}_1`$ integration and the absence of spin-polarization in the initial state.
4. Infrared singularities of $`G_0\overline{q}qg`$. — Although it is of order $`\alpha _S^3`$, $`\mathrm{\Gamma }(G_0\overline{q}qg)`$ is not chirally suppressed and may therefore be larger than $`\mathrm{\Gamma }(G_0\overline{s}s)`$, which is of order $`\alpha _S^2\times m_s^2/m_G^2`$. Setting $`m_q=0`$ we evaluated the 13 Feynman diagrams using the helicity spinor method with numerical evaluation of the 9 dimensional integral:
$$\begin{array}{cccc}\mathrm{\Gamma }(G_0\overline{q}qg)& =& _{h_3,h_4,\lambda _5}_{\mathrm{PS}}|(G_0\overline{q}_3q_4g_5)|^2\hfill & \\ & =& \frac{f_0^2}{16\pi ^2}_{h_3,h_4,\lambda _5}_{\mathrm{PS}}𝑑\mathrm{\Omega }_1𝑑\mathrm{\Omega }_1^{}ϵ_{5\alpha }^{}X_{\mu \nu }^{\mu \nu \alpha }(g_1g_2\overline{q}_4q_3g_5)\hfill & \\ & & \times ϵ_{5\beta }X_{\sigma \rho }^{}^{\sigma \rho \beta }(g_1^{}g_2^{}\overline{q}_4q_3g_5)^{}.\hfill & \hfill (14)\end{array}$$
Details will be presented elsewhere. We focus here on the infrared singularities, which provide a consistency check at one loop order that $`(G_0\overline{q}q)`$ vanishes in the chiral limit.
In general there could be soft IR divergences for $`E_q,E_{\overline{q}},E_g0`$ and collinear divergences for $`\theta _{qg},\theta _{\overline{q}g},\theta _{\overline{q}q}0`$. In fact, only the $`\overline{q}q`$ collinear divergence occurs, as can be seen from the distributions in figure 1, obtained by imposing only the cut $`\theta _{\overline{q}q}>0.1`$ in the $`G_0`$ rest frame: neither $`dN/dE_g`$ nor $`dN/dE_q`$ diverge at low energy, and only $`dN/\theta _{\overline{q}q}`$ diverges at $`\theta _{\overline{q}q}0`$. Instead $`dN/dE_g`$ diverges at the maximum energy, $`E_g=m_{G_0}/2`$, and $`dN/d\theta _{qg}`$ diverges for $`\theta _{qg}\pi `$. Both of these divergences are kinematical reflections of the collinear singularity at $`\theta _{\overline{q}q}0`$, for which the $`\overline{q}q`$ pair with $`m_{\overline{q}q}=0`$ recoils with half of the available energy against the gluon in the opposite hemisphere.
This is precisely the pattern of divergences required if $`G_0\overline{q}q`$ is chirally suppressed to all orders and, in particular, at one loop. For if there were soft divergences in any of $`E_q,E_{\overline{q}},E_g`$ or collinear divergences in $`\theta _{qg}`$ and $`\theta _{\overline{q}g}`$, then the resulting singularities at $`m_{qg},m_{\overline{q}g}0`$ would have to be cancelled by virtual corrections to $`G_0\overline{q}q`$, such as gluon self energy contributions to the quark propagator. The absence of these singularities is a consistency check (i.e., a necessary condition) that $`G_0\overline{q}q`$ is chirally suppressed at one loop order. The collinear divergence for $`\theta _{\overline{q}q}0`$ is cancelled by quark loop contributions to the $`gggg`$ amplitude, which in the present context are one loop corrections to the $`G_0`$ wave function.
5. Discussion. — We have shown to all orders in perturbation theory and with a simple, nonperturbative physical argument that the ground state $`J=0`$ glueball has a chirally suppressed coupling to light quarks, $`(G_0\overline{q}q)m_q`$, with corrections of higher order in $`m_q/m_G`$. From equation (7) with $`m_u,m_d,m_s`$ varied within $`1\sigma `$ limits, $`\mathrm{\Gamma }(G_0\overline{s}s)`$ dominates $`\mathrm{\Gamma }(G_0\overline{u}u+\overline{d}d)`$ by a factor between 20 and 100. Flavor symmetry is reinstated for $`G_0\overline{q}qg`$ when the gluon is well separated from the $`q`$ and $`\overline{q}`$. For sufficiently heavy $`m_G`$ one can test this picture by measuring strangeness yield as a function of thrust or sphericity, with enhanced strangeness in high thrust or low sphericity events, but it is unclear if this is feasible for $`m_G1.7`$ GeV. It is more feasible for the ground state pseudoscalar glueball, which is expected to be heavier than the scalar and which we also expect to be subject to chiral suppression.
For light scalar glueballs, the best hope to see strangeness enhancement is the two body decays, $`G_0K^+K^{}/\pi ^+\pi ^{}`$. Since $`G_0\overline{q}qg`$ and $`G_0ggg`$ are not chirally suppressed, naive power counting suggests $`\mathrm{\Gamma }(G_0\overline{q}qg+ggg)\mathrm{\Gamma }(G_0\overline{s}s)`$, so that $`n3`$ parton decay amplitudes are probably the dominant mechanism for multihadron production. Then $`\overline{K}K`$ will dominate two body decays while multiparticle final states are approximately $`SU(3)_{\mathrm{Flavor}}`$ symmetric, up to phase space corrections favoring nonstrange final states.
Chiral suppression has a major impact on the experimental search for the ground state scalar glueball. Candidates cannot be ruled out because they decay preferentially to strange final states, especially $`\overline{K}K`$, and mixing with $`\overline{s}s`$ mesons may be enhanced. This picture of a chirally suppressed $`G_0`$ fits nicely with the known properties of the $`f_0(1710)`$ meson. It is copiously produced in radiative $`\psi `$ decay in the $`\psi \gamma \overline{K}K`$ channel and in the gluon-rich central rapidity region in $`pp`$ scattering, has a small $`\gamma \gamma `$ coupling, has a mass consistent with the prediction of quenched lattice QCD, and has a strong preference to decay to $`\overline{K}K`$, with $`B(\pi \pi )/B(\overline{K}K)<0.11`$ at 95% CL. As a rough estimate of the stickiness, we combine the $`\gamma \gamma `$ 95% CL upper limit with central values for $`\psi `$ radiative decay, with the result $`S(f_0(1710)):S(f_2^{}(1525)):S(f_2(1270))(>36):12:1`$. A more complete discussion of the experimental situation will be given elsewhere — see also .
The interpretation of $`f_0(1710)`$ as the chirally suppressed scalar glueball can be tested both theoretically and experimentally. Lattice QCD can test the prediction that $`G_0\overline{K}K`$ is enhanced for the ground state $`J=0`$ glueball but not for $`J>0`$. With an order of magnitude more $`J/\psi `$ decays than BES II, experiments at BES III and CESR-C will extend partial wave analysis to rarer two body decays and to multiparticle decays. They could confirm $`B(f_0\overline{K}K)/B(f_0\pi \pi )1`$ and, if, as is likely, the rate for multiparticle decays is big, the lower bound on $`B(\psi \gamma f_0(1710)`$ will increase beyond its already appreciable value from $`\overline{K}K`$ alone. A large inclusive rate $`B(\psi \gamma f_0)`$, a large ratio $`B(f_0\overline{K}K)/B(f_0\pi \pi )`$, and approximately flavor symmetric couplings to multiparticle final states would support the identification of $`f_0(1710)`$ as the chirally suppressed, ground state scalar glueball.
Note added: Cornwall and Soni observed that the trace anomaly implies chiral suppression of the $`G_0`$ coupling to $`\pi \pi `$ and $`\overline{K}K`$ at zero four-momentum; however, the extrapolation to the $`G_0`$ mass-shell, e.g., at $`1700`$ MeV, is not under control and could be large. Similarly, in the conjectured AdS/CFT approach to QCD the scalar glueball is a dilaton, which therefore has chirally suppressed two body couplings at zero momentum. I thank A. Soni, H. Murayama, and M. Schwartz for bringing this work to my attention.
Acknowledgements: I wish to thank M. Golden and S. Sharpe for helpful discussions.
This work was supported in part by the Director, Office of Science, Office of High Energy and Nuclear Physics, Division of High Energy Physics, of the U.S. Department of Energy under Contract DE-AC03-76SF00098
|
warning/0506/hep-ex0506054.html
|
ar5iv
|
text
|
# Electroweak and QCD Results from DØ
## 1 Introduction
The DØ experiment is a multi-purpose collider detector located at the Fermilab Tevatron proton-antiproton collider. After undergoing a substantial upgrade, the experiment is now taking data at $`\sqrt{s}=1.96\mathrm{TeV}`$. There are many new results based on this Run II data in the areas of Quantum Chromodynamics (QCD) and Electroweak physics.
## 2 QCD Results
A study has been performed of the angular correlations in jets produced in the DØ detector . At leading order in QCD, jets are expected to be produced back-to-back in azimuth ($`\varphi `$). One would then expect that the difference in azimuthal angle between two jets is $`\mathrm{\Delta }\varphi =\pi `$. However, higher order effects, such as additional soft radiation in the event, will cause this angular difference to be less than $`\pi `$. The distribution of $`\mathrm{\Delta }\varphi `$ is sensitive to higher order effects. Additionally, this measurement does not rely on measuring the energy of the jet and hence does not suffer from energy scale systematic uncertainties. Figure 1 shows the $`\mathrm{\Delta }\varphi `$ distribution for jets in four different transverse momentum bins compared to the predictions from Monte Carlo generators.
The inclusive jet cross section and the dijet cross section have also been measured. These distributions are sensitive to both the strong coupling constant and to the parton density functions. Furthermore, many new physics models predict enhancements in the dijet mass cross section at large values of invariant mass. Figure 2 shows both the inclusive jet cross section and the dijet cross section distributions. These distributions are consistent with the next-to-leading order perturbative QCD theoretical predictions.
## 3 Electroweak Results
DØ has performed a measurement of the cross section times branching fraction for $`Z`$ bosons decaying to tau pairs . This measurement is a verification of the tau identification abilities for the experiment as well as a test of lepton universality. Furthermore, some physics models predict final states that would result in an excess of tau pair events over Standard Model predictions.
The analysis requires one tau to decay as a muon. The second tau is identified using a neural network that has been trained using the different tau decay topologies. 2008 candidate events are selected in 226 pb<sup>-1</sup> of data. Approximately 55% of these events are estimated to be background events. This results in a measurement of $`\sigma \mathrm{Br}(Z\tau \tau )=237\pm 15_{stat}\pm 18_{sys}\pm 15_{lum}\mathrm{pb}`$. This is consistent with the Standard Model prediction of $`242\pm 9\mathrm{p}\mathrm{b}`$.
DØ also has a variety of results on diboson production. At the Tevatron, pairs of gauge bosons can be produced through $`t`$ or $`u`$ channel quark exchange, or can be produced through an $`s`$-channel triple gauge boson vertex. The strength of these triple gauge boson vertices is an important test of the Standard Model. Diboson signatures are also important backgrounds for Higgs and new physics searches.
The cross section for the production of $`p\overline{p}W\gamma +X`$ has been measured. This analysis requires the $`W`$ boson to decay to either an electron or a muon and a neutrino. The photon is identified by its signature in the calorimeter and the absence of a matching track in the central tracking system. In the electron channel, 112 events are selected in 162 pb<sup>-1</sup> of data. In the muon channel, 161 events are identified in 134 pb<sup>-1</sup> of data. In both channels, the background is estimated to be approximately half the number of events in data. The combination of both channels results in a cross section measurement for $`W\gamma Xl\nu X`$ of $`14.8\pm 1.6_{stat}\pm 1.0_{sys}\pm 1.0_{lum}\mathrm{pb}`$. This is in agreement with the Standard Model prediction of $`16.0\pm 0.4\mathrm{pb}`$.
A measurement of the cross section times branching fraction has been performed for events with a photon and a $`Z`$ boson, with the $`Z`$ boson decaying to electron or muon pairs . In the $`ee\gamma `$ channel, 33 data events are present in 177 pb<sup>-1</sup> of data, with an estimated background of $`4.7\pm 0.7`$ events. In the $`\mu \mu \gamma `$ channel, 68 data events are present in 144 pb<sup>-1</sup> of data, with an estimated background of $`10.1\pm 1.3`$ events. Figure 3 shows the three body invariant mass versus two body invariant mass for the candidate events. The combined cross section times branching fraction for both channels is $`3.90\pm 0.51_{stat+sys}\pm 0.25_{lum}\mathrm{pb}`$, which is in good agreement with the expected value of 4.3 pb.
A search for pairs of $`W`$ bosons has also been performed . This analysis selects events with two opposite sign leptons (electrons or muons) and missing energy. A total of 25 events are selected in between 224 and 252 pb<sup>-1</sup> of data (depending on the channel). The background has been estimated to be $`8.1\pm 0.6_{stat}\pm 0.6_{sys}\pm 0.5_{lum}`$. The data represents a 5.2$`\sigma `$ excess over the background prediction.
A search for $`WZ`$ events has been performed by selecting events that have three leptons (electrons or muons) and missing energy . The missing transverse energy versus dilepton invariant mass distribution is shown in Figure 4. A total of three candidate events are observed in data in approximately 300 pb<sup>-1</sup> of data. There are $`2.04\pm 0.13`$ expected signal events and $`0.71\pm 0.08`$ expected background events. A 95% confidence level limit on the cross section has been set at 13.3 pb.
Limits have also been set on anomalous $`WW\gamma `$ and $`WWZ`$ couplings. These couplings can be parameterized in an effective Lagrangian, with a scale factor $`\mathrm{\Lambda }`$. Figure 5 shows the limits set on the anomalous coupling parameters. All of these anomalous coupling limit parameters are predicted to be zero in the Standard Model.
|
warning/0506/hep-th0506087.html
|
ar5iv
|
text
|
# References
In a series of papers Dijkgraaf and Vafa argued that for a wide class of $`𝒩=1`$ $`U(N)`$ supersymmetric gauge theories the effective superpotential, thought as a function of the chiral glueball superfield $`S`$, could be computed by means of a simple matrix model whose action is the tree-level superpotential of the gauge theory. Their proposal was the result of a detailed study of the various dualities between $`U(N)`$ supersymmetric gauge theories, B-model topological strings and matrix models in the large $`N`$ limit . This kind of limit was actually an old idea due to ’t Hooft who had shown that perturbation theory in the large $`N`$ limit singles out only the planar Feynman diagrams. What Diikgraaf and Vafa discovered was that a leading-order perturbative calculation via a matrix model could capture completely an exact quantity in the corresponding gauge theory, namely the effective superpotential which describes the effects of gaugino condensation. In particular they showed that for a theory with a matter field in the adjoint representation of the gauge group the effective superpotential is always of the form:
$$W_{eff}(S)=N\frac{_{\chi =2}(S)}{S}$$
(1)
where $`_{\chi =2}`$ represents the planar free energy from diagrams with the topology of the sphere. Then this result was obtained directly using a perturbative field theory approach , and through the analysis of the generalized Konishi anomaly . Many extensions of this idea have been studied, with the aim to include also quark fields in the (anti)fundamental representation . In particular in it was realized that such a generalization could be achieved by taking into account the contributions from planar diagrams with the topology of the disk ($`_{\chi =1}`$), i.e. the diagrams with one quark-loop boundary . In this case the relation (1) was modified as follows:
$$W_{eff}(S)=N\frac{_{\chi =2}(S)}{S}+_{\chi =1}(S)$$
(2)
The validity of this approach was tested considering a theory whose lagrangian contains a mass term for the quark fields $`q`$ $`(\stackrel{~}{q})`$, a Yukawa interaction with the field in the adjoint $`\varphi `$ and a quadratic tree-level superpotential for $`\varphi `$. This model has been studied also for different gauge groups and a generalized Yukawa coupling .
In this letter we focus on a generalization of this kind of models, with a $`𝒩=1`$ $`U(N)`$ gauge theory with $`N_f`$ flavors and an *arbitrary* tree-level polynomial superpotential for $`\varphi `$, i.e.
$$W(\varphi )=\underset{k=1}{\overset{n+1}{}}\frac{1}{k}g_k\varphi ^k$$
(3)
We want to obtain the matter contribution $`_{\chi =1}(S)`$ for this general case. Following we study the maximally-confining phase of the theory and compute $`_{\chi =1}`$ explicitly as a power series in $`S`$, i.e.
$$_{\chi =1}(S)=\underset{j}{}P_jS^j$$
(4)
We find that the coefficients $`P_j`$ for given $`j`$ depend only on the $`g_k`$ couplings in the potential with $`k<2j`$.
Our general derivation naturally includes the results obtained by Gomez-Reino , who solves a $`SU(N)`$ matrix model using the properties of the factorization of the Seiberg-Witten curve. This is a direct consequence of the fact that for a maximally confining $`SU(N)`$ theory only the moduli carrying even indices contribute to the factorization of the Seiberg-Witten curve . Therefore restricting our result to the case of an even polynomial tree-level superpotential we recover the results in .
We consider a $`𝒩=1`$ $`U(N)`$ supersymmetric gauge theory obtained by softly breaking $`𝒩=2`$ supersymmetry via the introduction of a tree-level superpotential. The action is given by
$$S=S_{matter}+S_{gauge}+S_{break}$$
(5)
with
$`S_{matter}`$ $`=`$ $`{\displaystyle d^4xd^2\theta d^2\overline{\theta }\left(e^V\overline{\varphi }e^V\varphi +\overline{q}e^Vq+\stackrel{~}{q}e^V\overline{\stackrel{~}{q}}\right)}`$
$`S_{gauge}`$ $`=`$ $`2\pi i\tau {\displaystyle d^4xd^2\theta Tr(W^\alpha W_\alpha )}`$
$`S_{break}`$ $`=`$ $`{\displaystyle d^4xd^2\theta Tr\left(W_{tree}(\varphi ,q,\stackrel{~}{q})\right)}`$ (6)
and
$$W_{tree}(\varphi ,q,\stackrel{~}{q})=W(\varphi )+m\stackrel{~}{q}q\stackrel{~}{q}\varphi q$$
(7)
where $`\varphi `$ is the matter field in the adjoint representation of $`U(N)`$, $`q`$ and $`\stackrel{~}{q}`$ $`N_f`$ pairs of quark fields (with mass $`m`$) in the fundamental and anti-fundamental and $`W^\alpha `$ the field-strength of the theory. The superpotential $`W(\varphi )`$ has the general form given in (3).
At the quantum level the vacua of the theory are determined by the appearance of a gaugino condensate described by a chiral superfield
$$S=\frac{1}{32\pi ^2}Tr(W^\alpha W_\alpha )$$
(8)
It is the effective superpotential $`W_{eff}(S)`$ which encodes the vacuum structure of the theory.
We follow the matrix model approach of with the generalization to include fundamental matter fields , so that we replace $`\varphi `$ with a $`N\times N`$ hermitian matrix $`\mathrm{\Phi }_b^a`$ ($`a,b=1,\mathrm{},N`$), $`q`$ with a $`N\times N_f`$ matrix $`Q_\alpha ^a`$ and $`\stackrel{~}{q}`$ with a $`N_f\times N`$ matrix $`\stackrel{~}{Q}_a^\alpha `$ ($`\alpha =1,\mathrm{},N_f`$) and write the matrix integral
$$Z=\frac{1}{Vol(U(N))}𝑑\mathrm{\Phi }𝑑Q𝑑\stackrel{~}{Q}e^{\frac{1}{g_s}W_{tree}(\mathrm{\Phi },Q,\stackrel{~}{Q})}$$
(9)
where $`Vol(U(N))`$ is the volume of the gauge group and $`g_s`$ is the string coupling.
It is well known that in the ’t Hooft limit, where we let $`N\mathrm{},`$ $`g_s0`$ while keeping $`Ng_s`$ fixed, the matrix model partition function $`Z`$ receives contributions from planar diagrams both with the topology of the sphere ($`\chi =2`$) and the topology of the disk ($`\chi =1`$) corresponding respectively to 0 and 1 quark boundary. Thus, if we call $``$ the free energy of the model, we have
$$Z=e^{}e^{\frac{_{\chi =2}}{g_s^2}+\frac{_{\chi =1}}{g_s}+\mathrm{}}$$
(10)
Now, if we interpret $`Ng_s=S`$ as the glueball chiral superfield of the gauge theory (5), we are led to the expression in (2.) The first term in (2) is exactly the glueball superpotential conjectured by Dijkgraaf-Vafa for theories with fields only in the adjoint representation while the second one comes from the extension to include quark fields in the (anti)fundamental.
We want to evaluate this matter contribution $`_{\chi =1}`$ for an arbitrary polynomial superpotential for $`\mathrm{\Phi }`$ of the form (3). As shown in $`_{\chi =1}`$ can be written in terms of an hyperelliptic curve $`y(x)`$ as follows:
$$_{\chi =1}=\frac{1}{2}N_f_m^{\mathrm{\Lambda }_0}(W^{}(x)y(x))𝑑x$$
(11)
where $`m`$ is the mass of the quarks and $`\mathrm{\Lambda }_0`$ a regularization cut-off. In order to integrate the hyperelliptic curve $`y(x)`$ in (11) here we will follow the approach of . We study the so-called maximally confining phase of the theory so that the hyperelliptic curve degenerates as
$$y(x)=\sqrt{W^{}(x)^2f_{n1}(x)}=P_{n1}(x)\sqrt{(x\sigma )^2\mu ^2}$$
(12)
where $`P_{n1}(x)`$ is a polynomial of degree $`n1`$ and $`\sigma `$ is a parameter that can be shifted to 0 in a $`U(N)`$ theory. The crucial point is to perform the change of variable
$$x\frac{\mu }{2}(\xi +\xi ^1)$$
(13)
in order to expand $`W(x)`$ and $`W^{}(x)`$ in series of $`\xi `$ and $`\xi ^1`$:
$`W(x)`$ $`=`$ $`W\left({\displaystyle \frac{\mu }{2}}(\xi +\xi ^1)\right)=b_0+{\displaystyle \underset{k=1}{\overset{n+1}{}}}b_k(\xi ^k+\xi ^k)`$
$`W^{}(x)`$ $`=`$ $`W^{}\left({\displaystyle \frac{\mu }{2}}(\xi +\xi ^1)\right)=c_0+{\displaystyle \underset{k=1}{\overset{n}{}}}c_k(\xi ^k+\xi ^k)`$ (14)
In this way one obtains (see for details)
$$y(x)=P_{n1}(x)\sqrt{x^2\mu ^2}=\underset{k=1}{\overset{n}{}}c_k(\xi ^k\xi ^k)$$
(15)
which allows to solve for $`S`$
$$S_\mu ^{+\mu }y(x)𝑑x=\mu \frac{c_1}{2}$$
(16)
and gives also
$$y(x)𝑑x=\mu c_1log(\xi )+b_0+\underset{k=1}{\overset{n+1}{}}b_k\left(\xi ^k\xi ^k\right)$$
(17)
Now we need to compute the coefficients $`b_k`$ and $`c_k`$ in (14). Performing the change of variable (13) we obtain
$$W(x)=\underset{j=1}{\overset{n+1}{}}\frac{g_j}{j}\left(\frac{\mu }{2}\right)^j\left(\begin{array}{c}j\\ \frac{j}{2}\end{array}\right)+\underset{k=1}{\overset{n+1}{}}\left(\xi ^k+\xi ^k\right)\underset{j=k}{\overset{n+1}{}}\frac{g_j}{j}\left(\frac{\mu }{2}\right)^j\left(\begin{array}{c}j\\ \frac{jk}{2}\end{array}\right)$$
(18)
with the following convention:
$$\left(\begin{array}{c}j\\ \frac{2m+1}{2}\end{array}\right)0\left(\begin{array}{c}j\\ 2m\end{array}\right)\frac{j!}{(2m)!(j2m)!}$$
(19)
From (14) and (18) we learn that:
$`b_0`$ $`=`$ $`{\displaystyle \underset{j=1}{\overset{n+1}{}}}{\displaystyle \frac{g_j}{j}}\left({\displaystyle \frac{\mu }{2}}\right)^j\left(\begin{array}{c}j\\ \frac{j}{2}\end{array}\right)`$ (22)
$`b_k`$ $`=`$ $`{\displaystyle \underset{j=k}{\overset{n+1}{}}}{\displaystyle \frac{g_j}{j}}\left({\displaystyle \frac{\mu }{2}}\right)^j\left(\begin{array}{c}j\\ \frac{jk}{2}\end{array}\right)`$ (25)
Following an analogous procedure we obtain the value of $`c_1`$ which is needed in (16):
$$c_1=\underset{j=1}{\overset{n}{}}g_{j+1}\left(\frac{\mu }{2}\right)^j\left(\begin{array}{c}j\\ \frac{j1}{2}\end{array}\right)$$
(26)
As shown in one finds
$`_{\chi =1}`$ $`=`$ $`{\displaystyle \frac{1}{2}}N_f\{W(m)+\mu c_1log\left({\displaystyle \frac{\xi (\mathrm{\Lambda }_0)}{\xi (m)}}\right)+b_0`$ (27)
$`+{\displaystyle \underset{k=1}{\overset{n+1}{}}}b_k\left({\displaystyle \frac{m}{\mu }}\right)^k[(1+\sqrt{1\left({\displaystyle \frac{\mu }{m}}\right)^2})^k(1\sqrt{1\left({\displaystyle \frac{\mu }{m}}\right)^2})^k]\}`$
where from (13) one can see that:
$`\xi (\mathrm{\Lambda }_0)`$ $`=`$ $`{\displaystyle \frac{\mathrm{\Lambda }_0}{\mu }}\left(1+\sqrt{1\left({\displaystyle \frac{\mu }{\mathrm{\Lambda }_0}}\right)^2}\right){\displaystyle \frac{2\mathrm{\Lambda }_0}{\mu }}\mathrm{\Lambda }_0\mathrm{}`$
$`\xi (m)`$ $`=`$ $`{\displaystyle \frac{m}{\mu }}\left(1+\sqrt{1\left({\displaystyle \frac{\mu }{m}}\right)^2}\right)`$ (28)
Using the following identity
$$(1+a)^k(1a)^k=2\underset{m=0}{\overset{[\frac{k1}{2}]}{}}\left(\begin{array}{c}k\\ 2m+1\end{array}\right)a^{2m+1}$$
(29)
we can rewrite (27) as
$`_{\chi =1}`$ $`=`$ $`{\displaystyle \frac{1}{2}}N_f[W(m)+Slog\left({\displaystyle \frac{2\mathrm{\Lambda }_0}{m}}\right)+b_02Slog\left({\displaystyle \frac{1+\sqrt{1(\mu /m)^2}}{2}}\right)`$ (32)
$`+2{\displaystyle \underset{k=1}{\overset{n+1}{}}}b_k\left({\displaystyle \frac{m}{\mu }}\right)^k{\displaystyle \underset{l=0}{\overset{[\frac{k1}{2}]}{}}}\left(\begin{array}{c}k\\ 2l+1\end{array}\right)(1\left({\displaystyle \frac{\mu }{m}}\right)^2)^{\frac{2l+1}{2}}]`$
This equation can be easily expanded in series of $`\mu ^2`$. Using
$`log\left({\displaystyle \frac{1+\sqrt{1(\mu /m)^2}}{2}}\right)`$ $`=`$ $`{\displaystyle \underset{j1}{}}{\displaystyle \frac{1}{2j}}{\displaystyle \frac{1}{2^{2j}}}\left(\begin{array}{c}2j\\ j\end{array}\right)\left({\displaystyle \frac{\mu }{m}}\right)^{2j}`$ (35)
$`\left(1\left({\displaystyle \frac{\mu }{m}}\right)^2\right)^{\frac{2l+1}{2}}`$ $`=`$ $`{\displaystyle \underset{j0}{}}()^j\left(\begin{array}{c}\frac{2l+1}{2}\\ j\end{array}\right)\left({\displaystyle \frac{\mu }{m}}\right)^{2j}`$ (38)
and the expansion in (22), finally we obtain
$`_{\chi =1}`$ $`=`$ $`{\displaystyle \frac{1}{2}}N_f[{\displaystyle \underset{j=1}{\overset{n+1}{}}}{\displaystyle \frac{g_j}{j}}m^j+Slog\left({\displaystyle \frac{2\mathrm{\Lambda }_0}{m}}\right)+{\displaystyle \underset{j=1}{\overset{[\frac{n+1}{2}]}{}}}{\displaystyle \frac{g_{2j}}{2j}}\left({\displaystyle \frac{\mu }{2}}\right)^{2j}\left(\begin{array}{c}2j\\ j\end{array}\right)`$ (53)
$`+\left({\displaystyle \underset{j=1}{\overset{[\frac{n+1}{2}]}{}}}g_{2j}\left({\displaystyle \frac{\mu }{2}}\right)^{2j}\left(\begin{array}{c}2j\\ j\end{array}\right)\right)\left({\displaystyle \underset{j1}{}}{\displaystyle \frac{1}{2j}}{\displaystyle \frac{1}{2^{2j}}}\left(\begin{array}{c}2j\\ j\end{array}\right)\left({\displaystyle \frac{\mu }{m}}\right)^{2j}\right)`$
$`+2{\displaystyle \underset{k=1}{\overset{n+1}{}}}\left({\displaystyle \underset{i=k}{\overset{n+1}{}}}{\displaystyle \frac{g_i}{i}}\left({\displaystyle \frac{\mu }{2}}\right)^i\left(\begin{array}{c}i\\ \frac{ik}{2}\end{array}\right)\right){\displaystyle \underset{l=0}{\overset{[\frac{k1}{2}]}{}}}\left(\begin{array}{c}k\\ 2l+1\end{array}\right)\left({\displaystyle \underset{j0}{}}()^j\left(\begin{array}{c}\frac{2l+1}{2}\\ j\end{array}\right)\left({\displaystyle \frac{\mu }{m}}\right)^{2jk}\right)]`$
Our final aim is to write $`_{\chi =1}`$ as a power series in $`S`$. In order to do so we have to work on (53) showing that it can be drastically simplified.
Let us consider a matrix model with $`N_f`$ flavors and a general superpotential for the chiral superfield in the adjoint representation of $`U(N)`$, given by the sum of an even polynomial $`W_{2n}`$ and an odd one $`W_{2n+1}`$ :
$$W(x)=W_{2n}(x)+W_{2n+1}(x)=\underset{j=1}{\overset{n}{}}\frac{g_{2j}}{2j}x^{2j}+\underset{j=1}{\overset{n}{}}\frac{g_{2j+1}}{2j+1}x^{2j+1}$$
(55)
We notice that, as explained in , the case $`W(x)=W_{2n}(x)`$ corresponds to a $`SU(N)`$ theory (for which only the even terms contribute to the glueball superpotential). The $`SU(N)`$ theory has been considered in , where the effective superpotential was computed order by order in $`S`$ using factorization properties of the Seiberg-Witten curve. Solving the general case in (55) we will be able to compare our results with those in .
First we rewrite (53) separating the even part from the odd one:
$`_{\chi =1}={\displaystyle \frac{1}{2}}N_f[{\displaystyle \underset{j=1}{\overset{n}{}}}{\displaystyle \frac{g_{2j}}{2j}}m^{2j}{\displaystyle \underset{j=1}{\overset{n}{}}}{\displaystyle \frac{g_{2j+1}}{2j+1}}m^{2j+1}+Slog\left({\displaystyle \frac{2\mathrm{\Lambda }_0}{m}}\right)+`$
$`+{\displaystyle \underset{j=1}{\overset{n}{}}}{\displaystyle \frac{g_{2j}}{2j}}\left({\displaystyle \frac{\mu }{2}}\right)^{2j}\left(\begin{array}{c}2j\\ j\end{array}\right)+\left({\displaystyle \underset{i=1}{\overset{n}{}}}g_{2i}\left({\displaystyle \frac{\mu }{2}}\right)^{2i}\left(\begin{array}{c}2i\\ i\end{array}\right)\right)\left({\displaystyle \underset{j1}{}}{\displaystyle \frac{1}{2j}}{\displaystyle \frac{1}{2^{2j}}}\left(\begin{array}{c}2j\\ j\end{array}\right)\left({\displaystyle \frac{\mu }{m}}\right)^{2j}\right)+`$ (62)
$`+2{\displaystyle \underset{k=1}{\overset{n}{}}}\left({\displaystyle \underset{i=k}{\overset{n}{}}}{\displaystyle \frac{g_{2i}}{2i}}\left({\displaystyle \frac{\mu }{2}}\right)^{2i}\left(\begin{array}{c}2i\\ ik\end{array}\right)\right){\displaystyle \underset{l=0}{\overset{k1}{}}}\left(\begin{array}{c}2k\\ 2l+1\end{array}\right)\left({\displaystyle \underset{j0}{}}()^j\left(\begin{array}{c}\frac{2l+1}{2}\\ j\end{array}\right)\left({\displaystyle \frac{\mu }{m}}\right)^{2(jk)}\right)+`$ (69)
$`+{\displaystyle \underset{k=0}{\overset{n}{}}}\left({\displaystyle \underset{i=k}{\overset{n}{}}}{\displaystyle \frac{g_{2i+1}}{2i+1}}\left({\displaystyle \frac{1}{2}}\right)^{2i}\left(\begin{array}{c}2i+1\\ ik\end{array}\right)\right){\displaystyle \underset{l=0}{\overset{k}{}}}\left(\begin{array}{c}2k+1\\ 2l+1\end{array}\right)`$ (74)
$`\left({\displaystyle \underset{j0}{}}()^j\left(\begin{array}{c}\frac{2l+1}{2}\\ j\end{array}\right)m^{2k2j+1}\mu ^{2(jk+i)}\right)]`$ (77)
In order to obtain $`_{\chi =1}`$ as a power series of $`S`$ we proceed in two steps. First we look for an expansion in $`\mu ^2`$. Then we will express $`\mu ^2`$ as a power series in $`S`$ by means of the formulas (16) and (26). To this end we organize eq. (S0.Ex7) in the following way:
$`_{\chi =1}={\displaystyle \frac{1}{2}}N_f[{\displaystyle \underset{j=1}{\overset{n}{}}}{\displaystyle \frac{g_{2j}}{2j}}m^{2j}{\displaystyle \underset{j=1}{\overset{n}{}}}{\displaystyle \frac{g_{2j+1}}{2j+1}}m^{2j+1}+Slog\left({\displaystyle \frac{2\mathrm{\Lambda }_0}{m}}\right)+`$
$`+{\displaystyle \underset{j=1}{\overset{n}{}}}A_j(\mu ^2)^j+{\displaystyle \underset{j2}{}}B_j(\mu ^2)^j+{\displaystyle \underset{j0}{}}C_j(\mu ^2)^j+{\displaystyle \underset{j0}{}}\stackrel{~}{C}_j(\mu ^2)^j]`$ (78)
where the seven terms correspond to the seven terms in (S0.Ex7).
The $`A_j`$ coefficients are immediately identified
$$A_j\frac{g_{2j}}{2j}\frac{1}{2^{2j}}\left(\begin{array}{c}2j\\ j\end{array}\right).$$
(79)
The $`B_j`$ coefficients are not too difficult to compute and one finds
$$B_j\underset{\begin{array}{c}i=1\\ in\end{array}}{\overset{j1}{}}\frac{g_{2i}}{2^{2j}}\left(\begin{array}{c}2i\\ i\end{array}\right)\left(\begin{array}{c}2(ji)\\ ji\end{array}\right)\frac{1}{2(ji)m^{2(ji)}}$$
(80)
On the other hand the computation of $`C_j`$ and $`\stackrel{~}{C}_j`$ is a more difficult task. Let us concentrate on them. In the last two terms in (S0.Ex7) we perform the change of index $`ii+k`$ and a shift $`jji`$. In this way we obtain
$`C_j`$ $``$ $`{\displaystyle \underset{k=1}{\overset{n}{}}}{\displaystyle \underset{\begin{array}{c}i=k\\ in\end{array}}{\overset{j+k}{}}}{\displaystyle \frac{g_{2i}}{2i}}\left(\begin{array}{c}2i\\ ik\end{array}\right){\displaystyle \frac{1}{2^{2i1}}}{\displaystyle \frac{()^{ji+k}}{m^{2(ji)}}}\left[{\displaystyle \underset{l=0}{\overset{k1}{}}}\left(\begin{array}{c}2k\\ 2l+1\end{array}\right)\left(\begin{array}{c}\frac{2l+1}{2}\\ ji+k\end{array}\right)\right]`$ (87)
$`\stackrel{~}{C}_j`$ $``$ $`{\displaystyle \underset{k=0}{\overset{n}{}}}{\displaystyle \underset{\begin{array}{c}i=k\\ in\end{array}}{\overset{j+k}{}}}{\displaystyle \frac{g_{2i+1}}{2i+1}}\left(\begin{array}{c}2i+1\\ ik\end{array}\right){\displaystyle \frac{()^{ji+k}}{2^{2i}m^{2(ji)1}}}\left[{\displaystyle \underset{l=0}{\overset{k}{}}}\left(\begin{array}{c}2k+1\\ 2l+1\end{array}\right)\left(\begin{array}{c}\frac{2l+1}{2}\\ ji+k\end{array}\right)\right]`$ (94)
The sum over $`l`$ in (87) and (94) can be rewritten in a nice form:
$$\underset{l=0}{\overset{k1}{}}\left(\begin{array}{c}2k\\ 2l+1\end{array}\right)\left(\begin{array}{c}\frac{2l+1}{2}\\ js+k\end{array}\right)2^{2(sj)}\frac{k}{(js+k)!}\frac{\mathrm{\Gamma }(k+sj)}{\mathrm{\Gamma }(2s2j+1)}$$
(95)
$$\underset{l=0}{\overset{k1}{}}\left(\begin{array}{c}2k+1\\ 2l+1\end{array}\right)\left(\begin{array}{c}\frac{2l+1}{2}\\ js+k\end{array}\right)\frac{2^{2k1}\sqrt{\pi }}{(js+k)!(2k)!}\frac{\mathrm{\Gamma }(2k+2)\mathrm{\Gamma }(2j2s1)}{\mathrm{\Gamma }(kj+s+\frac{3}{2})\mathrm{\Gamma }(2j2s2k1)}$$
(96)
We start computing the lowest terms:
$`C_0`$ $`=`$ $`{\displaystyle \underset{k=1}{\overset{n}{}}}{\displaystyle \frac{m^{2k}}{2^{2k1}}}{\displaystyle \frac{g_{2k}}{2k}}{\displaystyle \underset{l=0}{\overset{k1}{}}}\left(\begin{array}{c}2k\\ 2l+1\end{array}\right)={\displaystyle \underset{k=1}{\overset{n}{}}}m^{2k}{\displaystyle \frac{g_{2k}}{2k}}`$ (99)
$`\stackrel{~}{C}_0`$ $`=`$ $`{\displaystyle \underset{k=0}{\overset{n}{}}}{\displaystyle \frac{m^{2k+1}}{2^{2k}}}{\displaystyle \frac{g_{2k+1}}{2k+1}}{\displaystyle \underset{l=0}{\overset{k}{}}}\left(\begin{array}{c}2k+1\\ 2l+1\end{array}\right)={\displaystyle \underset{k=0}{\overset{n}{}}}m^{2k+1}{\displaystyle \frac{g_{2k+1}}{2k+1}}`$ (102)
In fact these coefficients are exactly cancelled by the first two terms in (S0.Ex8). Moreover we want to show that all the other coefficients $`C_j`$ and $`\stackrel{~}{C}_j`$, $`j0`$ receive contributions only from the first $`2j`$ and $`2j1`$ terms of the polynomial superpotential respectively.
Therefore we need to show that every term in (87) and (94) with an index $`i>j`$ does not contribute. Let us consider $`i=s`$ for some $`s>j`$. It is not hard to see that the contribution in (87) proportional to $`g_{2s}`$ can be written as a sum over $`k=sj,\mathrm{},s`$
$$g_{2s}\left[\underset{k=sj}{\overset{s}{}}\frac{1}{s}\left(\begin{array}{c}2s\\ sk\end{array}\right)\frac{1}{2^{2s}}\frac{()^{js+k}}{m^{2(js)}}\left(\underset{l=0}{\overset{k1}{}}\left(\begin{array}{c}2k\\ 2l+1\end{array}\right)\left(\begin{array}{c}\frac{2l+1}{2}\\ js+k\end{array}\right)\right)\right]$$
(103)
In the same way the contribution in (94) proportional to $`g_{2s+1}`$ becomes
$$g_{2s+1}\left[\underset{k=sj}{\overset{s}{}}\frac{1}{2s+1}\left(\begin{array}{c}2s+1\\ sk\end{array}\right)\frac{1}{2^{2s}}\frac{()^{js+k}}{m^{2(js)1}}\left(\underset{l=0}{\overset{k1}{}}\left(\begin{array}{c}2k+1\\ 2l+1\end{array}\right)\left(\begin{array}{c}\frac{2l+1}{2}\\ js+k\end{array}\right)\right)\right]$$
(104)
Now, setting $`tsj`$, $`t>0`$ (103) becomes proportional to
$`{\displaystyle \underset{k=0}{\overset{j}{}}}\left(\begin{array}{c}2(j+t)\\ jk\end{array}\right)()^k(k+t){\displaystyle \frac{\mathrm{\Gamma }(k+2t)}{\mathrm{\Gamma }(k+1)}}=`$ (107)
$`={\displaystyle \frac{4^jt\mathrm{\Gamma }(j+1/2)\mathrm{\Gamma }(2t)}{\sqrt{\pi }j\mathrm{\Gamma }(j)}}{\displaystyle \frac{\mathrm{\Gamma }(2j+1)\mathrm{\Gamma }(2t+1)}{2j^2\mathrm{\Gamma }^2(j)}}=`$
$`={\displaystyle \frac{2jt\mathrm{\Gamma }(2t)}{\sqrt{\pi }j^2\mathrm{\Gamma }^2(j)}}\left(2^{2j1}\mathrm{\Gamma }(j)\mathrm{\Gamma }(j+1/2)\sqrt{\pi }\mathrm{\Gamma }(2j)\right)0`$ (108)
In a similar manner from (104) we obtain
$$\underset{k=sj}{\overset{s}{}}\frac{()^k}{2^{2k}}\frac{\mathrm{\Gamma }(2k+2)}{(js+k)!\mathrm{\Gamma }(kj+s+\frac{3}{2})}\frac{\mathrm{\Gamma }(2j2s1)}{\mathrm{\Gamma }(2j2s2k1)(2k)!}$$
(109)
which is proportional to
$$4^j\mathrm{\Gamma }(j+\frac{1}{2})\mathrm{\Gamma }^2(j+1)2\mathrm{\Gamma }(j)\left[j\sqrt{\pi }(j2s2)\mathrm{\Gamma }(2j)+4^j(1+s)\mathrm{\Gamma }(j+\frac{1}{2})\mathrm{\Gamma }(j+1)\right]0$$
(110)
We have used repeatedly the following property of the $`\mathrm{\Gamma }`$ matrices (for integer $`j`$):
$$\mathrm{\Gamma }(j+1/2)\frac{\sqrt{\pi }}{2^{2j1}}\frac{\mathrm{\Gamma }(2j)}{\mathrm{\Gamma }(j)}$$
(111)
This completes our proof.
At this point we are left with:
$$C_j=\underset{s=1}{\overset{j}{}}c_{js}g_{2s}\stackrel{~}{C}_j=\underset{s=1}{\overset{j1}{}}\stackrel{~}{c}_{js}g_{2s+1}$$
where
$$c_{js}\underset{k=1}{\overset{s}{}}\frac{()^{js+k}}{2^{2s}sm^{2(js)}}\left(\begin{array}{c}2s\\ sk\end{array}\right)\underset{l=0}{\overset{k1}{}}\left(\begin{array}{c}2k\\ 2l+1\end{array}\right)\left(\begin{array}{c}\frac{2l+1}{2}\\ js+k\end{array}\right)$$
(112)
and
$$\stackrel{~}{c}_{js}\underset{k=1}{\overset{s}{}}\frac{()^{js+k}}{2^{2s}(2s+1)m^{2(js)}}\left(\begin{array}{c}2s+1\\ sk\end{array}\right)\underset{l=0}{\overset{k}{}}\left(\begin{array}{c}2k+1\\ 2l+1\end{array}\right)\left(\begin{array}{c}\frac{2l+1}{2}\\ js+k\end{array}\right)$$
(113)
Using (95) and (96) we find:
$`c_{jj}={\displaystyle \frac{1}{2^{2j}2j}}\left(\begin{array}{c}2j\\ j\end{array}\right)`$ (116)
$`c_{js}={\displaystyle \frac{1}{2^{2j1}jsm^{2(js)}}}{\displaystyle \frac{\mathrm{\Gamma }(2s)}{\mathrm{\Gamma }^2(s)}}\left(\begin{array}{c}2(js)1\\ js\end{array}\right)s<j`$ (119)
$`\stackrel{~}{c}_{js}={\displaystyle \frac{1}{2^{2j1}js^2m^{2(js)1}}}{\displaystyle \frac{\mathrm{\Gamma }(2s+1)}{\mathrm{\Gamma }^2(s)}}\left(\begin{array}{c}2(js)2\\ js1\end{array}\right)s<j`$ (122)
Note that the $`A_j`$ terms are exactly cancelled by $`c_{jj}`$ for $`j=1,\mathrm{}n`$ computed in (116), leaving no term linear in $`\mu ^2`$ (then no term linear in $`S`$ except for the standard piece $`Slog(2\mathrm{\Lambda }_0/m)`$). So we can rewrite (S0.Ex8) as follows:
$$_{\chi =1}=\frac{1}{2}N_f\left[Slog\left(\frac{2\mathrm{\Lambda }_0}{m}\right)+\underset{j2}{}(B_j+C_j^n+\stackrel{~}{C}_j)(\mu ^2)^j\right]$$
(123)
where we have defined
$$C_j^n\{\begin{array}{c}C_jg_{2j}c_{jj}jn\\ C_jj>n\end{array}$$
(124)
We give explicitly the form of the lowest terms for the coefficients $`B_j`$:
$`B_2`$ $`=`$ $`+{\displaystyle \frac{1}{8}}{\displaystyle \frac{g_2}{m^2}}`$
$`B_3`$ $`=`$ $`+{\displaystyle \frac{3}{64}}{\displaystyle \frac{g_2}{m^4}}+{\displaystyle \frac{3}{32}}{\displaystyle \frac{g_4}{m^2}}`$
$`B_4`$ $`=`$ $`+{\displaystyle \frac{5}{192}}{\displaystyle \frac{g_2}{m^6}}+{\displaystyle \frac{9}{256}}{\displaystyle \frac{g_4}{m^4}}+{\displaystyle \frac{5}{64}}{\displaystyle \frac{g_6}{m^2}}`$ (125)
and for the coefficients $`C_j`$ and $`\stackrel{~}{C}_j`$:
$`C_1`$ $`=`$ $`{\displaystyle \frac{g_2}{4}}`$
$`C_2`$ $`=`$ $`{\displaystyle \frac{g_2}{16m^2}}{\displaystyle \frac{3}{32}}g_4`$
$`C_3`$ $`=`$ $`{\displaystyle \frac{3}{96}}{\displaystyle \frac{g_2}{m^4}}{\displaystyle \frac{3}{96}}{\displaystyle \frac{g_4}{m^2}}{\displaystyle \frac{5}{96}}g_6`$
$`C_4`$ $`=`$ $`{\displaystyle \frac{5}{256}}{\displaystyle \frac{g_2}{m^6}}{\displaystyle \frac{9}{512}}{\displaystyle \frac{g_4}{m^4}}{\displaystyle \frac{5}{256}}{\displaystyle \frac{g_6}{m^2}}{\displaystyle \frac{35}{1024}}g_8`$ (126)
and
$`\stackrel{~}{C}_1`$ $`=`$ $`0`$
$`\stackrel{~}{C}_2`$ $`=`$ $`{\displaystyle \frac{g_3}{8m}}`$
$`\stackrel{~}{C}_3`$ $`=`$ $`{\displaystyle \frac{g_3}{24m^3}}{\displaystyle \frac{g_5}{16m}}`$
$`\stackrel{~}{C}_4`$ $`=`$ $`{\displaystyle \frac{3}{128}}{\displaystyle \frac{g_3}{m^5}}{\displaystyle \frac{3}{128}}{\displaystyle \frac{g_5}{m^3}}{\displaystyle \frac{5}{128}}{\displaystyle \frac{g_7}{m}}`$ (127)
As a final step we want to reexpress $`_{\chi =1}`$ in a power series of $`S`$. From (16) and (26) we know that
$$S=\frac{1}{2}\underset{j=1}{\overset{[\frac{n+1}{2}]}{}}g_{2j}\left(\frac{\mu }{2}\right)^{2j}\left(\begin{array}{c}2j\\ j\end{array}\right)$$
(128)
We want to invert this relation in order to obtain $`\mu ^2`$ in terms of $`S`$. We look for an expression of the form
$$\mu ^2=\underset{m>0}{}a_mS^m$$
(129)
where $`a_m`$ are the coefficients to be determined. Inserting (129) into (128) we obtain
$`S`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \underset{j=1}{\overset{[\frac{n+1}{2}]}{}}}{\displaystyle \frac{g_{2j}}{2^{2j}}}\left(\begin{array}{c}2j\\ j\end{array}\right)\left({\displaystyle \underset{m>0}{}}a_mS^m\right)^j`$ (132)
$`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \underset{j=1}{\overset{[\frac{n+1}{2}]}{}}}{\displaystyle \frac{g_{2j}}{2^{2j}}}\left(\begin{array}{c}2j\\ j\end{array}\right){\displaystyle \underset{\begin{array}{c}p_1,p_2,\mathrm{}=0\\ p_1+p_2+\mathrm{}=j\end{array}}{\overset{j}{}}}M(j;p_1,p_2,\mathrm{})a_1^{p_1}a_2^{p_2}\mathrm{}S^{p_1+2p_2+3p_3+\mathrm{}}`$ (135)
where we have defined the coefficient:
$$M(j;p_1,p_2,\mathrm{},p_t,\mathrm{})\frac{j!}{p_1!p_2!\mathrm{}p_t!\mathrm{}}$$
(136)
With a bit of labor one can argue that the general coefficient $`a_m`$ can be expressed recursively in terms of $`a_1,a_2,\mathrm{},a_{m1}`$:
$`a_1`$ $`=`$ $`{\displaystyle \frac{4}{g_2}}`$
$`a_m`$ $`=`$ $`{\displaystyle \underset{\begin{array}{c}p_1,\mathrm{},p_{m1}=0\\ p_1+2p_2+\mathrm{}+(m1)p_{m1}=m\\ p_1+p_2+\mathrm{}+p_{m1}[\frac{n+1}{2}]\end{array}}{\overset{m}{}}}{\displaystyle \frac{2}{g_2}}\left(\begin{array}{c}2(p_1+\mathrm{}+p_{m1})\\ p_1+\mathrm{}+p_{m1}\end{array}\right)a_1^{p_1}\mathrm{}a_{m1}^{p_{m1}}`$ (140)
$`{\displaystyle \frac{(p_1+\mathrm{}+p_{m1})!}{(p_1)!\mathrm{}(p_{m1})!}}{\displaystyle \frac{g_{2(p_1+\mathrm{}+p_{m1})}}{2^{2(p_1+\mathrm{}+p_{m1})}}}`$
Here we give the expressions of the first coefficients:
$`a_2`$ $`=`$ $`12{\displaystyle \frac{g_4}{g_2^3}}`$
$`a_3`$ $`=`$ $`+72{\displaystyle \frac{g_4^2}{g_2^5}}40{\displaystyle \frac{g_6}{g_2^4}}`$
$`a_4`$ $`=`$ $`540{\displaystyle \frac{g_4^3}{g_2^7}}+600{\displaystyle \frac{g_4g_6}{g_2^6}}140{\displaystyle \frac{g_8}{g_2^5}}`$
$`a_5`$ $`=`$ $`+4536{\displaystyle \frac{g_4^4}{g_2^9}}7560{\displaystyle \frac{g_4^2g_6}{g_2^8}}+2520{\displaystyle \frac{g_4g_8}{g_2^7}}+1200{\displaystyle \frac{g_6^2}{g_2^7}}504{\displaystyle \frac{g_{10}}{g_2^6}}`$ (141)
Now, using (129) and (S0.Ex19) we are able to give the expansion of $`_{\chi =1}`$ in series of $`S`$:
$$_{\chi =1}=\frac{1}{2}N_f\left[Slog\left(\frac{2\mathrm{\Lambda }_0}{m}\right)+\underset{k2}{}\left(\underset{j=2}{\overset{k}{}}(B_j+C_j^n+\stackrel{~}{C}_j)\underset{\begin{array}{c}p_1\mathrm{}p_k=0\\ p_1+\mathrm{}+p_k=j\\ p_1+\mathrm{}+kp_k=k\end{array}}{\overset{j}{}}M(j;p_1,\mathrm{},p_k)a_1^{p_1}\mathrm{}a_k^{p_k}\right)S^k\right]$$
$`=`$ $`N_f[{\displaystyle \frac{1}{2}}Slog\left({\displaystyle \frac{2\mathrm{\Lambda }_0}{m}}\right)+({\displaystyle \frac{1}{2g_2m^2}}{\displaystyle \frac{g_3}{g_2^2m}})S^2`$ (142)
$`+\left({\displaystyle \frac{1}{2g_2^2m^4}}{\displaystyle \frac{4}{3}}{\displaystyle \frac{g_3}{g_2^3m^3}}{\displaystyle \frac{g_4}{g_2^3m^2}}+6{\displaystyle \frac{g_3g_4}{g_2^4m}}2{\displaystyle \frac{g_5}{g_2^3m}}\right)S^3`$
$`+({\displaystyle \frac{5}{6}}{\displaystyle \frac{1}{g_2^3m^6}}3{\displaystyle \frac{g_3}{g_2^4m^5}}+{\displaystyle \frac{9}{2}}{\displaystyle \frac{g_4^2}{g_2^5m^2}}{\displaystyle \frac{9}{4}}{\displaystyle \frac{g_4}{g_2^4m^4}}+12{\displaystyle \frac{g_3g_4}{g_2^5m^3}}45{\displaystyle \frac{g_3g_4^2}{g_2^6m}}`$
$`+18{\displaystyle \frac{g_4g_5}{g_2^5m}}3{\displaystyle \frac{g_5}{g_2^4m^3}}{\displaystyle \frac{5}{2}}{\displaystyle \frac{g_6}{g_2^4m^2}}+20{\displaystyle \frac{g_3g_6}{g_2^5m}}5{\displaystyle \frac{g_7}{g_2^4m}})S^4+\mathrm{}]`$
If we extract from (142) the even contributions we reproduce exactly what has been computed in .
In order to have the full expression of the glueball superpotential (2) we have to add the $`\chi =2`$ contribution to (142). One can find the implicit solution to this problem in where the $`\chi =2`$ contribution is given in terms of some parameters which are non-linear functions of $`S`$. After some algebra we found that the explicit solution can be written as
$$N\frac{_{\chi =2}}{S}=NSlog\left(\frac{S}{g_2\mathrm{\Lambda }_0^2}\right)N\underset{j>0}{}(W_j\stackrel{~}{W}_j)S^j$$
(143)
where
$$W_j\underset{m=1}{\overset{j1}{}}\frac{()^{m+1}}{mD_1^m}\underset{\begin{array}{c}p_1,\mathrm{},p_{j1}=0\\ p_1+\mathrm{}+p_{j1}=m\\ p_1+\mathrm{}+(j1)p_{j1}=j1\end{array}}{\overset{m}{}}M(m;p_1,\mathrm{},p_{j1})D_2^{p_1}\mathrm{}D_j^{p_{j1}}$$
(144)
and
$`\stackrel{~}{W}_j{\displaystyle \underset{p=2}{\overset{2j}{}}}{\displaystyle \frac{g_p}{p}}{\displaystyle \underset{q=0}{\overset{[p/2]}{}}}\left(\begin{array}{c}p\\ 2q\end{array}\right)\left(\begin{array}{c}2q\\ q\end{array}\right){\displaystyle \underset{\begin{array}{c}p_1,\mathrm{},p_{j1}=0\\ p_1+\mathrm{}+p_{j1}=p2q\end{array}}{\overset{p2q}{}}}M(p2q;p_1,\mathrm{},p_{j1})z_1^{p_1}\mathrm{}z_{j1}^{p_{j1}}`$ (149)
$`{\displaystyle \underset{\begin{array}{c}q_1,\mathrm{},q_j=0\\ q_1+\mathrm{}+q_j=q\\ (p_1+q_1)+\mathrm{}+(j1)(p_{j1}+q_{j1})+jq_j=j\end{array}}{\overset{q}{}}}M(q;q_1,\mathrm{},q_j)D_1^{q_1}\mathrm{}D_j^{q_j}`$ (150)
are defined in terms of some parameters $`D_j`$ and $`z_j`$ which can be computed recursively
$$D_j\underset{k>0}{}d_k\underset{\begin{array}{c}p_1,\mathrm{},p_j=0\\ p_1+\mathrm{}+p_j=k\\ p_1+\mathrm{}+jp_j=j\end{array}}{\overset{k}{}}M(k;p_1,\mathrm{},p_j)z_1^{p_1}\mathrm{}z_j^{p_j}$$
(151)
and
$`z_1=2{\displaystyle \frac{g_3}{g_2^2}}`$
$`z_j{\displaystyle \frac{1}{g_2d_1}}{\displaystyle \underset{p=2}{\overset{2j}{}}}{\displaystyle \frac{g_p}{p}}{\displaystyle \underset{q=1}{\overset{[p/2]}{}}}q\left(\begin{array}{c}p\\ 2q\end{array}\right)\left(\begin{array}{c}2q\\ q\end{array}\right){\displaystyle \underset{\begin{array}{c}p_1,\mathrm{},p_j=0\\ p_1+\mathrm{}+p_j=q\end{array}}{\overset{q}{}}}M(q;p_1,\mathrm{},p_j)d_1^{p_1}\mathrm{}d_j^{p_j}`$ (156)
$`{\displaystyle \underset{\begin{array}{c}q_1,\mathrm{},q_{j1}=0\\ q_1+\mathrm{}+q_{j1}=p2q+p_1+\mathrm{}+jp_j\\ q_1+\mathrm{}+(j1)q_{j1}=j\end{array}}{\overset{p2q+p_1+\mathrm{}+jp_j}{}}}M(p2q+p_1+\mathrm{}+jp_j;q_1,\mathrm{},q_{j1})z_1^{q_1}\mathrm{}z_{j1}^{q_{j1}}`$ (157)
The $`d_j`$ coefficients which appear in (69) and (70) are also given recursively:
$`d_1={\displaystyle \frac{g_2}{2g_3}}`$
$`d_j{\displaystyle \frac{1}{2g_3}}{\displaystyle \underset{p=3}{\overset{2j}{}}}g_{p+1}{\displaystyle \underset{q=0}{\overset{[p/2]}{}}}\left(\begin{array}{c}p\\ 2q\end{array}\right)\left(\begin{array}{c}2q\\ q\end{array}\right){\displaystyle \underset{\begin{array}{c}p_1,\mathrm{},p_{j1}=0\\ p_1+\mathrm{}+p_{j1}=q\\ p2q+p_1+\mathrm{}+(j1)p_{j1}=j\end{array}}{\overset{q}{}}}M(q;p_1,\mathrm{},p_{j1})d_1^{p_1}\mathrm{}d_{j1}^{p_{j1}}`$ (162)
Explicitly to the fourth order in $`S`$ we obtain
$`N{\displaystyle \frac{_{\chi =2}}{S}}=NS(log\left({\displaystyle \frac{S}{g_2\mathrm{\Lambda }_0^2}}\right)1)N[(2{\displaystyle \frac{g_3^2}{g_2^3}}{\displaystyle \frac{3}{2}}{\displaystyle \frac{g_4}{g_2^2}})S^2+({\displaystyle \frac{32}{3}}{\displaystyle \frac{g_3^4}{g_2^6}}24{\displaystyle \frac{g_3^2g_4}{g_2^5}}+`$
$`+{\displaystyle \frac{9}{2}}{\displaystyle \frac{g_4^2}{g_2^4}}+12{\displaystyle \frac{g_3g_5}{g_2^4}}{\displaystyle \frac{10}{3}}{\displaystyle \frac{g_6}{g_2^3}})S^3+({\displaystyle \frac{280}{3}}{\displaystyle \frac{g_3^6}{g_2^9}}340{\displaystyle \frac{g_3^4g_4}{g_2^8}}+270{\displaystyle \frac{g_3^2g_4^2}{g_2^7}}{\displaystyle \frac{45}{2}}{\displaystyle \frac{g_4^3}{g_2^6}}+`$
$`+200{\displaystyle \frac{g_3^3g_5}{g_2^7}}180{\displaystyle \frac{g_3g_4g_5}{g_2^6}}+18{\displaystyle \frac{g_5^2}{g_2^5}}100{\displaystyle \frac{g_3^2g_6}{g_2^6}}+30{\displaystyle \frac{g_4g_6}{g_2^5}}+40{\displaystyle \frac{g_3g_7}{g_2^5}}{\displaystyle \frac{35}{4}}{\displaystyle \frac{g_8}{g_2^4}})S^4+\mathrm{}]`$ (164)
Let us notice that the $`S^k`$ term in this expression depends only on the low order couplings $`g_3,g_4,\mathrm{},g_{2k}`$, in a way parallel to what we have found for the case of fundamental matter (see (142)). When only an adjoint matter field is present this behavior follows immediately from the perturbative analysis in . There it was found that contributions to the $`S^k`$ term come from a planar diagram with exactly $`k`$ loops, which can contain vertices with at most $`2k`$ external legs.
What we found in (142) is the corresponding situation where now the graphs have an extra loop of fundamental matter. It is very easy to write down the graphs corresponding to every single term in the expressions (142) and (S0.Ex29) since the factors of $`g_2`$ and $`m`$ in the denominators count the number of adjoint and fundamental propagators respectively.
To summarize, we have computed the effective glueball superpotential for a $`𝒩=1`$ $`U(N)`$ gauge theory with matter in the fundamental ($`N_f`$ flavors) and a field in the adjoint with a general polynomial tree-level superpotential. This has been done by using the technology developed in for computing the contribution to the matrix model partition function coming from diagrams with the disk topology. The full contribution to the glueball superpotential is given in the expressions (142) and (143).
Our results generalize the ones obtained in since as discussed in the $`SU(N)`$ theory considered there is equivalent to a $`U(N)`$ model with an even polynomial superpotential for the adjoint field.
It would be interesting to extend these results to the case of a superpotential admitting several vacua.
Acknowledgements
This work has been partially supported by INFN, MURST and the European Commission RTN program MRTN-CT-2004-005104 in which the authors are associated to the University of Milano-Bicocca.
|
warning/0506/astro-ph0506136.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
In a scattering medium, the Cherenkov photons emitted along the trajectory of a relativistic charged particle (or resulting from electromagnetic or hadronic showers of neutrino neutral and charged current interactions) and propagating to some distance are delayed relatively to the direct geometrical path. The expression for the time delay distribution of a photon reaching a detecting device is the essential ingredient in reconstructing events from high energy neutrino telescopes and is used to calculate the likelihood of a given event hypothesis.
In this letter, we consider a monochromatic and isotropic point-like emitter-receiver problem: given a uniform medium characterized by scattering and absorbing optical properties and a gaussian time measurement uncertainty, we derive the arrival time delay distribution of the detected photons. The result is presented in a closed analytical form.
In the following section, we review the photon propagation function in a uniform medium and show that it satisfies the expected limit of a zero time delay when there is neither scattering nor absorption, or when the emitter-receiver distance is zero.
In section $`\mathrm{𝟑}`$, we derive the probability density function (PDF) for a realistic setup, i.e. the detected photons are measured with a finite time resolution. We also discuss the properties of this function, useful in a numerical implementation.
We conclude this letter with a discussion, presented in section $`\mathrm{𝟒}`$.
## 2 Photon propagation function $`p`$
Scattering in a medium causes a photon arrival time delay $`t`$ defined as a difference between the actual arrival time at the receiver, $`t_{arrival}`$, and $`t_{geom}`$, the arrival time from the path of a photon propagating in a medium without scattering:
$$tt_{arrival}t_{geom}=t_{arrival}\frac{R}{c}$$
(1)
where $`R`$ is the distance between the emitter and the detection point, and $`c`$ is the light speed in the medium.
Our goal is to determine the PDF describing the distribution of time delay $`t`$ of photons detected at the distance $`R`$ from the emitter by the device with a finite time resolution.
Instead of attempting the impregnable task of deriving the PDF from first principles, one might exploit the symmetry of the problem. This suggests the following equation for the photon propagation function $`p(R,t)`$ :
$$p(R_1+R_2,t)=_0^t𝑑t^{}p(R_1,t^{})p(R_2,tt^{})$$
(2)
which describes the picture for $`p`$, analogous to the Huygens principle for waves in optics. Eq. (2) relies on the spherical symmetry of the medium and of the light source. $`p(R_1+R_2,t)`$ is the PDF corresponding to a photon propagation with a time delay $`t`$ to a distance $`R_1+R_2`$ from the source (located at the origin). It can be expressed as the convolution of the PDF to propagate to the shell at distance $`R_1`$ from the origin with the PDF to propagate from this shell up to the shell at distance $`R_1+R_2`$ from the origin. Once multiplied with an exponential damping factor, accounting for absorption and depending on the time delay, the normalized solution of Eq. (2) can be presented as
$$p(\xi ,\rho ,t)=\frac{\rho ^\xi t^{\xi 1}}{\mathrm{\Gamma }(\xi )}e^{\rho t},_0^{\mathrm{}}𝑑tp(\xi ,\rho ,t)=\mathrm{\hspace{0.17em}1}$$
(3)
where
$$\xi \frac{R}{\lambda },\rho \frac{1}{\tau }+\frac{c}{\lambda _a}$$
(4)
and $`\mathrm{\Gamma }(\xi )`$ is the gamma function . In (4), $`\lambda _a`$ is the absorption length, $`\lambda `$ and $`\tau `$ are phenomenological parameters. When the photon propagation function $`p`$ is derived from first principles, $`\lambda `$ and $`\tau `$ are expressed in terms of the optical parameters of the medium and the differential cross section of the microscopic photon-medium interaction. As they stand in (3, 4), these parameters must be extracted from the experimental data.
Clearly, for $`R0`$, or for any distance without scattering, there should be no time delay, i.e. $`p`$ must be concentrated near $`t=0`$. These two regimes can be realized with the single requirement $`\xi 0`$, provided that in (3, 4) $`\lambda `$ is the effective scattering length in the medium (which is related to the mean free path and to the scattering profile in a way that when the effective scattering length increases, so does the mean free path).
Examining the case $`\xi 0`$ shows that $`lim_{\xi 0}p(\xi ,\rho ,t)`$ does not exist - $`p`$ is recognized as a generalized function with the kernel $`\gamma _+(x)x_+^{\xi 1}/\mathrm{\Gamma }(\xi )`$ . As is well known, generalized functions and their limits are not well defined for any values of their variables and parameters; the self-consistent approach is to consider $`(p,\phi )p\phi `$ instead of $`p`$, where $`\phi (t)`$ is a test function falling off at $`t\pm \mathrm{}`$ fast enough, ensuring the finiteness of $`(p,\phi )`$ .
In terms of $`(p,\phi )`$, the condition $`t=0`$ for no scattering and/or zero distance can be readily verified: for any analytic function $`\phi (t)`$, we have
$$\underset{\xi 0}{lim}(p,\phi )=\underset{\xi 0}{lim}_0^{\mathrm{}}𝑑tp(\rho ,\xi ,t)\phi (t)=\underset{\xi 0}{lim}\underset{n=0}{\overset{\mathrm{}}{}}\frac{\phi ^{(n)}(0)}{n!}_0^{\mathrm{}}𝑑tp(\xi ,\rho ,t)t^n=\phi (0)$$
(5)
meaning that $`p`$ has the limit (in the sense of generalized functions )
$$\underset{\xi 0}{lim}p(\xi ,\rho ,t)=\delta _+(t)$$
(6)
In (6), $`\delta _+(t)`$ is a generalized function, filtering out the point $`t=+0`$ \- origin on the $`t`$ -axis, approached from the region $`t>\mathrm{\hspace{0.17em}0}`$ (propagation time delay cannot be negative - $`p`$ is defined on the right semi-axis $`t0`$). Eq. (6) supports our interpretation of $`\lambda `$ as the effective scattering length in a medium.
The function $`p`$ describes just the propagation in the medium, and can be interpreted as the PDF for the measured photon arrival time assuming an ideal detecting device.
## 3 Photon arrival time PDF $`_\sigma `$
A PDF for realistic signals should account not only for the photon propagation and the properties of the medium, but also for the features of the detector. Here, we consider the finite time resolution of the detector when measuring a signal. The function $`p`$ described above can be interpreted as a limit corresponding to an ideal measurement. The realistic PDF is thus constructed in terms of $`p`$ and $`\phi `$, the function simulating the detector time resolution distribution.
We choose a gaussian distribution as a “detector function” $`\phi (t)`$. The width $`\sigma 0`$ of the gaussian distribution represents the time resolution and should be extracted from the characteristics of the detector. $`\sigma `$ is usually a composite from numerous sources affecting the final detector time resolution (which further motivates the use of a gaussian).
Considering a detector with a finite time resolution instead of an ideal one is satisfactory from the physical point of view. We will see that this also removes the divergence at $`t=0`$: the PDF accounting for a finite time resolution is an ordinary function which exists for all $`R,t`$.
### 3.1 Definition of $`_\sigma `$
From the fact that a realistic setup will allow only the observation of the sum of the two independent random variables, the photon time delay and the detector time resolution, it follows that the photon time residual PDF $`_\sigma `$ is the convolution of the photon propagation function $`p`$ with the detector time jitter function $`g_\sigma `$:
$$_\sigma (\xi ,\rho ,t)_0^{\mathrm{}}𝑑t^{}p(\xi ,\rho ,t^{})g_\sigma (tt^{})$$
(7)
where
$$g_\sigma (t)=\frac{\mathrm{exp}(t^2/2\sigma ^2)}{\sqrt{2\pi \sigma ^2}}$$
(8)
incorporates the finite time resolution $`\sigma 0`$.
Since (7) is of the form $`(p,\phi )`$, $`_\sigma `$ is not a generalized function, i.e. it exists for any $`R`$, $`t`$.
It should be mentioned that when the time resolution is accounted for, the definition of the time residual $`t`$ is slightly changed: it does not refer anymore to the photon arrival time but to the device trigger time - the arrival time corrected with a time jitter. Therefore $`t<0`$ is acceptable for $`_\sigma `$, in contrast to $`p`$, where the very definition of the time delay demands to consider only $`t0`$.
Now, the time residual $`t`$ can also be negative, and the normalization condition
$$_{\mathrm{}}^{\mathrm{}}𝑑t_\sigma (\rho ,\xi ,t)=_0^{\mathrm{}}𝑑t^{}p(\rho ,\xi ,t^{})=\mathrm{\hspace{0.17em}1}$$
(9)
is readily verified.
From the physical point of view, negative time residuals do not contradict causality if $`_\sigma `$ decreases faster than $`g_\sigma `$, when $`t<0`$. In other words, causality is preserved when $`g_\sigma `$ envelops $`_\sigma `$ for negative $`t`$. In the next subsection, we show that this is the case.
The integration in (7) can be carried out and we obtain:
$$_\sigma (\xi ,\rho ,t)=\frac{(\rho \sigma )^\xi }{\sqrt{2\pi \sigma ^2}}\mathrm{exp}\left(\frac{\rho ^2\sigma ^2}{4}\frac{\rho t}{2}\frac{t^2}{4\sigma ^2}\right)D_\xi (\eta )$$
(10)
where $`D`$ is the function of parabolic cylinder and
$$\eta \rho \sigma t/\sigma $$
For implementation purposes, it is convenient to express $`_\sigma `$ in terms of the confluent hypergeometric function $`{}_{1}{}^{}F_{1}^{}`$ :
$`_\sigma (\xi ,\rho ,t)`$ $`=`$ $`{\displaystyle \frac{(\rho \sigma )^\xi \mathrm{\hspace{0.17em}2}^{(\xi 3)/2}}{\sqrt{\pi }\mathrm{\Gamma }(\xi )}}{\displaystyle \frac{e^{t^2/2\sigma ^2}}{\sigma }}\left[\mathrm{\Gamma }\right({\displaystyle \frac{\xi }{2}}\left)_1F_1\right({\displaystyle \frac{\xi }{2}},{\displaystyle \frac{1}{2}};{\displaystyle \frac{1}{2}}\eta ^2)`$ (11)
$``$ $`\sqrt{2}\eta \mathrm{\Gamma }\left({\displaystyle \frac{\xi +\mathrm{\hspace{0.17em}1}}{2}}\right)_1F_1({\displaystyle \frac{\xi +\mathrm{\hspace{0.17em}1}}{2}},{\displaystyle \frac{3}{2}};{\displaystyle \frac{1}{2}}\eta ^2)]`$ (12)
### 3.2 Properties and moments of $`_\sigma `$
The confluent hypergeometric function $`{}_{1}{}^{}F_{1}^{}`$ is implemented in GSL, the GNU Scientific Library . Here we cite a few useful properties of $`_\sigma `$:
$``$ Large positive time residuals: $`t\sigma +\rho \sigma ^2`$. From the asymptotic for $`{}_{1}{}^{}F_{1}^{}`$ , we obtain:
$`_\sigma `$ $`=`$ $`e^{\rho ^2\sigma ^2/2}p(\xi ,\rho ,t)(1{\displaystyle \frac{\rho \sigma ^2}{t}})^{\xi 1}[1+{\displaystyle \frac{\sigma ^2(1\xi )(2\xi )}{t^2}}(1{\displaystyle \frac{\rho \sigma ^2}{t}})^2+`$ (13)
$`+`$ $`𝒪\left({\displaystyle \frac{\sigma ^4}{t^4}}(1{\displaystyle \frac{\rho \sigma ^2}{t}})^4\right)]`$ (14)
$``$ Behavior for negative $`t<\mathrm{\hspace{0.17em}0}`$. From the expansion of the function of parabolic cylinder $`D`$ , we obtain for $`|t|\sigma `$:
$$_\sigma (\rho ,\xi ,t)\frac{(\rho \sigma )^\xi }{\sqrt{2\pi \sigma ^2}}\left(\rho \sigma \frac{t}{\sigma }\right)^\xi \mathrm{exp}(t^2/2\sigma ^2)\left[1+𝒪\left(\rho \sigma \frac{t}{\sigma }\right)^1\right]$$
(15)
Now, we are in the position to verify that causality is not violated: considering the ratio of $`_\sigma `$ to $`g_\sigma `$, we find that indeed $`_\sigma `$ decreases faster for negative time residual than the detector time jitter function:
$$\frac{_\sigma (t)}{g_\sigma (t)}\left(1\frac{t}{\rho \sigma ^2}\right)^\xi \mathrm{\hspace{0.17em}1}$$
(16)
$``$ $`\sigma \mathrm{\hspace{0.17em}1}`$: using the relation $`lim_{\sigma 0}\mathrm{exp}(x^2/2\sigma ^2)/\sqrt{2\pi \sigma ^2}=\delta (x)`$, we recover $`p`$ in the zero time resolution limit:
$$\underset{\sigma 0}{lim}_\sigma (\xi ,\rho ,t)=p(\xi ,\rho ,t)$$
(17)
$``$ $`_\sigma `$ is finite for $`t=0`$ and any $`R`$; e.g. for $`R=0`$ ($`\xi =0`$)
$$_\sigma (0,\rho ,\mathrm{\hspace{0.17em}0})=\frac{1}{\sqrt{2\pi \sigma ^2}};\underset{\xi 0,\sigma 0}{lim}_\sigma (\xi ,\rho ,t)=\delta (t)$$
(18)
$``$ Long distance: for $`\xi \mathrm{\hspace{0.17em}1}`$ we have:
$$_\sigma (\xi ,\rho ,t)=\frac{(\rho \sigma )^\xi \xi ^{\xi /2}}{2\sigma \sqrt{\pi }}exp\left(\frac{\rho ^2\sigma ^2}{4}\frac{\rho t}{2}\frac{t^2}{4\sigma ^2}+\frac{\xi }{2}\eta \sqrt{\xi }\right)\left[1+𝒪(\xi ^{1/2})\right]$$
(19)
$``$ Short distance: in the limit $`\xi \mathrm{\hspace{0.17em}1}`$, the receiver time response function survives:
$`_\sigma (\xi ,\rho ,t)={\displaystyle \frac{\mathrm{exp}(t^2/2\sigma ^2)}{\sqrt{2\pi \sigma ^2}}}\left[1+{\displaystyle \frac{\xi }{2}}\left(2\sqrt{\pi }\mathrm{ln}(\rho \sigma )+\gamma _E+2\mathrm{ln}2\pi erfi\left({\displaystyle \frac{\eta }{\sqrt{2}}}\right)\right)+𝒪(\xi ^2)\right]`$ (20)
where $`erfi(z)`$ is an error function of the imaginary argument, $`erfi(z)\frac{2}{\sqrt{\pi }}_0^z𝑑xe^{x^2}`$, and $`\gamma _E0.57721`$ is the Euler’s constant .
These properties are useful to implement algorithms and to avoid numerical overflows. E.g., for large $`t`$, the calculation speed, as implemented in GSL , of $`{}_{1}{}^{}F_{1}^{}(a,b;x)`$ slows down with increasing $`x`$ and numerical overflow may occur. This can be circumvented by switching to the asymptotic behavior of $`_\sigma `$ for large $`t`$. The case of large $`\xi `$ can be treated using (19) to avoid numerical overflow caused by $`\mathrm{\Gamma }(\xi )`$, appearing in (12) and in the expression for the cumulative of $`_\sigma `$ (see below).
Also interesting are the moments of $`_\sigma `$:
$$t^N_{\mathrm{}}^{\mathrm{}}𝑑tt^N_\sigma (\rho ,\xi ,t)$$
(21)
which can be calculated in a closed analytical form; we explicitly present here the first two - the time residual average and the corresponding dispersion:
$$t=\frac{\xi }{\rho },(\mathrm{}t)^2t^2t^2=\sigma ^2+\frac{\xi }{\rho ^2}$$
(22)
As expected, only the detector time resolution survives in the variance at $`\xi =0`$.
### 3.3 Cumulative of $`_\sigma `$
$`_\sigma `$ is a PDF for a single detected photon. Another object of interest is the time delay probability density of the first photon in a sequence of $`N`$ detected photons. It is given by
$$_\sigma ^{(N)}(\rho ,\xi ,t)=N_\sigma (\rho ,\xi ,t)\left(1𝒞_{}(\rho ,\xi ,t)\right)^{N1}$$
(23)
where $`𝒞_{}`$ is the cumulative probability:
$$𝒞_{}(\rho ,\xi ,t)_{\mathrm{}}^t𝑑t_\sigma (\rho ,\xi ,t)$$
(24)
This function appears in the expression for the probability $`\omega (\rho ,\xi ,\sigma ,t;ϵ)`$ that the time residual lies between $`t`$ and $`t+ϵ`$:
$$\omega (\rho ,\xi ,\sigma ,t;ϵ)_t^{t+ϵ}𝑑t^{}_\sigma (\rho ,\xi ,t^{})=𝒞_{}(\rho ,\xi ,t+ϵ)𝒞_{}(\rho ,\xi ,t)$$
(25)
Note that since $`_\sigma `$ is a smooth function defined everywhere, for $`ϵt`$ the probability above can be approximated by $`_\sigma `$:
$$\omega (\rho ,\xi ,\sigma ,t;ϵ)ϵ_\sigma (\rho ,\xi ,t+ϵ/2)$$
(26)
This prescription cannot be used in the case $`\sigma =0`$. Instead, the analytical solution for the cumulative distribution $`𝒞_p`$, given in terms of incomplete gamma function $`\mathrm{\Gamma }(\xi ,\rho t)`$, must be used:
$$𝒞_p_0^t𝑑t^{}p(\rho ,\xi ,t^{})=1\mathrm{\Gamma }(\xi ,\rho t)/\mathrm{\Gamma }(\xi )$$
The analytical expression for $`𝒞_{}`$ with an arbitrary $`\xi `$ cannot be obtained.
From (12), it follows that the cumulative distribution can be expressed as an integral involving the error function:
$`𝒞_{}(\rho ,\xi ,t)={\displaystyle \frac{1}{2}}+{\displaystyle \frac{\rho ^\xi }{2\mathrm{\Gamma }(\xi )}}{\displaystyle _0^{\mathrm{}}}𝑑yy^{\xi 1}e^{\rho y}erf\left({\displaystyle \frac{ty}{\sqrt{2\sigma ^2}}}\right)`$ (27)
When $`\xi =\mathrm{\hspace{0.17em}1}`$ the result is
$$𝒞_{}(\rho ,\xi ,t)|_{\xi =1}=\frac{1}{2}+\frac{1}{2}\left[erf\left(\frac{t}{\sqrt{2\sigma ^2}}\right)\mathrm{exp}(\rho ^2\sigma ^2/2\rho t)\left(1erf\left(\rho \sqrt{\frac{\sigma ^2}{2}}\frac{t}{\sqrt{2\sigma ^2}}\right)\right)\right]$$
(28)
The case of integer $`\xi `$ can be handled using
$$_0^{\mathrm{}}𝑑yy^ne^{\rho y}erf(cy+b)=(1)^n\frac{^n}{\rho ^n}\frac{1}{\rho }\left[erf(b)+\mathrm{exp}((\rho ^2+4bc)/4c^2)\left(1erf\left(b+\frac{\rho }{2c}\right)\right)\right]$$
(29)
An analytic continuation for a fractional order derivative is possible , but not practical as higher hypergeometric functions emerge. In the Appendix, we give a simple semi-numerical approximation valid for any $`\xi 0`$.
In , further examples involving the cumulative distribution are presented.
## 4 Discussion
In this letter, we suggested and calculated the arrival time probability density for a photon at some distance from an isotropic point-like source in a uniform scattering medium for a realistic detector. $`_\sigma `$, accounting for a finite detector time resolution $`\sigma 0`$, is free from the spurious singularities which appear in the generalized function $`p`$ (the unrealistic zero time resolution limit of $`_\sigma `$) and is defined for all $`t`$ and $`R0`$. The expression for $`_\sigma `$ has been obtained in a closed analytic form (10, 12) and contains parameters characterizing the medium and the detector. We further discussed properties of $`_\sigma `$, useful for a numerical implementation. Finally, we calculated semi-analytically the corresponding cumulative probability entering multi-photon PDF’s in order to provide a usable numerical implementation scheme.
Acknowledgments. This research was supported by the National Science Foundation (NSF-G067771) and the Swiss National Foundation (PA002-104970). We would like to thank A. Bouchta, T. Castermans, F. Grard, P. Herquet, R. Porrata and C. Wiebusch for valuable comments.
## Appendix: Numerical integration scheme for $`𝒞_{}`$
We sketch the procedure for the calculation of the cumulative $`𝒞_{}`$.
To avoid numerical integration from $`0`$ to $`\mathrm{}`$ in (27), we use the asymptotic properties of the error function , and split the integral on three subintervals: $`(0,\mathrm{})=(0,y_{\mathrm{min}})(y_{\mathrm{min}},y_{\mathrm{max}})(y_{\mathrm{max}},\mathrm{})`$. The parameter $`\alpha `$ in $`y_{\mathrm{min}}=\mathrm{max}(0,t\alpha \sqrt{2\sigma ^2})`$ and $`y_{\mathrm{max}}=\mathrm{max}(0,t+\alpha \sqrt{2\sigma ^2})`$ is chosen from the condition $`\frac{|ty|}{\sqrt{2\sigma ^2}}>\alpha `$, leading to $`erf((ty)/\sqrt{2\sigma ^2})\pm 1`$. E.g., choosing $`\alpha =\mathrm{\hspace{0.17em}2.5}`$ results in an error less than $`0.041`$%.
The integration on the first and third subintervals, where $`erf=\pm 1`$, can be carried out and we are left with an integration over the finite interval $`(y_{\mathrm{min}},y_{\mathrm{max}})`$:
$$𝒞_{}(\rho ,\xi ,t)=1\frac{\mathrm{\Gamma }(\xi ,\rho y_{\mathrm{max}})+\mathrm{\Gamma }(\xi ,\rho y_{\mathrm{min}})}{2\mathrm{\Gamma }(\xi )}+J;J\frac{\rho ^\xi }{2\mathrm{\Gamma }(\xi )}_{y_{\mathrm{min}}}^{y_{\mathrm{max}}}𝑑yy^{\xi 1}e^{\rho y}erf\left(\frac{ty}{\sqrt{2\sigma ^2}}\right)$$
where $`\mathrm{\Gamma }(a,x)_x^{\mathrm{}}𝑑zz^{a1}e^z`$, the incomplete gamma function , is implemented in GSL .
Integral with finite limits $`J`$ is finite and can in principle be calculated numerically, but when $`\xi <\mathrm{\hspace{0.17em}1}`$, numerical integration algorithms may become unstable when $`y_{\mathrm{min}}0`$. This can be circumvented by handling $`J`$ separately for the two cases:
1. $`\xi >\mathrm{\hspace{0.17em}1}`$.
$`J`$ is calculated numerically. Since $`y_{\mathrm{min}}`$ is positive and $`\xi >\mathrm{\hspace{0.17em}1}`$, there is no problem with the convergence; in this case, a standard integration algorithm can be applied.
2. $`\xi <1`$.
When $`y_{\mathrm{min}}\mathrm{\hspace{0.17em}1}`$, $`y^\xi `$ is an integrable singularity in the region $`yy_{\mathrm{min}}`$. In order to isolate explicitly the terms $`y_{\mathrm{min}}^\xi `$, a small finite $`\epsilon `$, $`y_{\mathrm{min}}<\epsilon <y_{\mathrm{max}}`$, is introduced and the integration is performed analytically in the range $`y_{\mathrm{min}}<y<\epsilon `$. Expanding $`\mathrm{exp}(\rho y)`$ and $`erf((ty)/\sqrt{2\sigma ^2})`$ in $`y`$, we obtain:
$`J`$ $`=`$ $`{\displaystyle \frac{\rho ^\xi }{2\mathrm{\Gamma }(\xi )}}\left[{\displaystyle _{y_{\mathrm{min}}}^ϵ}+{\displaystyle _\epsilon ^{y_{\mathrm{max}}}}\right]dyy^{\xi 1}e^{\rho y}erf\left({\displaystyle \frac{ty}{\sqrt{2\sigma ^2}}}\right)=`$
$`=`$ $`{\displaystyle \frac{\rho ^\xi }{2\mathrm{\Gamma }(\xi )}}[{\displaystyle _\epsilon ^{y_{\mathrm{max}}}}dyy^{\xi 1}e^{\rho y}erf\left({\displaystyle \frac{ty}{\sqrt{2\sigma ^2}}}\right)+`$
$`+`$ $`{\displaystyle _{y_{\mathrm{min}}}^ϵ}dyy^{\xi 1}(1\rho y)(erf\left({\displaystyle \frac{t}{\sqrt{2\sigma ^2}}}\right)ye^{t^2/2\sigma ^2}\sqrt{{\displaystyle \frac{2}{\pi \sigma ^2}}})]=`$
$`=`$ $`{\displaystyle \frac{\rho ^\xi }{2\mathrm{\Gamma }(\xi )}}[{\displaystyle _\epsilon ^{y_{\mathrm{max}}}}dyy^{\xi 1}e^{\rho y}erf\left({\displaystyle \frac{ty}{\sqrt{2\sigma ^2}}}\right)+`$
$`+`$ $`erf\left({\displaystyle \frac{t}{\sqrt{2\sigma ^2}}}\right){\displaystyle \frac{\epsilon ^\xi y_{\mathrm{min}}^\xi }{\xi }}(\sqrt{{\displaystyle \frac{2}{\pi \sigma ^2}}}e^{t^2/2\sigma ^2}+`$
$`+`$ $`\rho erf\left({\displaystyle \frac{t}{\sqrt{2\sigma ^2}}}\right)){\displaystyle \frac{\epsilon ^{\xi +1}y_{\mathrm{min}}^{\xi +1}}{\xi +1}}+O(\epsilon ^{\xi +2})]`$ (31)
and the remaining integral
$$_\epsilon ^{y_{\mathrm{max}}}𝑑yy^{\xi 1}e^{\rho y}erf((ty)/\sqrt{2\sigma ^2})$$
can be calculated numerically using standard integration algorithms.
|
warning/0506/cond-mat0506026.html
|
ar5iv
|
text
|
# Viscoelasticity and metastability limit in supercooled liquids
(June 1, 2005)
## Abstract
A supercooled liquid is said to have a kinetic spinodal if a temperature $`T_{\mathrm{sp}}`$ exists below which the liquid relaxation time exceeds the crystal nucleation time. We revisit classical nucleation theory taking into account the viscoelastic response of the liquid to the formation of crystal nuclei and find that the kinetic spinodal is strongly influenced by elastic effects. We introduce a dimensionless parameter $`\lambda `$, which is essentially the ratio between the infinite frequency shear modulus and the enthalpy of fusion of the crystal. In systems where $`\lambda `$ is larger than a critical value $`\lambda _c`$ the metastability limit is totally suppressed, independently of the surface tension. On the other hand, if $`\lambda <\lambda _c`$ a kinetic spinodal is present and the time needed to experimentally observe it scales as $`\mathrm{exp}[\omega /(\lambda _c\lambda )^2]`$, where $`\omega `$ is roughly the ratio between surface tension and enthalpy of fusion.
When a liquid is cooled below its freezing point without forming a crystal, it enters a metastable equilibrium phase known as supercooled skripov . A supercooled liquid is squeezed in an uncomfortable time region: if we are too fast in measuring its properties, the system cannot thermalize and an off-equilibrium glass is formed; on the other hand, if we are too slow, the system has the time to nucleate the solid, and we obtain an off-equilibrium polycrystal, and, eventually, a thermodynamically stable crystal kauzmann48 ; turnbull69 . What is the maximum degree of supercooling a metastable equilibrium liquid can reach ?
The tricky point about this question is that it mixes aspects of the experimental protocol, with intrinsic properties of the system. If we stick to cooling the system linearly in time, there is a minimum cooling velocity below which the system is bound to crystallize turnbull69 . This velocity is inversely proportional to the minimum nucleation time as a function of temperature nose : we cannot cool slower than this minimum cooling rate, otherwise crystallization occurs. On the other hand, as we cool, the relaxation time increases steeply, and therefore the system necessarily leaves the supercooled phase, and becomes a glass, at the temperature where the relaxation time becomes too large for this minimum cooling rate.
To penetrate deeper in the supercooled region, one can use an ad hoc nonlinear cooling protocol: cool fast close to the temperature where crystallization is a concern nose , and where relaxation time is still small; and slow down at lower temperatures, to cope with the increasing relaxation time, once the nucleation time starts raising again. Therefore, we may think that the unique limitation to the extent of supercooling is given by our capability of cooling slow and fast enough a sample, and that in principle there is no bound to supercooling a metastable equilibrium liquid.
In fact, our experimental capability is not the only limitation to supercooling a system. If at a certain temperature the relaxation time of the liquid $`\tau _\mathrm{R}`$ exceeds the nucleation time of the crystal $`\tau _\mathrm{N}`$, no equilibrium measurements can be performed on the liquid sample and the supercooled phase does not exist anymore. Such a temperature is called kinetic spinodal $`T_{\mathrm{sp}}`$, and it marks the metastability limit of the supercooled phase kauzmann48 ; skripov . Such a metastability limit does not depend on the cooling protocol, and is rather an intrinsic property of the sample. Below $`T_{\mathrm{sp}}`$ the only equilibrium phase (either stable or metastable) is the crystal, whereas glassy and polycrystalline off-equilibrium configurations can be obtained if we cool fast or slow enough, respectively. The aim of this work is to answer the following question: what is the main mechanism determining whether or not a metastability limit is present in a given liquid ?
Our answer to the question above is that this mechanism is viscoelasticity. The central idea is that on time scales shorter than the structural relaxation time, deeply supercooled liquids exhibit a solid-like response to strains dyre . In particular, the inclusion of a crystalline nucleus in the liquid matrix produces a strain zonko02 ; zonko03 , and thus a long-range elastic response roi78 ; cah84 ; bus03 ; review . Therefore, the thermodynamic drive for crystal nucleation gets depressed by an elastic contribution, which depends on the ratio between the relaxation time $`\tau _\mathrm{R}`$ and the time scale $`\tau _\mathrm{N}`$ over which nucleation occurs. In this way a self-consistent equation involving nucleation and relaxation times is obtained.
We will show that liquids can be classified according to the magnitude of their elastic response relative to their bare thermodynamic drive to crystallization (that is the liquid-crystal Gibbs free energy density difference $`\delta G`$). We find that elastic effects can be large enough to suppress the metastability limit, opening the possibility to supercool a liquid down to the point where its entropy reaches the entropy of the solid, the so-called Kauzmann temperature $`T_\mathrm{k}`$ kauzmann48 . Moreover, even if a metastability limit is present, the viscoelastic response determines how large is the relaxation time at the kinetic spinodal, and it thus establishes whether the metastability limit is accessible within the maximum experimental time, or is hidden by the off-equilibrium glassy phase.
According to classical nucleation theory (CNT), the Gibbs free energy barrier to nucleation is given by turnbull49 ,
$$\mathrm{\Delta }G=\frac{a\sigma ^d}{[\delta G(T)]^{d1}},$$
(1)
where $`d`$ is the dimension, $`\sigma `$ is the surface tension, $`\delta G`$ is the (positive) Gibbs free energy difference per unit volume between the two phases, and $`a`$ is a numerical constant (for a spherical nucleus in $`d=3`$, $`a=16\pi /3`$).
When the nucleus of the stable phase forms in an elastic background, there is an extra energetic price that has to be paid roi78 ; cah84 ; bus03 ; review . Because of the long-range nature of the elastic strains, the elastic price is proportional to the nucleus volume, and thus it corrects the bare thermodynamic drive $`\delta G`$,
$$\mathrm{\Delta }G=\frac{a\sigma ^d}{\left[\delta G(T)E_{\mathrm{elastic}}\right]^{d1}}.$$
(2)
In a perfect isotropic solid, assuming that the elastic parameters are the same in the two phases, the elastic energy density is given by $`E_{\mathrm{elastic}}=ϵ_0G_{\mathrm{}}`$ where $`G_{\mathrm{}}`$ is the (constant) elastic shear modulus, and where review ,
$$ϵ_0=\frac{(d1)K}{dK+2(d1)G_{\mathrm{}}}\left(\frac{\delta v}{v}\right)^2,$$
(3)
is a dimensionless constant: $`K`$ is the bulk modulus and $`\delta v/v`$ is the stress-free volume misfit between the two phases. In a viscoelastic liquid we have a stress-relaxation function, or time-dependent shear modulus, $`G(t)`$ ferry . In this case, Schmelzer and coworkers have shown that the time-dependent elastic contribution can be written as,
$$E_{\mathrm{elastic}}=ϵ_0G_{\mathrm{}}f(t)$$
(4)
where the infinite frequency shear modulus is, $`G_{\mathrm{}}=G(t=0)`$, and the dimensionless function $`f`$ is zonko02 ; zonko03 ,
$$f(t)=\frac{1}{t}_0^t𝑑t^{}G(t^{})/G_{\mathrm{}}.$$
(5)
For a Maxwell liquid ferry $`G(t)=G_{\mathrm{}}e^{t/\tau _\mathrm{R}}`$ and thus, $`f(t/\tau _\mathrm{R})=(1e^{t/\tau _\mathrm{R}})\tau _\mathrm{R}/t`$. We shall not assume the Maxwell form, but we shall still make the hypothesis that the stress-relaxation function $`G`$ depends on the ratio $`t/\tau _\mathrm{R}`$, where $`\tau _\mathrm{R}`$ is the liquid’s structural relaxation time. In this case also $`f=f(t/\tau _\mathrm{R})`$. In log scale the function $`f`$ typically has a step-like shape, going from $`1=f(0)`$ to $`0=f(\mathrm{})`$ with a sharp drop at $`f(1)`$. The physical meaning is clear: for times much larger than the relaxation time the liquid is able to relax the stress, and there is no elastic contribution, while for times much smaller than $`\tau _\mathrm{R}`$ it responds as a solid, with finite shear modulus.
What is the time $`t`$ we have to plug into $`E_{\mathrm{elastic}}(t/\tau _\mathrm{R})`$ ? The viscoelastic contribution to the nucleation barrier is due to the formation of a crystal nucleus, and this happens on a time scale of the order of the nucleation time. Therefore we must set $`t=\tau _\mathrm{N}`$, and write,
$$\tau _\mathrm{N}=\tau _\mathrm{N}^0\mathrm{exp}\left\{\frac{a\sigma ^d}{k_\mathrm{B}T\left[\delta G(T)E_{\mathrm{elastic}}(\tau _\mathrm{N}/\tau _\mathrm{R})\right]^{d1}}\right\}$$
(6)
From this formula we clearly see that the main effect of the elastic contribution is to increase the nucleation time. This effect becomes dramatic in solid-solid phase transitions, where nucleation may be totally suppressed, even with a nonzero thermodynamic drive $`\delta G`$, whenever $`\delta G(T)<E_{\mathrm{elastic}}=ϵ_0G_{\mathrm{}}`$ roi78 ; cah84 ; review ; bus03 . In liquids, however, the stress is always relaxed for long enough times, and eq.(6) always admits a solution $`\tau _\mathrm{N}(T)`$, as long as $`\tau _\mathrm{R}<\mathrm{}`$.
In order to solve eq. (6), we have to discuss the $`T`$-dependence of the various quantities. For the relaxation time we shall assume a Vogel-Fulcher-Tamman (VFT) form, which is known to describe very well a wide variety of supercooled liquids on a large range of times angell88 ,
$$\tau _\mathrm{R}=\tau _\mathrm{R}^0\mathrm{exp}\left(\frac{\mathrm{\Delta }}{TT_\mathrm{k}}\right).$$
(7)
This form of the relaxation time implies a singularity at $`T_\mathrm{k}`$, which is normally identified with the Adam-Gibbs thermodynamic transition adam65 , and it is approximately equal to the temperature where the liquid-crystal entropy difference $`\delta S`$ vanishes. The Gibbs free energy difference is often fitted to a linear form, $`\delta G=(1T/T_\mathrm{m})\delta h/\nu `$, where $`T_\mathrm{m}`$ is the melting temperature, $`\delta h`$ is the molar enthalpy of fusion, and $`\nu `$ is the molar volume of the crystal kiselev01 ; weimberg02 . However, recalling that $`\delta G=(\delta S)/T`$, consistency with (7) requires that $`\delta G`$ must reach its maximum at $`T_\mathrm{k}`$. Therefore, we shall write,
$$\delta G(T)=g(T/T_\mathrm{m})\left(1T_\mathrm{k}/T_\mathrm{m}\right)\delta h/\nu $$
(8)
where the dimensionless function $`g`$ satisfies the relations: $`g(1)=0`$ and $`g(T_\mathrm{k}/T_\mathrm{m})=1`$. The surface tension $`\sigma `$ will be assumed to be a constant, which is a reasonable hypothesis within CNT kiselev01 .
The prefactor of nucleation $`\tau _\mathrm{N}^0`$ in (6) is significantly $`T`$-dependent. According to CNT, $`\tau _\mathrm{N}^0`$ is proportional to the product of an elementary thermal time scale, $`h_{\mathrm{plank}}/k_\mathrm{B}T`$, and an Arrhenius term due to the barrier $`B`$ a particle has to overcome to cross the nucleus interface, $`\mathrm{exp}(B/k_\mathrm{B}T)`$ turnbull49 . It is customary to approximate such term by $`1/D`$ ($`D`$ is the self-diffusion coefficient), and by using the Stokes-Einstein (SE) relation, to write $`\tau _\mathrm{N}^01/D\tau _\mathrm{R}`$ turnbull69 . However, as the system becomes viscous, the SE relation breaks down rossler90 ; fujara92 ; cicerone96 , and typically $`1/D\tau _\mathrm{R}`$. Indeed, in the extreme case of solids, even though the shear viscosity (and thus $`\tau _\mathrm{R}`$) is infinite, $`D`$ remains nonzero. Thus, the prefactor $`\tau _\mathrm{N}^0`$ remains much smaller than the relaxation time at low temperatures, and it cannot by itself grant the fact that $`\tau _\mathrm{N}\tau _\mathrm{R}`$.
A second delicate point about $`\tau _\mathrm{N}^0`$ is its dependence on the sample’s volume. The quantity studied by CNT is not directly the nucleation time, but the nucleation rate $`J`$, that is the number of nuclei per unit time, per unit volume. $`J`$ is a constant for large enough samples and therefore the nucleation time of a given sample of volume $`V`$ scales as $`1/V`$ turnbull69 . Taking very small samples is indeed a well known trick to increase nucleation time. However, there is a limit to this procedure, given by the volume $`v_0`$ below which the $`1/V`$ scaling breaks down: below $`v_0`$ the nucleation rate is not anymore constant, either because surface effects become dominant over bulk properties, or because the sample size is smaller than the critical nucleus paolo . The theoretical metastability limit of a liquid can be defined as the point where the largest possible nucleation time is surpassed by the relaxation time, and therefore we set $`\tau _\mathrm{N}=1/(v_0J)`$. Anyhow, we recall that the results are rather insensitive to the choice of the reference volume, since this enters only in the prefactor $`\tau _\mathrm{N}^0`$, and not in the most relevant exponential factor.
We measure all times in equation (6) in units of $`\tau _\mathrm{N}^0`$, and define $`x=\mathrm{log}(\tau _\mathrm{N}/\tau _\mathrm{N}^0)`$ and $`y=\mathrm{log}(\tau _\mathrm{R}/\tau _\mathrm{N}^0))`$. In this way equation (6) can be rewritten as,
$$T^{}x=\omega \left[g(T^{})\lambda f\left(e^x/e^y\right)\right]^{1d}$$
(9)
where $`T^{}=T/T_\mathrm{m}`$ is the reduced temperature, and,
$`\omega `$ $`=`$ $`{\displaystyle \frac{a\sigma ^d}{k_\mathrm{B}T_\mathrm{m}}}\left[{\displaystyle \frac{\nu }{(1T_\mathrm{k}/T_\mathrm{m})\delta h}}\right]^{d1}`$ (10)
$`\lambda `$ $`=`$ $`{\displaystyle \frac{ϵ_0\nu G_{\mathrm{}}}{(1T_\mathrm{k}/T_\mathrm{m})\delta h}}.`$ (11)
Two more dimensionless parameters are present in the theory: they are the rescaled Kauzmann temperature $`T_\mathrm{k}^{}=T_\mathrm{k}/T_\mathrm{m}`$ and the rescaled VFT barrier $`\mathrm{\Delta }^{}=\mathrm{\Delta }/T_\mathrm{m}`$. They both enter in the relaxation time, and thus in $`y=y(T^{}/T_\mathrm{k}^{},\mathrm{\Delta }^{})`$. If we set $`\lambda =0`$, $`T_\mathrm{k}=0`$ and $`g=(1T/T_\mathrm{m})`$, we recover the classic expression for the nucleation time, where $`\mathrm{exp}(\omega )`$ gives the scale of the minimum nucleation time (see, for example, kiselev01 ). On the other hand, for $`\lambda 0`$ we have elastic corrections to the nucleation time.
Equation (9) can be easily studied numerically once the forms of $`g`$ and $`f`$ are specified. However, the key physical features can be worked out in full generality. First, we note that for $`TT_\mathrm{m}`$, $`g0`$ whereas $`y`$ remains finite, and thus the solution $`x`$ of (9) diverges. As expected, $`\tau _\mathrm{N}\mathrm{}`$ at the melting point. In the opposite limit $`TT_\mathrm{k}`$ a graphical study of equation (9) shows that for $`\lambda <1`$ the nucleation time remains finite, while $`y\mathrm{}`$, and thus a metastability limit $`T_{\mathrm{sp}}>T_\mathrm{k}`$ exists. On the other hand, for $`\lambda >1`$, $`\tau _\mathrm{N}\mathrm{}`$ for $`TT_\mathrm{k}`$: in this case both relaxation and nucleation times diverge at $`T_\mathrm{k}`$, but they still may cross at a kinetic spinodal. To check this last possibility we set $`x=y`$ in equation (9), and see whether there is a solution $`T_{\mathrm{sp}}>T_\mathrm{k}`$ of this equation,
$$T^{}\left(\mathrm{\Omega }+\frac{\mathrm{\Delta }^{}}{T^{}T_\mathrm{k}^{}}\right)=\omega \left[g(T^{})\lambda f(1)\right]^{1d}.$$
(12)
In this formula the factor $`\mathrm{\Omega }=\mathrm{log}(\tau _\mathrm{R}^0/\tau _\mathrm{N}^0)`$ is not relevant, since it is large in modulus only at low temperatures, where however the factor $`(T^{}T_\mathrm{k}^{})^1`$ dominates. We recall that, by construction, $`f(1)`$ is a number close to $`1/2`$ (in Maxwell’s case $`f(1)=11/e=0.63`$). The l.h.s. of equation (12) diverges at $`T_\mathrm{k}^{}`$, whereas the r.h.s. is finite at $`T_\mathrm{k}^{}`$ and diverges at a temperature larger than $`T_\mathrm{k}^{}`$, defined by $`g(T^{})\lambda f(1)=0`$. Given that $`g(T_\mathrm{k}^{})=1`$, we conclude that for $`\lambda <1/f(1)`$ there is a solution $`T_{\mathrm{sp}}^{}`$, and thus a metastability limit, while for $`\lambda >1/f(1)`$ there is no solution in the physical interval $`g(T^{})\lambda f(1)0`$.
We recall that $`\lambda `$ is, up to factors of order one, the ratio between the characteristic scales, $`G_{\mathrm{}}`$ and $`\delta h`$, of the two competing mechanisms at work: the elastic forces, depressing nucleation, and the thermodynamic drive to crystallization, enhancing nucleation. What we have found is a critical value,
$$\lambda _c=1/f(1),$$
(13)
separating systems with a kinetic spinodal, from systems without a kinetic spinodal. In those liquids where $`\lambda <\lambda _c`$ the metastability limit may be shifted at lower $`T`$ by elasticity, but it still exists, while for $`\lambda >\lambda _c`$ elastic effects are strong, and the nucleation time is amplified enough to completely suppress the kinetic spinodal. This distinction is, of course, only valid to logarithmic accuracy: whenever $`\lambda \lambda _c`$, the nucleation time may be only slightly larger than the relaxation time, and it is wise to assume that the liquid is not really stable against crystallization.
In Fig.1 we show solutions of eq.(9) for $`\lambda >\lambda _c`$ (left panel), and $`\lambda <\lambda _c`$ (right panel). In the first case, there is no spinodal, and thus the three phases, glass, supercooled liquid, and polycrystal, are well defined. In particular, the off-equilibrium glassy and polycrystalline states are well separated from each other by the ‘buffer’ liquid phase. When $`\lambda <\lambda _c`$, however, the situation is rather different. The equilibrium liquid phase is still well defined as long as it exists, that is for $`T>T_{\mathrm{sp}}`$; however, it is hard to say what is the physical difference between the two off-equilibrium phases, i.e. glass and polycrystal. Although it may be operationally meaningful to distinguish glass from polycrystal when $`TT_{\mathrm{sp}}`$, there seems to be no obvious qualitative difference when $`TT_{\mathrm{sp}}`$, or, even worse, when $`T<T_{\mathrm{sp}}`$. Close to the spinodal point, whether the system falls out of equilibrium by crossing the broken or the full line, is irrelevant. Glass and polycrystal blend into a single off-equilibrium phase cavagna03 .
In systems with $`\lambda >\lambda _c`$ the supercooled phase is well defined down to $`T_\mathrm{k}`$, where the relaxation time diverges. In this case, a solution of Kauzmann paradox (for example, in terms of a thermodynamic transition at $`T_\mathrm{k}`$) must be found. On the other hand, when a spinodal is present the supercooled liquid ceases to exist at $`T_{\mathrm{sp}}`$, which is Kauzmann’s resolution of the paradox kauzmann48 .
When a kinetic spinodal exists, is it within the boundaries of experimental observation ? To answer this question we have to define a spinodal time, $`\tau _{\mathrm{sp}}=\tau _\mathrm{R}(T_{\mathrm{sp}})`$, and compare it to the maximum time of the experiment, $`\tau _{\mathrm{exp}}`$. From eq. (12) we obtain (for $`d=3`$),
$$\tau _{\mathrm{sp}}\mathrm{exp}\left[\frac{\omega }{(1\lambda /\lambda _c)^2}\right].$$
(14)
Therefore, as expected, the position of the kinetic spinodal depends also on the parameter $`\omega `$: a large value of $`\omega `$ implies a large value of the surface tension $`\sigma `$ compared to the drive to crystallization $`\delta h`$, and thus a larger nucleation time turnbull69 . Therefore, when $`\omega `$ is significantly larger than $`(1\lambda /\lambda _c)^2`$, the spinodal time may be too large compared to the experimental time, and thus the metastability limit be experimentally unaccessible tigi .
Conventionally, a good glass-former is a system which does not crystallize easily under a linear cooling, i.e. a system whose minimum cooling rate $`r_{\mathrm{min}}`$ is not too large. Given that $`r_{\mathrm{min}}1/\tau _{\mathrm{min}}`$, where $`\tau _{\mathrm{min}}`$ is the minimum nucleation time as a function of $`T`$ nose , the larger $`\tau _{\mathrm{min}}`$, the better the glass-former. To what extent $`\tau _{\mathrm{min}}`$ depends on $`\lambda `$ ? When $`\lambda \lambda _c`$ the kinetic spinodal typically takes place at a point where the nucleation time is still a decreasing function of $`T`$ (Fig. 1, inset of right panel). In such case $`\tau _{\mathrm{min}}`$ is effectively given by the spinodal point. From (14) we therefore have $`r_{\mathrm{min}}\mathrm{exp}[\omega /(1\lambda /\lambda _c)^2]`$, and the glass-forming capability of the system is strongly dependent on $`\lambda `$. On the other hand, when $`\lambda \lambda _c`$ or $`\lambda >\lambda _c`$, $`\tau _{\mathrm{min}}`$ is given as usual by the minimum of the nucleation curve. In this case $`\tau _{\mathrm{min}}\mathrm{exp}(\omega )`$, the glass-forming properties depend almost exclusively on $`\omega `$, whereas their dependence on $`\lambda `$ is weak.
In this work we studied the effects of viscoelastic response on the metastability limit of supercooled liquids. We defined a parameter $`\lambda `$ whose value rules whether a kinetic spinodal exists or not. This parameter encodes the competition between elastic response and thermodynamic drive to crystallization: for values of $`\lambda `$ below $`\lambda _c`$ elasticity is relatively weak, the drive to crystallization always wins, and a metastability limit is present. For $`\lambda >\lambda _c`$ the elastic response is so large that nucleation in inhibited compared to relaxation, and a metastability limit is not present.
We thank A. Angell, G. Biroli, C. Di Castro, I. Giardina, T.S. Grigera, and P. Verrocchio for many useful comments and clarifying discussions.
|
warning/0506/hep-th0506233.html
|
ar5iv
|
text
|
# Contents
## 1 Introduction
There has been much progress recently in constructing solutions of the supergravity equations describing rotating and charged black holes in $`n`$-dimensional anti-de Sitter backgrounds . A primary motivation for this work was the elucidation of the thermodynamics of these black holes, with a view to comparing it with that of the dual conformal field theories on the boundary of the spacetime, which approaches AdS with radius of curvature $`l`$.<sup>1</sup><sup>1</sup>1In $`n=5`$ dimensions, the dual boundary conformal field theory is $`𝒩=4`$ supersymmetric $`SU(N)`$ Yang-Mills theory, where $`N^2=\pi c^3l^3/(2\mathrm{}G_5)`$. In particular the correct energies and angular momenta for Kerr-AdS black holes, measured with respect to a frame that is non-rotating at infinity, were calculated in all dimensions in , where it was also demonstrated that these quantities satisfy the first law of thermodynamics,
$$dE=TdS+\mathrm{\Omega }^idJ_i.$$
(1.1)
This resolved some of the apparent ambiguities in earlier work, that had focused on energies and angular velocities measured with respect to a particular frame rotating at infinity, which we shall denote with primes throughout this paper.<sup>2</sup><sup>2</sup>2Not all quantities measured in the rotating coordinate system are changed from their values in the asymptotically static frame, but for clarity we denote all quantities measured in the rotating frame with primes in the present discussion. As shown in , these do not satisfy the first law of thermodynamics:
$$dE^{}T^{}dS^{}+\mathrm{\Omega }_{}^{i}{}_{}{}^{}dJ_i^{},$$
(1.2)
since the asymptotic rotation rate in this frame depends on the black-hole rotation parameters.
In a recent paper, Cai et al. noticed that by passing from the bulk quantities $`(E^{},J_i^{})`$ to the dual CFT quantities $`(e^{},j_i^{},\mathrm{})`$ on the boundary, and by including an additional pressure term $`p^{}`$ and suitably-defined volume term $`v^{}`$, the equation
$$de^{}=t^{}ds^{}+\omega _{}^{i}{}_{}{}^{}dj_i^{}p^{}dv^{}$$
(1.3)
holds. They interpreted this equation as the first law for the dual CFT, and furthermore they made the surprising claim that no such analogous CFT thermodynamic variables can be introduced that are dual to the unprimed bulk quantities, and which satisfy the first law.
$$de=tds+\omega ^idj_ipdv.$$
(1.4)
One purpose of this paper is to refute this surprising claim, and on the contrary to demonstrate that following the perfectly standard transcription rules relating bulk and boundary quantities, the first law (1.4) does indeed hold. We also raise questions as to whether the interpretation of the primed quantities given in is physically correct.
A remarkable feature of , following earlier work in , is the observation that the primed bulk quantities satisfy a Cardy-Verlinde type formula,
$`S^{}`$ $`=`$ $`{\displaystyle \frac{2\pi l}{n2}}\sqrt{E_c^{}(2E^{}E_c^{})},`$
$`E_c^{}`$ $`=`$ $`(n1)E^{}(n2)(T^{}S^{}+\mathrm{\Omega }_{}^{i}{}_{}{}^{}J_i^{}),`$ (1.5)
whereas the unprimed quantities measured with respect to a non-rotating frame at infinity do not. This has motivated us to re-examine the old question of whether or not AdS black holes satisfy some sort of possibly modified Cardy-Verlinde formula, and a Bekenstein-type bound of the form<sup>3</sup><sup>3</sup>3Note that the Bekenstein bound does not contain Newton’s constant, and so it makes sense for any thermodynamic quantity in an AdS background. However, in this paper our concern is solely with black holes.
$$E\frac{(n2)S}{2\pi l}.$$
(1.6)
The existence of the Bekenstein bound is a necessary but not sufficient condition for the validity of a Cardy-Verlinde formula. In fact, we find that while there appears to be no universal formulation of a modified Cardy-Verlinde formula that will cover all of the charged and rotating AdS black holes that we know of, we do find that a Bekenstein bound holds in all cases.
In fact, the Bekenstein bound follows from a stronger and more fundamental inequality, the cosmic censorship bound, which takes the form
$$E\frac{(n2)A}{16\pi l}\left[l\left(\frac{A}{𝒜_{n2}}\right)^{{\scriptscriptstyle \frac{1}{n2}}}+\frac{1}{l}\left(\frac{A}{𝒜_{n2}}\right)^{{\scriptscriptstyle \frac{1}{n2}}}\right],$$
(1.7)
where $`A`$ is the area of the event horizon, or more generally, in time-dependent cases, the area of the outermost apparent horizon, and $`𝒜_{n2}`$ is the volume of the unit $`(n2)`$-sphere. We show that the recently-constructed exact solutions for rotating and charged AdS black holes give strong support for the conjectured cosmic censorship bound.
As well as lower bounds for the energy in terms of the entropy, it is well known that there are interesting upper bounds for the temperature as a function of entropy, for black holes in asymptotically flat spacetimes. It turns out that these may be generalised to the static anti-de Sitter case, taking the form
$$4\pi T(n3)\left(\frac{A}{𝒜_{n2}}\right)^{{\scriptscriptstyle \frac{1}{n2}}}+(n1)l^2\left(\frac{A}{𝒜_{n2}}\right)^{{\scriptscriptstyle \frac{1}{n2}}},$$
(1.8)
in $`n`$ dimensions. That is, the temperature is never greater than the value it would have in the Schwarzschild-AdS solution with the same entropy. We investigate whether the bound extends to stationary black holes, finding that it is obeyed by all Kerr-AdS black holes in four and five dimensions, but not in dimensions six or higher.
The uncharged rotating black hole solutions are of course valid also if the cosmological constant is taken to be positive, corresponding to sending $`l^2l^2`$. In this case, an additional, cosmological, horizon is present. We verify that these solutions support the general conjecture that the area $`A_C`$ of the cosmological horizon satisfies
$$A_C𝒜_{n2}l^2,$$
(1.9)
and the black hole horizon satisfies the inequality
$$A_H𝒜_{n2}l^{n2}\left(\frac{n3}{n1}\right)^{{\scriptscriptstyle \frac{n2}{2}}}.$$
(1.10)
The plan of the paper is as follows. In section 2 we establish a general equivalence between bulk and boundary thermodynamics, and explain our disagreement with some of the work in . In section 3, we review the Cardy-Verlinde formula and its consequence, the AdS-Bekenstein bound, explaining why it is saturated at the Hawking-Page phase transition. We show that a simple modification holds for Reissner-Nordström-AdS black holes, and gives rise to a strengthened form of the Bekenstein bound. We also show that in all dimensions rotating black holes without charge satisfy the AdS-Bekenstein bound. In section 4, we examine a large number of examples of rotating and/or charged black holes, finding that despite the failure in general of the Cardy-Verlinde formula, the AdS-Bekenstein bound, or its strengthened electrostatic form, holds. In section 5 we discuss how the AdS-Bekenstein bound may be regarded as a consequence of the AdS cosmic censorship bound, and demonstrate in all the examples we have checked that the AdS cosmic censorship bound does indeed hold, in some cases strengthened by an electrostatic contribution. Section 6 discusses upper bounds for the temperature of AdS black holes, in terms of their entropy. We find that rotating black holes in four and five dimensions always have a temperature that is less than that of the Schwarzschild-AdS solution with the same area. However, rotating black holes of dimension six or higher do not satisfy such a bound. Section 7 contains a brief discussion of the areas of the cosmological and black-hole horizons for rotating black holes with positive cosmological constant. We collect for the reader’s convenience, in an appendix, the pertinent formulae for Kerr-AdS black holes in arbitrary dimensions. Our conclusions are contained in section 8.
## 2 Bulk and Boundary Thermodynamics
The purpose of this section is to show that there is a universal rule allowing one to pass between bulk and boundary quantities in such a way that if one set of quantities satisfies the first law of thermodynamics, then so will the other.
### 2.1 Tolman or UV/IR scaling transformations
It is quite generally true that if an arbitrary thermodynamic system satisfies the first law of thermodynamics without a pressure term,
$$dE=TdS+\mathrm{\Omega }^idJ_i+\mathrm{\Phi }_idQ_i,$$
(2.1)
then associated with it is a second system, with pressure equal to the energy density divided by the spatial dimension, which satisfies the first law with pressure term:
$$de=tds+\omega ^idj_i+\varphi _idq_ipdv.$$
(2.2)
Actually, since the first system need have no natural dimension associated to it, the spatial dimension $`(n2)`$ of the second system can be arbitrary. The thermodynamic quantities of the second system, denoted by lower-case letters, are related to those of the first system by
$$\overline{)\begin{array}{ccc}\hfill e& =& \frac{l}{y}E,\omega ^i=\frac{l}{y}\mathrm{\Omega }^i,\varphi _i=\frac{l}{y}\mathrm{\Phi }_i,t=\frac{l}{y}T,\hfill \\ \hfill s& =& S,j_i=J_i,q_i=Q_i,\hfill \end{array}}$$
(2.3)
with
$$v=𝒜_{n2}y^{n2},p=\frac{e}{(n2)v},$$
(2.4)
where $`𝒜_{n2}`$ is the volume of the unit $`(n2)`$-sphere. Here, $`y`$ is to be thought of as the radius of the second system, and $`l`$ is an arbitrary constant, which in our application is related to the cosmological constant, so that $`R_{\mu \nu }=(n1)l^2g_{\mu \nu }`$. Note that in (2.3) the intensive quantities $`(T,\mathrm{\Omega }^i,\mathrm{\Phi }_i)`$ are scaled, as is the energy $`E`$, whilst the extensive quantities $`(S,J_i,Q_i)`$ are not scaled.
As it stands, the above result is a mathematical triviality. However, in the case we are considering, where the first system is a rotating charged black hole in an AdS<sub>n</sub> background, it allows us to relate the bulk thermodynamic quantities associated with the black hole to the boundary quantities associated with the dual conformal field theory. The quantities $`(E,T,S,\mathrm{\Omega }^i,J_i,\mathrm{\Phi }_i,Q_i)`$ are all evaluated with respect to a coordinate frame $`(t,y,\widehat{\mu }_i,\phi _i)`$ that is non-rotating and asymptotically spherical at infinity.<sup>4</sup><sup>4</sup>4The time coordinate $`t`$ should not be confused with the CFT temperature $`t`$ — it should be clear from the context which is which. In these coordinates, the metric of the rotating black hole approaches the AdS metric
$$d\overline{s}^2=(1+y^2l^2)dt^2+\frac{dt^2}{1+y^2l^2}+y^2\underset{k=1}{\overset{N+ϵ}{}}(d\widehat{\mu }_k^2+\widehat{\mu }_k^2d\phi _k^2),$$
(2.5)
in $`n=2N+ϵ+1`$ dimensions, where $`ϵ=(n1)`$ mod 2 (see , and appendix A, for a more detailed discussion). Note that the radial coordinate $`y`$ is related to the Fefferman-Graham coordinate $`zl^2/y`$ for which the metric asymptotes to $`l^2dt^2/z^2+l^2dz^2/z^2+l^4/z^2d\mathrm{\Omega }_{n2}^2`$.
The Killing vector $`/t`$ is normalised so that near infinity
$$g(\frac{}{t},\frac{}{t})\frac{y^2}{l^2},$$
(2.6)
and this fixes the normalisation of the quantities $`(E,T,\mathrm{\Omega }^i,\mathrm{\Phi }_i)`$. A boundary conformal field theory living on a surface $`y=`$constant will thus occupy the volume $`v`$ given in (2.4). The intensive quantities $`(t,\omega ^i,\varphi _i)`$ of the CFT are then given by the standard Tolman redshifting factor, or, in the language of the AdS/CFT correspondence, the UV/IR connection, which coincides with our formulae in (2.3). The pressure $`p`$ is that expected of a conformally-invariant system, the trace of whose energy-momentum tensor should vanish. One reason why the extensive quantities $`S`$, $`J_i`$ and $`Q_i`$ cannot scale under the UV/IR connection is that they are subject to quantisation conditions, and are given by integers.
The upshot of the above discussion is that the introduction of the pressure term is a triviality, which ensures that if the first law of thermodynamics holds in the bulk, then it holds also in the boundary CFT.
### 2.2 Relation to earlier work
Although for us the introduction of the pressure term is, as we have explained above, a triviality, because our bulk quantities satisfy the first law of thermodynamics, it is less transparent if bulk thermodynamic variables are chosen that do not satisfy the first law. As we showed in , the way to obtain bulk thermodynamic quantities for black holes that satisfy the first law is by calculating them with respect to a frame that it non-rotating at infinity. The energy measured in this frame can be derived using the Ashtekar-Magnon-Das conformal definition of mass in AdS backgrounds . It has also been shown that the same expression can be derived from the superpotential of Katz, Bičák and Lynden-Bell . A further calculation leading to the same expression for the energy was given recently in .
There are, of course, infinitely many frames one could choose that do rotate, with different rotation rates, at infinity. One popular choice is the asymptotically-rotating coordinate system in which Carter first wrote the Kerr-AdS black hole in four dimensions . Analogous rotating frames were introduced in five dimensions by Hawking, Hunter and Taylor-Robinson , and in all higher dimensions in . In these papers, the metrics are given in a coordinate system which is rotating with angular velocity
$$\mathrm{\Omega }_{\mathrm{}}^i=\frac{a_i}{l^2}$$
(2.7)
with respect to an asymptotically static frame, where $`a_i`$ are the rotation parameters. (See appendix A for a summary of the salient details of the Kerr-AdS metrics. In the appendix, the metrics are given in an asymptotically static coordinate system.)
The geometrical significance of this particular rotating frame is that with respect to it the Kerr-Schild congruence, which was used to construct the solution, is non-rotating at infinity. However, this in itself does not appear to endow it with any privileged dynamical significance. Nevertheless one can certainly, as has been done in some of the literature, associate with it energies and angular velocities, which we shall denote by primes, that are given in terms of the unprimed non-rotating thermodynamic quantities by
$$E^{}=E\frac{a_i}{l^2}J_i,\mathrm{\Omega }_{}^{i}{}_{}{}^{}=\mathrm{\Omega }^i\frac{a_i}{l^2},$$
(2.8)
with all the other quantities being the same in the primed and the unprimed frame.<sup>5</sup><sup>5</sup>5The reason why $`J_i=J_i^{}`$ is that “passing to the rotating frame” means in effect choosing a new timelike Killing field from which $`E^{}`$ is constructed, but retaining the same angular Killing fields from which the $`J_i`$ are constructed. In other words, one introduces the new rotating coordinates $`(t^{},\phi _i^{})`$, related to the asymptotically static coordinates $`(t,\phi _i)`$ by $`t^{}=t`$, $`\phi _i^{}=\phi _ia_il^2t`$, and associates the energy $`E^{}`$ with the Killing vector $`/t^{}=/t+a_il^2/\phi _i`$, as opposed to the energy $`E`$ associated with $`/t`$. Thus passing to a rotating frame is not the same as performing an asymptotic $`SO(n1,2)`$ transformation; it is merely picking a new basis for the Lie algebra $`𝔰𝔬(n1,2)`$. Note that
$$E^{}\mathrm{\Omega }_{}^{i}{}_{}{}^{}J_i^{}=E\mathrm{\Omega }^iJ_i=E^{}\mathrm{\Omega }_{}^{i}{}_{}{}^{}J_i.$$
(2.9)
Throughout the rest of this paper, we shall use the symbols $`E^{}`$ and $`\mathrm{\Omega }_{}^{i}{}_{}{}^{}`$ to denote energies and angular velocities measured with respect to the asymptotically rotating frames for which (2.7) holds.
Although $`E^{}`$ appears to have no special physical significance, it turns out that it provides a useful bound for the true energy $`E`$, in other words
$$EE^{},$$
(2.10)
with equality if and only if the black hole is non-rotating.
As we noted in ,
$$dE^{}TdS+\mathrm{\Omega }_{}^{i}{}_{}{}^{}dJ_i.$$
(2.11)
However, Klemm et al. discussed the thermodynamics of rotating AdS black holes with a single non-vanishing rotation parameter, and obtained an extended system involving a chemical potential $`\mu `$ and number $`N`$, satisfying the first law. More recently, Cai et al. have introduced thermodynamic quantities which in our notation we shall write as $`e^{}`$, $`t^{}`$, $`s^{}`$, $`\omega _{}^{i}{}_{}{}^{}`$, $`j_i^{}`$, $`p^{}`$, $`v^{}`$, given by
$$v^{}=\frac{𝒜_{n2}r^{n2}}{_j\mathrm{\Xi }_j},p^{}=\frac{e^{}}{(n2)v^{}},$$
(2.12)
$$\begin{array}{ccc}\hfill e^{}& =& \frac{l}{r}E^{},\omega _{}^{i}{}_{}{}^{}=\frac{l}{r}\mathrm{\Omega }_{}^{i}{}_{}{}^{},t^{}=\frac{l}{r}T,s^{}=S,j_i^{}=J_i.\hfill \end{array}$$
(2.13)
and they have shown that these satisfy the first law
$$de^{}=t^{}ds^{}+\omega _{}^{i}{}_{}{}^{}dj_i^{}p^{}dv^{}.$$
(2.14)
Note that the formula (2.12) for $`v^{}`$ is not as we defined in (2.4), but rather has the additional factor $`_i\mathrm{\Xi }_i`$ in the denominator. This is needed in order to get the first law (2.14) in the primed variables to work out, to compensate for the failure of the bulk primed quantities to satisfy the first law. It is also not difficult to see that if one chooses a different rotation rate at infinity, replacing the right-hand side of (2.7) with some general functions of the rotations $`a_i`$, then one cannot in general find a formula for $`v^{}`$ of the form (2.12) with the factor $`_i\mathrm{\Xi }_i`$ replaced by a suitable function of the $`a_i`$. To that extent, the fact that the primed CFT quantities satisfy the first law (2.14) is no entirely fortuitous. Nevertheless, it seems to us that it is the unprimed CFT quantities given by (2.3) and (2.4) that most closely correspond to the physical situation that motivated the work in . In other words, the relevant CFT should rotate, relative to a frame non-rotating at infinity, with the same angular velocity as that of the black hole in the bulk theory.
One could pass to a frame that is co-rotating with the black hole, i.e. one whose angular velocity with respect to the frame that is non-rotating at infinity is equal to $`\mathrm{\Omega }^i`$, given by (A.7). The associated Killing vector
$$K=\frac{}{t}+\mathrm{\Omega }^i\frac{}{\phi ^i}$$
(2.15)
(expressed using the asymptotically non-rotating coordinates in (A.2)) coincides on the horizon with its null generator, and is, provided $`|a_i|<l`$, everywhere timelike outside the horizon. This has the desirable feature that local energy densities measured with respect to this Killing vector are everywhere positive . However, it has the distinct disadvantage that when considering any energy exchange between the rotating black hole and its environment, one must change to a new rotating frame because in general $`\mathrm{\Omega }^i`$ changes. It is for this reason that the first law of thermodynamics does not hold with respect to the primed quantities. More generally, one could consider a Killing vector of the form
$$\stackrel{~}{K}=\frac{}{t}+\stackrel{~}{\mathrm{\Omega }}^i\frac{}{\phi ^i}.$$
(2.16)
A simple calculation shows that on the horizon,
$$g(\stackrel{~}{K},\stackrel{~}{K})=g_{ij}(\stackrel{~}{\mathrm{\Omega }}^i\mathrm{\Omega }^i)(\stackrel{~}{\mathrm{\Omega }}^j\mathrm{\Omega }^j),$$
(2.17)
where $`g_{ij}=g(/\phi _i,/\phi _j)`$, and thus we see that for any angular velocity $`\stackrel{~}{\mathrm{\Omega }}^i`$ that differs from $`\mathrm{\Omega }^i`$, the associated Killing vector $`\stackrel{~}{K}`$ is spacelike on (and therefore in the neighbourhood of) the horizon. In particular, this applies to the choice $`\stackrel{~}{\mathrm{\Omega }}^i=\mathrm{\Omega }_{\mathrm{}}^i=a_i/l^2`$. Thus to use the primed frame would neither achieve positivity of the local energy density nor a simple form for the first law. It seems, therefore, that neither it, nor any other frame that is rotating at infinity (other than, possibly, the frame that is rotating with the angular velocity of the black hole) has any particular merit or advantage over the frame that is non-rotating at infinity. Of course physical results cannot depend upon an arbitrary choice of frame.<sup>6</sup><sup>6</sup>6It should be emphasised that the different expressions for the energy and angular velocity in different frames is not the result of the coordinate transformation per se, but of using the transformed time coordinate when defining the energy and angular velocity. It is clear that we can describe a rigidly-rotating gas either as being at rest in a rotating frame, or moving in a non-rotating frame. The choice which seems to us the most straightforward and simple is the latter. Similarly, the calculations in need not have been done using the primed quantities, and we disagree with the assertion in that it is necessary to use the primed quantities in order “to extract data useful to a dual CFT.” As we have seen above, this is a trivial matter using the non-rotating frame.
Using the primed energy $`E^{}`$ is precisely analogous to using the kinetic energy of a particle with respect to a rotating frame, such as that of the earth. It can be done, but it is then necessary to consider the contributions to the energy and the equations of motion due to the centrifugal and Coriolis forces. If the particle is freely moving on the rotating platform, and the rotation rate is changed, the kinetic energy with respect to the rotating frame will obviously change, while it will clearly be constant with respect to an inertial frame. There would seem to be no merit in introducing an artificially time-dependent rotating frame merely to describe straight-line motion in inertial coordinates. If instead of free particles we considered a gas in a state of rigid rotation, we would have a situation rather more analogous to that of the CFT. The gas would exert a pressure on its container, which in principle could be measured in any rotating frame, but the two that are most relevant are surely the rigid rotating frame co-rotating the gas, or the one that is non-rotating with respect to an inertial coordinate system. As we have explained earlier, if the rotation rate changes with time, it is the latter which is more convenient. Choosing to use the energy $`E^{}`$ in the rotating black-hole problem is the equivalent of using a frame that is neither non-rotating at infinity nor is it rotating at the angular velocity of the black hole horizon. Furthermore, in previous work where the energy $`E^{}`$ has nevertheless been used, the necessary corrections to compensate for the changes in the rotation rate of the primed frame have been omitted.
In , the CFT is assumed to be on a surface of large $`r`$ in Boyer-Lindquist coordinates, and the spatial volume is supposed to be the volume of that surface. It should be noted that although this spatial surface has the topology of an $`(n2)`$-sphere it does not have an $`SO(n1)`$ isometry group even asymptotically at large $`r`$. If one nevertheless chooses this $`r=`$constant boundary one must face up to the fact that the temperature will be space dependent and there will be no conventional thermodynamic interpretation where a global temperature is well defined.
For example, in four dimensions the Kerr-AdS metric at large $`r`$ approaches the form
$$ds_4^2=\frac{r^2\mathrm{\Delta }_\theta }{l^2\mathrm{\Xi }}\left[dt^2+l^2\left\{\frac{\mathrm{\Xi }d\theta ^2}{\mathrm{\Delta }_\theta ^2}+\frac{\mathrm{sin}^2\theta }{\mathrm{\Delta }_\theta }(d\varphi +al^2dt)^2\right\}\right],$$
(2.18)
where
$$\mathrm{\Delta }_\theta =1\frac{a^2}{l^2}\mathrm{cos}^2\theta ,\mathrm{\Xi }=1\frac{a^2}{l^2}.$$
(2.19)
Defining a new coordinate $`\widehat{\theta }`$ by $`\mathrm{tan}\widehat{\theta }=(\mathrm{tan}\theta )/\sqrt{\mathrm{\Xi }}`$, it can be seen that the 2-metric enclosed in braces is nothing but the standard unit 2-sphere, whose volume is of course $`4\pi `$. The metric in the square brackets is that of a three-dimensional rotating Einstein universe. The CFT metric is in fact conformal to this, and with respect to this metric the spatial volume is
$$\frac{2\pi r^2}{\mathrm{\Xi }}_0^\pi 𝑑\widehat{\theta }\mathrm{sin}\widehat{\theta }\mathrm{\Delta }_\theta =\frac{4\pi r^2l}{a\sqrt{\mathrm{\Xi }}}\mathrm{arcsin}(a/l).$$
(2.20)
This is not equal to $`4\pi r^2/\mathrm{\Xi }`$, which is the value given in . We are thus unsure as to precisely which spatial volume the quantity $`4\pi r^2/\mathrm{\Xi }`$ in is supposed to be. Similar remarks apply to all the higher-dimensional expressions for $`v^{}`$ given in and reproduced in (2.12).
A striking feature of the work in is the finding that the quantities $`e^{}`$, $`t^{}`$, $`s^{}`$, $`\omega _{}^{i}{}_{}{}^{}`$, $`j_i^{}`$, $`p^{}`$ and $`v^{}`$ satisfy a suggested formula of E. Verlinde , which itself was based on an attempt to incorporate Bekenstein’s conjecture of some sort of bound relating entropy, energy and radius. This has motivated us to look in more detail at the general question of such formulae and bounds, which we shall do in the next section.
## 3 The Cardy-Verlinde Formula and the Bekenstein Bound
### 3.1 The ideal Cardy-Verlinde gas
According to a proposal of E. Verlinde, a CFT living on an $`(n1)`$-dimensional Einstein Static Universe (ESU) of radius $`y`$ and hence metric
$$ds^2=dt^2+y^2d\mathrm{\Omega }_{n2}^2,$$
(3.1)
where $`d\mathrm{\Omega }_{n2}^2`$ is the canonical round metric on $`S^{n2}`$ should have
* A pressure, energy and volume related by
$$p=\frac{1}{n2}\frac{e}{v},v=𝒜_{n2}y^{n2},$$
(3.2)
* An entropy $`s`$ given by
$$s=\frac{2\pi y}{n2}\sqrt{e_c(2ee_c)},$$
(3.3)
where $`e_c`$ is a measure of the extent to which the energy $`e`$ is non-extensive and given by
$$e_c:=(n2)(ets\omega j\varphi q+pv).$$
(3.4)
Subsequent work showed that for free field theory, the Cardy-Verlinde formula does not hold exactly, but it agrees, up to a constant factor, in the high-temperature limit. That is, at large $`T`$ one has $`E_cT^{n3}`$, and $`ET^{n1}`$, and hence $`ST^{n2}`$. However the factor of proportionality is wrong. Nevertheless, it is still possible that it, or some modified form, may hold in the strongly-interacting case that is relevant for the AdS/CFT correspondence.
In this limit, it is more convenient to discuss bulk, rather than boundary, quantities. As we have emphasised in section 2.1, one can freely translate back and forth between the two descriptions. Using the Tolman redshifting formula, or the UV/IR relation between lower case and capital letter quantities, one sees that (3.3) may be re-written in terms of bulk quantities as
$$S=\frac{2\pi l}{n2}\sqrt{E_c(2EE_c)},$$
(3.5)
where $`E_c`$ is given by
$$\overline{)E_c:=(n2)[(1+\frac{1}{n2})ETS\mathrm{\Omega }J\mathrm{\Phi }Q].}$$
(3.6)
and we identify $`E`$ as the conformal generator of $`J_{0n}𝔰𝔬(n1,2)`$ associated to the asymptotically-static Killing field $`\frac{}{t}`$,
$$E=l^{n3}J_{0n}.$$
(3.7)
For later purposes, we re-write (3.6) as<sup>7</sup><sup>7</sup>7The symbol $`:=`$ means that the quantity on the left-hand side is defined by the expression on the right-hand side. In particular we shall, for the sake of clarity, always stick with this primary definition of $`e_c`$, and the analogous blue shifted quantities $`E_c`$. If we need to modify the definition we will indicate any modified quantity by a circumflex, thus $`\widehat{e}_c`$. The importance of this cannot be over-emphasised. If one does not stick to the primary definition (3.8), then the question of the existence or non-existence of such a formula becomes completely meaningless, since one could always define $`e_c`$ in such a way that the Cardy-Verlinde formula became a trivial identity!
$$E_c=(n1)E(n2)\left[TS+\mathrm{\Omega }J\mathrm{\Phi }Q\right].$$
(3.8)
From now on we shall primarily be concerned with the Cardy-Verlinde formula in terms of “bulk,” that is black hole, quantities. However, it is worth remarking that in some papers the anti-de Sitter radius $`l`$ is replaced by $`r_+`$, where typically $`r_+`$ is the radius of the horizon in Schwarzschild or, in the rotating case, Boyer-Lindquist coordinates. In effect this amounts to setting $`y=r_+`$ in the redshifted CFT form. However, unless $`r_+>>l`$, this cannot be regarded as a legitimate application of the UV/IR relation, since if $`r_+l`$ then $`g_{tt}`$ is not well approximated by $`y/l`$.
### 3.2 Anti-de Sitter Bekenstein bound and Hawking-Page transition
The (strict) Cardy-Verlinde formula (3.5) may be regarded as a formula for the energy $`E`$ in terms of the entropy $`S`$ and non-extensive energy $`E_c`$:
$$E=\frac{1}{2}E_c+\frac{1}{2E_c}\left[\frac{(n2)S}{2\pi l}\right]^2.$$
(3.9)
Minimization with respect to $`E_c`$ leads to the lower bound
$$\overline{)2\pi lE(n2)S.}$$
(3.10)
which we shall refer to as the Anti-de Sitter Bekenstein Bound, regardless of whether it arises from a Cardy-Verlinde formula. Clearly the anti-de Sitter Bekenstein bound is a necessary, but not sufficient condition for the existence of a Cardy-Verlinde formula.
This lower bound for the energy in terms of the entropy, or alternatively upper bound for the entropy in terms of the energy, is attained when
$$E=E_c=\frac{(n2)S}{2\pi l}.$$
(3.11)
Note that for $`E>\frac{(n2)S}{2\pi l}`$ there are two values of $`E_c`$ satisfying the Cardy-Verlinde formula (3.9), whilst if $`E<\frac{(n2)S}{2\pi l}`$, there are none.
The calculation above may be re-organised as follows. The thermodynamic potential $`\mathrm{\Psi }`$ of the bulk black hole is given by
$$\mathrm{\Psi }=ETS\mathrm{\Omega }J\mathrm{\Phi }Q.$$
(3.12)
Thus
$$E_c=E+(n2)\mathrm{\Psi },2EE_c=E(n2)\mathrm{\Psi }.$$
(3.13)
The Cardy-Verlinde formula can be cast in the Pythagorean form
$$\frac{E^2}{(n2)^2}=\frac{S^2}{4\pi ^2l^2}+\mathrm{\Psi }^2.$$
(3.14)
We see from (3.14) that the Bekenstein bound is attained if and only if the thermodynamic potential vanishes,
$$\mathrm{\Psi }=0.$$
(3.15)
If one accepts the quantum statistical relation between thermodynamic potential and Euclidean action $`I`$,
$$\mathrm{\Psi }=TI,$$
(3.16)
then the Bekenstein bound is attained when the Euclidean action vanishes. In the black hole case, this indicates that the Euclidean black hole solution with large energy $`E`$ no longer has smaller action than that of flat space, and a type of phase transition is indicated, as was first discussed by Hawking and Page in the case of AdS<sub>4</sub> black holes, and by Witten in the case of AdS<sub>5</sub> black holes.
The bound (3.10) resembles the controversial universal bound suggested by Bekenstein for systems in flat Minkowski spacetime, except that Bekenstein’s putative universal Minkowski bound contains an undefined radius. In the AdS-Bekenstein Bound (3.10), this radius is taken to be that of AdS<sub>n</sub>. However, for large radius we can use the Tolman redshifting formulae (which are of course valid only for large radius since we are using an approximate form for the metric near the boundary), and we may just as well write
$$2\pi ye(n2)s,$$
(3.17)
where now the radius is that of $`S^{n2}`$. For clarity, we shall call this latter bound the Spherical Bekenstein Bound or the $`S^{n2}`$-Bekenstein bound.
Note that neither the AdS<sub>n</sub> Bekenstein bound nor the $`S^{n2}`$ Bekenstein bound, like that in Bekenstein’s original and rather imprecise Minkowski-spacetime bound, contain Newton’s constant or make any specific reference to gravity. Moreover, the AdS<sub>n</sub> Bekenstein bound (3.10) reduces, in the Minkowski limit $`l\mathrm{}`$, to the undemanding requirement that the energy be non-negative.
### 3.3 Non-rotating Reissner-Nordström black holes
Remarkably, the Cardy-Verlinde formula is satisfied by Schwarzschild-AdS black holes in arbitrary dimensions. However, it is violated by Reissner-Nordström-AdS black holes. Nevertheless, a simple minimal modification does hold, namely
$$\frac{(n2)S}{2\pi l}=\sqrt{E_c\left(2EE_c\mathrm{\Phi }Q\right)},$$
(3.18)
or in Pythagorean form,
$$(E\frac{1}{2}\mathrm{\Phi }Q)^2=\left(\frac{(n2)S}{2\pi l}\right)^2+\left(\frac{\mathrm{\Psi }}{(n2}+\frac{1}{2}\mathrm{\Phi }Q\right)^2.$$
(3.19)
The minimally modified Bekenstein bound becomes
$$E\frac{1}{2}\mathrm{\Phi }Q+\frac{(n2)S}{2\pi l},$$
(3.20)
with equality if and only if
$$(n2)\mathrm{\Psi }+\frac{1}{2}\mathrm{\Phi }Q=0,$$
(3.21)
or
$$(n2)TI+\frac{1}{2}\mathrm{\Phi }Q=0.$$
(3.22)
Because $`\mathrm{\Phi }Q0`$, we see that the AdS<sub>n</sub> Bekenstein bound holds for Reissner-Nordström-AdS black holes.
### 3.4 Rotating black holes without charge
As observed in , the Cardy-Verlinde formula does not hold for rotating black holes in any dimension $`n4`$, if one uses the thermodynamic quantities defined with respect to a non-rotating frame at infinity. Remarkably, however, is was found in (see for earlier discussions) that in all dimensions it does hold if one uses the quantities defined with respect to a frame that rotates with angular velocities $`a_i/l^2`$ at infinity. Because $`EE^{}`$, one obtains an inequality,
$$EE^{}=\frac{1}{2}E_c+\frac{1}{2E_c}\left[\frac{(n2)S}{2\pi l}\right]^2,$$
(3.23)
whence
$$E\frac{(n2)S}{2\pi l}.$$
(3.24)
In other words, although the conserved quantities measured with respect to a frame non-rotating at infinity do not satisfy the Cardy-Verlinde formula, they do satisfy the Bekenstein bound.
In fact, one does not need to pass to the quantities $`E^{}`$ and $`\mathrm{\Omega }_{}^{i}{}_{}{}^{}`$ to establish that rotating AdS black holes satisfy the AdS<sub>n</sub> Bekenstein bound. From (A.12), we have
$$S=\frac{𝒜_{n2}ml}{4(_j\mathrm{\Xi }_j)}\frac{r_+/l}{1+r_+^2/l^2}$$
(3.25)
and hence, since $`x/(1+x^2)\frac{1}{2}`$, we have
$$S\frac{𝒜_{n2}ml}{4(_j\mathrm{\Xi }_j)},$$
(3.26)
with equality if and only if $`r_+=l`$. From results in , the Euclidean action for the $`n`$-dimensional Kerr-AdS black hole is given by
$$I=\frac{\beta 𝒜_{n2}m}{8\pi (_i\mathrm{\Xi }_i)}\frac{l^2r_+^2}{l^2+r_+^2},$$
(3.27)
where $`\beta `$ is the inverse Hawking temperature. Thus we see that $`r_+=l`$ corresponds to the Hawking-Page transition, where the Euclidean action vanishes.
From (A.10) and (A.11) we have
$$E\frac{𝒜_{n2}m(n2)}{8\pi (_j\mathrm{\Xi }_j)},$$
(3.28)
since $`\mathrm{\Xi }_i1`$, with equality if and only if the black hole is non-rotating, i.e. if and only if all $`a_i=0`$. Combining (3.26) and (3.28) gives the AdS<sub>n</sub> Bekenstein bound (3.24), with equality if and only if the black hole is non-rotating, and at the Hawking-Page transition.
## 4 Further Examples
From the previous work, it is natural to wonder whether a simple modification of the Cardy-Verlinde formula, involving the use of $`E^{}`$, and the electrostatic term $`\mathrm{\Phi }Q`$, continues to hold for more complicated black holes with, for example, more than one charge, or the recently-constructed solutions representing rotating black holes with one or more charge, and one or more rotation parameters. In this section we shall study the various cases, and find that while no universal simple modified Cardy-Verlinde formula appears to exist that covers all these cases, we do find in many cases we have studied that
$$E_c^{}(2E^{}E_c^{}\mathrm{\Phi }_iQ_i)\left(\frac{(n2)S}{2\pi l}\right)^2,$$
(4.1)
which implies the electrostatic form of the AdS<sub>n</sub> Bekenstein bound,
$$E\frac{1}{2}\mathrm{\Phi }_iQ_i+\frac{(n2)S}{2\pi l}.$$
(4.2)
### 4.1 Non-rotating black holes with multiple charges
Modifications of the Cardy-Verlinde formula for multi-charge non-rotating black holes in gauged supergravities have been discussed in . Here, we give a related discussion, focussing in particular on the electrostatic AdS-Bekenstein bound, which we show to be satisfied in all the four, five and seven-dimensional examples that we consider.
#### 4.1.1 Four-dimensional multi-charge black holes
The general four-charge solutions in four-dimensional gauged supergravity are given by
$`ds_4^2`$ $`=`$ $`({\displaystyle \underset{i}{}}H_i)^{1/2}fdt^2+({\displaystyle \underset{i}{}}H_i)^{1/2}[f^1d\rho ^2+\rho ^2d\mathrm{\Omega }_2^2],`$
$`A_i`$ $`=`$ $`(1H_i^1)\sqrt{{\displaystyle \frac{q_i+\mu }{q_i}}}dt,X_i=({\displaystyle \underset{j}{}}H_j)^{1/4}H_i^1,`$
$`H_i`$ $`=`$ $`1+{\displaystyle \frac{q_i}{\rho }},f=1{\displaystyle \frac{\mu }{\rho }}+g^2\rho ^2{\displaystyle \underset{i}{}}H_i,`$ (4.3)
where $`X_i=e^{{\scriptscriptstyle \frac{1}{2}}\stackrel{}{a}_i\stackrel{}{\phi }}`$, where $`\stackrel{}{\phi }`$ denotes the three canonically-normalised scalar fields, the $`\stackrel{}{a}_i`$ are constant vectors satisfying $`\stackrel{}{a}_i\stackrel{}{a}_j=4\delta _{ij}1`$, and the Lagrangian is given by
$$e^1=\frac{1}{16\pi }R\frac{1}{8\pi }(\phi )^2\frac{1}{16\pi }\underset{i}{}X_i^2F_i^2+\frac{g^2}{16\pi }\underset{i<j}{}X_iX_j.$$
(4.4)
(We use units where $`G=1`$. See for a more extensive discussion of the notation and conventions that we are using.) Note that in order for the solution to be real, and free of naked singularities, we must have $`q_i0`$.
Straightforward calculations give
$`E`$ $`=`$ $`\frac{1}{2}\mu +\frac{1}{4}{\displaystyle \underset{i}{}}q_i,Q_i=\sqrt{q_i(\mu +q_i)},`$
$`E_c`$ $`=`$ $`\frac{1}{4}{\displaystyle \underset{i}{}}(\rho _++q_i),`$
$`2EE_c{\displaystyle \underset{i}{}}\mathrm{\Phi }_iQ_i`$ $`=`$ $`\frac{1}{4}\rho _+(\mu \rho _+){\displaystyle \underset{i}{}}{\displaystyle \frac{1}{(\rho _++q_i)}},`$
$`S`$ $`=`$ $`{\displaystyle \frac{\pi }{g}}\sqrt{\rho _+(\mu \rho _+)},`$ (4.5)
where $`\rho _+`$ is the radius of the horizon, i.e. the largest root of $`f(\rho )=0`$.
One observes that unless the charges are equal, $`q_i=q`$, the minimally modified Cardy-Verlinde formula (3.18) fails. However, using the fact that the arithmetic mean is never less than the harmonic mean, one finds that
$$E\frac{1}{2}\underset{i}{}Q_i\mathrm{\Phi }_i\frac{1}{2}E_c+\frac{g^2S^2}{2\pi ^2E_c}.$$
(4.6)
Thus although the Cardy-Verlinde formula is violated,
$$S\frac{\pi }{g}\sqrt{E_c(2EE_c\underset{i}{}\mathrm{\Phi }_iQ_i)},$$
(4.7)
nevertheless, the electrostatic AdS-Bekenstein bound still holds,
$$E\frac{1}{2}\underset{i}{}Q_i\mathrm{\Phi }_i+\frac{gS}{\pi },$$
(4.8)
despite the fact that the scalar fields, and hence also the potential $`g^2_{i<j}X_iX_j`$, are space dependent. In these cases $`1/g`$ is only the asymptotic value of the anti-de Sitter radius.
#### 4.1.2 Five-dimensional multi-charge black holes
The general three-charge solutions in five-dimensional gauged supergravity are given by
$`ds_5^2`$ $`=`$ $`({\displaystyle \underset{i}{}}H_i)^{2/3}fdt^2+({\displaystyle \underset{i}{}}H_i)^{1/3}[f^1d\rho ^2+\rho ^2d\mathrm{\Omega }_3^2],`$
$`A_i`$ $`=`$ $`(1H_i^1)\sqrt{{\displaystyle \frac{q_i+\mu }{q_i}}}dt,X_i=({\displaystyle \underset{j}{}}H_j)^{1/3}H_i^1,`$
$`H_i`$ $`=`$ $`1+{\displaystyle \frac{q_i}{\rho ^2}},f=1{\displaystyle \frac{\mu }{\rho ^2}}+g^2\rho ^2{\displaystyle \underset{i}{}}H_i,`$ (4.9)
where $`X_i=e^{{\scriptscriptstyle \frac{1}{2}}\stackrel{}{a}_i\stackrel{}{\phi }}`$, where $`\stackrel{}{\phi }`$ denotes the two canonically-normalised scalar fields, and the relevant Lagrangian is given by
$$e^1=\frac{1}{16\pi }R\frac{1}{8\pi }(\phi )^2\frac{1}{16\pi }\underset{i}{}X_i^2F_i^2+\frac{g^2}{4\pi }\underset{i}{}X_i^1.$$
(4.10)
(We again use units where $`G=1`$.) Again we must have $`q_i0`$ in order to have a real solution with no naked singularities.
After straightforward calculation, we find that
$`E`$ $`=`$ $`\frac{3}{8}\pi \mu +\frac{1}{4}\pi {\displaystyle \underset{i}{}}q_i,Q_i=\frac{1}{4}\pi \sqrt{q_i(\mu +q_i)},`$
$`E_c`$ $`=`$ $`\frac{1}{4}\pi {\displaystyle \underset{i}{}}(\rho _+^2+q_i),`$
$`2EE_c{\displaystyle \underset{i}{}}\mathrm{\Phi }_iQ_i`$ $`=`$ $`\frac{1}{4}\pi \rho _+^2(\mu \rho _+^2){\displaystyle \underset{i}{}}{\displaystyle \frac{1}{\rho _+^2+q_i}},`$
$`S`$ $`=`$ $`{\displaystyle \frac{\pi ^2}{2g}}\rho _+(\mu \rho _+^2)^{1/2},`$ (4.11)
where $`\rho _+`$ is the largest root of $`f(\rho )=0`$. Again we see that the minimally-modified Cardy-Verlinde formula (3.18) fails unless the $`q_i`$ are all equal. Again, using the fact that the arithmetic mean is never less than the harmonic mean, we obtain the inequality
$$E_c(2EE_c\underset{i}{}\mathrm{\Phi }_iQ_i)\left(\frac{3gS}{2\pi }\right)^2,$$
(4.12)
and hence we derive the minimally-modified AdS-Bekenstein bound
$$E\frac{1}{2}\underset{i}{}\mathrm{\Phi }_iQ_i\frac{3gS}{2\pi }.$$
(4.13)
Note that again, $`1/g`$ is only the asymptotic AdS radius.
#### 4.1.3 Seven-dimensional multi-charge black holes
The non-rotating multi-charge black-hole solutions of maximal gauged supergravity in seven dimensions take the form
$`ds_7^2`$ $`=`$ $`(H_1H_2)^{4/5}fdt^2+(H_1H_2)^{1/5}[f^1d\rho ^2+\rho ^2d\mathrm{\Omega }_5^2],`$
$`A_i`$ $`=`$ $`(1H_i^1)\sqrt{{\displaystyle \frac{q_i+\mu }{q_i}}}dt,X_i=(H_1H_2)^{2/5}H_i^1,`$
$`H_i`$ $`=`$ $`1+{\displaystyle \frac{q_i}{\rho ^4}},f=1{\displaystyle \frac{\mu }{\rho ^2}}+g^2\rho ^2H_1H_2,`$ (4.14)
with the charges carried by the $`U(1)\times U(1)`$ gauge fields in the abelian subgroup of $`SO(5)`$. Straightforward calculations show that
$`E`$ $`=`$ $`\frac{5}{16}\pi ^2\mu +\frac{1}{4}\pi ^2{\displaystyle \underset{i}{}}q_i,Q_i=\frac{1}{4}\pi ^2\sqrt{q_i(\mu +q_i)},`$
$`E_c`$ $`=`$ $`\frac{1}{8}\pi ^2(5\rho _+^4+{\displaystyle \underset{i}{}}q_i),`$
$`2EE_c{\displaystyle \underset{i}{}}\mathrm{\Phi }_iQ_i`$ $`=`$ $`\frac{1}{8}\pi ^2\left[5(\mu \rho _+^4)+3{\displaystyle \underset{i}{}}q_i2{\displaystyle \underset{i}{}}{\displaystyle \frac{q_i(\mu +q_i)}{\rho _+^4+q_i}}\right],`$
$`S`$ $`=`$ $`{\displaystyle \frac{\pi ^3}{4g}}\rho _+^2(\mu \rho _+^4)^{1/2}.`$ (4.15)
The verification that these solutions satisfy the electrostatic AdS-Bekenstein bound is slightly more complicated than for the four-dimensional and five-dimensional cases. This is presumably related to the fact that unlike in $`n=4`$ and $`n=5`$, in these seven-dimensional solutions the scalar fields are non-constant even when the charges are set equal. In fact the easiest way to verify the Bekenstein bound is by performing a direct calculation of
$$XE\frac{1}{2}\underset{i}{}\mathrm{\Phi }_iQ_i\frac{5Sg}{2\pi },$$
(4.16)
and verifiying that $`X`$ is non-negative. We find that
$$X=\frac{1}{\rho _+^2}\left[5\rho _+^6(g\rho _+1)^2+q_1q_2g^2+3(q_1+q_2)\rho _+^2(1+g^2\rho _+^2)10\sqrt{(\rho _+^4+q_1)(\rho _+^4+q2)}\right].$$
(4.17)
Using the Maclaurin-Cauchy inequality
$$\underset{i}{}b_i^{q_i}\underset{i}{}b_iq_i,\text{where}\underset{i}{}q_i=1,$$
(4.18)
we may deduce that
$$\sqrt{(\rho _+^4+q_1)(\rho _+^4+q2)}\frac{1}{2}(2\rho _+^4+q_1+q_2),$$
(4.19)
and hence we see that
$$X\frac{1}{\rho _+^2}\left[5\rho _+^6(g\rho _+1)^2+q_1q_2g^2+(q_1+q_2)\rho _+^2[3(g\rho _+\frac{5}{6})^2+\frac{11}{12}]\right],$$
(4.20)
which proves that $`X0`$ and hence the Bekenstein bound is satisfied.
#### 4.1.4 Relation to previous work
A more complicated modification of the Cardy-Verlinde formula has been proposed, in terms of the parameters $`q_i`$ and $`\mu `$ which appear in the multi-charge metrics . This modification incorporates the idea that the pressure of the associated conformal field theory should be reduced by its electrostatic self-repulsion. However, when re-expressed in terms of the fundamental thermodynamic variables $`E`$, $`\mathrm{\Phi }_i`$, $`Q_i`$, $`T`$ and $`S`$, the modification takes on a rather complicated form, which appears to be a somewhat ad hoc construction designed to ensure the continued validity of the Cardy-Verlinde formula in these particular examples.
The modified Cardy-Verlinde formula in is given by
$$S=\frac{2\pi }{(n2)g}\sqrt{\widehat{E}_c(2E2E_q\widehat{E}_c)},$$
(4.21)
where
$$\widehat{E}_c=E_cE_q,E_q=\frac{𝒜_{n2}(n3)}{16\pi }\underset{i}{}q_i.$$
(4.22)
This implies
$$E=\frac{1}{2}\widehat{E}_c+\frac{1}{2\widehat{E}_c}\left(\frac{(n2)gS}{2\pi }\right)^2+E_q,$$
(4.23)
and hence, minimising with respect to $`\widehat{E}_c`$, one again obtains the AdS-Bekenstein bound
$$EE_q+\frac{(n2)gS}{2\pi }\frac{(n2)gS}{2\pi }.$$
(4.24)
It is interesting to compare $`E_q`$, given in (4.22), with $`_i\mathrm{\Phi }_iQ_i`$, which is given by
$$\frac{𝒜_{n2}(n3)}{16\pi }\underset{i}{}\frac{q_i(\mu +q_i)}{\rho _+^{n3}+q_i},$$
(4.25)
from which it follows that $`\mathrm{\Phi }_iQ_iE_q`$, since $`\rho _+^{n3}\mu `$.
### 4.2 Rotating charged black holes
Various solutions for charged rotating black holes in gauged supergravities have been obtained. In this section, we study the generalised Bekenstein bounds for these cases.
#### 4.2.1 Four-dimensional Kerr-Newman-AdS black holes
One might have hoped that, at least for the four-dimensional Kerr-Newman-AdS solution, a simple modification of the Cardy-Verlinde formula would work. However, one finds that the minimally-modified Cardy-Verlinde formula, even in in terms of of quantities measured with respect to the canonically rotating frame, fails. Explicitly
$$E_c^{}\left(2E^{}E_c^{}\mathrm{\Phi }Q\right)=\left(\frac{S}{\pi l}\right)^2+\frac{16a^2Q^2}{l^2},$$
(4.26)
where $`a`$ is given in terms of the extensive quantities by
$$a=\sqrt{\frac{E_{}^{}{}_{}{}^{2}l^4}{4J^2}+l^2}\frac{E^{}l^2}{2|J|}.$$
(4.27)
Note that
$$E_c^{}\left(2E^{}E_c^{}\mathrm{\Phi }Q\right)\left(\frac{S}{\pi l}\right)^2$$
(4.28)
whence
$$E^{}\frac{1}{2}\mathrm{\Phi }Q+\frac{1}{2}E_c^{}+\frac{S^2}{2\pi ^2l^2E_c^{}}\frac{1}{2}\mathrm{\Phi }Q+\frac{S}{\pi l}.$$
(4.29)
But since $`EE^{}`$, we have
$$E\frac{1}{2}\mathrm{\Phi }Q+\frac{S}{\pi l}.$$
(4.30)
In other words, once again, despite the fact that the Cardy-Verlinde formula fails to hold, the electrostatic AdS<sub>4</sub>-Bekenstein bound still holds.
#### 4.2.2 Four-dimensional rotating black holes with pair-wise equal charges
The solution four rotating black holes in four-dimensional gauged supergravity with four charges that are set pairwise equal was given in . The thermodynamic quantities were evaluated in , where it was shown that the conserved energy, angular momentum and charges are given by
$`E`$ $`=`$ $`{\displaystyle \frac{2m+q_1+q_2}{2\mathrm{\Xi }^2}},J={\displaystyle \frac{a(2m+q_1+q_2)}{2\mathrm{\Xi }^2}},`$
$`Q_1`$ $`=`$ $`Q_2={\displaystyle \frac{\sqrt{q_1(2m+q_1)}}{4\mathrm{\Xi }}},Q_3=Q_4={\displaystyle \frac{\sqrt{q_2(2m+q_2)}}{4\mathrm{\Xi }}},`$ (4.31)
where the parameters $`q_i`$ are related to the boost parameters $`\delta _i`$ in by $`q_i=2m\mathrm{sinh}^2\delta _i`$.
We find that
$$E_c^{}\left(2E^{}E_c^{}\underset{i}{}\mathrm{\Phi }_iQ_i\right)=\left(\frac{S}{\pi l}\right)^2+\frac{Xg^2}{4\mathrm{\Xi }r_+},$$
(4.32)
where
$`X`$ $`=`$ $`2a^2(q_1+q_2)(a^2+a^2g^2q_1q_2+g^2q_1^2q_2^2)`$ (4.33)
$`+[q_1q_2(q_1q_2)^2+a^4g^2(q_1^2+6q_1q_2+q_2^2)`$
$`+a^2((q_1q_2)^2+2q_1^2+2q_2^2+3g^2q_1^3q_2+3g^2q_2^3q_1+10g^2q_1^2q_2^2)]r_+`$
$`+(q_1+q_2)(2a^2+2a^4g^2+(q_1q_2)^2+a^2g^2(q_1^2+q_2^2)+10a^2g^2q_1q_2]r_+^2`$
$`+[(q_1q_2)^2+3a^2g^2(q_1^2+q_2^2)+10a^2g^2q_1q_2]r_+^3`$
$`+2a^2g^2(q_1+q_2)r_+^4.`$
Since the parameters $`q_1`$ and $`q_2`$ must be positive, the positivity of $`X`$ is manifest. Hence we obtain the Bekenstein bound
$$E\frac{1}{2}\underset{i}{}\mathrm{\Phi }_iQ_i+\frac{Sg}{\pi }.$$
(4.34)
#### 4.2.3 Five-dimensional rotating black holes with charges
The solution for a rotating black hole in five-dimensional minimal gauged supergravity, with independent rotation parameters in the two orthogonal planes in the transverse space, was obtained recently . The solution can equivalently be viewed as a solution of $`SO(6)`$-gauged supergravity, with three equal charges carried by the $`U(1)^3`$ abelian subgroup. The thermodynamic quantities were also evaluated in , and it was shown that the energy, angular momenta and charge are given in terms of the parameters $`m`$, $`a`$ and $`b`$ in the metric by by
$`E`$ $`=`$ $`{\displaystyle \frac{\pi m(2\mathrm{\Xi }_a+2\mathrm{\Xi }_b\mathrm{\Xi }_a\mathrm{\Xi }_b)+2\pi qabg^2(\mathrm{\Xi }_a+\mathrm{\Xi }_b)}{4\mathrm{\Xi }_a^2\mathrm{\Xi }_b^2}},`$
$`J_a`$ $`=`$ $`{\displaystyle \frac{2\pi ma+\pi qb(1+a^2g^2)}{4\mathrm{\Xi }_a^2\mathrm{\Xi }_b}},`$
$`J_b`$ $`=`$ $`{\displaystyle \frac{2\pi mb+\pi qa(1+b^2g^2)}{4\mathrm{\Xi }_b^2\mathrm{\Xi }_a}},`$
$`Q`$ $`=`$ $`{\displaystyle \frac{\pi \sqrt{3}q}{4\mathrm{\Xi }_a\mathrm{\Xi }_b}},`$ (4.35)
where $`g=1/l`$. We find that
$$E_c^{}(2E^{}E_c^{}\mathrm{\Phi }Q)=\left(\frac{3Sg}{2\pi }\right)^2+\frac{\pi ^2g^2qX}{16\mathrm{\Xi }_a^2\mathrm{\Xi }_b^2[(r_+^2+a^2)(r_+^2+b^2)+abq]r_+},$$
(4.36)
where $`X`$ is given by
$`X`$ $`=`$ $`q^3a^2b^2+q^2ab(18a^2b^2+15(a^2+b^2)r_+^2+4a^2b^2r^2+9r^4)`$ (4.37)
$`+q(r_+^2+a^2)(r_+^2+b^2)(18a^2b^2+9(a^2+b^2)r_+^2+10a^2b^2g^2r_+^2)`$
$`+6ab(r_+^2+a^2)(r_+^2+b^2)^2(1+g^2r_+^2).`$
Since $`X`$ is manifestly positive, at least when $`q`$, $`a`$ and $`b`$ are positive, we therefore obtain the Bekenstein bound
$$E\frac{1}{2}\mathrm{\Phi }Q+\frac{3Sg}{2\pi }.$$
(4.38)
The bound is saturated if $`a=b=q=0`$.
A further solution for charged rotating five-dimensional black holes was obtained in , which corresponds to a case where the three charges carried by the $`U(1)^3SO(6)`$ gauge fields are still all non-zero, but with two of them being equal, and the third related to the first two in a specific way. Thus the solutions in have four independent parameters, namely the mass, the two rotations, and a parameter characterising the charges. For these solutions we find
$$E_c(2E^{}E_c\underset{i}{}\mathrm{\Phi }_iQ_i)=\left(\frac{3Sg}{2\pi }\right)^2+\frac{\pi ^2g^2qX}{16\mathrm{\Xi }_a^2\mathrm{\Xi }_b^2r_+^2[(r_+^2+a^2)(r_+^2+b^2)+qr_+^2]},$$
(4.39)
where $`X`$ is given by
$`X`$ $`=`$ $`3(a^2+b^2)(r_+^2+a^2)(r_+^2+b^2)(1+g^2r_+^2)+q^3r_+^2(3a^2b^2g^2+2r_+^2+(a^2+b^2)g^2r_+^2)`$ (4.40)
$`+q(r_+^2+a^2)(r_+^2+b^2)[3a^2b^2+(a^2+b^2)r_+^2(8+7g^2r_+^2)+g^2r_+^2(a^4+b^4+11a^2b^2)+2r_+^4]`$
$`+q^2r_+^2[5a^2b^2+(a^2+b^2)(7r_+^2+4a^2b^2g^2+5g^2r_+^4)+2g^2r_+^2(a^4+b^4+8a^2b^2)].`$
This is manifestly positive when $`q`$ is positive, and so again the AdS-Bekenstein bound (4.38) is satisfied.
## 5 The Cosmic Censorship Bound
So far, we have established that for many of the stationary black hole solutions we have examined, the electrostatic AdS<sub>n</sub> Bekenstein bound (4.2) is satisfied. In this section, we shall propose that this is a consequence of a more basic and more general lower bound for the energy $`E`$ of any initial data set for the Einstein equations with negative cosmological constant, coupled to a matter system that satisfies the dominant energy condition. This Cosmic Censorship Bound is expressed in terms of the area $`A`$ of the outermost apparent horizon of that initial data set. In the case that there is no charge, the postulated lower bound reads
$$E\frac{(n2)A}{16\pi l}\left[l\left(\frac{A}{𝒜_{n2}}\right)^{{\scriptscriptstyle \frac{1}{n2}}}+\frac{1}{l}\left(\frac{A}{𝒜_{n2}}\right)^{{\scriptscriptstyle \frac{1}{n2}}}\right].$$
(5.1)
Some consequences of this bound, which is a more global extension of Hawking’s variational principle for black holes , are:
* If $`l\mathrm{}`$, then (5.1) reduces to a bound first proposed in $`n=4`$ dimensions by Penrose, who observed that it is a necessary condition for the cosmic censorship hypothesis . (See .)
* The proposed bound (5.1) implies a generalisation of the AdS<sub>n</sub> Bekenstein bound, to the non-stationary case. Noting that the quantity in square brackets in (5.1) must be greater than or equal to 2, we have a generalisation of Bekenstein’s bound to the time-dependent case when one may no longer equate entropy with $`1/4`$ of the area of an apparent horizon:
$$E\frac{(n2)A}{8\pi l}.$$
(5.2)
* The cosmic censorship bound (5.1) is attained for the case of a Schwarzschild-anti-de Sitter black hole, and we propose that this is the only case for which it is saturated.
The strongest physical argument in favour of (5.1) is as follows. Consider an initial data set with total energy $`E_{\mathrm{initial}}`$, a single outermost apparent horizon of area $`A_{\mathrm{initial}}`$, and vanishing total angular momentum and charge. According to standard lore, this should settle down to a stationary state described by a Schwarzschild-de Sitter black hole with total energy $`E_{\mathrm{final}}`$ and event-horizon area $`A_{\mathrm{final}}`$, where
$$E_{\mathrm{final}}=\frac{(n2)A_{\mathrm{final}}}{16\pi l}\left[l\left(\frac{A_{\mathrm{final}}}{𝒜_{n2}}\right)^{{\scriptscriptstyle \frac{1}{n2}}}+\frac{1}{l}\left(\frac{A_{\mathrm{final}}}{𝒜_{n2}}\right)^{{\scriptscriptstyle \frac{1}{n2}}}\right].$$
(5.3)
Assuming cosmic censorhip, the apparent horizon lies inside the event horizon, and since, in the time symmetric case, the apparent horizon is a minimal surface, its area gives a lower bound for the area of the event horzion. Now applying Hawking’s theorem stating that the area of the horion is non-decreasing, we obtain
$$A_{\mathrm{final}}A_{\mathrm{initial}},$$
(5.4)
In anti-de Sitter spacetime, unlike in asymptotically-flat spacetimes, the total energy is constant, and therefore
$$E_{\mathrm{final}}=E_{\mathrm{initial}}.$$
(5.5)
It follows that
$$E_{\mathrm{initial}}\frac{(n2)A_{\mathrm{initial}}}{16\pi l}\left[l\left(\frac{A_{\mathrm{initial}}}{𝒜_{n2}}\right)^{{\scriptscriptstyle \frac{1}{n2}}}+\frac{1}{l}\left(\frac{A_{\mathrm{initial}}}{𝒜_{n2}}\right)^{{\scriptscriptstyle \frac{1}{n2}}}\right].$$
(5.6)
If the initial value set had non-vanishing charge or angular momentum, it would be expected to settle down to the relevant stationary solution carrying those charges or angular momenta. However, he energy due to the charge or angular momentum could be extracted by dropping particles carrying charge or angular momentum into the black hole. In this process, the area of the event horizon cannot decrease, but the energy may. Thus we expect the energy of the black hole with charge or angular momentum to be less than that of a Schwarzschild-AdS black hole with the same event-horizon area.
We shall now review some of the additional evidence for this form of the cosmic censorship bound, and provide some further support for it.
In the case of $`n=4`$ dimensions and time-symmetric initial data, Jang and Wald’s extension of Geroch’s suggested method of proof of the positive mass theorem for asymptotically-flat metrics using the inverse mean-curvature flow may be extended to cover the case of asymptotically anti-de Sitter metrics. Furthermore, if one does so one obtains precisely the proposed lower bound (5.1). The Geroch-Jang-Wald proposed method of proof has been made into a rigorous theorem by Huisken and Ilmanen . It seems plausible, but there is as yet no rigorous proof, that their methods will extend to the anti-de Sitter case.
There is no general proof of the original asymptotically-flat cosmic censorship inequality in higher dimensions. In , it is shown to hold in the case of a collapsing shell, using the obvious generalisation of the four-dimensional calculations in .
There exists a natural generalisation of the inverse mean-curvature flow to higher-dimensional time-symmetric initial-value sets . This might yield a proof of the higher-dimensional inequality if on each level surface
$$[\overline{R}(n2)(n3)]𝑑A0,$$
(5.7)
where $`\overline{R}`$ is the Ricci scalar of the $`(n2)`$-dimensional metric on the level surface, and the integration is over this surface.
In the static spherically-symmetric case, the inequality (5.1) may be proved as follows. We write the metric as
$$ds^2=e^{2\nu (r)}\left(1\frac{2𝔪(r)}{r^{n3}}\right)dt^2+\left(1\frac{2𝔪(r)}{r^{n3}}\right)^1dr^2+r^2d\mathrm{\Omega }_{n2}^2.$$
(5.8)
There is an horizon of area $`A=𝒜_{n2}r_+^{n2}`$ at $`r=r_+`$, where
$$r_+^{n3}=2𝔪(r_+).$$
(5.9)
The Einstein equations
$$R_{\mu \nu }\frac{1}{2}Rg_{\mu \nu }=8\pi T_{\mu \nu }$$
(5.10)
(where we include the contribution of the cosmological constant in $`T_{\mu \nu }`$) imply
$$\frac{d𝔪}{dr}=\frac{8\pi }{n2}r^{n2}T_{\widehat{t}\widehat{t}},$$
(5.11)
where the hats indicate components in an orthonormal frame. We have
$$T_{\mu \nu }=T_{\mu \nu }^{\mathrm{cosmic}}+T_{\mu \nu }^{\mathrm{matter}},$$
(5.12)
where $`T_{\mu \nu }^{\mathrm{cosmic}}`$ is the contribution from the cosmological term. As $`r`$ tends to infinity,
$$𝔪(r)m\frac{r^{n1}}{2l^2},$$
(5.13)
where $`l`$ is the asymptotic de Sitter radius. We may integrate (5.11) from the horizon to infinity, to obtain
$$m=\frac{1}{2}r_+^{n3}+\frac{r_+^{n1}}{2l^2}+\frac{8\pi }{n2}_{r_+}^{\mathrm{}}𝑑rr^{n2}T_{\widehat{0}\widehat{0}}^{\mathrm{matter}}+_{r_+}^{\mathrm{}}𝑑rr^{n2}\left[\frac{n1}{2l^2}+\frac{8\pi }{n2}T_{\widehat{t}\widehat{t}}^{\mathrm{cosmic}}\right].$$
(5.14)
If $`T_{\widehat{t}\widehat{t}}^{\mathrm{cosmic}}`$ is constant and $`T_{\widehat{t}\widehat{t}}^{\mathrm{matter}}`$ satisfies the positive-energy condition, we obtain the cosmic censorship bound, which will be saturated if and only if $`T_{\widehat{t}\widehat{t}}^{\mathrm{matter}}=0`$, i.e. for the Schwarzschild-de Sitter metric.
If $`T_{\widehat{t}\widehat{t}}^{\mathrm{cosmic}}`$ is not constant, because of the presence of varying scalar fields, we still obtain a lower bound for $`m`$ if the integrand is positive. Unfortunately, in the gauged supergravities we have considered the integrand is in fact negative, because the potential is in general more negative than its negative value at its vanishing-scalar stationary point. However, even in this case the cosmic censorship bound may continue to hold, because kinetic energy term for the scalars is positive. Indeed, this is what happens in the examples we have examined.
### 5.1 Cosmic censorship for Kerr-AdS black holes in arbitrary dimension
It is straightforward to show that the general Kerr-AdS black holes in arbitrary spacetime dimension $`n`$ satisfy the cosmic censorship bound (5.1). First, we note that one can characterise the Kerr-AdS metrics by their rotation parameters $`a_i`$, together with the radius $`r_+`$ of the outer horizon. The mass parameter $`m`$ appearing in the metric (A.2) is then solved for using $`V(r_+)=m`$. It is then helpful to introduce a new parameterisation in terms of $`y_i`$ and $`z`$ instead of $`a_i`$ and $`r_+`$, where
$$y_i\frac{r_+^2+a_i^2}{r_+^2\mathrm{\Xi }_i},z=\frac{r_+}{l}.$$
(5.15)
Clearly we must have have $`y_i1`$, $`z0`$, and
$$r_+=zl,a_i^2=\frac{(y_i1)z^2l}{(1+z^2y_i)},\mathrm{\Xi }_i=\frac{1+z^2}{1+z^2y_i}.$$
(5.16)
We begin by considering the case when $`n=2N+1`$ is odd. From (A.4), (A.10) and (A.12), the energy $`E`$ is given by
$$E=\frac{r_+^{n3}𝒜_{n2}(\underset{j}{}y_j)}{16\pi }\left(n2+2z^2\underset{i}{}y_i\right),$$
(5.17)
whilst the right-hand side of (5.1) is given by
$$\frac{(n2)r_+^{n2}(\underset{j}{}y_j)}{16\pi l}\left[z(\underset{i}{}y_i)^{{\scriptscriptstyle \frac{1}{n2}}}+\frac{1}{z}(\underset{i}{}y_i)^{{\scriptscriptstyle \frac{1}{n2}}}\right],$$
(5.18)
and so to show that the cosmic censorship bound is satisfied, we must show that
$$(n2)\left[1(\underset{i}{}y_i)^{{\scriptscriptstyle \frac{1}{n2}}}\right]+z^2\left[2\underset{i}{}y_i1(n2)(\underset{i}{}y_i)^{{\scriptscriptstyle \frac{1}{n2}}}\right]0.$$
(5.19)
The first bracketed term in (5.19), i.e. the term independent of $`z`$, is manifestly positive since $`y_i1`$. For the terms at order $`z^2`$ we may use the Maclaurin-Cauchy inequality (4.18) to show that
$$(\underset{i}{}y_i)^{{\scriptscriptstyle \frac{1}{n2}}}\frac{1}{n2}\underset{i}{}y_i+\frac{n3}{2(n2)}.$$
(5.20)
Substituting this into the second bracketed term in (5.19) shows that
$$2\underset{i}{}y_i1(n2)(\underset{i}{}y_i)^{{\scriptscriptstyle \frac{1}{n2}}}\underset{i=1}{\overset{N}{}}(y_i1)0,$$
(5.21)
and hence the cosmic censorship bound is proved.
In the case of even dimensions $`n=2N+2`$, an analogous calculation shows that cosmic censorship is satisfied if
$$(n2)\left[1(\underset{i}{}y_i)^{{\scriptscriptstyle \frac{1}{n2}}}\right]+z^2\left[2\underset{i}{}y_i(n2)(\underset{i}{}y_i)^{{\scriptscriptstyle \frac{1}{n2}}}\right]0.$$
(5.22)
Again using (4.18), we can show that
$$(\underset{i}{}y_i)^{{\scriptscriptstyle \frac{1}{n2}}}\frac{1}{n2}\underset{i}{}y_i+\frac{1}{2}$$
(5.23)
and so since $`y_i1`$, the inequality (5.22) can indeed be seen to hold.
### 5.2 Cosmic censorship for four-dimensional charged rotating black holes
The cosmic censorship bound (5.1) can be generalised in the case of charged black-hole solutions . In four dimensions, it becomes
$$E\frac{A}{8\pi l}\left[l\left(\frac{A}{4\pi }\right)^{1/2}+\frac{1}{l}\left(\frac{A}{4\pi }\right)^{1/2}+lQ^2\left(\frac{A}{4\pi }\right)^{3/2}\right],$$
(5.24)
with equality being attained for the Reissner-Nordström-AdS solution. Calculating $`E^2`$ minus the square of the right-hand side of (5.24) for the Kerr-Newman-AdS solution, we find
$$E^2(\mathrm{RHS})^2=\frac{4(1+\mathrm{g}^2\mathrm{r}_+^2)[\mathrm{a}^2+\mathrm{q}^2+\mathrm{r}_+^2+\mathrm{g}^2\mathrm{r}_+^2(\mathrm{r}_+^2+\mathrm{a}^2)]^2}{\mathrm{\Xi }^4\mathrm{r}_+^2(\mathrm{r}_+^2+\mathrm{a}^2)},$$
(5.25)
which is manifestly positive, thus demonstrating that the inequality (5.24) is obeyed in this case.
### 5.3 Cosmic censorship for five-dimensional charged rotating black holes
For solutions of five-dimensional minimal gauged supergravity, the generalised cosmic censorship bound can be written as
$$\frac{8E}{3\pi }\left(\frac{A}{2\pi ^2}\right)^{{\scriptscriptstyle \frac{2}{3}}}+g^2\left(\frac{A}{2\pi ^2}\right)^{{\scriptscriptstyle \frac{4}{3}}}+\frac{16Q^2}{3\pi ^2}\left(\frac{A}{2\pi ^2}\right)^{{\scriptscriptstyle \frac{2}{3}}},$$
(5.26)
with equality being attained in the case of the non-rotating Reissner-Nordström-AdS black hole. From the results obtained in , we find that the inequality (5.26) translates into the requirement that
$`\frac{2}{3}h[3+z^2(y_1+y_2)]+\frac{1}{3}y_1y_2[3z^2+2z^2(y_1+y_2)]`$
$`+{\displaystyle \frac{h^2(1+z^2)}{3(y_11)(y_21)}}[3z^2+2z^2(y_1+y_2)](y_1y_2+h)^{{\scriptscriptstyle \frac{2}{3}}}z^2(y_1y_2+h)^{{\scriptscriptstyle \frac{4}{3}}}`$
$`{\displaystyle \frac{h^2(y_1y_2+h)^{2/3}}{(y_11)(y_21)}}(1+z^2y_1)(1+z^2y_2)0,`$ (5.27)
where
$$y_1=\frac{r_+^2+a^2}{r_+^2\mathrm{\Xi }_a},y_2=\frac{r_+^2+b^2}{r_+^2\mathrm{\Xi }_b},z=gr_+,h=\frac{abq}{r_+^4\mathrm{\Xi }_a\mathrm{\Xi }_b},$$
(5.28)
with $`y_i1`$, $`z>0`$, $`h0`$. We have studied (5.27) numerically and find that it appears to be satisfied for all allowed values of the parameters $`(y_1,y_2,z,h)`$, and thus it appears that the generalised cosmic censorship bound is obeyed by the five-dimensional charged rotating AdS black holes obtained in .
## 6 An Upper Bound for the Temperature?
It is well known that the presence of matter with a positive energy density tends to reduce the temperature of a black hole, because of the redshift produced by the gravitational field of the matter. It is also well known that charged or rotating black holes tend to have a smaller temperature for the same entropy than their neutral or non-rotating versions. A general explanation for this observation was provided by Visser in the static spherically-symmetric case in four dimensions with no cosmological term . In this section we shall generalise Visser’s observation, and apply it to the Hawking-Page transition.
For the spherically-symmetric static metric (5.8), the Einstein equations imply
$`{\displaystyle \frac{d𝔪}{dr}}`$ $`=`$ $`{\displaystyle \frac{8\pi r^{n2}T_{\widehat{t}\widehat{t}}}{(n2)}}`$ (6.1)
$`{\displaystyle \frac{d\nu }{dr}}`$ $`=`$ $`{\displaystyle \frac{8\pi r^{n2}(T_{\widehat{t}\widehat{t}}+T_{\widehat{r}\widehat{r}})}{(n2)[r^{n3}2𝔪(r)]}},`$ (6.2)
where the hats indicate components in an orthonormal frame.
The surface gravity is
$$\kappa =2\pi T=\frac{1}{2r_+}e^{\nu (r_+)}(n32r^{4n}\frac{d𝔪}{dr})).$$
(6.3)
This becomes
$$\kappa =2\pi T=\frac{1}{2r_+}e^{\nu (r_+)}\left[(n3)\frac{16\pi r_+^2}{(n2)}(T_{\widehat{t}\widehat{t}}^{\mathrm{cosmic}}+T_{\widehat{t}\widehat{t}}^{\mathrm{matter}})\right].$$
(6.4)
If $`T_{\widehat{t}\widehat{t}}^{\mathrm{cosmic}}`$ is constant, then (6.4) becomes
$$\kappa =2\pi T=\frac{1}{2r_+}e^{\nu (r_+)}\left[(n3)+(n1)g^2r_+^2\frac{16\pi r_+^2}{(n2)}T_{\widehat{t}\widehat{t}}^{\mathrm{matter}}\right].$$
(6.5)
If the matter satisfies the dominant energy condition then
$$T_{\widehat{t}\widehat{t}}^{\mathrm{matter}}|T_{\widehat{r}\widehat{r}}^{\mathrm{matter}}|0.$$
(6.6)
Moreover
$$\nu (r)=_r^{\mathrm{}}\frac{8\pi r_{}^{}{}_{}{}^{n2}(T_{\widehat{t}\widehat{t}}^{\mathrm{matter}}+T_{\widehat{r}\widehat{r}}^{\mathrm{matter}})}{(n2)[r_{}^{}{}_{}{}^{n3}2𝔪(r^{})]}𝑑r^{}.$$
(6.7)
Thus $`\nu (r)`$ will be non-positive and
$$\overline{)4\pi T\frac{(n3)}{r_+}+(n1)g^2r_+.}$$
(6.8)
If $`T_{\widehat{t}\widehat{t}}^{\mathrm{cosmic}}`$ is not constant, one might expect the kinetic term for the scalars to compensate for any extra positive contribution from $`T_{\widehat{t}\widehat{t}}^{\mathrm{cosmic}}`$, and the inequality to continue to hold.
In the Schwarzschild-AdS case the inequality (6.8) becomes an equality. The minimum value of the right-hand side occurs at
$$r_+=\frac{1}{g}\sqrt{\frac{n3}{n1}},$$
(6.9)
at which
$$T=\frac{g\sqrt{(n1)(n3)}}{2\pi }.$$
(6.10)
This lower bound for the temperature is associated with the Hawking-Page phase transition. Below this temperature, there is no black-hole solution, whilst above it, there are two. The Hawking-Page transition itself occurs at $`r_+=1/g`$, for which $`T=(n2)g/(2\pi )`$. For temperatures greater than (6.10) but smaller than than $`(n2)g/(2\pi )`$, both Schwarzschild-AdS solutions have larger Euclidean action than that of anti-de Sitter spacetime.
The general inequality (6.8) for spherically-symmetric black holes may be recast in the form
$$4\pi T(n3)\left(\frac{A}{𝒜_{n2}}\right)^{{\scriptscriptstyle \frac{1}{n2}}}+(n1)g^2\left(\frac{A}{𝒜_{n2}}\right)^{{\scriptscriptstyle \frac{1}{n2}}},$$
(6.11)
where $`A`$ is the area of the outer horizon. Equality is achieved in the case of Schwarzschild-AdS black holes.
One might think that when the inequality is expressed in the form (6.11), it would continue to hold for rotating as well as non-rotating black holes. In other words, one might conjecture that the minimum temperature, as a function of entropy, is always less than or equal to the minimum temperature of the Schwarzschild-AdS case. In fact, we find that the bound (6.11) is obeyed bay all Kerr-AdS black holes in $`n=4`$ and $`n=5`$ diemnsions. However, counterexamples can be found for Kerr-AdS black holes in all dimensions greater than or equal to 6.
To discuss the situation in arbitrary dimensions, it is again helpful to use the parameterisation introduced in (5.15). In odd dimensions $`n=2N+1`$, showing that the inequality (6.11) is obeyed is equivalent to showing that
$$(n3)(\underset{i}{}y_i)^{{\scriptscriptstyle \frac{1}{n2}}}2\underset{i}{}y_i^1+2+(n1)[(\underset{i}{}y_i)^{{\scriptscriptstyle \frac{1}{n2}}}1]z^20.$$
(6.12)
This must hold for the $`z^0`$ and $`z^2`$ terms independently. It is clearly true for the $`z^2`$ terms, since $`y_i1`$, and so checking the temperature bound for odd-dimensional Kerr-AdS black holes amounts to checking whether
$$(N1)(\underset{i=1}{\overset{N}{}}y_i)^{{\scriptscriptstyle \frac{1}{2N1}}}\underset{i=1}{\overset{N}{}}y_i^11$$
(6.13)
for all $`y_i1`$. It is straightforward to see that in five dimensions, for which $`N=2`$, the function
$$(y_1y_2)^{1/3}\frac{1}{y_1}\frac{1}{y_2}+1$$
(6.14)
is non-negative for all $`y_i1`$, since it can be written in the manifestly non-negative form
$$(y_1y_2)^1\left\{(y_11)(y_21)+[(y_1y_2)^{2/3}1]\right\}.$$
(6.15)
This shows that all Kerr-AdS black holes in five dimensions obey the temperature bound (6.11). However, if $`N3`$ it is clear that the inequality in (6.13) can be violated for valid choices of the parameters $`y_i`$. For example, we can take $`y_1=y_2=1`$, thus ensuring that the right-hand side of (6.13) is at least 1, and then choose the remaining $`y_i`$ large enough so that the left-hand side of (6.13) is less than 1. Clearly if $`z`$, which is independently specifiable and subject only to the restriction $`z0`$, is chosen to be sufficiently small, then the order $`z^2`$ terms in (6.12) will not be sufficiently positive to overwhelm the negative contribution from the terms at order $`z^0`$, and so (6.12) will be violated.
In even dimensions $`n=2N+2`$, the analogous calculation shows that for these Kerr-AdS black holes the temperature inequality (6.11) is equivalent to
$$(n3)(\underset{i}{}y_i)^{{\scriptscriptstyle \frac{1}{n2}}}2\underset{i}{}y_i^1+1+(n1)[(\underset{i}{}y_i)^{{\scriptscriptstyle \frac{1}{n2}}}1]z^20.$$
(6.16)
Again, the terms at order $`z^2`$ are clearly positive, and so showing that (6.16) is satisfied is equaivalent to showing that
$$(N\frac{1}{2})(\underset{i=1}{\overset{N}{}}y_i)^{{\scriptscriptstyle \frac{1}{2N}}}\underset{i=1}{\overset{N}{}}y_i^1\frac{1}{2}.$$
(6.17)
Clearly this inequality is always obeyed in four dimensions, corresponding to $`N=1`$, since the function
$$\frac{1}{2}y_1^{1/2}\frac{1}{y_1}+\frac{1}{2}=\frac{1}{2}y_1^1[(y_11)+(y_1^{1/2}1)]$$
(6.18)
is manifestly non-negative for all $`y_11`$. Thus all Kerr-AdS black holes in four dimensions obey the temperature bound (6.11). It is clear, however, that the inequality (6.17) can be violated for valid choices of the parameters, $`y_i1`$, if $`N`$ is greater than or equal to 2 (i.e. in even dimensions $`n`$ greater than or equal to 6). For example, we could take $`y_1=1`$, and then by taking the remaining $`y_i`$ large enough, the left-hand side of (6.17) can be made arbitrarily small, while the right-hand side exceeds $`\frac{1}{2}`$. By also taking $`z`$ sufficiently small, this means that (6.16) can be violated when $`N2`$.
More generally, one can see that in all dimensions $`n6`$, there exist regions in the $`(y_i,z)`$ parameter space for which the inequalities (6.12) or (6.16) are violated, and using (5.16) these can be translated back into regions in the parameter space for $`(a_i,r_+)`$ for which the temperature inequality (6.11) is not obeyed.
It is also worth remarking that similar conclusions are obtained if we consider asymptotically flat, rather than asymptotically AdS, rotating black holes. From (5.15) we see that the asymptotically flat case, which arises when $`g=0`$, corresponds to taking $`z`$ to zero. We saw above that in the asymptotically AdS case there were terms in the inequality that were of order $`z^2`$, and terms of order $`z^0`$. The former were always consistent with the inequality, and it was the $`z^0`$ terms, which are the ones that survive in the $`g0`$ limit, that had to be investigated in more detail. Thus the conclusions for asymptotically-flat rotating black holes are the same as those for asymptoticallty-AdS rotating black holes, namely that violations of the temperature inequality (6.11) can occur in all dimensions 6 and higher, in cases where some of the rotations are small and some are large.
Finally, we should emphasise that our finding of violations of the inequality (6.11) does not contradict or threaten any cherished beliefs. The inequality was derived for static solutions, and, although commonly such considerations can lead to conjectured inequalities that have a wider range of applicability, as in the case of the cosmic censorship bound (5.1), there is no a priori reason why it should do so in this case. The result could, perhaps, be viewed as a salutary reminder that a conjecture that holds up well in low dimensions may run into trouble in higher dimensions.
## 7 Cosmological Event Horizons
In this section we take the cosmological constant to be positive, thus
$$R_{\mu \nu }=\frac{n1}{l^2}g_{\mu \nu }.$$
(7.1)
In order to obtain the necessary formulae one makes the substitution $`l^2l^2`$.
In the case of pure de Sitter spacetime, $`dS_n`$, one has $`m=a_i=0`$ and there is a cosmological horizon at $`r=l`$. If $`m>0`$, this is at
$$r=r_Cl.$$
(7.2)
Inside the cosmological horizon there will, in general, be a black hole horizon, at $`r=r_H`$ say.
If the spacetime dimension $`n`$ is even, then the area of the cosmological horizon $`A_C`$ is easily seen to be bounded above by the value in pure $`dS_n`$,
$$A_C𝒜_{n2}l^{n2}.$$
(7.3)
(For the four-dimensional case, see .) In some sense, $`S_{\mathrm{max}}=\frac{1}{4}𝒜_{n2}l^{n2}`$ represents the largest amount of information that can ever be lost through the cosmological horizon.
By manipulations similar to those in the case of a negative cosmological constant, one may convince oneself that
$$(\frac{A_C}{𝒜_{n2}})^{\frac{n3}{n2}}\left(1\frac{1}{l^2}(\frac{A_C}{𝒜_{n2}})^{\frac{2}{n2}}\right)(\frac{A_H}{𝒜_{n2}})^{\frac{n3}{n2}}\left(1\frac{1}{l^2}(\frac{A_H}{𝒜_{n2}})^{\frac{2}{n2}}\right)$$
(7.4)
with equality only for the Kottler, i.e. Schwarzschild-de Sitter, solution. In the Reissner-Nordström-de Sitter case, one can do more, and obtain
$$(\frac{A_C}{𝒜_{n2}})^{\frac{n3}{n2}}\left(1\frac{1}{l^2}(\frac{A_C}{𝒜_{n2}})^{\frac{2}{n2}}\right)\mathrm{\Phi }_CQ(\frac{A_H}{𝒜_{n2}})^{\frac{n3}{n2}}\left(1\frac{1}{l^2}(\frac{A_H}{𝒜_{n2}})^{\frac{2}{n2}}\right)\mathrm{\Phi }_HQ,$$
(7.5)
where $`\mathrm{\Phi }_C`$ and $`\mathrm{\Phi }_H`$ are the electrostatic potentials of the cosmological horizon and black hole horizon. Actually, only the potential difference between the two horizons enters the inequality, as must be the case by gauge invariance.
An interesting question is whether there is an upper bound to the area of a black hole in a background de Sitter spacetime . For the Schwarzschild-de Sitter solution, there is such an upper bound, which occurs when the two horizons coincide. This happens when
$$r_C=r_H=l\sqrt{\frac{n3}{n1}}.$$
(7.6)
It is natural therefore to conjecture that more generally,
$$A_H𝒜_{n2}l^{n2}\left(\frac{n3}{n1}\right)^{{\scriptscriptstyle \frac{1}{2}}(n2)}.$$
(7.7)
It is easy to check that this is true for the Reissner-Nordström-de Sitter solution in any dimension. The radius $`r`$ at which the two horizons coincide is easily seen to be less than $`l\sqrt{\frac{n3}{n1}}`$ and so the area of a charged black hole in a background de Sitter spacetime is indeed never greater than $`𝒜_{n1}l^{n2}\left(\frac{n3}{n1}\right)^{n2}`$.
In the rotating case the black hole and cosmological horizons coincide when $`m`$, considered as a function of $`r`$, has a vanishing derivative. One may check that this happens at
$$r_C=r_H<l\sqrt{\frac{n3}{n1}}.$$
(7.8)
We have verified that the inequality (7.7) is satisfied for rotating black holes in a variety of cases. These include the general Kerr-de Sitter metrics in four and six dimensions; the Kerr-de Sitter metrics with equal angular momenta in five and seven dimensions, and the Kerr-de Sitter metrics with equal angular momenta in all even dimensions.
Here, we shall just present the proof for the case of Kerr-de Sitter metrics with equal angular momenta in all even dimensions $`n=2N+2`$. From the formulae collected in the appendix, and setting $`a_i=a`$, we can show that the condition for double root $`r_C=r_H`$ can be expressed as
$$l^2=\frac{r_H^2[(n1)r_H^2a^2]}{(n3)r_H^2a^2},$$
(7.9)
whilst the area of the horizon is given by
$$A_H=𝒜_{n2}l^{n2}\left(\frac{r_H^2+a^2}{l^2+a^2}\right)^{{\scriptscriptstyle \frac{n2}{2}}}.$$
(7.10)
It is straightforward to see that
$$\frac{r_H^2+a^2}{l^2+a^2}=\frac{n3}{n1}\frac{2a^2}{(n1)[(n1)r_H^2a^2]}\frac{n3}{n1},$$
(7.11)
thus proving that these Kerr-AdS black holes indeed satisfy the bound (7.7).
## 8 Conclusions
In this paper, we have studied the relation between the thermodynamics of the bulk variables describing rotating black holes in gauged supergravities, and the corresponding variables in the boundary CFT. We have shown that by using the standard UV/IR connection between the bulk and the boundary, bulk quantities that satisfy the first law of thermodynamics are mapped into boundary quantities that likewise satisfy the first law of thermodynamics. An important point when considering rotating AdS black holes is that the natural conformal boundary at large distance is defined with respect to a coordinate frame in which the metric is asymptotically static and asymptotically spherical.
Our results have clarified some previous puzzling claims in the literature, including in particular the assertion that to get boundary quantities that satisfy the first law one must start from bulk quantities that do not. This assertion was based on calculations performed in a specific frame that is rotating and non-spherical at infinity, with an angular velocity that depends on the rotation parameters of the black hole. In our opinion this is not a convenient or natural frame to use, and we believe that this is why it led to apparently puzzling conclusions.
In this context, it is perhaps worth remarking that in much of the literature on the subject of rotating AdS black holes, there is a tendency to refer to just two choices of frame, namely the frame that is asymptotically static, and the frame with rotation rates given by (2.7) at infinity. In our opinion, the discussion of whether the thermodynamic quantities such as energy should be defined with respect to the former or the latter frame is misplaced. In reality there are infinitely many different frames that could be chosen, with arbitrary choices of asymptotic rotation rates. Asymptotically static frames enjoy a preferred status, and, as we showed in , the quantities defined in an asymptotically static frame satisfy the first law of thermodynamics. Frames whose asymptotic rotation rates depend upon the black-hole rotation parameters (such as the frame specified by (2.7)) seem to be particularly unnatural from the point of view of thermodynamic discussions, since one would need to include extra terms to compensate for the changing centrifugal and Coriolis contributions to the energy. Furthermore, if physical results (such as the energy) depend upon the choice of frame (in the sense that they depend upon the choice of timelike Killing vector used to define the energy, etc.), then a justification is called for as to why some specific frame, rather than one with some other rate of rotation, has been chosen. We have argued that for thermodynamic discussions, at least, the asymptotically static frame is the physically natural one.
Some of the results in show that in a different context, namely the discussion of the Cardy-Verlinde formula, there is a significant merit to considering the energy function $`E^{}`$ defined with respect to the frame with asymptotic angular velocity given by (2.7). It was shown in five dimensions in , and in higher dimensions in , that the Cardy-Verlinde formula (3.5) holds for rotating AdS black holes, provided that one uses energies and angular velocities measured with respect to the frame with angular velocities given by (2.7) at infinity. As far as we are aware, there is no a priori reason why a frame with this particular angular velocity should be singled out in this context, but the observation is certainly an interesting one.
Results had also been obtained for modifications to the Cardy-Verlinde formula when applied to non-rotating charged black holes . The recent construction of black holes that have both rotation and charge has provided a wider spectrum of examples where the Cardy-Verlinde formula can be tested, and we have reported some results in the present paper. It seems that there is no natural and universal modification which encompasses all the cases. Nevertheless, as we have shown, there is a closely related and physically more significant result that does always hold for all the rotating charged black holes, namely the existence of an AdS-Bekenstein bound (3.10), and its electrostatic generalisation (4.2). The AdS-Bekenstein bound is itself a consequence of a more fundamental cosmic censorship bound, and we have explicitly demonstrated for many of the rotating and charged black holes that this bound is indeed satisfied.
We have also examined the question of whether there is an upper bound for the temperature as a function of entropy for black holes in AdS backgrounds. In four and five dimensions, we found that the temperature of a rotating AdS black hole is always less than that of the Schwarzschild-AdS black hole of the same entropy. In six or more dimensions, by contrast, we find that for certain choices of the rotation parameters, the rotating AdS black hole can have a higher temperature than the Schwarzschild-AdS black hole of the same entropy. Finally, we discussed area inequalities for rotating black holes with a positive cosmological constant, for which there is a cosmological horizon as well as a black hole horizon.
Acknowledgements
G.W.G. thanks the Centre for Mathematical Sciences, Zheijiang University, Hangzhou, and C.N.P. thanks the Relativity and Cosmology group, Cambridge, for hospitality during the course of this work.
## Appendix A General Kerr-AdS Black Hole in Arbitrary Dimensions
In this appendix, we collect some general results on rotating asymptotically AdS black holes in arbitrary dimension $`n`$. The solution was obtained in $`n=4`$ in , in $`n=5`$ in , and in $`n6`$ in . Results on the thermodynamics of the arbitrary-dimension rotating AdS black holes were obtained in .
The metrics have $`N[(n1)/2]`$ independent rotation parameters $`a_i`$ in $`N`$ orthogonal 2-planes. We have $`n=2N+1`$ when $`n`$ is odd, and $`n=2N+2`$ when $`n`$ is even. Defining $`ϵ(n1)`$ mod 2, so that $`n=2N+1+ϵ`$, the metrics can be described by introducing $`N`$ azimuthal angles $`\varphi _i`$, and $`(N+ϵ)`$ “direction cosines” $`\mu _i`$ obeying the constraint
$$\underset{i=1}{\overset{N+ϵ}{}}\mu _i^2=1.$$
(A.1)
In Boyer-Lindquist type coordinates that are asymptotically non-rotating, the metrics are given by
$`ds^2`$ $`=`$ $`W(1+r^2l^2)dt^2+{\displaystyle \frac{2m}{U}}\left(Wdt{\displaystyle \underset{i=1}{\overset{N}{}}}{\displaystyle \frac{a_i\mu _i^2d\phi _i}{\mathrm{\Xi }_i}}\right)^2+{\displaystyle \underset{i=1}{\overset{N}{}}}{\displaystyle \frac{r^2+a_i^2}{\mathrm{\Xi }_i}}\mu _i^2d\phi _i^2`$ (A.2)
$`+{\displaystyle \frac{Udr^2}{V2m}}+{\displaystyle \underset{i=1}{\overset{N+ϵ}{}}}{\displaystyle \frac{r^2+a_i^2}{\mathrm{\Xi }_i}}d\mu _i^2{\displaystyle \frac{l^2}{W(1+r^2l^2)}}\left({\displaystyle \underset{i=1}{\overset{N+ϵ}{}}}{\displaystyle \frac{r^2+a_i^2}{\mathrm{\Xi }_i}}\mu _id\mu _i\right)^2,`$
where
$`W`$ $``$ $`{\displaystyle \underset{i=1}{\overset{N+ϵ}{}}}{\displaystyle \frac{\mu _i^2}{\mathrm{\Xi }_i}},Ur^ϵ{\displaystyle \underset{i=1}{\overset{N+ϵ}{}}}{\displaystyle \frac{\mu _i^2}{r^2+a_i^2}}{\displaystyle \underset{j=1}{\overset{N}{}}}(r^2+a_j^2),`$ (A.3)
$`V`$ $``$ $`r^{ϵ2}(1+r^2l^2){\displaystyle \underset{i=1}{\overset{N}{}}}(r^2+a_i^2),\mathrm{\Xi }_i1a_i^2l^2.`$ (A.4)
They satisfy $`R_{\mu \nu }=(n1)l^2g_{\mu \nu }`$.
The constant-$`r`$ spatial surfaces at large distance are inhomogeneously distorted $`(n2)`$-spheres. Making the coordinate transformations
$$\mathrm{\Xi }_iy^2\widehat{\mu }_i^2=(r^2+a_i^2)\mu _i^2,$$
(A.5)
where $`_i\widehat{\mu }_i^2=1`$, the metrics at large $`y`$ approach the standard AdS form
$$d\overline{s}^2=(1+y^2l^2)dt^2+\frac{dt^2}{1+y^2l^2}+y^2\underset{k=1}{\overset{N+ϵ}{}}(d\widehat{\mu }_k^2+\widehat{\mu }_k^2d\phi _k^2),$$
(A.6)
with round $`(n2)`$-spheres of volume $`𝒜_{n2}y^{n2}`$ at radius $`y`$, where $`𝒜_{n2}`$ is the volume of the unit $`(n2)`$-sphere.
The angular velocities of the horizon, measured relative to the frame that is non-rotating at infinity, are given by
$$\mathrm{\Omega }^i=\frac{(1+r_+^2l^2)a_i}{r_+^2+a_i^2},$$
(A.7)
and the angular momenta are
$$J_i=\frac{ma_i𝒜_{n2}}{4\pi \mathrm{\Xi }_i(_j\mathrm{\Xi }_j)},$$
(A.8)
where
$$𝒜_{n2}=\frac{2\pi ^{(n1)/2}}{\mathrm{\Gamma }[(n1)/2]}$$
(A.9)
is the volume of the unit $`(n2)`$-sphere. As shown in , the energy of the black hole, again measured in the asymptotically static frame, is given by
$`n\text{= odd}:`$ $`\overline{)E={\displaystyle \frac{m𝒜_{n2}}{4\pi (_j\mathrm{\Xi }_j)}}\left({\displaystyle \underset{i=1}{\overset{N}{}}}{\displaystyle \frac{1}{\mathrm{\Xi }_i}}{\displaystyle \frac{1}{2}}\right)}`$ (A.10)
$`n\text{= even}:`$ $`\overline{)E={\displaystyle \frac{m𝒜_{n2}}{4\pi (_j\mathrm{\Xi }_j)}}{\displaystyle \underset{i=1}{\overset{N}{}}}{\displaystyle \frac{1}{\mathrm{\Xi }_i}}}`$ (A.11)
The area of the event horizon is given by
$$A=𝒜_{n2}r_+^{ϵ1}\underset{i}{}\frac{r_+^2+a_i^2}{\mathrm{\Xi }_i}.$$
(A.12)
The Euclidean action was also calculated in , and found to be given by
$$I=\frac{\beta 𝒜_{n2}}{8\pi _i\mathrm{\Xi }_i}\left(mr_+^ϵl^2\underset{j}{}(r_+^2+a_j^2)\right),$$
(A.13)
where $`\beta `$ is the inverse of the Hawking temperature, which is given by
$`n=\text{ odd}:`$ $`2\pi T=r_+(1+r_+^2l^2){\displaystyle \underset{i}{}}{\displaystyle \frac{1}{r_+^2+a_i^2}}{\displaystyle \frac{1}{r_+}},`$ (A.14)
$`n=\text{ even}:`$ $`2\pi T=r_+(1+r_+^2l^2){\displaystyle \underset{i}{}}{\displaystyle \frac{1}{r_+^2+a_i^2}}{\displaystyle \frac{1r_+^2l^2}{2r_+}}.`$ (A.15)
The traditional asymptotically-rotating Boyer-Lindquist coordinate system, where the angular velocities at infinity are given by (2.7), is related to the coordinates in (A.2) by defining
$$\phi _i^{}=\phi _ia_il^2t,t^{}=t.$$
(A.16)
The energy $`E^{}`$ calculated in the asymptotically-rotating frame, i.e. using the timelike Killing vector $`/t^{}`$, is given by
$$E^{}=E\frac{1}{l^2}\underset{i}{}a_iJ_i=\frac{(n2)m𝒜_{n2}}{8\pi (_j\mathrm{\Xi }_j)}.$$
(A.17)
|
warning/0506/quant-ph0506263.html
|
ar5iv
|
text
|
# Demonstration of an optical quantum controlled-NOT gate without path interference
## Abstract
We report the first experimental demonstration of an optical quantum controlled-NOT gate without any path interference, where the two interacting path interferometers of the original proposals (Phys. Rev. A 66, 024308 (2001), Phys. Rev. A 65, 012314 (2002)) have been replaced by three partially polarizing beam splitters with suitable polarization dependent transmittances and reflectances. The performance of the device is evaluated using a recently proposed method (Phys. Rev. Lett. 94, 160504 (2005)), by which the quantum process fidelity and the entanglement capability can be estimated from the 32 measurement results of two classical truth tables, significantly less than the 256 measurement results required for full quantum tomography.
preprint: APS/123-QED
Quantum computing promises to solve problems such as factoring large integers SHOR97 and searching over a large database GRO97 efficiently. One of the greatest challenges is to implement the basic elements of quantum computation in a reliable physical system and to evaluate the performance of the operation in a sufficient manner. In one of the earliest proposals for implementing quantum computationMIL88 , each qubit was encoded in a single photon existing in two optical modes. The main advantage of the photonic implementation of qubits is the robustness against decoherence and the availability of one-qubit operations. However, the difficulty of realizing the nonlinear interactions between photons that are needed for the implementation of two-qubit operations has been a major obstacle. In recent work, Knill, Laflamme and Milburn (KLM) KLM01 have shown that this obstacle can be overcome by using linear optics, single photon sources and photon number detectors. By now, various controlled-NOT (CNOT) gates for photonic qubits using linear optics have been proposed KYI01 ; Pit01 ; Hof01 ; Ral01 ; Ral02 ; Nie04 and demonstrated Pit02 ; San02 ; Pit03 ; Bir03 ; Bir04 ; Zhao05 .
In particular, it has been shown in Hof01 ; Ral01 that a ‘compact’ CNOT gate can be realized by interaction at a single beam splitter and post-selection of the output. Since this gate requires no ancillary photon inputs or additional detectors, it should be especially useful for experimental realizations of optical quantum circuits. However, there have been two crucial difficulties. In the original schemeHof01 ; Ral01 , the polarization sensitivity of the operation was achieved by separating the paths of the orthogonal polarizations, essentially creating two interacting two-path interferometers. Therefore, the initial experimental realizations based the original proposal Bir03 ; Bir04 are very sensitive to the noisy environment (thermal drifts and vibrations), making it necessary to control and to stabilize nanometer order path-length differences. In addition to these problems, perfect mode-matching is required in each output of the interferometer. Thus, it is very difficult to construct quantum circuits using devices based on those experimental setups. Another difficulty is the evaluation of experimental errors in multi qubit gates. In order to obtain the most complete evaluation of gate performance possible, quantum process tomography has been used in the previous experiment Bir04 . However, 256 different measurement setups are required to evaluate only one CNOT device. When we have to evaluate even more complicated quantum devices realized by a combination of gates, the number of measurements required for tomography rapidly increases as the number of input and output qubits increases.
In this paper, we present an experimental realization of the ‘compact’ optical CNOT gate Hof01 ; Ral01 without any path interference. We show that the CNOT gate can be implemented using three partially polarizing beam splitters (PPBSs) with suitable polarization dependent transmittances and reflectances, where the essential interaction is realized by a single intrinsic PPBS, while the other two supplemental PPBSs act as local polarization compensators on the input qubits. The gate operation can then be obtained directly from the polarization dependence of the reflectances of the PPBSs, removing the need for interference between different paths for orthogonal polarizations. We have evaluated the device operation using a recently proposed method Hof04 , by which we can determine the lower and upper bounds of the process fidelity from measurements of only 32 input-output combinations. We can thus characterize the gate operation with 1/8 the number of input-output measurements required for complete quantum process tomography. We hope that these results will open a door to the realization of more complex quantum circuits for quantum computing.
Figure 1(a) shows the previously proposed optical circuit for the CNOT gate Hof01 ; Ral01 . The beam splitter sitting in the center of the circuit is the essential one which realizes the quantum phase gate operation by flipping the phase of the state $`|H;H`$, where both photons are horizontally polarized, to $`|H;H`$ due to two-photon interference. Since this operation attenuates the amplitudes of horizontally polarized components by a factor of $`1/\sqrt{3}`$, the other two beam splitters with reflectivity 1/3 are inserted in each of the interferometer paths in order to also attenuate the amplitudes of vertically polarized components $`|V`$, so that the total amplitude of any two photon input is uniformly attenuated to $`1/3`$. If we now define the computational basis of the gate as $`|0_z_C|V_C`$, $`|1_z_C|H_C`$ for the control qubit, and $`|0_z_T1/\sqrt{2}(|V_T+|H_T)`$, $`|1_z_T1/\sqrt{2}(|V_T|H_T)`$ for the target qubit, the gate performs the unitary operation $`\widehat{U}_{\text{CNOT}}`$ of the quantum CNOT on the input qubits.
The difficulty in the original proposal is that we have to stabilize the two interferometers of the horizontally and vertically polarized paths by controlling the length of four optical paths with an accuracy on the order of nanometers in order to achieve a reliable operation of the device. In addition, the modes in each path have to be aligned precisely at the output ports of the polarizing beam splitters. Such difficulties have been crucial obstacles for the future realization of optical quantum circuits consisting of several CNOT gates.
Fig. 1(b) shows our solution to this problem. We use one intrinsic PPBS (PPBS-A) and two supplemental PPBSs (PPBS-B) in the optical circuit. The intrinsic PPBS-A, which corresponds to the central beam splitter in Fig. 1(a), implements the quantum phase gate operation by reflecting vertically polarized light perfectly and reflecting (transmitting) 1/3 (2/3) of horizontally polarized light. The two supplemental PPBS-Bs are inserted to adjust the amplitudes of the local horizontal and vertical components of the photonic qubits by transmitting (reflecting) 1/3 (2/3) of vertically polarized light and transmitting horizontally polarized light perfectly. As Fig. 1(b) shows, the use of PPBSs allows us to reduce the four optical paths in Fig. 1(a) to only two optical paths, and path-interferometers are no longer required for the implementation of the ‘compact’ quantum CNOT gate.
In the following experimental demonstration, we used a simple polarization compensation instead of the two supplemental PPBS-Bs. The only purpose of the PPBS-Bs is to reduce the amplitude of the vertical component in the input to $`1/\sqrt{3}`$ of the original input value, while leaving the horizontal component unchanged. Therefore, we can easily simulate the function of the supplemental PPBS-Bs by using compensated input states whose vertical component is reduced to $`1/\sqrt{3}`$. To simulate a general input state $`|\psi _{\text{effective}}=c_H|H+c_V|V`$, we thus use a compensated input state of $`|\psi _{\text{comp.}}=c_H|H+(c_V/\sqrt{3})|V`$. For example, the input for the target qubit state $`|0_z_T`$ becomes $`|0_z^{}_T=(\sqrt{3}|H+|V)/\sqrt{6}`$, which can be easily prepared just by changing the angle of linear polarization by rotating the $`\lambda `$/2 plate (HWP) in the target input.
The schematic of our experimental setup is shown in Fig.2. We used a pair of photons generated through spontaneous parametric down conversion for our input. A beta barium borate (BBO) crystal cut for the type-II twin-beam condition Tak01 was pumped by an argon ion laser at a wavelength of 351.1 nm. The pump beam was focused in the BBO crystal using a convex lens to increase the photon fluxKur01 . Pairs of twin photons are emitted at 702.2 nm with orthogonal polarizations. Gland-Thompson polarizers are used to increase the extinction ratio. After removing the scattered pump light using bandpass filters (IF, center wavelength 702.2nm, FWMH 0.3nm), each of the photons was guided to polarization maintaining single-mode fibers (PMFs) via objective lens and then delivered to the CNOT verification setup. After the collimation lens for the output of the PMFs, the polarization of photons are controlled by HWPs mounted in rotating stages. The timing of the two entangled photons injected to the PPBS-A was controlled using an optical delay. A $`\lambda /4`$ plate (QWP) was inserted to compensate the phase change between horizontal and vertical polarization in the optical delay. After the quantum interference which occurs at PPBS-A, the polarization of output photons was analyzed using HWPs and PBSs. Finally, those photons are coupled into single mode fibers (SMFs) and counted by the single photon counters (SPCM-AQ-FC, Perkin Elmer).
Our setup permits us to select various linear polarizations for the input, and to detect the corresponding linear polarizations in the output. As has been shown in Hof04 , it is possible to characterize the essential quantum properties of the gate operation by using the computational $`ZZ`$-basis given above and the complementary linearly polarized $`XX`$-basis given by $`|0_x_C1/\sqrt{2}(|V_C+|H_C)`$, $`|1_x_C1/\sqrt{2}(|V_C|H_C)`$ for the control qubit, and $`|0_x_T|V_T`$, $`|1_x_T|H_T`$ for the target qubit. The operation of the gate on this input basis also corresponds to a CNOT, with the target qubit acting on the control qubit (reverse CNOT).
The measurement result of the input-output probabilities of our CNOT gate in the $`ZZ`$ basis and in the $`XX`$ basis are shown in tables 1(a) and 1(b), respectively. We measured the coincidence counts between two SPCMs by appropriately setting the HWPs for 16 different combinations of input and output states. In order to convert the coincidence rates to probabilities, we normalize them with the sum of coincidence counts obtained for the respective input state.
The three dimensional bar graphs of table 1 are shown in Fig. 3. The fidelity $`F_{zz}`$ of the CNOT operation in the $`ZZ`$ basis, defined as the probability of obtaining the correct output averaged over all four possible inputs, is $`0.85`$. Similarly, the fidelity $`F_{xx}`$ of the reverse CNOT operation in the $`XX`$ basis is $`0.87`$.
As discussed in detail in Hof04 , the two complementary fidelities $`F_{zz}`$ and $`F_{xx}`$ define an upper and lower bound for the quantum process fidelity $`F_{\mathrm{process}}`$ of the gate with
$$F_{zz}+F_{xx}1F_{\mathrm{process}}\mathrm{min}\{F_{zz},F_{xx}\}$$
(1)
Thus, our experimental results show that the process fidelity of our experimental quantum CNOT gate is
$$0.72F_{\mathrm{process}}0.85.$$
(2)
The lower bound of the process fidelity also defines a lower bound of the entanglement capability of the gate, since the fidelity of entanglement generation is at least equal to the process fidelity. In terms of the concurrence $`C`$ that the gate can generate from product state inputs, the minimal entanglement capability is therefore given by Hof04
$$C2F_{\text{process}}1.$$
(3)
Since our experimental results show that the minimal process fidelity of the gate is $`0.72`$, the lower bound of the entanglement capability is
$$C0.44.$$
(4)
The experimental results shown in table 1 are therefore sufficient to confirm the entanglement capability of our gate.
In order to gain a better understanding of the noise effects in our experimental quantum gate, we can analyze the errors in the classical operations shown in table 1 and associate them with quantum errors represented by elements of the process matrix. For this purpose, it is useful to classify the errors according to the bit flip errors in the output of the operations in the $`ZZ`$ and the $`XX`$ basis, using 0 for the correct gate operation, C for a flip of the control bit output, T for a flip of the target bit output, and B for a flip of both outputs. It is then possible to expand the process matrix in terms of 16 orthogonal unitary operations $`\widehat{U}_i`$, where the index $`i=\{`$ 00, C0, B0, T0, 0C, CC, TC, BC, 0T, CT, TT, BT, 0B, CB, TB, BB $`\}`$ defines the pair of error syndromes in the complementary operations in the $`ZZ`$ and the $`XX`$ basis, e.g. $`\widehat{U}_{\text{TC}}`$ for a target flip error in the $`ZZ`$ operation and a control flip error in the $`XX`$ operation and $`\widehat{U}_{00}=\widehat{U}_{\text{CNOT}}`$ for the ideal gate operation. The process matrix describing the noisy gate operation is given by the operator sum representation of the relation between an arbitrary input density matrix $`\widehat{\rho }_{\text{in}}`$ and its output density matrix $`\widehat{\rho }_{\text{out}}`$,
$$\widehat{\rho }_{\text{out}}=\underset{i,j}{}\chi _{i,j}\widehat{U}_i\widehat{\rho }_{\text{in}}\widehat{U}_j^{}.$$
(5)
Since each operation $`\widehat{U}_i`$ describes a well defined combination of errors in the $`ZZ`$ and $`XX`$ operations, it is now possible to relate the error probabilities observed in table 1 to sums over the diagonal elements $`\chi _{i,i}`$ of the process matrix, as shown in table 2.
Since the correlations between errors in $`ZZ`$ and errors in $`XX`$ are unknown, a range of different distributions of diagonal elements $`\chi _{i,i}`$ is consistent with the experimental data. It is possible to illustrate this range of possibilities by considering the scenarios of lowest and highest process fidelity. The matrix elements of these cases are shown in tables 3(a) and 3(b).
In the worst case scenario of a process fidelity of $`0.72`$ shown in table 3(a), each error syndrome is only observed in either the $`ZZ`$ or the $`XX`$ basis. Therefore, the errors observed in the $`ZZ`$ operation can be identified directly with $`i=`$C0,T0,B0, and the errors observed in the $`XX`$ operation can be identified with $`i=`$0C,0T,0B. In the optimal case of a process fidelity of $`0.85`$ shown in table 3(b), the diagonal elements of $`i=`$C0,T0,B0 for errors only in $`ZZ`$ are all zero. The remaining errors have been distributed over the other diagonal matrix elements within the constraints given by table 2. Note that the distribution of matrix elements in the optimal case is far more homogeneous than the worst case scenario. The actual process fidelity is therefore likely to be closer to the upper limit of $`0.85`$ than to the lower limit of $`0.72`$.
In conclusion, we have demonstrated the first experimental realization of an optical quantum CNOT gate without any path interference by measuring the fidelities of the classical CNOT operations in the computational $`ZZ`$ basis and in the complementary $`XX`$ basis. The performance of both operations by the same quantum gate at fidelities of $`0.85`$ and $`0.87`$ indicates that our device has a quantum process fidelity of $`0.72F_{\text{process}}0.85`$ and an entanglement capability of $`C0.44`$. Since the gate presented in this paper requires no path length adjustments, it should be ideal for the construction of quantum circuits using multiple gates. The present work may therefore provide an important first step towards the realization of optical quantum computation with larger numbers of qubits.
We would like to thank H. Fujiwara, K. Tsujino and D. Kawase for technical suggestions. This work was supported in part by Core Research for Evolutional Science and Technology, Japan Science and Technology Agency, Grant-in-Aid of Japan Science Promotion Society and 21st century COE program.
|
warning/0506/astro-ph0506561.html
|
ar5iv
|
text
|
# The VIMOS VLT Deep SurveyBased on data obtained with the European Southern Observatory Very Large Telescope, Paranal, Chile
## 1 Introduction
The understanding of how matter structures grow via gravitational instability in an expanding Universe is quite well developed and has led to a successful and predictive theoretical framework (e.g., Peebles, 1980; Davis et al., 1985).
One of the most critical problems, however, is to understand the complex mechanisms which, on various cosmological scales, regulate the formation and the evolution of luminous structures within the underlying dark-matter distribution. Its solution ultimately relies on the comprehension of the “microscopic” physics which describes how the baryons fall, heat-up, virialize, cool and form stars in the potential wells generated by the dominant mass component of the Universe, i.e., the non-baryonic dark matter (e.g., White & Rees, 1978). A zero-order, minimal approach to this investigation consists in characterizing “macroscopically” the cosmological matter fluctuations in terms of a reduced set of fundamental quantities, essentially their positions and mass scales, and in studying how the respective spatial clustering and density amplitudes relate to the corresponding statistics computed for light fluctuations. This comparison scheme is generally referred to as matter-galaxy biasing (e.g., Dekel & Lahav, 1999).
An operational definition of bias is conventionally given in terms of continuous density fields by assuming that the local density fluctuation pattern traced by galaxies ($`\delta _g`$) and mass ($`\delta `$) are deterministically related via the “linear biasing scheme”
$$\delta _g(\delta )=b\delta $$
(1)
where the constant “slope” b is the biasing parameter (Kaiser, 1984).
This specific formulation, however, represents a very crude approximation which is not based on any theoretical or physical motivation. It is obvious, for example, that such a model cannot satisfy the physical requirement $`\delta _g(1)=1`$ for any arbitrary value $`b1`$. In particular the biasing process could be non local (e.g., Catelan et al., 1998), stochastic (e.g., Dekel & Lahav, 1999) and non linear (e.g., Mo & White, 1996). Moreover, both theory and numerical simulations predict that the bias grows monotonically from the present cosmic epoch to high redshifts (e.g., Dekel & Rees, 1987; Fry, 1996; Mo & White, 1996; Tegmark & Peebles, 1998; Basilakos & Plionis, 2001).
From a theoretical perspective, light does not follow the matter distribution on sub-galactic scales, where nearly 90$`\%`$ of dense, low-mass dark matter fluctuations (M$``$ 10<sup>7</sup>-10<sup>8</sup>M) failed to form stars and to become galaxies (e.g., Klyplin et al., 1999; Moore et al., 1999; Dalal & Kochanek, 2002). A difference in the spatial distribution of visible and dark matter is predicted also on galactic scales, since the radial scaling of density profiles of dark matter halos (Navarro, Frenk & White, 1997) differs from the three-dimensional radial distributions of light (Sersic and Freeman laws). Galaxy biasing is theoretically expected also on cosmological scales. In particular, simulations of the large-scale structure predict the existence of a difference in the relative distribution of mass and dark halos (e.g., Cen & Ostriker, 1992; Bagla, 1998; Kravtsov & Klypin, 1999) or galaxies (e.g., Evrard, Summers, & Davis, 1994; Blanton et al., 2000; Kayo, Taruya & Suto, 2001). Various physical mechanisms for biasing have been proposed, such as, for example, the peaks-biasing scheme (Kaiser, 1984; Bardeen et al., 1986), the probabilistic biasing approach (Coles, 1993), or the biasing models derived in the context of the extended Press & Schechter approximation (Mo & White, 1996; Matarrese et al., 1998).
Turning to the observational side, the fact that, in the local Universe, galaxies cluster differently according to morphological type (Davis & Geller, 1976), surface brightness (Davis & Djorgovski, 1985), luminosity (Maurogordato & Lachièze-Rey, 1987), or internal dynamics (White, Tully & Davis, 1988) implies that not all can simultaneously trace the underlying distribution of mass, and that galaxy biasing not only exists, but might also be sensitive to various physical processes. As a matter of fact, redshift information which recently became available for large samples of galaxies have significantly contributed to better shaping our current understanding of galaxy biasing, at least in the local Universe. The analysis of the power spectrum (Lahav et al., 2002) and bi-spectrum (Verde et al., 2002) of the 2dF Galaxy Redshift Survey (2dFGRS Colless et al., 2001) consistently shows that a flux-limited sample of local galaxies ( z $`<`$0.25), optically selected in the $`b_J`$-band ($`b_J19.4`$), traces the mass, i.e., it is unbiased, on scales 5$`<`$R($`h^1`$Mpc )$`<`$30.
The galaxy correlation function has been measured up to redshift $``$1 by the CFRS (Le Fèvre et al., 1996), and by the CNOC (Carlberg et al., 2000) surveys giving conflicting evidence on clustering amplitude and bias evolution (see Small et al., 1999). More recently, the analysis of the first season DEEP2 data (Coil et al., 2004) seems to indicate that a combined R-band plus color selected sample is unbiased at z$``$1. On the contrary, measurements of the clustering (Steidel et al., 1998; Giavalisco et al., 1998; Foucaud et al., 2003) or of the amplitude of the count-in-cell fluctuations (Aldeberger et al., 1998) of Lyman-break galaxies (LBGs) at z $``$3 suggest that these objects are more highly biased tracers of the mass density field than are galaxies today. Higher redshift domains have been probed by using photometric redshift information (Arnouts et al., 1999), or compilation of heterogeneous samples (Magliocchetti et al., 2000). Again, the clustering signal appears to come from objects which are highly biased with respect to the underlying distribution of mass.
While there is general observational consensus on the broad picture, i.e., that biasing must decrease with cosmic time, the elucidation of the finer details of this evolution as well as any meaningful comparison with specific theoretical predictions is still far from being secured. Since clustering depends on morphology, color and luminosity, and since most high redshift samples have been selected according to different colors or luminosity criteria, it is not clear, for example, how the very different classes of objects (Ly-break galaxies, extremely red objects or ultraluminous galaxies), which populate different redshift intervals, can be considered a uniform set of mass tracers across different cosmic epochs. Furthermore, one must note that the biasing relation is likely to be nontrivial, i.e., non-linear and scale dependent, especially at high redshift (e.g., Somerville et al., 2001).
Only large redshift surveys defined in terms of uniform selection criteria and sampling typical galaxies (or their progenitors), rather than small subclasses of peculiar objects, promise to yield a more coherent picture of biasing evolution. In particular, the 3D spatial information provided by the VIMOS-VLT Deep Survey (VVDS, Le Fèvre et al., 2005a, hereafter Paper I) should allow us to investigate the mass and scale dependence, as well as to explore the time evolution of the biasing relation between dark matter and galaxies for a homogeneous, flux-limited (I$``$24) sample of optically selected galaxies.
The intent of this paper is to provide a measure, on some characteristic scales R, over the continuous redshift interval 0.4$`<`$z $`<`$1.5, of the local, non-linear, deterministic biasing function
$$b=b(z,\delta ,R).$$
(2)
The goal is to provide an observational benchmark to theories predicting the efficiency of structure formation, or semi-analytical simulations of galaxy evolution. The problem, however, is to find an optimal strategy to evaluate the biasing function in the quasi-pencil-beam geometry of the first-epoch VVDS survey. At present, the angular size of the first-epoch VVDS redshift cone ($`0.5`$deg<sup>2</sup>) does not allow us to constrain the biasing function using high order moments of the galaxy distribution (for example the 3-point correlation function). Moreover, we cannot determine the biasing function simply regressing the galaxy fluctuations ($`\delta _g`$) versus mass fluctuations ($`\delta `$). The fundamental limitation preventing such an intuitive comparison is evident: it is easy to “pixelize” the survey volume and to measure the galaxy fluctuations in each survey cell, but, since the VVDS is not a matter survey, it is much less straightforward to assign to each cell a mass density value. Thus, in this paper, we take an orthogonal approach and we infer the biasing relation $`\delta _g=\delta _g(\delta )`$ between mass and galaxy overdensities from their respective probability distribution functions (PDFs) $`f(\delta )`$ and $`g(\delta _g)`$: assuming a one-to-one mapping between mass and galaxy overdensity fields, conservation of probability implies
$$\frac{d\delta _g(\delta )}{d\delta }=\frac{f(\delta )}{g(\delta _g)}.$$
(3)
The advantage over other methods is that we can explore the functional form of the relationship $`\delta _g=b(z,\delta ,R)\delta `$ over a wide range in mass density contrasts, redshift intervals and smoothing scales R without specifying any a-priori parametric functional form for the biasing function.
In pursuing our approach, we assume that the PDF of matter overdensities $`f(\delta )`$ is satisfactorily described by theory and N-body simulations. What we will try to assess explicitly, is the degree at which the measured PDF of the VVDS overdensities $`g(\delta _g)`$ reproduces the PDF of the underlying parent population of galaxies. The large size and high redshift sampling rate of the VVDS spectroscopic sample, together with the multi-color information in the B, V, R, I filters of the parent photometric catalog and the relatively simple selection functions of the survey, allow us to check for the presence of observational systematics in the data. In principle, this analysis helps us to constrain the range of the parameter space where first-epoch VVDS data can be analyzed in a statistically unbiased way and results can be meaningfully interpreted.
The outline of the paper is the following: in §2 we briefly describe the first-epoch VVDS data sample. In §3 we introduce the technique applied for reconstructing the three-dimensional density field traced by VVDS galaxies, providing details about corrections for various selection effects. In §4 we outline the construction of the PDF of galaxy overdensities and test its statistical representativity. We then derive the PDF of VVDS density contrasts and analyze its statistical moments. In §5 we review the theoretical properties of the analogous statistics for mass fluctuations. Particular emphasis is given to the problem of projecting the mass PDF derived in real space into redshift-perturbed comoving coordinates in the high redshift Universe. The method for computing the biasing function is introduced and tested against possible systematics in §6. VVDS results are presented and discussed in §7. and compared to theoretical models of biasing evolution in §8. Conclusions are drawn in §9.
The coherent cosmological picture emerging from independent observations and analysis motivate us to frame all the results presented in this paper in the context of a $`\mathrm{\Lambda }`$CDM cosmological model with $`\mathrm{\Omega }_m=0.3`$ and $`\mathrm{\Omega }_\mathrm{\Lambda }=0.7`$. Throughout, the Hubble constant is parameterized via $`h=H_0/100`$. All magnitudes in this paper are in the AB system (Oke & Gunn 1983), and from now on we will drop the suffix AB.
## 2 The First-Epoch VVDS Redshift Sample
The primary observational goal of the VIMOS-VLT Redshift Survey as well as the survey strategy and first-epoch observations in the VVDS-0226-04 field (from now on simply VVDS-02h) are presented in Paper I.
Here it is enough to stress that, in order to minimize selection biases, the VVDS survey in the VVDS-02h field, has been conceived as a purely flux-limited ($`17.5I24`$) survey, i.e., no target pre-selection according to colors or compactness is implemented. Stars and QSOs have been a-posteriori removed from the final redshift sample. Photometric data in this field are complete and free from surface brightness selection effects, up to the limiting magnitude I=24 (Mc Cracken et al., 2003).
First-epoch spectroscopic observations in the VVDS-02h field were carried out using the VIMOS multi-object spectrograph (Le Fèvre et al., 2003) during two runs between October and December 2002 (see Paper I). VIMOS observations have been performed using 1 arcsecond wide slits and the LRRed grism which covers the spectral range $`5500<\lambda (\AA )<9400`$ with an effective spectral resolution $`R227`$ at $`\lambda =7500\AA `$. The accuracy in redshift measurements is $``$275 km/s. Details on observations and data reduction are given in Paper I, and in Le Fèvre et al. (2004).
The first-epoch VVDS-02h data sample extends over a sky area of 0.7$`\times `$0.7 deg (which was targeted according to a 1, 2 or 4 passes strategy, i.e., giving to any single galaxy in the field 1, 2 or 4 chances to be targeted by VIMOS masks (see fig. 12 of paper I)) and has a median depth of about z $``$0.76. It contains 6582 galaxies with secure redshifts (i.e., redshift determined with a quality flag$``$2 (see Paper I)) and probes a comoving volume (up to z =1.5) of nearly $`1.510^6h^3`$Mpc<sup>3</sup> in a standard $`\mathrm{\Lambda }`$CDM cosmology. This volume has transversal dimensions $``$ 37x37 $`h^1`$Mpc at z =1.5 and extends over 3060 $`h^1`$Mpc in radial direction.
For this study we define a sub-sample (VVDS-02h-4) with galaxies having redshift z $`<`$1.5 and selected in a continuous sky region of 0.4$`\times `$0.4 deg which has been homogeneously targeted four times by VIMOS slitmasks. Even if we measure redshifts up to z $``$5 and in a wider area, the conservative angular and redshift limits bracket the range where we can sample in a denser way the underlying galaxy distribution and, thus, minimize biases in the reconstruction of the density field (see the analysis in §4.1). The VVDS-02h-4 subsample contains 3448 galaxies with secure redshift (3001 with $`0.4<z<1.5`$) and probes one-third of the total VVDS-02h volume. This is the main sample used in this study.
## 3 The Density Field Reconstruction Scheme
The first ingredient we need in order to derive the biasing relation
$$\delta _g=b(z,\delta ,R)\delta $$
(4)
is a transformation scheme for diluting an intrinsic point-like process, such as the galaxy distribution in a redshift survey, into a continuous 3D overdensity field (see the review by Strauss & Willick, 1995). We write the dimensionless density contrast at the comoving position r, smoothed over a typical dimension $`R`$ as
$$\delta _g(𝐫,R)=\frac{\rho _g(𝐫,R)\overline{\rho _g}}{\overline{\rho _g}}.$$
(5)
and we define (e.g., Hudson, 1993) the smoothed number density of galaxies above the absolute magnitude threshold $`^c`$ as the convolution between Dirac’s delta functions and some arbitrary filter
$$\rho _g(𝐫,R,<^c)=\underset{i}{}\frac{\delta ^D(𝐫𝐫_𝐢)F\left(\frac{|𝐫𝐫_𝐢|}{R}\right)}{S(r_i,^c)\mathrm{\Phi }_z(m)}$$
(6)
where the sum is taken over all the galaxies in the sample, $`S(r,^c)`$ is the distance-dependent selection (or incompleteness) function of the sample (see §3.2), $`\mathrm{\Phi }_z(m)`$ is the redshift sampling function (see §3.3) and $`F(|𝐫|/R)`$ is a smoothing kernel of width $`R`$. In this paper, the smoothing window F is modeled in terms of a normalized Top-Hat (TH) filter
$$F\left(\frac{|𝐫|}{R}\right)=\frac{3}{(4\pi R^3)}\mathrm{\Theta }\left(1\frac{|𝐫|}{R}\right),$$
(7)
where $`\mathrm{\Theta }`$ is the Heaviside function, defined as $`\mathrm{\Theta }(x)=1`$ for 0 $``$ x $``$ 1, and $`\mathrm{\Theta }(x)=0`$ elsewhere.
Within this weighting scheme, shot-noise errors are evaluated by computing the variance of the galaxy field
$$ϵ(𝐫)=\frac{1}{\overline{\rho _g}}\left[\underset{i}{}\left(\frac{F\left(\frac{|𝐫𝐫_𝐢|}{R}\right)}{S(r_i,^c)\mathrm{\Phi }_z(m_i)}\right)^2\right]^{1/2}$$
(8)
Note that all coordinates are comoving and that the mean density $`\overline{\rho _g}`$ depends on cosmic time. Since we observe an evolution of nearly a factor of two in the mean density of galaxies brighter than $`_B^{}(z=0)`$ from redshift z=0 up to z=1 (see Ilbert et al. (2005), hereafter Paper II), we compute the value of $`\overline{\rho _g}`$ at position r (corresponding to some look-back time t) with eq. (6) by simply averaging the galaxy distribution in survey slices $`r\pm R_s`$ where $`R_s=400`$ $`h^1`$Mpc. We note that conclusions drawn in this paper depend very weakly on the choice of $`R_S`$ in the interval 200$`<R_s`$( $`h^1`$Mpc )$`<`$600.
In this paper, the density field is evaluated at positions r in the VVDS-02h volume that can be either random or regularly displaced on a 3D grid (see section 3.4). Even if we correct for the different sampling rate in the VVDS-02h field (by weighting each galaxy by the inverse of redshift sampling function ($`\mathrm{\Phi }_z(m)`$)) we always select, for the purposes of our analysis, only the density fluctuations recovered in spheres having at least 70$`\%`$ of their volume in the denser 4-passes volume. This in order to minimize the Poissonian noise due to the sparser redshift sampling outside the VVDS-02h-4 field.
We also consider only volumes above the redshift threshold z <sub>t</sub> where the transversal dimension L of the first-epoch VVDS-02h cone is L$`(z_t)>`$2R. As an example, for TH windows of size R=5(10) $`h^1`$Mpc we have z $`{}_{t}{}^{}`$ 0.4(0.7). Within the redshift range 0.4$`<`$z $`<`$1.5 the VVDS-02h field contains 5252 galaxies (of which 3001 are in the four passes region.)
Note that we have characterized the galaxy-fluctuation field in terms of the number density contrast, instead of the luminosity density contrast, because the former quantity is expected to show a time-dependent variation which is more sensitive to the galaxy evolution history (formation and merger rates, for example). Moreover, as described in section §3.4, a robust description of the density field and a reliable determination of the PDF shape can be obtained only minimizing the shot noise component of the scatter; this is more easily done by considering galaxy number densities rather than galaxy light densities.
The most critical elements of the smoothing process are directly readable in eq. (6): we must first evaluate galaxy absolute magnitudes at each redshift in the most reliable way, then specify the selection function and the redshift sampling rate of the VVDS survey. In the next sections we will describe how these quantities have been evaluated.
### 3.1 The K-correction
The absolute magnitude is defined as:
$$M^r=m^o5\mathrm{log}d_L(z,\mathrm{\Omega })K(z,\mathrm{SED})$$
(9)
where the suffixes $`r`$ and $`o`$ designate respectively the rest-frame band in which the absolute magnitude is computed and the band where the apparent magnitude is observationally measured, and $`d_L`$ is the luminosity distance evaluated in a given a-priori cosmology (i.e., using an appropriate set of cosmological parameters $`\mathrm{\Omega }[\mathrm{\Omega }_m,\mathrm{\Omega }_\mathrm{\Lambda }])`$.
The correction factor $`K`$, which depends on redshift and the spectral energy distribution (SED), accounts for the fact that the system response in the observed frame corresponds to a narrower, bluer rest-frame passband, depending on the redshift of the observed object. A complete description of the application of this transformation technique to VVDS galaxies is detailed in Paper II.
The estimate of the galaxy absolute luminosity is thus affected by the uncertainties introduced by probing redshift regimes where the $`kcorrection`$ term cannot be neglected. Anyway, using mock catalogs simulating the VVDS survey, we have shown (see Fig. A.1 in Paper II) that the errors in the recovered absolute magnitude are significantly smaller in the B band ($`\sigma _B=0.08`$) than in the I band ($`\sigma _I=0.17`$). Thus, in what follows we will use absolute luminosities determined in the B-band rest-frame.
### 3.2 The Radial Selection Function
Since our sample is limited at bright and faint apparent magnitudes (17.5$``$I$``$24), at any given redshift we can only observe galaxies in a specific, redshift-dependent, absolute magnitude range. It is usual to describe the sample radial incompleteness by defining the selection function in terms of the galaxy luminosity function $`\phi ()`$
$$S(r,^c)=\frac{_{_b(r)}^{_f(r)}\phi ()𝑑}{_{\mathrm{}}^^c\phi ()𝑑}.$$
(10)
Here $`_b(r)`$ and $`_f(r)`$ are the B-band absolute magnitudes which correspond, at distance r, to the I-band limiting apparent magnitudes $`m_b=17.5`$ and $`m_f=24`$ respectively (see discussion at the end of the section). Since eq. (9) also depends on galaxy colors, we compute its mean value at distance r by a weighted average over the population mix observed at distance r.
The VVDS luminosity function (LF) has been derived in Paper II and is characterized by a substantial degree of evolution over the redshift range 0$`<`$z $`<`$1.5. Therefore, we estimate $`\phi ()`$ in the B band at any given position in the redshift interval \[0, 1.5\] by interpolating, with a low order polynomial function, the Schechter shape parameters $`\alpha `$ and $`^{}`$ given in Table 1 of Paper II.
Assuming $`_B^c=15`$ in eq. 10, which corresponds to the limiting absolute magnitude over which the LF of the VVDS-02h sample can be robustly constrained in the lowest redshift bins, the selection function exponentially falls by nearly 2 orders of magnitude in the redshift range up to z $`<`$1.5. Thus, the density field reconstruction strongly depends on the radial selection function used especially at high redshifts, where eq. 10 can be affected by possible systematics in the determination of the LF or in the measurements of faint magnitudes. Therefore, we will also analyze volume-limited sub-samples, which are essentially free from these systematics.
Since in a magnitude limited survey progressively brighter galaxies are selected as a function of redshift, a volume-limited sample also allows us to disentangle spurious luminosity-dependent effects from the measurement of the redshift evolution of the biasing function.
### 3.3 The Redshift Sampling Rate
As for most redshift surveys, the VVDS does not target spectroscopically all the galaxies which satisfy the given flux limit criteria in the selected field of view (see Paper I). Because of the sparse sampling strategy, we have to correct the density estimator with a sampling rate weight $`\mathrm{\Phi }_z`$ in order to reconstruct the real underlying galaxy density field in a statistically unbiased way.
The VVDS redshift sampling rate is the combination of two effects: i) only a fraction of the galaxies ($`40\%`$, see Paper I) in the photometric sample is targeted (target sampling rate), ii) and only a fraction of the targeted objects ($`80\%`$ see Paper I) yield a redshift (spectroscopic success rate). We can model this correction term, by assuming, in first approximation, that the sampling rate depends only on the apparent magnitude. Since the VVDS targeting strategy is optimized to maximize the number of slits on the sky, the selection of faint objects is systematically favored. Inversely, the ability of measuring a redshift degrades progressively towards fainter magnitudes, i.e., for spectra having lower signal-to-noise ratios (the spectroscopic success rate decreases from $`>90\%`$ at I$``$22 down to $`60\%`$ at I$``$24.) These two opposite effects conspire to produce the magnitude-dependent sampling rate function shown in Fig. 1. Clearly, with such an approximation, we neglect any possible dependence of the sampling rate from other important parameters such as for example surface brightness, spectral type or redshift. However, in in §3.2 of Paper II we showed, using photometric redshifts, that any systematic sampling bias introduced by a possible redshift dependence of the spectroscopic success rate is expected to affect only the tails of our observed redshift distribution ( z $`<`$0.5 and z $`>`$1.5) i.e., redshift intervals not considered in this study (see §3).
We describe the VVDS sampling completeness $`\mathrm{\Phi }_z(m)`$ at a given magnitude $`m`$, as the fraction of objects with measured redshifts $`N_z`$ over the number $`N`$ of objects detected in the photometric catalog.
$$\mathrm{\Phi }_z(m)=\frac{_{i=1}^{N_z}w(mm_i,dm)}{_{i=1}^Nw(mm_i,dm)}$$
(11)
with the window function $`w`$ defined as:
$$w(mm_i,dm)=\{\begin{array}{cc}\hfill 1& \text{if }|mm_i|<dm/2\hfill \\ \hfill 0& \text{otherwise}\hfill \end{array}$$
(12)
Working in low resolution mode (i.e allowing up to 4 galaxies to have spectra aligned along the same dispersion direction with a typical sky separation of 2 arcmin) and observing the same region of sky 4 times for a total of about 16 hours (four-passes strategy), we can measure redshifts for nearly 30$`\%`$ of the parent photometric population, as shown in Fig. 1. In other terms, on average, nearly one over three galaxies with magnitude I$``$24 has a measured redshift in the four-passes VVDS region. This high spatial sampling rate is a critical factor for minimizing biases in the reconstruction of the 3D density field of galaxies. In particular we note that our I$``$24, z $`<`$1.5 VVDS-02h-4 sample is characterized by an effective mean inter-particle separation in the redshift range \[0,1.5\] ($`r5.1`$ $`h^1`$Mpc ) which is smaller than that of the original CFA sample ($`r5.5`$$`h^1`$Mpc) used by Davis & Huchra (1981) to reconstruct the 3D density field of the local Universe (i.e., out to $``$ 80 $`h^1`$Mpc ). At the median depth of the survey, i.e., in the redshift interval 0.75$`\pm 0.05`$, the mean inter-particle separation is 4.4 $`h^1`$Mpc, a value nearly equal to the mean inter-particle separation at the median depth of the 2dFGRS. We finally note that in the redshift range \[0.7,1.35\], which is also covered by the DEEP2 survey, the VVDS mean-inter-particle separation is 5.5 $`h^1`$Mpc compared to the value $``$ 6.5 $`h^1`$Mpc inferred from the values quoted by Coil et al. (2004) for their most complete field, currently covering 0.32 deg<sup>2</sup>.
By dividing the VVDS-02h-4 field in smaller cells and repeating the analysis, we conclude that the sampling rate does not show appreciable variations, i.e., the angular selection function can be considered constant for the purposes of our analysis. This corresponds to the fact that the success rate in redshift measurement in each VIMOS quadrant (i.e., the spectroscopic success rate per mask) is, in good approximation, constant and equal to $``$80$`\%`$ (see Paper I).
### 3.4 Shot Noise
In a flux-limited sample, the shot noise in the density field is an increasing function of distance (see eq. (8)). One may correct for the increase of the mean VVDS inter-particle separation as a function of redshift (and thus the increase of the variance of the density field) by opportunely increasing the length of the smoothing window (e.g., Strauss & Willick, 1995). However, since we are interested in comparing the fluctuations recovered on the same scale at different redshifts in a flux limited survey, we take into account the decreased sampling sensitivity of the survey at high redshift in an alternative way.
We deconvolve the signature of this noise from the density maps by applying the Wiener filter technique (cf. Press et al., 1992) which provides the minimum variance reconstruction of the smoothed density field, given the map of the noise and the a priori knowledge of the underlying power spectrum (e.g., Lahav et al., 1994). The application of the Wiener denoising procedure to the specific geometry of the VVDS sample is described in detail in appendix A.
Here we note that the Wiener filter requires a model for the underlying 3D power spectrum P(k,z) which we compute, over the frequency scales where the correlation function of VVDS galaxies is well constrained ($`0.06k10`$), as (see eq. (48) in Appendix A):
$$P(k,z)=4\pi \frac{r_0(z)^{\gamma (z)}}{k^{3\gamma (z)}}\mathrm{\Gamma }(2\gamma (z))\mathrm{sin}\frac{\gamma (z)\pi }{2},$$
(13)
where the normalization $`r_0(z)`$ and the slope $`\gamma (z)`$ of the correlation function at redshift z have been derived by interpolating the values measured in various redshift intervals of the VVDS-02h volume by Le Fèvre et al. (2005b)(hereafter Paper III).
The variance of the galaxy distribution on a 8 $`h^1`$Mpc scale in the VVDS-02h sample can be obtained by integrating eq. (13) using a TH window of radius 8 $`h^1`$Mpc and the ($`r_0,\gamma `$) parameters of the VVDS correlation function
$$\sigma _8^2=\frac{1}{2\pi ^2}P(k)\stackrel{~}{F}_k^2k^2𝑑k$$
(14)
where $`\stackrel{~}{F}_k`$ is the Fourier transform of the TH filter (see eq. (45) in Appendix A). Note that, by integrating the power spectrum down to $`k0`$, i.e., extrapolating the power law shape of eq. (13) beyond $`L100`$ $`h^1`$Mpc ($`k<`$0.06), one would revise upwards the value of $`\sigma _8`$ by $`2\%`$(at z=0.35) and by $`4\%`$ at z=1.4. Since, however, the amplitude of the power spectrum on large scales is expected to downturn and to be systematically lower than predicted by eq. (13), we safely conclude that, with our computation scheme, the inferred $`\sigma _8`$ value should be biased low by no more than $`2\%`$ and $`4\%`$ in the first and last redshift bin, respectively.
Projecting the results from real-space into the redshift-distorted space (see §5), i.e., implementing the effects of large-scale streaming motions, we infer that the rms of galaxy fluctuations are $`\sigma _8`$=\[0.55$`\pm `$0.10, 0.62$`\pm `$0.12, 0.71$`\pm `$0.11, 0.69$`\pm `$0.14, 0.75$`\pm `$0.14, 0.92$`\pm `$0.20\] at redshift z=\[0.35,0.6,0.8,1.2,1.2,1.65\] (see Fig. 2). Thus, the amplitude of $`\sigma _8`$ for a flux-limited I$``$24 sample increases as a function of redshift by nearly $`70\%`$ between z$``$0.3 and z$``$1.7.
### 3.5 The 3-Dimensional VVDS Density Field
The VVDS-02h galaxy density field reconstructed on a scale R= 5 $`h^1`$Mpc and in the redshift bin 0.8$`<`$z $`<`$1.1 is visually displayed in the left most panel of Fig. 3. Note that the chosen smoothing scale nearly corresponds to the mean inter-galaxy separation in the VVDS-02h sample at z $``$ 1. Fig. 3 shows the regular patterns traced by over- and under-dense regions in the selected redshift interval. Specifically, we note that, in this redshift slice, there are over- and under-dense regions which extend over characteristic scales as large as $``$ 100 $`h^1`$Mpc. A more complete discussion of the “cartography” in such deep regions of the Universe is presented by Le Fèvre et al. (2005c).
In the same figure, we also display the density field reconstructed in an analogous redshift range, using the GALICS simulation (Galaxies in Cosmological Simulations, Hatton et al. (2003)). GALICS combines cosmological simulations of dark matter with semi-analytic prescriptions for galaxy formation to produce a fully realistic deep galaxy sample. In particular we plot the density field of the I$``$24 flux-limited simulation as well as the density field recovered after applying to the pure flux-limited simulation all the VVDS target selection criteria (see §4.1). No qualitative difference between the density fields reconstructed before and after applying to the simulation all the survey systematics is seen.
Clearly, a more quantitative assessment of the robustness and reliability of the VVDS overdensity field can be done by studying its PDF.
## 4 The PDF of Galaxy Fluctuations
Once the three-dimensional field of galaxy density contrasts $`\delta _g`$ has been reconstructed on a given scale R, one can fully describe its properties by deriving the associated PDF $`g(\delta _g)`$. This statistical quantity represents the normalized probability of having a density fluctuation in the range $`(\delta _g,\delta _g+d\delta _g)`$ within a region of characteristic length R randomly located in the survey volume.
While the shape of the PDF of mass fluctuations at any given cosmic epoch is theoretically well constrained from CDM simulations (see next section), little is known about the observational PDF of the general population of galaxies in the high redshift Universe. Even locally, this fundamental statistics has been often overlooked (but see Ostriker et al. (2003)). Notwithstanding, the shape of the galaxy PDF is strongly sensitive to the effects of gravitational instability and galaxy biasing, and its redshift dependence encodes valuable information about the origin and evolution of galaxy density fluctuations.
The shape of the PDF can be characterized in terms of its statistical moments. In particular the variance of a zero-mean field (such as the overdensity field we are considering) is
$$\sigma _R^2=\delta _g^2_R_0^{\mathrm{}}\delta _g^2g_R(\delta _g)𝑑\delta _g.$$
(15)
Higher moments can be straightforwardly derived (see Bernardeau et al., 2002, for a review). In the following, we will drop the suffix R, unless we need to emphasize it.
### 4.1 Estimating Reconstruction Systematics Using Mock Catalogs
For the purposes of our analysis, it is imperative to check that the various instrumental selection effects as well as the VVDS observing strategy are not compromising the determination of the PDF of galaxy fluctuations. In this section, we explore the region of the parameter space (essentially redshift and smoothing scales) where the PDF of VVDS overdensities traces in a statistically unbiased way the underlying parent distribution.
Possible systematics can be hidden in the reconstructed PDF essentially because i) the VVDS redshift sampling rate is not unity, ii) the slitlets are allocated on the VIMOS masks with different constraints along the dispersion and the spatial axis, and iii) the VIMOS field of view is splitted in four different rectangular quadrants separated by a vignetting cross.
We have addressed point iii) by designing a specific telescope pointing strategy which allows us to cover in a uniform way the survey sky region (see the telescope pointing strategy shown in Fig 1 of Paper I). With the adopted survey strategy, we give to each galaxy in the VVDS-02h-4 field four chances to be targeted by VIMOS, thus increasing the survey sampling rate (nearly 1 galaxy over 3 with magnitude I$``$24 has a measured redshift.)
Concerns about points i) and ii) can be directly addressed using galaxy simulations covering a cosmological volume comparable to the VVDS one. Thanks to the implementation of the Mock Map Facility (MoMaF Blaizot et al., 2003), it is possible to convert the GALICS 3D mocks catalog into 2D sky images, and handle the 2D projection of the simulation as a pseudo-real imaging survey. Pollo et al. (2005), have then built a set of 50 fully realistic mock VVDs surveys from the GALICS simulations to which the whole observational pipeline and biases has been applied. By comparing specific properties of the resulting pseudo-VVDS sample with the true underlying properties of the pseudo-real Universe from which the sample is extracted, we can directly explore the robustness, as well as the limits, of the particular statistical quantities we are interested in.
In brief these include addeding to the 3D galaxy mocks a randomly simulated distribution of stars to mimic the same star contamination affecting our survey. Next, we have masked the sky mocks using the VVDS photometric masks, i.e., we have implemented the same geometrical pattern of excluded regions with which we avoid to survey sky regions contaminated by the presence of bright stars or photometric defects. Then, we have extracted the spectroscopic targets by applying the target selection code (VMMPS, Bottini et al., 2005) to the simulated 2D sky distribution. To each GALICS redshift, which incorporates the Doppler contribution due to galaxy peculiar velocities, we have added a random component to take into account errors in z measurements. Finally, we have processed the selected objects implementing the same magnitude-distribution of failures in redshift measurements which characterizes the first-epoch observations of the VVDS survey (see Paper I and Fig 1).
Since GALICS galaxies have magnitudes simulated in the same 4 bands surveyed by VVDS (B,V,R,I), we have applied the K-correction to obtain rest frame absolute magnitudes and we have empirically re-derived all the selection functions for the mock catalogs according to the scheme presented in §3. In this way we can also check the robustness of the techniques we apply for computing absolute magnitudes and selection functions (see Paper II).
The PDF of galaxy overdensities obtained from the mock samples has been finally compared to the PDF of the parent population. For brevity, in what follows, we will call s-samples (survey-samples) the mocks simulating the VVDS redshift survey and p-sample (parent sample) the whole GALICS simulation flux-limited at 17.5$``$I$``$24.
The density contrasts have been calculated as described in §3. In the following, we will restrict our analysis to the set of smoothing scales in the interval R =(5,10) $`h^1`$Mpc. The choice of these particular limits is motivated by the fact that 5 $`h^1`$Mpc is the minimum smoothing scale for which the reconstructed density field is unbiased over a substantial redshift interval (see discussion below). Note, also, that below this typical scale, linear regimes approximations, largely used in this paper, do not hold anymore. The upper boundary is constrained by the transverse comoving dimensions covered by the first-epoch VVDS data (see §2), which is still too small for being partitioned using bigger scale-lengths without introducing significant noise in the reconstructed PDF (see the transverse comoving dimension of the VVDS-02h field quoted in §2).
The PDF of the galaxy density contrasts computed using the s-sample is compared to the parent distribution inferred from the p-sample in Fig. 4. We conclude that the distribution of galaxy overdensities of the s-samples for $`R=8`$ and $`10`$ $`h^1`$Mpc scales is not biased with respect to the underlying distribution of p-sample galaxy fluctuations. Thus, the VVDS density field reconstructed on these scales is not affected by the specific VVDS observational strategy.
It is evident in Fig. 4 that the VVDS redshift sampling rate is not high enough to map in an unbiased way the low density regions of the Universe ($`\mathrm{log}(1+\delta _g)<0.5`$) when the galaxy distribution is smoothed on scales of 5 $`h^1`$Mpc. Fig. 5 shows that incompleteness in underdense regions is a function of redshift, with the bias in the low-density tail of the PDF developing and increasing as the redshift increases. As a rule of thumb, the PDF of the s-sample starts to deviate significantly from the parent PDF when the mean inter-galactic separation $`r`$ of the survey sample is larger than the scale R on which the field is reconstructed. Since we measure $`r(z=1)5`$ $`h^1`$Mpc ($`r(z=1.5)8`$ $`h^1`$Mpc ), the PDF of the density field recovered using a TH filter of radius 5 $`h^1`$Mpc is effectively unbiased (at least over the density range we are interested in, i.e., $`log(1+\delta )>1`$) only if the sample is limited at z $`\mathit{}\mathit{<}1`$. Therefore, in the following, results obtained for R=5 $`h^1`$Mpc are quoted only up to z=1.
On scales R$`>8`$$`h^1`$Mpc , the agreement between the PDFs of s- and p-samples holds true also for volume-limited subsamples Specifically, the 2nd and 3rd moments of the PDF of overdensities recovered using volume-limited s-samples on these scales are within 1$`\sigma `$ of the corresponding values computed for the parent, volume-limited, p-samples in each redshift interval of interest up to z =1.5.
To summarize, the results of simulated VVDS observations presented in this section show that, at least on scales R$``$8 $`h^1`$Mpc, the VVDS PDF describes in an unbiased way the general distribution properties of a sample of I=24 flux-limited galaxies up to redshift 1.5. In other terms, the VVDS density field sampled in this way is essentially free from selection systematics in both low- and high-density regions, and can be meaningfully used to infer the physical bias in the distribution between galaxy and matter. Obviously, the representativeness of the measured PDF of VVDS overdensities with respect to the “universal” PDF up to z =1.5 is a different question. Since the volume probed is still restricted to a limited region of space in one field, the shape and moments of the galaxy PDF derived from the VVDS-02h are expected to deviate from the “universal” PDF of galaxies at this redshift because of cosmic variance. Reducing the cosmic variance is one of the main goals of extending the VVDS to 4 independent fields. Anyway, our 50 mock realisations of the VVDS-02h sample allow us to estimate realistic errors that include the contribution from cosmic variance.
### 4.2 The PDF of VVDS Galaxies: Results
Let us then investigate the evolution, as a function of the lookback time, of the observed PDF of VVDS galaxy fluctuations.
In an apparent magnitude-limited survey such as the VVDS, only brighter galaxies populate the most distant redshift bins, whereas fainter galaxies are visible only at
lower redshifts. As more luminous galaxies tend to cluster more strongly than fainter ones (e.g., Hamilton, 1988; Croton et al., 2004), the PDF of galaxy fluctuations is expected to be systematically biased as a function of redshift.
This effect is clearly seen in the first correlation analysis of the VVDS (paper III), and can be minimized by selecting a volume-limited sample. Therefore, we have defined a subsample with absolute magnitude brighter than $`_B^c=20+5\mathrm{log}h`$ in the rest frame B band ($`1350`$ galaxies with 0.7$`<`$z $`<`$1.5 in the VVDS-02h field, $`800`$ of which are in the VVDS-02h-4 region).
This threshold corresponds to the faintest galaxy luminosity visible at redshift z =1.5 in a I=24 flux-limited survey and it is roughly 0.6 magnitudes brighter(fainter) than the value of $`_B^{}`$ recovered at z $``$0($``$1.5) using the VVDS data (see Paper II). The median absolute magnitude for this volume-limited sample is $`20.4`$.
We note, however, that the populations of galaxies with the same luminosity at different redshifts may actually be different. As we have shown in Paper II, we measure a substantial degree of evolution in the luminosity of galaxies, and, as a consequence, with our absolute magnitude cut we are selecting $`^{}+0.6`$ galaxies at z =1.5, but $`^{}0.6`$ galaxies at z =0. Thus, the clustering signal at progressively earlier epochs may not be contributed by the progenitors of the galaxies that are sampled at later times in the same luminosity interval.
The PDF of density fluctuations, in various redshift intervals, and traced on scales of 8 and 10 $`h^1`$Mpc by VVDS galaxies brighter than $`_B^c`$, is presented in Fig. 6. Note that the analysis of the previous section guarantees that, on these scales, the VVDS PDF fairly represents the PDF of the real underlying population of galaxies up to z =1.5. Fig. 6 shows how the shape of the measured galaxy PDF changes across different cosmic epochs.
A Kolmogorov-Smirnov test confirms that the PDFs at different cosmic epochs are statistically different (i.e., the null hypothesis that the three distributions are drawn from the same parent distribution is rejected with a confidence $`P_{ks}>110^6`$).
In particular the peak of the PDF in the lowest redshift interval is shifted towards smaller values of the density contrast $`\delta _g`$ when compared to the peak of the PDF in the highest redshift bin. Moreover, the shape of the distribution, also shows a systematic “deformation”. Specifically, we observe the development of a low-$`\delta `$ tail in the PDF as a function of time on both scales investigated. In other terms the probability of having low density regions increases as a function of cosmic time. For example, underdense regions, defined as the regions where the galaxy density field is $`\mathrm{log}(1+\delta _g)<0.5`$ on a R= 8 $`h^1`$Mpc scale, occupy a fraction of nearly 35$`\%`$ of the VVDS volume at redshift 0.7$`<`$z $`<`$1.0, but only a fraction of about 25$`\%`$ at redshift 1.25$`<`$z $`<`$1.5. Similar trends are observed when lowering the absolute luminosity threshold of the volume limited sample (and consequently lowering the upper limit of the redshift interval probed) or when modifying the binning in redshift space.
If galaxies are faithful and unbiased tracers of the underlying dark matter field, then both this effects, the peak shift and the development of a low density tail may be qualitatively interpreted as a direct supporting evidence for the paradigm of the evolution of gravitational clustering in an expanding Universe. At variance with overdense regions which collapse, a net density deficit ($`\delta <0`$) in an expanding Universe brings about a sign reversal of the effective gravitational force: a density depression is a region that induces an effective repulsive peculiar gravity (Peebles 1980). If gravity is the engine which drives clustering in an expanding Universe, we thus expect that, as time goes by, low density regions propagate outwards and a progressively higher portion of the cosmological volume becomes underdense.
The observed evolution in the PDF could also indicate the presence of a time-dependent biasing between matter and galaxies. As a matter of fact, it can be easily shown that a monotonic bias, increasing with redshift, offers a natural mechanism to re-map the galaxy PDF into progressively higher intervals of density contrasts.
We can better discriminate the physical origin of the observed trends, i.e., if they are purely induced by gravitation or strengthened by the collateral and cooperative action of biasing, by studying the evolution of the PDF moments. In Fig. 7 the redshift evolution of the rms ($`\sigma `$) and skewness ($`S_3`$) of the overdensity fields (see table 1) for the $`_B<20+5\mathrm{log}h`$ sample are shown and compared to local measurements.
Following the standard convention within the hierarchical clustering model, we define the skewness $`S_3`$ in terms of the volume-averaged two- and three-point correlation functions ($`S_3\overline{\xi }_3/\overline{\xi }_2^2`$) noting that in the case of a continuous $`\delta _g`$-field with zero mean this expression reduces to $`S_3\delta _g^3/\delta _g^2^2`$. We do not derive the moments $`<\delta _g^2>`$ and $`<\delta _g^3>`$ of the PDF by directly applying the computation scheme given in eq. (15), but by correcting the count-in-cells statistics for discreteness effects using the Poissonian shot-noise model (e.g., Peebles, 1980; Fry, 1985, cfr. eqs. (374) and (375) of Bernardeau et al. 2002, possible biases introduced by this estimation technique are discussed by Hui & Gaztañaga 1999.). The corresponding values of $`\sigma `$ and $`S_3`$ for the local Universe (in redshift space) have been derived by Croton et al. 2004 using the 2dFGRS. Here we plot the values corresponding to their $`21<5\mathrm{log}h<20`$ subsample, which actually brackets the median luminosity of our volume limited sample.
We can see that the rms amplitude of fluctuations of the VVDS density field, on scales 8 $`h^1`$Mpc, is with good approximation constant over the full redshift baseline investigated, with a mean value of 0.94$`\pm `$0.07 over 0.7$`<`$z $`<`$1.5.
While the strength of clustering of galaxies brighter than $`<20+5\mathrm{log}h`$ does not change much in this redshift interval, each VVDS measurement is lower than the value inferred at z $``$0 by Croton et al. 2004. In particular, our mean value is $``$10$`\%`$ smaller than the 2dFGRS value and the difference is significant at $`2\sigma `$ level.
The skewness $`S_3`$, which measures the tendency of gravitational clustering to create asymmetries between underdense and overdense regions, decreases as a function of redshift. We observe a systematic decrement not only internally to the VVDS sample, but also when we compare our measurements with the z =0 estimate. This trend is caused by the development of the low-$`\delta `$ tail in the PDF as a function of time on both the R=8,10 $`h^1`$Mpc scales and reflects the fact that the probability of having underdense regions is greater at present epoch than it was at z $``$1.5 (where its measured value is $`2\sigma `$ lower.)
The amplitudes of the rms and skewness of galaxy overdensities show an evolutionary trend dissimilar from that predicted in first and second order perturbation theory for the gravitational growth of dark matter fluctuations (see Bernardeau et al. 2002 for a review). According to linear perturbation theory the amplitude of the rms of mass fluctuations scales with redshift as in eq. (18) while second order perturbation theory predicts that, on the scales where the quasi-linear approximation holds, the growth rate of $`<\delta _g^3>`$ and variance $`<\delta _g^2>^2`$ are syncronized so that the skewness $`S_3`$ of an initially Gaussian fields should remain constant (Peebles, 1980; Juszkiewicz, Bouchet & Colombi, 1993; Bernardeau, 1993). <sup>1</sup><sup>1</sup>1Note that the observed redshift evolution of the skewness is just the opposite of what is expected also in generic dimensional non-Gaussian models where $`S_3`$ is predicted to increase with redshift. Furthermore, in Le Fèvre et al. (2005c) we show that even the general shape of the galaxy PDF deviates from a lognormal distribution, i.e., from the profile in terms of which the mass PDF is generally approximated (see §5). Therefore, we conclude that the PDF evolution is not caused by gravity alone; the redshift scaling of its global shape and moments effectively indicates the presence of a time evolving bias.
We can deconvolve the purely gravitational signature and investigate properties and characteristics of the biasing between matter and galaxies by comparing the galaxy PDF to the corresponding statistics computed for mass fluctuations. Thus, we now turn to the problem of deriving the PDF of mass fluctuations.
## 5 The PDF of Mass Fluctuations in Redshift-Distorted Comoving Coordinates
The VVDS survey is providing a rich body of redshift data for mapping the galaxy density field in extended regions of space and over a wide interval of cosmic epochs. On the contrary, the direct determination of the underlying mass density field and its associated PDF is a less straightforward process. Nonetheless we may gain insight into the mass statistics by using simulations and theoretical arguments.
In the standard picture of gravitational instability, the PDF of the primordial cosmological mass density fluctuations is assumed to obey a random Gaussian distribution. Once the density fluctuations reach the non-linear stage, their PDF significantly deviates from the initial Gaussian profile and a variety of phenomenological models have been proposed to describe its shape (e.g., Saslaw, 1985; Lahav et al., 1993). In particular, it is well established in CDM models that when structure formation has reached the nonlinear regime, the density contrasts in comoving space $`f(\delta )`$ follow, to a good approximation, a lognormal distribution (Coles & Jones, 1991; Kofman et al., 1994; Taylor & Watts, 2000; Kayo, Taruya & Suto, 2001),
$$f(\delta )=\frac{(2\pi \omega ^2)^{1/2.}}{1+\delta }\mathrm{exp}\left\{\frac{[\mathrm{ln}(1+\delta )+\omega ^2/2]^2}{2\omega ^2}\right\}$$
(16)
This approximation becomes poor in the highly non-linear regime (Bernardeau & Kofman, 1995; Ueda & Yokoyama, 1996, e.g). The PDF of mass overdensities $`f(\delta )`$ is characterized by a single parameter ($`\omega `$) that is related to the variance of the $`\delta `$-field as
$$\omega ^2=\mathrm{ln}[1+\delta ^2]$$
(17)
At high redshifts, the variance $`\sigma _R`$ over sufficiently large scales R (those explored in this paper) may be easily derived using the linear theory approximation:
$$\sigma _R(z)=\sigma _R(z=0)D(z)$$
(18)
where D(z) is the linear growth rate of density fluctuations normalized to unity at z =0 (Heat, 1977; Hamilton, 2001).
The lognormal approximation formally describes the distribution of matter fluctuations computed in real comoving coordinates. On the contrary, the PDF of galaxies is observationally derived in redshift space. In order to map properly the mass overdensities into galaxy overdensities the mass and galaxy PDFs must be computed in a common reference frame. It has been shown by Sigad, Branchini & Dekel (2000) that an optimal strategy to derive galaxy biasing is to compare both mass and galaxy density fields directly in redshift space. Implicit in this approach is the assumption that mass and galaxies are statistically affected in the same way by gravitational perturbations, and thus, that there is no velocity bias in the motion of the two components.
A general model which allows the explicit computation of the statistical distortions caused by peculiar velocities has been proposed by Kaiser (1987). This applies in the linear regime (i.e., on large scales) and in the local Universe where redshift and distances are linearly related. At cosmological distances z, however, the mapping between real comoving coordinates ($`𝐱`$) and redshift comoving coordinates ($`𝐲`$), i.e., the pseudo-comoving coordinates inferred on the basis of the observed redshifts, is less trivial, and we proceed to obtain it in the following.
In an inhomogeneous Universe, galaxies have motions above and beyond their Hubble velocity (e.g., Giovanelli et al., 1998; Marinoni et al., 1998; Branchini et al., 2001). As a consequence, Doppler spectral shifts add to the cosmological signal and the observed redshift $`(\stackrel{~}{z})`$ is given by
$$\stackrel{~}{z}=z+\frac{U(𝐱)}{c}(1+z)$$
(19)
where z is the cosmological redshift in a uniform Friedman-Robertson-Walker metric and where $`U(𝐱)=𝐯(𝐱)\widehat{𝐫}=|𝐯(𝐱)|\mu `$ is the radial component of the peculiar velocity ($`\mu `$ is the cosine of the angle between the peculiar velocity vector and the line-of-sight versor $`\widehat{𝐫}`$).
The redshift comoving distance of a galaxy at the observed redshift $`\stackrel{~}{z}`$ is thus
$$y=\frac{c}{H_0}_0^{z+\frac{U}{c}(1+z)}\frac{1}{E(\chi )}𝑑\chi ,$$
(20)
where
$$E(z)=[\mathrm{\Omega }_m(1+z)^3+(1\mathrm{\Omega }_m\mathrm{\Omega }_\mathrm{\Lambda })(1+z)^2+\mathrm{\Omega }_\mathrm{\Lambda }]^{1/2}.$$
(21)
At high redshifts ($`z\frac{|U|}{c}`$), we can write
$$y=\frac{c}{H_0}\left[_0^z\frac{1}{E(\chi )}𝑑\chi +\frac{U}{c}(1+z)E(z)^1\right]$$
(22)
which, in turns, gives the coordinate transformation from real comoving space $`𝐱`$ to the redshift comoving space $`𝐲`$
$$𝐲=𝐱\left[1+p(z)\frac{U(𝐱)}{𝐱}\right].$$
(23)
In this mapping, the cosmological term $`p(z)`$,
$$p(z)=\frac{1+z}{H_0E(z)}$$
(24)
is a correcting factor which takes into account the fact that, at high redshifts, distances do not scale linearly with redshift, and, thus, that peculiar velocities cannot be simply added to redshift space positions as in the local Universe.
The galaxy density field in the redshift-distorted space is related to the galaxy density in real space by the Jacobian of the transformation between the two coordinate systems
$$\rho _y(𝐲)=\rho _x(𝐱)\left[1+p(z)\frac{U(𝐱)}{𝐱}\right]^2\left[1+p(z)\frac{dU(𝐱)}{d𝐱}\right]^1$$
(25)
At sufficiently large distances from the observer, neglecting the survey selection function (i.e., considering a volume-limited redshift survey) and at first order in perturbations we obtain
$$\delta _y(𝐲)=\delta _x(𝐱)p(z)\frac{dU(𝐱)}{d𝐱}.$$
(26)
The second term on the right hand side can be evaluated using linear-regime approximations and gravitational instability theory. In comoving coordinates it is given by
$$\frac{dU(𝐱)}{d𝐱}=\frac{\mu ^2\mathrm{f}(\mathrm{z})\mathrm{H}(\mathrm{z})}{1+z}\delta _x(𝐱)$$
(27)
where $`\mathrm{f}=\mathrm{d}\mathrm{ln}\mathrm{D}/\mathrm{d}\mathrm{ln}\mathrm{a}`$ is the logarithmic derivative of the linear growth rate of density fluctuations with respect to the expansion factor a(t). At redshift $`z`$ (corresponding to the comoving position x) a useful approximation is given by:
$$\mathrm{f}(\mathrm{z})\mathrm{\Omega }_\mathrm{m}^{3/5}\mathrm{E}(\mathrm{z})^{6/5}(1+\mathrm{z})^{9/5}$$
(28)
(see Martel (1991); Lahav et al. (1991)).
By combining the previous results we obtain
$$\delta _y(𝐲)=\delta _x(𝐱)[1+\mu ^2\mathrm{f}(\mathrm{z})].$$
(29)
which reduces to the Kaiser (1987) correction when z =0
The relation between the azimuthally averaged variances measured in real and redshift comoving space is
$$\sigma ^y(z)=\left[1+\frac{2}{3}\mathrm{f}(\mathrm{z})+\frac{1}{5}\mathrm{f}^2(\mathrm{z})\right]^{1/2}\sigma ^\mathrm{x}(\mathrm{z}).$$
(30)
We have tested the validity of eq. (30) in the high redshift domain using the Hubble volume N-body simulations carried out by the Virgo consortium (Colberg et al., 2000). This is a large numerical experiment which allows the simulation of mass surveys along the observer past light cone. The simulated mass distribution is computed in a $`\mathrm{\Lambda }`$CDM cosmogony with parameters $`\mathrm{\Omega }_m=0.3`$, $`\mathrm{\Omega }_\mathrm{\Lambda }=0.7`$ and $`H_0=70`$ km s<sup>-1</sup>Mpc<sup>-1</sup>. The volume covered by this N-body simulation is large enough that the mass survey extracted along the diagonal of the simulation cubes extends up to the redshift of interest i.e., the redshift covered by the first-epoch VVDS data ( z =1.5). In this simulation the mass-particle resolution is $`2.210^{12}h^1`$M and the present epoch is defined by a linear rms density fluctuation in a sphere of radius 8 $`h^1`$Mpc of $`\sigma _8=0.9`$.
The mass density contrasts in the redshift perturbed comoving coordinates $`\delta (𝐲_i,R)`$ have been calculated at random positions $`𝐲_i`$ in the simulation volume, by smoothing the particle distribution with a spherical top hat window of length $`R=8`$ $`h^1`$Mpc. Mass variances in different redshift bins are then derived using eq. (15). The result is compared to the prediction of eq. (30) in Fig. 8. Note that, even if it is clear that measurements suffer from cosmic variance due to the relatively small volume sampled at each redshift, the predictions of eq. (30) are in agreement with the observed scaling of the linear mass variance. The magnitude of the correction with respect to the unperturbed case is also evident; mass fluctuations recovered in redshift space on a 8 $`h^1`$Mpc scale, in the redshift comoving coordinates, at z =0.5(1.5) are $`25(35)\%`$ larger than in real comoving space (the correction factor is $`17\%`$ in the local Universe.) Fig. 9 shows that this apparent enhancement in the $`\mathrm{𝑟𝑚𝑠}`$ fluctuations results in a broadening of the mass PDF recovered in the redshift comoving space. Thus, the effect of peculiar velocities is to shrink overdense regions and to inflate underdense regions, enhancing the probability of having large density fluctuations (both positive and negative).
We finally compare, in various redshift intervals, the accuracy with which the lognormal mass PDF derived in the redshift comoving space (by using eq. (30) in 17) approximates the PDF directly inferred from the Hubble volume simulation (see Fig. 9). On a scale of 8 $`h^1`$Mpc, the agreement between the analytical and simulated mass PDFs is satisfactory at all redshifts. This holds true also when the mass PDFs recovered on $`R=5`$ and 10 $`h^1`$Mpc scales are compared.
Thus, with a good degree of confidence, we can use eq. (30) to predict the PDF of mass fluctuations in redshift distorted comoving coordinates (the same coordinates where the galaxy PDF is observed) and in a generic cosmological background. This allows us to speed up computation time and to frame the results about the biasing function in a generic cosmological model.
## 6 Measuring Galaxy Biasing
In this section we describe the method applied to determine the relationship between galaxy and mass overdensities. The galaxy overdensity field $`\delta _g`$ depends in principle on various astrophysical and cosmological parameters such as spatial position (r), underlying matter density fluctuations ($`\delta `$), scale R with which the density field is reconstructed, cosmological time (z), galaxy colors, local gas temperature, non-local environment, etc.
For the purposes of this study, we will rely on the following simplifying theoretical assumptions:
i) the efficiency of galaxy formation on a given cosmological scale is sensitive only to the underlying mass distribution. This means that the galaxy fluctuation field is in a reasonably tight one-to-one relationship with the underlying mass fluctuation field, and that the biasing scheme may be formally represented via the relationship $`\delta _g=b(z,\delta ,R)\delta `$. While such an approach represents a non-trivial step forward in understanding the properties of the biasing function b (if compared, for example, to constant parameterizations of the biasing relation), it is however evident that the biasing function could show, in principle, a more complex functional dependence.
ii) The current theoretical understanding of how clustering of DM proceeds via gravitational instability in the expanding Universe is well developed, i.e., the PDF of mass fluctuations of the real Universe can be safely derived via analytical models or N-body simulations (see discussion in §6) In particular, in what follows, we will consider a $`\mathrm{\Lambda }`$CDM background mass distribution locally normalized to $`\sigma _8(z=0)=0.9`$.
iii) The redshift distortions affect the densities of galaxies and mass in a similar way, i.e., there is no velocity bias between these two components, and galaxies follow the matter flow.
### 6.1 The Method
As described in §1, we derive the relationship between galaxy and mass overdensities in redshift space $`\delta _g=\delta _g(\delta )`$ as the one-to-one transformation which maps the theoretical mass PDF $`f(\delta )`$ into the observed galaxy PDF $`g(\delta _g)`$. A similar method to derive the biasing function has been proposed and tested using CDM simulations by Sigad, Branchini & Dekel (2000) (see also Szapudi & Pan (2004)). This same technique has been recently applied in different contexts by Marinoni & Hudson (2002) to derive the mass-to-light ($`M=M(L)`$) and the X-ray-to-optical ($`L_x=L_x(L)`$) functions for a wide mass range of virialized systems, and by Ostriker et al. (2003) to explore the void phenomena in the context of hydrodynamic simulations.
Using eq. 1,2 and 3, we obtain the biasing function $`b(\delta )`$ as the solution of the following differential equation
$$\{\begin{array}{c}\delta _g(1)=1\hfill \\ \\ b^{^{}}(\delta )\delta +b(\delta )=f(\delta )g(\delta _g)^1\hfill \end{array}$$
(31)
where the prime denotes the derivative with respect to $`\delta `$, $`f(\delta )`$ and $`g(\delta )`$ are the PDF of mass and galaxy fluctuations respectively, and the initial condition has been physically specified by requiring that galaxies cannot form where there is no mass.
With this approach, we loose information on a possible stochasticity characterizing the biasing function. The advantage is that we can provide a measure, on some characteristic scales R, of the local, non-linear, deterministic biasing function (eq. (2)) over the continuous redshift interval 0.4$`<`$z $`<`$1.5.
We have obtained the biasing function $`b(\delta )`$ by numerically integrating the differential equation (31), i) in different redshift intervals in order to follow the evolution of $`b(\delta )`$ as a function of cosmic time, and ii) using matter and galaxy PDFs obtained by smoothing the density fields on R=5,8, and 10 $`h^1`$Mpc in order to test the scale dependence of the galaxy biasing function.
The information contained in the non-linear function $`b(\delta )`$ can be compressed into a single scalar which may be easily compared to the constant values in term of which the biasing relation is usually parameterized (see eq 1). Since, by definition, $`b(\delta )\delta =0`$, the most interesting linear bias estimators are associated to the second order moments of the PDFs, i.e., the variance $`\delta _g^2`$ and the covariance $`\delta _g\delta `$. Following the prescriptions of Dekel & Lahav (1999), we characterize the biasing function as follows:
$$\widehat{b}\frac{b(\delta )\delta ^2}{\delta ^2}$$
(32)
and
$$b_L^2\frac{b^2(\delta )\delta ^2}{\delta ^2}$$
(33)
where the parameter $`\widehat{b}`$, measuring the slope of the linear regression of $`\delta _g`$ on $`\delta `$, is the natural generalization of the linear bias parameter defined in equation 1 and $`b_L^2`$ is an “unbiased estimator” of the linear biasing parameter defined as $`\xi _g=b^2\xi `$, when the bias relation is deterministic, i.e., non-stochastic. The ratio $`r=\widehat{b}/bl`$ is the relevant measure of nonlinearity in the biasing relation; it is unity for linear biasing, and it is either larger or smaller than unity for nonlinear biasing.
The errors in the measured values of the biasing parameter $`b_L`$ have been computed using independent mock catalogs which implement all the selection functions of the VVDS. This allows us to incorporate in our error estimates the uncertainties due to cosmic variance.
### 6.2 Testing the Method
Before applying the biasing computation scheme (eq. 31) to VVDS data, we have tested that the method can be meaningfully applied, i.e., it is free of systematics when implemented with samples of simulated galaxies which mimic all the observational systematics of our sample.
The procedure consists in computing the biasing function $`\delta _s=b(\delta _p)\delta _p`$ between the density field $`\delta _s`$ reconstructed using an s-sample (representing the pseudo-survey sample, see §4.1) and the density fluctuations $`\delta _p`$ of the corresponding p-sample (representing the pseudo-real Universe). We have already determined the range of redshift, density contrasts and smoothing scales where the sample simulating all the VVDS selection functions (s-sample) trace the underlying density of galaxies (p-sample). We thus expect, for consistency, that, in that range, the biasing between the two samples derived by applying our computation scheme (eq. 31, using the PDFs of the s- and p-samples) is independent of $`\delta _p`$ and equal to $`b(\delta _p)=1`$.
Results are presented in Fig. 10 for two different TH smoothing scales. Note that a log-log density plot is used in order to emphasize the behavior of the biasing function in underdense regions. We conclude that on scales R$``$8 $`h^1`$Mpc the density recovered by a “four-passes” VVDS-like survey is not biased with respect to the underlying distribution on any density scale and in any redshift interval up to z =1.5. As a matter of fact, the linear bias parameter with which information contained in the biasing function can be at first order approximated is $`b_L`$ 1 and the biasing relation does not show any significant deviation from linearity as indicated by the fact that the r parameter is also very close to unity.
If the density field is smoothed on 5 $`h^1`$Mpc , the effects of the incompleteness in low-density regions (already discussed in §4.1, see Figs. 4 and 5) become evident. Underdense regions ($`log(1+\delta _p)<0.5`$) in volumes at redshift greater than 1 are poorly sampled with the VVDS survey strategy.
In the same spirit, we have also solved eq. (31) for determining the biasing relation between the PDF of the Hubble volume mass fluctuations and the lognormal approximation given in eq. (16). The biasing relation between these two different descriptions of the mass density field is linear and consistent with the no-bias hypothesis between the two representations of the density field on the scales we are interested in ($`R5`$ $`h^1`$Mpc and $`\mathrm{log}(1+\delta )>1`$).
## 7 The Biasing Function up to $`z1.5`$
### 7.1 Results
The numerical solutions of eq. (31) for the $`_B^c=20+5\mathrm{log}h`$ volume-limited VVDS sample are plotted in various redshift slices, in Fig. 11 for the cases R=8 and 10 $`h^1`$Mpc. Note that a log-log density plot is used in order to emphasize the behavior of the biasing function in underdense regions (note that in these units linear biasing appears as a curved line).
The corresponding parameters $`b_L`$ and r (also computed for the whole flux-limited sample) are quoted in table 1, together with our estimates of the second moment of the galaxy PDF $`\sigma _R`$ and of the skewness parameter S<sub>3</sub>. Both these statistics have been computed as described in §5. Also note that the values of $`\sigma _8`$ measured for the flux-limited sample are consistent with the values independently derived in §3.4 on the basis of the results of the analysis of the clustering properties of VVDS galaxies (Paper III).
An empirical fit of the biasing function is obtained by using a formula similar to the one proposed by Dekel & Lahav (1999)
$$\delta _g(\delta )=\{\begin{array}{cc}(1+a_0)(1+\delta )^{a_1}1\hfill & \delta 0\hfill \\ a_0+a_2\delta +a_3\delta ^2\hfill & \delta >0\hfill \end{array}.$$
(34)
which best describes the behavior of biasing in underdense regions ($`\delta <0`$), either the second order Taylor expansion of the density contrast of dark matter (Fry& Gaztañaga 1993)
$$\delta _g=\underset{k=0}{\overset{2}{}}\frac{b_k}{k!}\delta ^k.$$
(35)
which allows an easier comparison of our results with other studies. The best fitting parameters of these non-linear approximations are quoted in table 2.
The dependence of the shape of the biasing function on galaxy luminosity is plotted in Fig. 12. Results are shown at the median depth of the VVDS sample (in the redshift bin 0.7$`<`$z $`<`$0.9) where faint objects ($`_B<17.7+5\mathrm{log}h`$) are still sampled.
In Fig. 13 we show the redshift evolution of the linear biasing parameter $`b_L`$ computed over the redshift interval $`0.4<z<1.5`$ for both the flux and volume limited samples. We can conclude that biasing is not changing with cosmic time for $`z<0.8`$, while there is a more pronounced evolution of biasing in the redshift interval \[0.8,1.5\]. In particular, the difference between the value of $`b_L`$ at redshift z$``$1.5 and z$``$ 0 for a population of galaxies with luminosity $`_B<20+5\mathrm{log}h`$ is $`\mathrm{\Delta }b_L=0.5\pm 0.14`$, thus significant at a confidence level greater than 3$`\sigma `$.
In Fig. 14 we show the dependence of the linear biasing parameter on galaxy luminosity. Intrinsically brighter galaxies are more strongly biased than less luminous ones at every redshift and the dependence of biasing on luminosity at z $``$0.8 is in good agreement with what is observed in the local Universe (Norberg et al., 2001).
Given the difference in the rest-frame colors of elliptical and irregular galaxies and the fact that the observed I band corresponds to bluer rest-frame bands at higher redshift, the relative fraction of early- and late-type galaxies in our I band limited survey will change as a function of redshift.
Specifically, the observed difference in the B-band luminosity function of early- and late-types (Zucca et al., 2005), implies that the VVDS survey selects preferentially late-type galaxies at higher redshift. It is known that at z =0 late-type galaxies cluster less strongly than early-types (e.g., Giovanelli, Haynes & Chincarini 1986, Guzzo et al. 1997, Giuricin et al. 2001, Madgwick et al. 2002, Zehavi et al. 2002), and, thus, we might observe a variation of the amplitude of density fluctuations at high redshifts just because the morphological composition of our sample changes.
In order to disentangle the spurious morphological contribution to the observed evolution of the global biasing function we have splitted our sample according to rest frame colors, selecting a red ($`(BI)_0>1.5`$; 849 galaxies in the 4-passes region with $`z>0.7`$) and a blue subsample of galaxies ($`(BI)_0<1`$; 1891 galaxies with $`z>0.7`$). These color cuts roughly correspond to selecting, respectively, morphological types $`II`$ and IV according to the classification scheme devised by Zucca et al. 2005 for the VVDS sample.
Clearly, this subsample selection does not correspond to the ideal case of a redshift survey sampling galaxies according to their rest-frame colors; however, useful information about differences in clustering between red and blue populations can still be inferred.
Note that the hypothesis on which the technique of comparing mass and galaxy density distributions is based (§6) can be straightforwardly generalized to compute the biasing between the density distributions of different galaxy types. In particular, we assume that the large scale velocities of late and early types are not dissimilar relative to each other (as it is effectively observed at z=0 e.g., Dekel, 1994; Marinoni et al., 1998) i.e., the two velocity fields are noisy versions of the same underlying field.
Results about the color dependence of biasing are summarized in table 3 and graphically presented in Fig. 15. The red sample is systematically a more biased tracer of mass than the blue one in every redshift interval investigated (i.e., $`b^r>b^b`$), but the relative biasing between the two populations is nearly constant ($`b^r/b^b1.4\pm 0.1`$)
### 7.2 Analysis and Discussion
#### 7.2.1 Biasing for the Global Galaxy Population
Here we examine and interpret the results derived in the previous section. We begin by discussing the general shape of the non-linear biasing function, for the global population, in different density regions. Our results can be summarized as follows:
i) in underdense regions ($`1+\delta <`$1) the local slope of the biasing function $`b(\delta )`$ is always larger than unity even when the global slope is $`b_L<1`$ (see for example Fig. 12). The fact that galaxies in low-mass density regions are always positively biased with respect to the mass distribution (i.e., locally $`b>1`$) is possibly physically caused by the fact that galaxies do not form in very low-density mass regions, i.e., below some finite mass underdensity
the galaxy formation efficiency drops to zero. Using the biasing relation given in eq. (34) the characteristic mass density threshold $`\delta _c`$ below which very few galaxies form ($`\delta _g0.9`$), can be approximated as
$$\mathrm{log}(1+\delta _c)\frac{1+\mathrm{log}(1+a_0)}{a_1}$$
(36)
There is evidence that this mass-density threshold, characterizing regions avoided by galaxies, increases as a function of redshift (see Fig. 11) and luminosity (see Fig. 12). If we consider R=8 $`h^1`$Mpc and the redshift bin 0.7$`<z<`$0.9 we see that while faint galaxies seem to be present even where the mass density contrast is very low
(left panel of Fig. 12, $`\mathrm{log}(1+\delta _c)=0.96\pm 0.09`$), brighter galaxies do not seem to form in deep mass underdensities (right panel of Fig. 12, $`\mathrm{log}(1+\delta _c)=0.73\pm 0.11`$). Therefore low-density regions are preferentially inhabited by low luminosity galaxies.
Moreover the mass-density threshold below which the formation of bright galaxies ($`_B<20+5\mathrm{log}h`$) seems to be inhibited increases, irrespective of the scale investigated (see Fig. 11) as a function of redshift. On a scale R=8 $`h^1`$Mpc, the threshold shifts from $`\mathrm{log}(1+\delta _c)=0.73\pm 0.10`$ at z=0.8 to $`\mathrm{log}(1+\delta _c)0.55\pm 0.07`$ at z=1.4 This suggests that galaxies of a given luminosity were tracing systematically higher mass overdensities in the early Universe, i.e, as time progresses, galaxy formation begins to take place also in lower density peaks.
ii) Even in regions where the mass density distribution is close to its mean value (1+$`\delta `$ 1) bright galaxies are not unbiased tracer of the mass-overdensity field (Fig. 11). This can also be seen by setting $`\delta =0`$ in eq. 34 and noting that $`\delta _g(\delta =0)=a_0>0`$ for both analyzed samples (flux- and volume-limited) in all redshift ranges (see table 2). This result is at variance with what is expected within the simple linear biasing picture, where, by construction, $`\delta _g(\delta =0)=0`$.
iii) In higher matter-density environments (1+$`\delta >1`$) galaxies were progressively more biased mass tracers in the past, i.e., the local slope $`b(\delta )`$ systematically increases with redshift on every scale investigated (Fig. 11). There is some indication that, at the upper tail of the mass density distribution, galaxies are anti-biased with respect to mass on all scales (i.e., the local slope is $`b(\delta )<1`$ for $`\delta 1`$). Antibiasing in overdense regimes is a feature actually observed in simulations (e.g., Sigad, Branchini & Dekel, 2000; Somerville et al., 2001) and expected in theoretical models (e.g., Taruya & Sato, 2000). Physically this could be due to the merging of galaxies which reduces the number density of visible objects in high density regions or because galaxy formation is inhibited in regions where the gas is too hot to collapse and form stars.
iv) In general the linear approximation offers a poor description of the richness of details encoded in the biasing function. As a matter of fact the linear biasing function (dotted line in Fig. 11 and 12) poorly describes, in many cases, the observed scaling of the biasing relation (solid line). At the comoving scales of R=5, 8 and 10 $`h^1`$Mpc, non-linearities in the biasing relation are typically $`(1r)<10\%`$ in the redshift range investigated. We find that the ratio $`b_2/b_1`$ between the quadratic and linear term of the series approximation given in eq. (35) is nearly constant in the redshift range $`0.7<z<1.5`$ and does not depend on luminosity (i.e. it is nearly the same for the flux- and volume-limited subsamples) or smoothing scale. We find that, on average, $`b_2/b_10.15\pm 0.04`$ for $`R=8`$ $`h^1`$Mpc and $`b_2/b_10.19\pm 0.04`$ for $`R=10`$ $`h^1`$Mpc. .
To facilitate comparison with other studies, which generally focus on the linear representation of biasing, we now discuss the properties of the linear approximation of our biasing function. The general characteristics of the linear parameter $`b_L`$ can be summarized as follows:
v) by inspecting table 1, we do not find any significant evidence that the global value of the linear biasing parameter $`b_L`$ depends on the smoothing scale. Any possible systematic variation, if present, is smaller than the amplitude of our errorbars ($``$0.15). This scale independence in the biasing relation extends into the high redshift regimes similar conclusions obtained in the local Universe by the 2dFGRS on scales $`>`$5 $`h^1`$Mpc (Verde et al., 2002). Moreover our results may be interpreted as a supporting evidence for theoretical arguments suggesting that bias is expected to be scale-independent on scales larger than a few $`h^1`$Mpc (e.g., Mann, Peacock & Heavens 1998, Weinberg et al. 2004).
Since we find no evidence of scale-dependent bias, and since with different R scales we are probing different redshift regimes, in Fig. 13 we have averaged the linear biasing parameters measured on 5,8, and 10 $`h^1`$Mpc scales (values quoted in Table 1) in order to follow, in a continuous way, the redshift evolution of the linear galaxy biasing over the larger redshift baseline 0.4$`<`$z $`<`$1.5. Fig. 13 shows that $`b_L`$ for galaxies brighter than $`_B=20+5\mathrm{log}h`$ changes from $`1.10\pm 0.18`$ at $`z0.55`$ to $`1.55\pm 0.12`$ at z$`1.4`$.
An even steeper variation is observed for the biasing of the flux-limited sample, indicating that biasing depends on galaxy luminosity. Fig. 13 shows that the ratio between the amplitude of galaxy fluctuations and the underlying mass fluctuations declines with cosmic time. This scaling is effectively predicted within the framework of the peaks-biasing theoretical model (Kaiser, 1984). At early times, galaxies are expected to form at the highest peaks of the density field since one needs a dense enough clump of baryons in order to start forming stars. Such high-$`\sigma `$ peaks are highly biased tracers of the underlying mass density field. According to this picture, as time progresses and the density field evolves, galaxy formation moves to lower-$`\sigma `$ peaks, nonlinear peaks become less rare events and thus galaxies become less biased tracers of the mass density field. Additional “debiasing” mechanisms may contribute to the observed scaling shown in Fig. 13. It is likely that the densest regions stop forming new galaxies because their gas becomes too hot, cannot cool efficiently, and thus cannot collapse and form stars (Blanton et al., 1999). As galaxy formation moves out of the hottest (and rarest) regions of the Universe, the biasing decreases. Finally, we also note that in order to derive the biasing function we have assumed that there is no difference in the velocity field of the luminous and matter components. After galaxies form, they are subject to the same gravitational forces as the dark matter, and thus they tend to trace the dark matter distribution more closely with time as shown by Dekel & Rees (1987); Fry (1996); Tegmark & Peebles (1998).
vi) In Fig. 13 we also show, for comparison, the value of the 2dFGRS linear biasing parameter inferred at z=0.17 (the effective depth of the survey) as the ratio between the $`\sigma _8`$ value measured by Croton et al. 2004 (in redshift-distorted space; see their Fig 3 and 4) for a sample of objects with $`21<_B5\mathrm{log}h<20`$ (which actually brackets the median luminosity of our volume limited sample $`2L^{}`$), and the rms of mass fluctuations (in redshift-distorted space) in a $`\mathrm{\Lambda }`$CDM background (see §6). This value ($`1.07\pm 0.06`$) is in excellent agreement with what one would independently obtain by combining the linear bias parameter measured by Verde et al. (2002) for the whole 2dFGRS ($`1.04\pm 0.11`$) with the bias scaling law recipe of Norberg et al. (2001), i.e., $`b(z=0.17,L=2L^{})=1.07\pm 0.13`$.
We can conclude that the time dependence of biasing is marginal ($`db/b7\pm 25\%`$) for z$`<`$0.8 while it is substantial ($`db/b33\pm 18\%`$) in the resdhift interval \[0.8-1.5\]. The observed time evolution of bias is well described by the simple scaling relationship $`b_L=1+(0.03\pm 0.01)(1+z)^{3.3\pm 0.6}`$ in the interval 0$`<z<`$1.5.
Assuming a linear biasing scheme, one may note that this result was already implicit in Fig. 7 of §4. The rms fluctuations of the mass density field on a 8 $`h^1`$Mpc scale decrease monotonically with redshift by a factor of $`22\%`$ and $`23\%`$ in the redshift intervals \[0.17-0.8\] and \[0.8-1.4\], respectively; thus, a nearly constant bias is predicted in the redshift range z=\[0.17-0.8\] because the rms fluctuations of the galaxy density field are also decreasing by a factor $`16\%`$ in this same interval. Since, instead, $`\sigma _8`$ of galaxies is marginally increasing in the range z=0.8-1.4 ($`d\sigma /\sigma 10\%`$, see table 1), over this redshift baseline the biasing evolves rapidly.
vii) Bright galaxies are more biased mass tracers than the general population (see Fig. 12). This result confirms and extends into the high redshift domain the luminosity dependence of biasing which is observed in local samples of galaxies (e.g., Benoist et al. 1996, Giuricin et al 2001, Norberg et al. 2001, Zehavi et al. 2002). Specifically, in Fig. 14 we show the dependence of galaxy biasing from luminosity measured in the redshift interval 0.4$`<`$z $`<`$0.9 using three different volume-limited VVDS subsamples (i.e., $`_B5\mathrm{log}h<17.7,<18.7`$ and $`<20`$ respectively) and compare their linear biasing parameters with those observed locally for a sample of objects having the same median luminosities of the VVDS subsamples (i.e., $`L/L^{}`$=0.52,0.82,2.0 respectively). The local estimates have been computed on the basis of the scaling relationship $`b/b^{}`$= 0.85 + 0.15 $`L/L^{}`$ derived by Norberg et al. (2001) using the 2dFGRS sample, assuming the $`b^{}`$ value given by Verde et al. (2002). As shown above for the volume-limited sample, no significant evolution is seen up to z $``$0.8 also when the dependence of bias from luminosity is analyzed.
Finally, we note that, as already discussed in §5, galaxies with the same luminosity at different redshifts may actually correspond to different populations. Since, as we have shown, biasing increases with luminosity also at high redshift, and since the measured value of $`^{}`$ for our sample at redshift z=0.4(1.5) (Paper II) is fainter(brighter) than the cut-off magnitude $`_B^c=20+5\mathrm{log}h`$, we can infer that $`b_L(z)`$ for a population of objects selected, at any given redshift, in a narrow luminosity range around $`(z)`$ should increase with redshift even more than what we have measured for our volume-limited sample (see Fig. 13). A more detailed analysis of the biasing for $`^{}(z)`$ galaxies will be presented in the future, when a larger VVDS data sample will be available.
#### 7.2.2 Biasing as a Function of Galaxy Color
Results summarized in table 3 and presented in Fig. 15 show that, on scales R=8 $`h^1`$Mpc, the red sample is a more biased tracer of mass than the blue one in every redshift interval. Similarly to what we have found for the global population, there is some indication of a systematic increase as a function of redshift of the biasing of bright red and blue objects even if, because of the large errorbars, this trend is not statistically significant.
We can compare our results to the biasing measured for extremely red objects (EROS), i.e., objects with extremely red colors ($`(RK)_{vega}>5`$). Using the results of the correlation analysis of Firth et al. (2002), we obtain, for their $`(IH)_{vega}>3,H_{vega}<20.5`$ sample (which has a median blue luminosity $`_B=20.3+5\mathrm{log}h`$), $`b_L^{EROS}(z1.2)2.3\pm 0.6`$. Considering the results of Daddi et al. (2001), who analyzed a sample of EROS with $`(RK)_{vega}>5`$ (which roughly corresponds to $`(IH)_{vega}>3`$), $`K_{vega}<19.2`$ sample, we conclude that $`b_L^{EROS}(z1.2)4\pm 1`$. These values for the galaxy biasing are respectively $`0.5`$ and 1.8$`\sigma `$ higher than that measured for our sample of bright ($`_B<20+5\mathrm{log}h`$) but moderately red galaxies ($`b_L^r(z1.2)=2\pm 0.5`$). One may interpret this results as an indication for the reddest objects being more strongly biased then moderately red galaxies of similar luminsity. Anyway, given the large errorbars, the evidence that, at $`z1.2`$, the biasing properties of these two differently selected populations are different is not statistically significant. As a matter of fact, the values quoted above are also consistent with an alternative hypothesis, i.e., the strength of the EROS fluctuations with respect to the mass fluctuations is not exceptional when compared to the density fluctuations observed in a sample of high redshift, moderately red galaxies of similar luminosity.
The specific values of the biasing parameter at each cosmic epoch are affected by large errors due to the sparseness of our volume-limited subsamples, and to the presence of cosmic variance. One way to bypass uncertainties due to cosmic variance consists in computing the relative biasing function $`b^{rel}(\delta )=b^r(\delta )/b^b(\delta )`$ between the red and blue subsamples. As the subsamples are drawn from the same volume, this ratio should be minimally affected by the finiteness of the volume probed by the first epoch VVDS data.
Results about the relative biasing between galaxy of different colors are graphically shown in the lower panel of Fig. 15, while estimates of the corresponding $`b_L`$ are quoted in table 3.
We do not observe any trend in the relative biasing between red and blue volume-limited subsamples in the redshift range 0.7$`<`$z $`<`$1.5. Moreover, our best estimate $`b_L^{rel}1.4\pm 0.1`$ is in excellent agreement with what is found for nearly the same color-selected populations both locally (Willmer et al. (1998) found that, on a scale R=8 $`h^1`$Mpc, $`b_L^{rel}b((BR)_{0,vega}>1.3)/b((BR)_{0,vega}<1.3)=1.4\pm 0.3`$, while Wild et al. 2005 using the 2dFGRS found on the same scale $`b_L^{rel}b((BR)_{0,vega}>1.07)/b((BR)_{0,vega}<1.07)=1.5\pm 0.07`$) and at z$``$1 (Coil et al. (2004) found, on a scale R=8 $`h^1`$Mpc, that $`b_L^{rel}b((BR)_0>0.7)/b((BR)_0<0.7)=1.41\pm 0.10`$). Thus, VVDS results suggest that there is no-redshift dependence for the relative biasing between red and blue objects up to $`z1.5`$. Possible systematics could conspire to produce the observed results; the linear approximation may not always captures, in an accurate way, all the information contained in the biasing function, and more importantly, a purely magnitude limited survey samples the red and blue populations at high redshift with a different efficiency (see discussion in §7.1).
In principle, the relative bias could be further studied as a function of scale. For example, locally, there is evidence of scale dependence in the relative bias with the bias decreasing as scale increases (Willmer et al. 1998, Madgwick et al. 2003, Wild et al. 2005). However, the sample currently available is not sufficiently large to obtain proper statistics on this effect, although this should be measurable from the final data set.
Finally, we note that no differences in the value of $`b_L^{rel}`$ are seen by comparing volume-limited subsamples with the flux-limited one in different redshift intervals (see table 3).
Thus, we can deduce that in each redshift bins $`b^R(_B<20)/b^Rb^B(_B<20)/b^B`$. In other terms the biasing between the most luminous objects of a particular color and the global population of objects of the same type appears to be independent of galaxy colors (see table 3).
## 8 Comparison with Theoretical Predictions
In this section we compare our results about the biasing of the $`_B^c=20+5\mathrm{log}h`$ volume-limited, global galaxy sample, with predictions of different theoretical models.
Since we have found that the distribution of galaxy and mass fluctuations are different and the bias was systematically stronger in the past, we can immediately exclude the scenario in which galaxies trace the mass at all cosmic epochs. We thus consider more complex theoretical descriptions of the biasing functions, in particular three different pictures based on orthogonal ideas of how evolution proceeds: the conserving, the merging, and the star forming biasing models (see e.g., Moscardini et al. (1998)).
In the first model the number of galaxies is conserved as a function of time (Dekel & Rees, 1987; Fry, 1996). This model does not assume anything about the distribution and mass of dark matter halos or their connection with galaxies. In this scheme one assumes that galaxies are biased at birth and then they follow the flow of matter without merging, in other terms they behave as test particles dragged around by the surrounding density fluctuations. Because the acceleration on galaxies is the same as that on the dark matter, the gravitational evolution after formation will tend to bring the bias closer to unity, as described by Fry (1996) and Tegmark & Peebles (1998).
The evolution of the bias is given by (e.g., Tegmark & Peebles, 1998)
$$b(z)=1+(b_f1)\frac{D(z_f)}{D(z)}$$
(37)
where $`b_f`$ is the bias at the formation time $`z_f`$.
An alternative picture for the bias evolution, which explicitly takes into account galaxy merging, has been proposed by Mo & White (1996) who gave analytical prescriptions for computing the bias of halos using the Press & Schechter formalism.
If we explicitly assume that galaxies can be identified with dark matter halos, an approximate expression for the biasing of all halos of mass $`>M`$ existing at redshift z (but which collapsed at redshift greater than the observation redshift, see discussion in Matarrese et al. 1997) is given by
$$b(M,z)=1+\frac{1}{\delta _c}\left(\frac{\delta _c^2}{\sigma ^2(M,z)}1\right)$$
(38)
where $`\delta _c1.69`$ is the linear overdensity of a sphere which collapses in an Einstein-de Sitter Universe and $`\sigma (M,z)`$ is the linear rms fluctuations on scales corresponding to mass M at the redshift of observation.
The third model is also framed within the peaks-biasing formalism. It assumes that the distribution of galaxies with luminosity $`>L`$ is well traced by halos with mass $`>`$M, and predicts the biasing of objects that just collapsed at the redshift of observation (e.g., Blanton et al. 2000). In this star forming model,
$$b(M,z)=1+\frac{\delta _c}{\sigma ^2(M,z)}$$
(39)
represents the biasing of galaxies that formed in a narrow time interval around redshift z (i.e., galaxies which experienced recent star formation at redshift z.)
Clearly the above models, are based on a set of theoretical ingredients which represent a crude approximation of the complex multiplicity of physical phenomena entering the cosmic recipe of galaxy biasing. In this context, our goal is to investigate the robustness of the simplifying assumptions on which theoretical models are based, and explore the validity or limits of their underlying physical motivations.
Theoretical predictions are compared to observations (VVDS data plus the local normalization derived from 2dFGRS data) in Fig. 16. The best fitting parameters for each model are evaluated using a $`\chi ^2`$ statistics and are quoted, together with the corresponding minimum $`\chi ^2`$ value of the fit, in table 4.
The best fitting galaxy conserving model is obtained when the bias at birth is $`b_f(z_f=1.4)=1.28\pm 0.03`$ and the corresponding normalized $`\chi ^2`$-value is $`\chi _N^2=2`$. As shown in Fig. 16 the redshift evolution predicted by this model is much weaker than suggested by data. Thus, the gravitational debiasing is a physical mechanism that alone may not fully explain the observed redshift evolution of the biasing, in the sense that it significantly underpredicts the rate of evolution.
The redshift evolution is more pronounced in the merging model (specifically, in Fig. 16, we show the bias evolution of galaxies hosted in halos having mass $`M>2.410^{12}h^1M_{}`$). While this model successfully describes the redshift dependence of the biasing of halos (Mo & White, 1996; Somerville et al., 2001) it poorly accounts for the redshift evolution of the bias of galaxies (with $`_B20+5\mathrm{log}h`$) between z =0 and z =1.5 which is slower than predicted ($`\chi _N^2=5.5`$ for the best fitting model). Thus, although merging is an important mechanism for describing the evolution of matter clustering, our result implies that merging processes affect galaxies in a less dramatic way than halos. Since in the Press & Schechter formalism halos are required to merge instantaneously in bigger units at the redshift of observation, our result would imply, also, that the merger time-scales of galaxies is different from that of halos. Moreover, selecting galaxies with a fixed luminosity threshold may not correspond, over such a wide z range as that investigated here, to selecting halos above a given fixed mass threshold. In this sense our result would be suggestive of evolution in the mass-to-light ratio as a function of time.
In Fig 16 we also show the expected redshift evolution for the star forming model for halos of $`M>3.210^{10}h^1`$M. In this case, the agreement between model and observations is better ($`\chi _N^2=0.7`$). Clearly this does not mean that we are analyzing a sample of objects that just collapsed and formed stars at the time they were observed; as a matter of fact the model cannot capture all the physical processes shaping the biasing relation. Moreover, the low value fitted for the mass threshold is somewhat unrealistic for the bright objects we are considering. Notwithstanding, Blanton et al. 2000 already noted that the prediction of this biasing model is not much different from the biasing evolution expected for the general population of galaxies in a hydrodynamical simulation of the large scale structure.
Our analysis seems to suggest the apparent need of more complex biasing models that better approximate the observed biasing evolution. Understanding our results completely, however, will require more discriminatory power in the data, and, thus, a larger VVDS sample.
## 9 Summary and Conclusions
Deep surveys of the Universe provide the basic ingredients needed to compute the probability distribution function of galaxy fluctuations and to constrain its evolution with cosmic time. The evolution of the galaxy PDF may shed light onto the general assumption that structures grows via gravitational collapse of density fluctuations that are small at early time. When this statistics is combined with analytical CDM predictions for the PDF of mass, useful insights into the biasing function relating mass and galaxy distributions can be obtained.
In this paper, we have explored the potentiality of this approach by analyzing the first-epoch data of the VVDS survey. This is the largest, purely flux-limited sample of spectroscopically measured galaxies, currently available in a continuously connected volume and with a robust sampling up to redshifts z$``$1.5. It is worth to emphasize that the VVDS is probing the high redshift domain at I$``$24 in the VVDS-02h-4 field with the same sampling rate of pioneer surveys of the local Universe such as the CFA (at z$``$0) and, more recently, the 2dFGRS (at z$``$0.1).
Particular attention has been paid to assess the completeness of the VVDS sample and to test the statistical reliability of the PDF of VVDS galaxy fluctuations. In particular:
a) by applying the VVDS observational selection functions to GALICS semi-analytical galaxy simulations we have explored the region of the parameter space where the PDF of VVDS-like densities traces in a statistically unbiased way the parent underlying PDF of the real distribution of galaxy overdensities.
b) we have reconstructed the VVDS galaxy density field on different scales R= $`h^1`$Mpc by correcting the density estimator for various VVDS selection functions. The final density map for the flux-limited sample has been Wiener filtered in order to minimize the shot-noise contribution
By studying the PDF of galaxies in the high redshift Universe we have found that the peak of the galaxy PDF systematically shifts to lower density contrasts as a function of redshift and that the probability of observing underdense regions is greater at z $``$0.7 than it was at z $``$1.5. Both these effects provide strong supporting evidence for the standard assumption that the large scale structure is the result of the gravitational growth of small primordial density fluctuations in an expanding universe.
First, within the paradigm of gravitational instability, the assembling process of the large-scale structures is thought to be regulated by the interplay of two competing effects: the tendency of local self-gravity to make overdense regions collapse and the opposite tendency of global cosmological expansion to move them apart. A key signature of gravitational evolution of density fluctuations in an expanding Universe is that underdense regions, experiencing the cosmological matter outflow, occupy a larger volume fraction at present epoch than in the early Universe. Secondly, both this effects, the peak shift and the development of a low density tail, could indicate the existence of a time-evolving biasing between matter and galaxies, since galaxy biasing, systematically increasing with redshift, offers a natural mechanism to re-map the galaxy PDF into progressively higher intervals of density contrasts.
This last interpretation is confirmed by our measurements of the evolution properties of the second and third moments of the galaxy PDF. We find that i) the rms amplitude of the fluctuations of bright VVDS galaxies is with good approximation constant over the full redshift baseline investigated. Specifically we have shown that, in redshift space, $`\sigma _8`$ for galaxies brighter than $`_B^c=20+5\mathrm{log}h`$ has a mean value of 0.94$`\pm 0.07`$ in the redshift interval 0.7$`<`$z $`<`$1.5; ii) the third moment of the PDF, i.e., the skewness, increases with cosmic time. Its value at $`z1.5`$ is nearly $`2\sigma `$ lower than measured locally by the 2dFGRS. Both these results, when compared to predictions of linear and second order perturbation theory, unambiguously indicate that galaxy biasing is an increasing function of redshift.
Exploiting the sensitivity of the galaxy PDF to the specific form of the mass-galaxy mapping, we have derived the redshift-, density-, and scale-dependent biasing function $`b(z,\delta ,R)`$ between galaxy and matter fluctuations in a $`\mathrm{\Lambda }`$CDM Universe, by analyzing the Jacobian transformation between their respective PDFs. Particular attention has been paid to devise an optimal strategy so that the comparison of the PDFs of mass and galaxies can be carried out in an objective and accurate way. Specifically, we have corrected the lognormal approximation, which describes the mass density PDF, in order to take into account redshift distortions induced by galaxy peculiar velocities at early cosmic epochs where the mapping between redshifts and comoving positions is not linear. In this way, theoretical predictions can be directly compared to observational quantities derived in redshift space.
Without a-priori parameterizing the form of the biasing function, we have shown its general non trivial shape, and studied its evolution as a function of cosmic epochs. Our main results about biasing in the high redshift Universe can be summarized as follows:
i) we detect non-linear effects in the biasing relation. The ratio between the quadratic and linear term of the biasing expansion (cfr eq. 35) is different from zero at a confidence level greater than 3$`\sigma `$ in all the redshift bins and for all the smoothing scales probed. This result confirms a general prediction of CDM-based hierarchical models of galaxy formation (e.g., Sigad, Branchini & Dekel, 2000; Somerville et al., 2001). Such non-linear distortions of the biasing function are not observed locally in the 2dFGRS sample (Verde et al., 2002, but see Baugh et al. 2004), although indirect evidences of a non-linear bias at $`z0`$ exist (Benoist et al. 1999, Baugh et al. 2004).
ii) The biasing function rises sharply in underdense regions (the local slope is $`b(\delta )>1)`$ indicating that below some finite mass density threshold the formation efficiency of galaxies brighter than $`_B<20+5\mathrm{log}h`$ drops to zero. This threshold shifts towards higher values of the mass density field as the luminosity or the redshift of the galaxy population increases.
iii) We do not observe the imprints of scale-dependency in the biasing function a behavior in agreement with results derived from more local surveys at z$``$ 0 (Verde et al., 2002).
iv) By representing the biasing function in linear approximation, we have found that the linear biasing parameter $`b_L`$ evolves with cosmic time: it appears that we live in a special epoch in which the galaxy distribution traces the underlying mass distribution on large scales ($`b_L1`$), while, in the past, the two fields were progressively dissimilar and the relative biasing systematically higher. The difference between the value of $`b_L`$ at redshift z$``$1.5 and z$``$ 0 for a population of galaxies with luminosity $`_B<20+5\mathrm{log}h`$ is significant at a confidence level greater than 3$`\sigma `$ ($`\mathrm{\Delta }b_L0.5\pm 0.14`$). In this interval, the essential characteristics of the time evolution of the linear bias are well described in terms of the phenomenological relationship $`b_L=1+(0.03\pm 0.01)(1+z)^{3.3+0.6}`$.
v) Over the redshift baseline investigated, the rate of biasing evolution is a function of redshift: z$``$0.8 is the characteristic redshift which marks the transition from a “minimum-evolution” late epoch to an early period where the biasing evolution for a population of $`_B<20+5\mathrm{log}h`$ galaxies is substantial ($`33\pm 18\%`$ between redshift 0.8 and 1.5).
vi) Brighter galaxies are more strongly biased than less luminous ones at every redshift and the dependence of biasing on luminosity at z $``$0.8 is in good agreement with what is observed in the local Universe.
vii) By comparing our results to predictions of theoretical models for the biasing evolution, we have shown that the galaxy conserving model (Fry, 1996) and halo merging (Mo & White, 1996) model offer a poor description of our data. This result could suggest that the gravitational debiasing and the hierarchical merging of halos may not be the only physical mechanisms driving the evolution of galaxy biasing across cosmic epochs. At variance with these results, the star forming model (Blanton et al. 2000) seems to describe better the observed redshift evolution of the linear biasing factor.
viii) After splitting the first-epoch redshift catalog into red and blue volume-limited subsamples, we have found that the red sample is systematically a more biased tracer of mass than the blue one in every redshift interval investigated, but the relative biasing between the two populations is nearly constant in the redshift range 0.7$`<`$z $`<`$1.5 ($`b^r/b^b1.4\pm 0.1`$), and comparable with local estimates. Moreover, we have found that the bright red subsample is biased with respect to the general red population in the same way as the bright sample of blue objects is biased with respect to the global blue population thus indicating, that biasing as a function of luminosity might be, at first order, independent of colors.
ix) Because the VVDS and various EROS samples are not yet large enough, the bias of our sample of bright and moderately red objects at z$``$1 is not statistically dissimilar from that expected for EROS of similar luminosity, even if the EROS biasing appears to be systematically larger.
One key aspect of this paper is the measure of evolution in the distribution properties of galaxy overdensities from a continuous volume sampled with the same selection function over a wide redshift baseline. As our volume sampled is still limited, errors on the analysis presented in this paper are dominated by cosmic variance. The technique presented here will be applied to a larger sample as the VVDS observational program progresses.
## Acknowledgments
We would like to acknowledge useful discussions with A. Dekel, R. Giovanelli, and L. Moscardini. We also thank S. Andreon and J. Afonso for their useful comments on the paper. This research has been developed within the framework of the VVDS consortium and it has been partially supported by the CNRS-INSU and its Programme National de Cosmologie (France), and by the Italian Ministry (MIUR) grants COFIN2000 (MM02037133) and COFIN2003 (num.2003020150). CM also acknowledges financial support from the Region PACA. The VLT-VIMOS observations have been carried out on guaranteed time (GTO) allocated by the European Southern Observatory (ESO) to the VIRMOS consortium, under a contractual agreement between the Centre National de la Recherche Scientifique of France, heading a consortium of French and Italian institutes, and ESO, to design, manufacture and test the VIMOS instrument. We thanks the GALICS group for privileged access to their semi-analytic simulations. The mass simulations used in this paper were carried out by the Virgo Supercomputing Consortium using computers based at the Computing Centre of the Max-Planck Society in Garching and at the Edinburgh parallel Computing Centre.
## Appendix A
Here we describe the application of the Wiener filtering technique to deconvolve the noise signature from the VVDS density map. We de-noise data in Fourier space, noting, however, that an equivalent filtering can be directly applied in real space (e.g., Rybicki & Press, 1992; Zaroubi et al., 1995).
Let us assume that the observed smoothed density field $`\delta _O(𝐱)`$, and the true underlying density field $`\delta _T(𝐱)`$, smoothed on the same scale, are related via
$$\delta _O(𝐱)=\delta _T(𝐱)+ϵ(𝐱),$$
(40)
where $`ϵ(𝐱)`$ is the local contribution from shot noise (see eq. (8)). The Wiener filtered density field, in Fourier space, is
$$\stackrel{~}{\delta }_F(𝐤)=(𝐤)\stackrel{~}{\delta }_O(𝐤),$$
(41)
$$(𝐤)=\frac{\stackrel{~}{\delta }_T^2(𝐤)}{\stackrel{~}{\delta }_T^2(𝐤)+(2\pi )^3P_ϵ(𝐤)}.$$
(42)
where brackets denote statistical averages and where $`P_ϵ(𝐤)=(2\pi )^3|\stackrel{~}{ϵ}^2(𝐤)|`$ is the power spectrum of the noise. Assuming ergodic conditions for the noise, we can derive its power spectrum as $`P_ϵ(𝐤)=(2\pi )^3|\stackrel{~}{ϵ}(𝐤)|^2.`$
The Power spectrum of the underlying theoretical density distribution of galaxies, smoothed with the window F and taking into account the VVDS geometrical constraints, can be derived from eq. (5). Specifically he theoretical overdensity field smoothed on a certain scale $`R`$, which is sampled by an idealized survey with no selection functions, is
$$\delta _g(𝐫,R)\frac{1}{\overline{\rho }}\underset{p}{}\delta ^D(𝐫𝐫_p)F\left(\frac{|𝐫𝐫_𝐩|}{R}\right)1$$
(43)
If we assume that this density field is periodic on same volume V, having, for example, the same geometry of the VVDS survey, its Fourier transform is
$$\stackrel{~}{\delta }_T(𝐤)=\frac{1}{\overline{\rho }}\underset{i}{}\frac{n_i}{V}e^{i𝐤𝐫_𝐢}\stackrel{~}{F}_k\frac{1}{V}e^{i𝐤𝐫}d^3𝐫$$
(44)
where the mean theoretical galaxy density $`\overline{\rho }`$ may be estimated averaging over sufficiently large volumes the VVDS data corrected for selection functions and sampling rate (see eq. (6)).
In eq. (44), $`n_i`$ represent the occupation numbers of the infinitesimal cells $`d𝐫_i`$ in which the VVDS volume can be partitioned ($`n_i`$=0 or 1) and the sum is intended over all the cells of the survey volume. The Fourier transform of the smoothing window function F with which the discontinuous galaxy density field is regularized is, in the case of a Top-Hat spherical smoothing filter,
$$\stackrel{~}{F}_k=3\left[\frac{sin(kR)}{(kR)^3}\frac{cos(kR)}{(kR)^2}\right].$$
(45)
In order to obtain an estimate of the quantity $`\stackrel{~}{\delta }_T^2(𝐤)`$ which enters the Wiener filter definition (eq. 42), we compute the ensemble average of the squares of the modulus of the Fourier amplitudes (eq. 44) and obtain
$`\stackrel{~}{\delta }_T(𝐤)\stackrel{~}{\delta }_T^{}(𝐤^{})`$ $`={\displaystyle \frac{1}{(\overline{\rho }V)^2}}[{\displaystyle \underset{i}{}}{\displaystyle \underset{j}{}}n_in_je^{i(𝐤𝐫_i𝐤^{}𝐫_j)}\stackrel{~}{F}_k\stackrel{~}{F}_k^{}^{}+`$
$`+W_kW_k^{}^{}\overline{\rho }V{\displaystyle \underset{i}{}}n_ie^{i𝐤𝐫_i}\stackrel{~}{F}_kW_k^{}^{}+`$
$`\overline{\rho }V{\displaystyle \underset{j}{}}n_je^{i𝐤^{}𝐫_j}\stackrel{~}{F}_k^{}^{}W_k]`$
where
$$W(𝐤)=\frac{1}{V}e^{i𝐤𝐫}d^3𝐫$$
(46)
The mean value of the occupation number is given by (Peebles 1980)
$$n_in_j=\overline{\rho }^2d^3𝐫_id^3𝐫_j[1+\xi (𝐫_i𝐫_j)]$$
$$n_i^2=n_i=\overline{\rho }d^3𝐫_i$$
where the correlation function may be expressed via the Fourier conjugates
$$\xi (𝐫_{ij})=\frac{1}{(2\pi )^3}d^3𝐤P(𝐤)e^{i𝐤𝐫_{ij}}$$
(47)
$$P(𝐤)=d^3𝐫\xi (𝐫)e^{i𝐤r}$$
(48)
We convert the sums into integrals over the occupation cells taking into account the specific VVDS geometry and obtain
$`\stackrel{~}{\delta }_T^2(𝐤)`$ $`=|\stackrel{~}{F}_k|^2{\displaystyle d^3𝐤^{}P(𝐤^{})G(𝐤𝐤^{})}+`$ (49)
$`+|\stackrel{~}{F}_k^2|(\overline{\rho }V)^1+|W_k|^2(1\stackrel{~}{F}_k)^2`$ (50)
were
$$G(𝐤𝐤^{})=\frac{1}{(2\pi )^3}|W(𝐤𝐤^{})|^2$$
(51)
Note that, in the idealized conditions of a survey of infinite spatial extension ($`V\mathrm{}`$) and no smoothing applied to data the previous expression reduces to
$$\stackrel{~}{\delta }_T^2(𝐤)=(2\pi )^3P(𝐤^{})\delta ^D(𝐤𝐤^{}).$$
(52)
Given the quasi pencil-beam nature of the first season VVDS data, however, the sample of the $`\delta `$ field can be described, with good approximation, as confined to a cylinder of length L (aligned along the redshift direction) and radius R. In this case the Fourier transform of the survey window function (cfr. eq. 46) is given by
$$W_k=j_0\left(\frac{L}{2}k_{}\right)\mathrm{\hspace{0.17em}2}\frac{J_1(k_{}R)}{k_{}R}$$
(53)
where $`j_0`$ and $`J_1`$ are the spherical and first kind Bessel functions and where $`k=\sqrt{k_{}^2+k_{}^2}`$.
The convolution integral on the right hand side of equation 50 can then be evaluated as follows (e.g., Kaiser & Peacock 1991, Fisher et al. 1993)
$$d^3𝐤^{}P(𝐤^{})G(𝐤𝐤^{})=\frac{2}{\pi V}_{\mathrm{}}^{\mathrm{}}𝑑yj_0^2\left(yk\frac{L}{2}\right)F(y)$$
where V=$`\pi R^2L`$ is the volume of the cylinder and where
$$F(x)=_0^{\mathrm{}}P\left(\frac{\sqrt{x^2+x^2}}{R}\right)\frac{J_1^2(x^{})}{x^{}}𝑑x^{}.$$
(54)
Note that the theoretically expected variance of the density field in the VVDS volume is derived in real comoving space. Thus, we have corrected the theoretical predictions (cfr. eq. 50) in order to take into account redshift space distortions induced by galaxy peculiar velocities (see discussion in §5).
We compute the galaxy density field on a regular Cartesian grid of spacing 0.5 $`h^1`$Mpc using the smoothing scheme presented in eq. (5). The resulting 3D density map is then Fourier transformed in redshift slices having line-of-sight dimensions d z =0.1. This partition strategy is implemented in order to describe consistently, using cylindrical approximations, the deep survey volume of the VVDS (whose comoving transversal dimensions are an increasing function of distance.) The Wiener filter at each wave-vector position is then computed by using eq. (50). Finally, as described in §3, we select only the Wiener filtered density fluctuations recovered in spheres having at least 70$`\%`$ of their volume in the 4-passes, VVDS-02h-4 field.
In Fig. 17 we use the GALICS semi-analytical simulation (see §4.1), to show the effect of the Wiener filter on the reconstructed galaxy density field. As Fig. 17 shows, the net effect of the correction is to shift towards low-values the density contrasts having low signal-to-noise ratio. By definition, in fact, F is always smaller than unity. Fig. 17 shows that i) at the same density, the effects of the correction are bigger at high redshift where the density field is noisier due to the increasing sparseness of a flux-limited sample, and ii) at the same redshift the Wiener filter mostly affects the low-density tail of the distribution where the counts within the TH window are small. It is also evident from Fig. 17 that, in the density interval $`1<\mathrm{log}(1+\delta _g)<1`$, the Wiener filtered distribution offers a better approximation of the underlying PDF, than the observed (uncorrected) overdensity distribution.
|
warning/0506/hep-ph0506273.html
|
ar5iv
|
text
|
# Nonequilibrium Effects and Baryogenesis
## I Introduction
It is evident that no concentration of antimatter exists within the Solar system and the Milky Way. The absence of annihilation radiation from the Virgo cluster indicates that antimatter can hardly be found within a 20 Mpc scale. A study of the contribution of annihilation radiation near the matter-antimatter boundaries to the cosmic diffuse gamma-ray background virtually excludes domains of antimatter in the visible Universe cohen .
Direct observations of luminous matter show that baryons constitute about a few percents of the total mass of the Universe. This gives a value of order $`10^{10}`$ for $`n_\mathrm{B}/s`$, the ratio of the baryon number density to the entropy density. Precise measurements of the abundance of primordial light elements predicted in big-bang nucleosynthesis combined with the cosmic microwave background experiments restrict this ratio in the range $`n_\mathrm{B}/s(510)\times 10^{11}`$ fields .
Many scenarios for baryon production in the early Universe have been proposed to explain this small baryon asymmetry, such as baryogenesis in grand unified theories, electroweak baryogenesis, leptogenesis, and Affleck-Dine (AD) baryogensis enqvist . Affleck and Dine afdine proposed a mechanism of baryogensis in supersymmetric models in which scalar quark and lepton fields obtain large vacuum expectation values along flat directions of the scalar potential. These coherent scalars or the condensate start to oscillate at a temperature of order 10 TeV when supersymmetry-breaking effects start to become important, and a net baryon number is developed and stored in the oscillating fields provided that $`C`$ and $`CP`$ symmetries are also violated. Subsequently, the scalar quark and lepton fields decay and produce a baryon asymmetry. In fact, the AD mechanism is too efficient and the resulting baryon asymmetry is usually too large. Several dilution processes have been considered to reduce the large baryon asymmetry to the observed value. They involve either introducing additional entropy releases after baryogenesis (e.g., by the decays of the inflaton Linde , massive field fluctuations kkk , or the dilaton dolgov ) or reducing the baryon production by invoking nonrenormalizable terms ng . In all of these studies, the scalar fermion fields have been treated as classical fields and obey the classical equations of motion. In this paper, we will be concerned with quantum fluctuations created from the dynamics of the Affleck-Dine scalar field and its back reaction effects. In particular, we study how these nonequilibrium processes affect the baryon production in the AD mechanism.
It has been proposed that in a certain AD flat direction the condensate is not the state of lowest energy but fragments to form metastable or stable Q-balls, depending on the shape of the radiative correction to the flat direction enqvist . For instance, this occurs along the flat direction with large mixtures of scalar top quark and light gaugino masses or in the $`H_uL`$-direction. In this paper we estimate the baryon asymmetry assuming that the AD condensate does not lead to this type of Q-ball formation. Our presentation is organized as follows. The model of Affleck-Dine baryogenesis is described in Sec. II. We introduce the method of nonequilibrium field theory in Sec. III. It is then applied to the Affleck-Dine model in Sec. IV. We show in Sec. V how to incorporate nonequilibrium effects in the calculation of the baryon asymmetry, and present the results of our numerical calculation and discuss their implications in Sec. VI. Our conclusions are offered in Sec. VII.
## II Affleck-Dine Baryogenesis
Affleck and Dine afdine have shown that, in a supersymmetric SU(5) grand unified model, there is a flat direction in the low-energy effective potential for the following set of vacuum states of the up ($`\stackrel{~}{u}`$), strange ($`\stackrel{~}{s}`$), and bottom ($`\stackrel{~}{b}`$) squark fields as well as the scalar muon field ($`\stackrel{~}{\mu }`$):
$$\stackrel{~}{u}_3^c=a,\stackrel{~}{u}_1=v;\stackrel{~}{s}_2^c=a,\stackrel{~}{\mu }=v;\stackrel{~}{b}_1^c=e^{i\xi }\sqrt{|v|^2+|a|^2},$$
(1)
with all other fields having vanishing vacuum expectation values. Here the superscript $`c`$ denotes charge conjugation, the subscripts denote color indices, $`a`$ and $`v`$ are arbitrary complex numbers, and $`\xi `$ is real. A large baryon asymmetry can be generated by the decays of the condensate associated with these vacuum expectation values via baryon-number violating dimension-four scalar couplings which appear after supersymmetry has been broken. To illustrate how the mechanism works, Affleck and Dine considered a toy model with a single complex scalar field $`\mathrm{\Phi }`$ described by the Lagrangian density
$$=\left(_\mu \mathrm{\Phi }^{}\right)\left(^\mu \mathrm{\Phi }\right)m_\mathrm{\Phi }^2\mathrm{\Phi }^{}\mathrm{\Phi }i\lambda \left(\mathrm{\Phi }^4\mathrm{\Phi }^{\mathrm{\hspace{0.17em}4}}\right),$$
(2)
which contains the mass term with $`m_\mathrm{\Phi }^2=M_S^2`$ and the baryon-number violating coupling, $`\lambda ϵM_S^2/M_G^2`$, where $`M_S`$ is the effective supersymmetry breaking scale, $`ϵ`$ is a real parameter characterizing $`CP`$ violation, and $`M_G`$ is the grand unification scale. For small $`\mathrm{\Phi }`$, the theory has an approximately conserved current
$$j_\mu =i\left(\mathrm{\Phi }^{}_\mu \mathrm{\Phi }(_\mu \mathrm{\Phi }^{})\mathrm{\Phi }\right)$$
(3)
due to the approximate global $`U(1)`$ symmetry: $`\mathrm{\Phi }e^{i\theta }\mathrm{\Phi }`$. The corresponding charge will be referred to as the baryon number (this will be the case if, for example, $`\mathrm{\Phi }`$ represents a scalar quark field).
Classically, in an expanding universe, $`\mathrm{\Phi }`$ obeys the equation of motion
$$\frac{d^2\mathrm{\Phi }}{dt^2}+3H\frac{d\mathrm{\Phi }}{dt}+m_\mathrm{\Phi }^2\mathrm{\Phi }=4i\lambda \mathrm{\Phi }^{\mathrm{\hspace{0.17em}3}},$$
(4)
where $`H`$ is the Hubble parameter. With the initial conditions at $`t=t_0`$:
$$\mathrm{\Phi }|_{t=t_0}=i\mathrm{\Phi }_0\mathrm{and}\dot{\mathrm{\Phi }}|_{t=t_0}=0,$$
(5)
where $`\mathrm{\Phi }_0`$ is real and $`\dot{\mathrm{\Phi }}=d\mathrm{\Phi }/dt`$, it was found that the baryon number per particle at large times ($`tm_\mathrm{\Phi }^1`$) in either a matter-dominated or a radiation-dominated universe is given by
$$r\lambda \mathrm{\Phi }_0^2/m_\mathrm{\Phi }^2.$$
(6)
Assuming $`m_\mathrm{\Phi }^2=M_S^2`$ and $`\lambda =ϵM_S^2/M_G^2`$, we have $`rϵ\mathrm{\Phi }_0^2/M_G^2`$, which can easily provide a large initial $`n_\mathrm{B}/s`$ and thus dilution processes have to be introduced to reduce it to the observed value. For example, taking $`ϵ=10^3`$, $`M_S=10^{16}M_P`$, $`M_G=10^1M_P`$, and $`\mathrm{\Phi }_0^2=10^3M_P^2`$, where $`M_P`$ is the Planck mass, we find $`\lambda =10^{33}`$ and $`r10^4`$. Note that the decay width of the condensate can be estimated as $`\mathrm{\Gamma }_\mathrm{\Phi }(\alpha _s/\pi )^2m_\mathrm{\Phi }^3/|\mathrm{\Phi }|^2`$ afdine , which is typically much smaller than the frequency of the oscillating scalar fields, i.e., $`\mathrm{\Gamma }_\mathrm{\Phi }m_\mathrm{\Phi }`$. Therefore, a net baryon number is developed and gets saturated in the oscillating scalar quark and lepton fields before they decay and produce a baryon asymmetry.
The above consideration did not take into account the fact that a dynamical background field could generate quantum fluctuations which could in turn influence its evolution boyan1 ; hu ; mottola ; lee . Substantial quantum fluctuations from the vacuum state may lead to a value of the generated baryon number density quite different from that obtained from purely classical arguments. In the following sections, we will study how the quantum fluctuation of the dynamical $`\mathrm{\Phi }`$ field and the back reaction on its evolution will affect the ratio $`r`$ for the model of Eq. (2). Nonperturbative Hartree factorizations will be implemented along with the method of nonequilibrium quantum field theory to self-consistently take account of quantum fluctuations and back-reaction effects on the background field. To be completely general, we will not fix the values of $`m_\mathrm{\Phi }`$, $`\lambda `$, and $`\mathrm{\Phi }_0`$, leaving them as free parameters.
## III Nonequilibrium Quantum Field Theory
Before we discuss the Affleck-Dine baryogenesis, let us review the field theoretical methods for treating the quantum fluctuation and back reaction of a dynamical quantum field. The basic methods for studying nonequilibrium phenomena were developed many years ago by Schwinger and Keldysh schwinger . In nonequilibrium situations we are interested in obtaining the equations of motion for the expectation values and correlation functions of the quantum fields in an evolving quantum state or density matrix. This is accomplished by tracking the time evolution of the density matrix, which corresponds to an initial valued problem for the Liouville equation.
In the Schrödinger picture the evolution of the density matrix $`\rho `$ is determined by
$$\rho (t)=U(t,t_0)\rho (t_0)U^1(t,t_0),$$
(7)
where $`U(t,t_0)`$ is the time evolution operator:
$$U(t,t_0)=𝒯\mathrm{exp}\left[i_{t_0}^t𝑑t^{^{}}H(t^{})\right]$$
(8)
for a time-dependent Hamiltonian $`H`$. The symbol $`𝒯`$ means to take the time-ordered product. The expectation value of an operator $`𝒪`$ in the Schrödinger picture is given by
$$𝒪(t)=\frac{\mathrm{Tr}\left[U(t,t_0)\rho (t_0)U^1(t,t_0)𝒪\right]}{\mathrm{Tr}\left[\rho (t_0)\right]},$$
(9)
and thus can be determined once the initial density matrix $`\rho (t_0)`$ is specified.
Consider the case in which the initial density matrix describes a system in equilibrium. When a perturbation is switched on at time $`t_0`$, the resulting time-dependent Hamiltonian can drive the initial state out of equilibrium. Just as in the usual perturbation theory, we write $`H=H_0+H_{\mathrm{int}}`$, where $`H_0`$ is the unperturbed quadratic Hamiltonian and $`H_{\mathrm{int}}`$ stands for the perturbations. For technical convenience of treating this initial valued problem, we can model the dynamics by the following Hamiltonian:
$$H(t)=\mathrm{\Theta }(t_0t)H_0(t_0)+\mathrm{\Theta }(tt_0)[H_0(t)+H_{\mathrm{int}}(t)],$$
(10)
where $`\mathrm{\Theta }(tt_0)`$ is the Heaviside step function. For $`t<t_0`$, the unperturbed Hamiltonian $`H_0(t_0)`$ is fixed at time $`t_0`$ from which we can specify the initial thermal state of the system, and for $`t>t_0`$, the perturbation is switched on in $`H_{\mathrm{int}}`$. Before the perturbation is switched on, the system is assumed to be in equilibrium at a temperature $`T=1/\beta `$ and is described by the initial density matrix
$$\rho (t_0)=\mathrm{exp}[\beta H_0(t_0)].$$
(11)
The zero temperature limit for an initial vacuum state can be obtained by taking $`T0`$, as we shall consider later.
Notice that the initial density matrix can be expressed in terms of the time evolution operator along imaginary time in the distant past: $`\rho (t_0)=U(\mathrm{}i\beta ,\mathrm{})`$. This allows us to write the expectation value in Eq.(9) as
$$𝒪(t)=\frac{\mathrm{Tr}\left[U(\mathrm{}i\beta ,\mathrm{})U^1(t,\mathrm{})𝒪U(t,\mathrm{})\right]}{\mathrm{Tr}\left[U(\mathrm{}i\beta ,\mathrm{})\right]},$$
(12)
where we have inserted $`U(t_0,\mathrm{})U^1(t_0,\mathrm{})=1`$ to the right of $`\rho (t_0)`$ in the numerator, commuted $`U(t_0,\mathrm{})`$ with $`\rho (t_0)`$, and used the following composition property of the evolution operator: $`U^1(t_0,\mathrm{})U^1(t,t_0)=[U(t,t_0)U(t_0,\mathrm{})]^1=U^1(t,\mathrm{})`$. Another insertion of $`1=U^1(\mathrm{},t)U(\mathrm{},t)`$ yields
$$𝒪(t)=\frac{\mathrm{Tr}\left[U(\mathrm{}i\beta ,\mathrm{})U^1(\mathrm{},\mathrm{})U(\mathrm{},t)𝒪U(t,\mathrm{})\right]}{\mathrm{Tr}\left[U(\mathrm{}i\beta ,\mathrm{})\right]}.$$
(13)
Although not apparent from the above expression, the time-dependent expectation value for $`𝒪`$ depends implicitly on $`t_0`$ through the perturbation $`H_{\mathrm{int}}`$ which is switched on at time $`t_0`$. Because there are both forward and backward \[given by $`U^1(\mathrm{},\mathrm{})`$\] time evolutions, and because the initial density matrix is written as the time evolution operator along imaginary time, one is led to consider the following generating functional
$$𝒵[j^+,j^{}]=\mathrm{Tr}\left[U(\mathrm{}i\beta ,\mathrm{})U(\mathrm{},\mathrm{},j^{})U(\mathrm{},\mathrm{},j^+)\right]$$
(14)
for deriving the nonequilibrium equations of motion for the expectation values and correlation functions of the quantum fields.
In anticipation of the subsequent application to the Affleck-Dine model (2) for baryogenesis, let us consider the case of a scalar field $`\phi `$ as an example. In this case the generating functional above has the following path-integral representation
$`𝒵[j^+,j^{}]={\displaystyle 𝑑\phi 𝑑\phi _1𝑑\phi _2𝒟\phi ^+𝒟\phi ^{}𝒟\phi ^\beta \mathrm{exp}\left(i_{\mathrm{}}^{\mathrm{}}𝑑td^3𝐱\left\{\left[\phi ^+\right]+j^+\phi ^+\right\}\right)}`$
$`\times \mathrm{exp}\left(i{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑t{\displaystyle d^3𝐱\left\{\left[\phi ^{}\right]+j^{}\phi ^{}\right\}}\right)\mathrm{exp}\left(i{\displaystyle _{\mathrm{}}^{\mathrm{}i\beta }}𝑑t{\displaystyle d^3𝐱\left\{_0\left[\phi ^\beta \right]\right\}}\right)`$
with the boundary conditions: $`\phi ^+(𝐱,\mathrm{})=\phi ^\beta (𝐱,\mathrm{}i\beta )=\phi (𝐱)`$, $`\phi ^+(𝐱,\mathrm{})=\phi ^{}(𝐱,\mathrm{})=\phi _1(𝐱)`$, and $`\phi ^{}(𝐱,\mathrm{})=\phi ^\beta (𝐱,\mathrm{})=\phi _2(𝐱)`$. The path integrals represented by the $`\phi ^+`$ and $`\phi ^{}`$ fields correspond to the forward and backward time evolution, respectively, while the one described by the $`\phi ^\beta `$ field corresponds to the time evolution along the path parallel to the imaginary time axis. The boundary conditions are in fact a result of the trace as well as the bosonic nature of the operators. Notice that the path integral corresponding to evolution in imaginary time involves only the unperturbed quadratic Lagrangian $`_0`$ \[see Eq. (10)\], while the other path integrals involve the full Lagrangian $`=_0+_{\mathrm{int}}`$, where $`_{\mathrm{int}}`$ stands for the interaction Lagrangian that will be treated as a perturbation.
The introduction of sources in Eq. (III) allows one to obtain expectation values involving the quantum field operators by taking functional derivatives with respect to the sources as follows:
$$\phi ^+i\frac{\delta }{\delta j^+},\phi ^{}i\frac{\delta }{\delta j^{}},$$
(16)
and then setting the sources to zero in the end. In fact, various Green functions can be obtained by taking functional derivatives of the generating function with respect to the appropriate sources. For instance, functional differentiation with respect to the sources $`j^+`$ and $`j^{}`$ can generate the time-ordered and anti-time-ordered Green functions, respectively.
As for computing the real-time correlation functions of interest, there is clearly no need to introduce any source along the path parallel to the imaginary time axis since the Lagrangian along this path involves only its unperturbed part $`_0`$. In the Schrödinger picture, the ensemble average we implement corresponds to an initial density matrix describing a state of thermal equilibrium with respect to the unperturbed Hamiltonian. Path integrals over $`_0`$ can be carried out exactly and we can obtain the generating functional in terms of the nonequilibrium propagators. The temperature dependence enters through the boundary conditions on the nonequilibrium propagators as we shall see later. Thus, the relevant generating functional for computing real-time correlation functions can be obtained as
$`𝒵[j^+,j^{}]=\mathrm{exp}\left\{i{\displaystyle d^4x\left(_{\mathrm{int}}\left[i\frac{\delta }{\delta j^+}\right]_{\mathrm{int}}\left[i\frac{\delta }{\delta j^{}}\right]\right)}\right\}`$
$`\times \mathrm{exp}\{{\displaystyle \frac{1}{2}}{\displaystyle }d^4x{\displaystyle }d^4x^{}[j^+(x)G^{++}(x,x^{})j^+(x^{})j^+(x)G^+(x,x^{})j^{}(x^{})`$
$`j^{}(x)G^+(x,x^{})j^+(x^{})+j^{}(x)G^{}(x,x^{})j^{}(x^{})]\},`$ (17)
where the nonequilibrium propagators corresponding to the forward and backward time branches are given by
$`G^{++}(𝐱,𝐱^{};t,t^{})`$ $`=`$ $`\phi ^+(𝐱,t)\phi ^+(𝐱^{},t^{})`$
$`=`$ $`G^>(𝐱,𝐱^{};t,t^{})\mathrm{\Theta }(tt^{})+G^<(𝐱,𝐱^{};t,t^{})\mathrm{\Theta }(t^{}t),`$
$`G^{}(𝐱,𝐱^{};t,t^{})`$ $`=`$ $`\phi ^{}(𝐱,t)\phi ^{}(𝐱^{},t^{})`$
$`=`$ $`G^>(𝐱,𝐱^{};t,t^{})\mathrm{\Theta }(t^{}t)+G^<(𝐱,𝐱^{};t,t^{})\mathrm{\Theta }(tt^{}),`$
$`G^+(𝐱,𝐱^{};t,t^{})`$ $`=`$ $`\phi ^+(𝐱,t)\phi ^{}(𝐱^{},t^{})`$
$`=`$ $`G^<(𝐱,𝐱^{};t,t^{})=\phi (𝐱^{},t^{})\phi (𝐱,t),`$
$`G^+(𝐱,𝐱^{};t,t^{})`$ $`=`$ $`\phi ^{}(𝐱,t)\phi ^+(𝐱^{},t^{})`$ (18)
$`=`$ $`G^>(𝐱,𝐱^{};t,t^{})=\phi (𝐱,t)\phi (𝐱^{},t^{}).`$
Since time translational invariance is lost for a system that is out of equilibrium, the Green functions defined above will in general depend on the time $`t`$ and $`t^{}`$ separately.
The functions $`G^>`$ and $`G^<`$ can be obtained from computing the 2-point correlation functions of the quantum field $`\phi `$, which can be expressed in terms of the usual creation and annihilation operators with mode functions that are solutions to the homogeneous equations of motion for the quadratic Lagrangian $`_0`$. The explicit derivation for them will be shown in the next section. As a result of the assumed initial thermal state mentioned above, they obey the boundary condition
$$G^<(𝐱,𝐱^{};\mathrm{},t^{})=G^>(𝐱,𝐱^{};\mathrm{}i\beta ,t^{}),$$
(19)
which corresponds to the Kubo-Martin-Schwinger (KMS) condition at a large negative time before the pertubation is switched on bellac .
In summary, the generating functional (17) can be used to obtain the Feynman rules that define the perturbative expansion for weak couplings. There are four sets of nonequilibrium propagators in the forms given in Eq. (18). Two sets of interaction vertices defined by the interaction Lagrangian $`_{\mathrm{int}}`$ involve fields in the forward branch and those in the backward branch. Notice that there exists a relative minus sign difference between these two types of vertices. The calculation of the combinatorial factors is the same as in the usual quantum field theory. This constitutes the basic tool for our studies.
## IV Complex Scalar Field in an Expanding Space-time
Since the interaction rate for scattering processes between the scalar fermion fields and the thermal bath in the early Universe is found afdine to be of the same order of magnitude as their decay rate $`\mathrm{\Gamma }_\mathrm{\Phi }^1`$ (see Sec. II), the scalar fermions are essentially decoupled from the bath for a time scale much longer than the oscillation period of the scalar fields: $`m_\mathrm{\Phi }^1t\mathrm{\Gamma }_\mathrm{\Phi }^1`$. We can therefore apply the method of nonequilibrium quantum field theory with zero temperature to study Affleck-Dine baryogenesis by considering the model of Eq. (2).
The relevant action can be written as
$$𝒮=d^4x\sqrt{g}\left[g_{\mu \nu }\left(^\mu \mathrm{\Phi }^{}\right)\left(^\nu \mathrm{\Phi }\right)m_\mathrm{\Phi }^2\mathrm{\Phi }^{}\mathrm{\Phi }i\lambda \left(\mathrm{\Phi }^4\mathrm{\Phi }^{\mathrm{\hspace{0.17em}4}}\right)\right].$$
(20)
The background geometry is governed by the spatially flat Robertson-Walker metric,
$$d^2s=d^2ta^2(t)d^2𝐱,$$
(21)
where $`a(t)`$ is a scale factor. It will be convenient to introduce the conformal time $`\eta `$ defined as $`d\eta =a^1dt`$ and the conformally rescaled field, $`\chi =a\mathrm{\Phi }`$, whereby the above action can be written as that of a complex scalar field in flat space-time with a time-dependent mass term. To see this, we express the complex scalar field $`\chi `$ in terms of its real and imaginary parts: $`\chi =(1/\sqrt{2})(\chi _1+i\chi _2)`$. The action becomes
$`𝒮`$ $`=`$ $`{\displaystyle 𝑑\eta d^3𝐱[\chi _1,\chi _2]}`$ (22)
$`=`$ $`{\displaystyle 𝑑\eta d^3𝐱\left[\frac{1}{2}\chi _1^{{}_{}{}^{}\mathrm{\hspace{0.17em}2}}\frac{1}{2}\left(\stackrel{}{}\chi _1\right)^2+\frac{1}{2}\chi _2^{{}_{}{}^{}\mathrm{\hspace{0.17em}2}}\frac{1}{2}\left(\stackrel{}{}\chi _2\right)^2U(\chi _1,\chi _2)\right]},`$
where $`\chi _i^{}\chi _i/\eta `$ and the new potential
$$U(\chi _1,\chi _2)=\frac{1}{2}m_\chi ^2(\eta )\left(\chi _1^2+\chi _2^2\right)2\lambda \chi _1\chi _2\left(\chi _1^2\chi _2^2\right)$$
(23)
has the time-dependent mass term with $`m_\chi ^2(\eta )=m_\mathrm{\Phi }^2a^2(a^{^{\prime \prime }}/a)`$.
Since we are interested in the effects of the quantum fluctuations of the scalar fields $`\chi _1`$ and $`\chi _2`$, it will be useful to express each of them in terms of its expectation value with respect to an initial density matrix that will be specified later and the fluctuation field around this average value:
$`\chi _1^\pm (𝐱,\eta )`$ $`=`$ $`\chi _1^0(\eta )+\stackrel{~}{\chi }_1^\pm (𝐱,\eta ),\chi _1^\pm (𝐱,\eta )=\chi _1^0(\eta );`$
$`\chi _2^\pm (𝐱,\eta )`$ $`=`$ $`\chi _2^0(\eta )+\stackrel{~}{\chi }_2^\pm (𝐱,\eta ),\chi _2^\pm (𝐱,\eta )=\chi _2^0(\eta ).`$ (24)
We have listed the fields belonging to the forward $`(+)`$ and backward $`()`$ time branches explicitly. The reason for shifting the ($`\pm `$) fields by the same configuration is that the field expectation values enter in the time evolution operator as c-number background fields, and the evolution forward and backward are considered in this background boyan1 . Note that $`\chi _1^0`$ and $`\chi _2^0`$ are time-dependent for the nonequilibrium problem. We will implement later the corresponding tadpole conditions, $`\stackrel{~}{\chi }_1^\pm (𝐱,\eta )=0`$ and $`\stackrel{~}{\chi }_2^\pm (𝐱,\eta )=0`$, to derive the equations of motion for these field expectation values. Expanding the action in terms of the fluctuation fields $`\stackrel{~}{\chi }_1(𝐱,\eta )`$ and $`\stackrel{~}{\chi }_2(𝐱,\eta )`$, the Lagrangian density becomes
$`[\chi _1^0+\stackrel{~}{\chi }_1^+,\chi _2^0+\stackrel{~}{\chi }_2^+]`$ $`=`$ $`[\chi _1^0,\chi _2^0]+[\stackrel{~}{\chi }_1^+,\stackrel{~}{\chi }_2^+]`$ (25)
$``$ $`\stackrel{~}{\chi }_1^+\left\{\chi _1^{0^{\prime \prime }}+m_\chi ^2(\eta )\chi _1^02\lambda \left[3\left(\chi _1^0\right)^2\chi _2^0\left(\chi _2^0\right)^3\right]\right\}`$
$``$ $`\stackrel{~}{\chi }_2^+\left\{\chi _2^{0^{\prime \prime }}+m_\chi ^2(\eta )\chi _2^0+2\lambda \left[3\chi _1^0\left(\chi _2^0\right)^2\left(\chi _1^0\right)^3\right]\right\}`$
$`+`$ $`6\lambda \chi _1^0\chi _2^0\left(\stackrel{~}{\chi }_1^{+\mathrm{\hspace{0.17em}2}}\stackrel{~}{\chi }_2^{+\mathrm{\hspace{0.17em}2}}\right)+6\lambda \stackrel{~}{\chi }_1^+\stackrel{~}{\chi }_2^+\left[\left(\chi _1^0\right)^2\left(\chi _2^0\right)^2\right]`$
$`+`$ $`2\lambda \left(\chi _2^0\stackrel{~}{\chi }_1^{+\mathrm{\hspace{0.17em}3}}\chi _1^0\stackrel{~}{\chi }_2^{+\mathrm{\hspace{0.17em}3}}+3\chi _1^0\stackrel{~}{\chi }_1^{+\mathrm{\hspace{0.17em}2}}\stackrel{~}{\chi }_2^+3\chi _2^0\stackrel{~}{\chi }_1^+\stackrel{~}{\chi }_2^{+\mathrm{\hspace{0.17em}2}}\right)`$
and similarly for $`[\chi _1^0+\stackrel{~}{\chi }_1^{},\chi _2^0+\stackrel{~}{\chi }_2^{}]`$.
We now introduce a Hartree factorization to take account of the quantum fluctuations. The Hartree factorization can be employed for both the $`(+)`$ and $`()`$ fields as follows:
$`\stackrel{~}{\chi }_1^33\stackrel{~}{\chi }_1^2\stackrel{~}{\chi }_1,\stackrel{~}{\chi }_2^33\stackrel{~}{\chi }_2^2\stackrel{~}{\chi }_2,`$
$`\stackrel{~}{\chi }_1^2\stackrel{~}{\chi }_22\stackrel{~}{\chi }_1\stackrel{~}{\chi }_2\stackrel{~}{\chi }_1+\stackrel{~}{\chi }_1^2\stackrel{~}{\chi }_2,\stackrel{~}{\chi }_1\stackrel{~}{\chi }_2^22\stackrel{~}{\chi }_1\stackrel{~}{\chi }_2\stackrel{~}{\chi }_2+\stackrel{~}{\chi }_2^2\stackrel{~}{\chi }_1,`$
$`\stackrel{~}{\chi }_1^3\stackrel{~}{\chi }_23\stackrel{~}{\chi }_1\stackrel{~}{\chi }_2\stackrel{~}{\chi }_1^2+3\stackrel{~}{\chi }_1^2\stackrel{~}{\chi }_1\stackrel{~}{\chi }_2,\stackrel{~}{\chi }_1\stackrel{~}{\chi }_2^33\stackrel{~}{\chi }_1\stackrel{~}{\chi }_2\stackrel{~}{\chi }_2^2+3\stackrel{~}{\chi }_2^2\stackrel{~}{\chi }_1\stackrel{~}{\chi }_2,`$ (26)
where the one-loop corrections to the 2-point Green functions of $`\stackrel{~}{\chi }_1`$ and $`\stackrel{~}{\chi }_2`$ are cancelled by the counterterms introduced above. We would like to point out that the Hartree approximation we adopt here is uncontrolled in the model of Eq. (2). It will be controlled, for example, in alternative models in which the scalar fields are in the vector representation of the symmetry group $`O(N)`$ as the Hartree approximation is then equivalent to the large-$`N`$ approximation. Our justification for using this approximation here is simply based on the virtue that it provides a nonperturbative framework to treat the quantum fluctuations by solving the evolution equations self-consistently.
After applying the above factorizations, the nonequilibrium Lagrangian density can be written as
$`[\chi _1^0+\stackrel{~}{\chi }_1^+,\chi _2^0+\stackrel{~}{\chi }_2^+][\chi _1^0+\stackrel{~}{\chi }_1^{},\chi _2^0+\stackrel{~}{\chi }_2^{}]`$ (27)
$`=`$ $`_0[\stackrel{~}{\chi }_1^+,\stackrel{~}{\chi }_2^+]_0[\stackrel{~}{\chi }_1^{},\stackrel{~}{\chi }_2^{}]+_{\mathrm{int}}[\stackrel{~}{\chi }_1^+,\stackrel{~}{\chi }_2^+]_{\mathrm{int}}[\stackrel{~}{\chi }_1^{},\stackrel{~}{\chi }_2^{}],`$
where
$`_0[\stackrel{~}{\chi }_1^+,\stackrel{~}{\chi }_2^+]`$ $`=`$ $`{\displaystyle \frac{1}{2}}\stackrel{~}{\chi }_1^{+^{}\mathrm{\hspace{0.17em}2}}{\displaystyle \frac{1}{2}}\left(\stackrel{}{}\stackrel{~}{\chi }_1^+\right)^2{\displaystyle \frac{1}{2}}_{\chi _1}^2(\eta )\stackrel{~}{\chi }_1^{+\mathrm{\hspace{0.17em}2}}`$
$`+`$ $`{\displaystyle \frac{1}{2}}\stackrel{~}{\chi }_2^{+^{}\mathrm{\hspace{0.17em}2}}{\displaystyle \frac{1}{2}}\left(\stackrel{}{}\stackrel{~}{\chi }_2^+\right)^2{\displaystyle \frac{1}{2}}_{\chi _2}^2(\eta )\stackrel{~}{\chi }_2^{+\mathrm{\hspace{0.17em}2}},`$ (28)
with
$`_{\chi _1}^2(\eta )`$ $`=`$ $`m_\chi ^2(\eta )12\lambda \chi _1^0\chi _2^012\lambda \stackrel{~}{\chi }_1\stackrel{~}{\chi }_2,`$
$`_{\chi _2}^2(\eta )`$ $`=`$ $`m_\chi ^2(\eta )+12\lambda \chi _1^0\chi _2^0+12\lambda \stackrel{~}{\chi }_1\stackrel{~}{\chi }_2,`$ (29)
and
$$_{\mathrm{int}}[\stackrel{~}{\chi }_1^+,\stackrel{~}{\chi }_2^+]=\alpha _1(\eta )\stackrel{~}{\chi }_1^+\alpha _2(\eta )\stackrel{~}{\chi }_2^++\alpha _{12}(\eta )\stackrel{~}{\chi }_1^+\stackrel{~}{\chi }_2^+,$$
(30)
with the couplings
$`\alpha _1(\eta )`$ $`=`$ $`\chi _1^{0^{\prime \prime }}+m_\chi ^2(\eta )\chi _1^02\lambda \left[3\left(\chi _1^0\right)^2\chi _2^0\left(\chi _2^0\right)^3\right]6\lambda \chi _2^0\left(\stackrel{~}{\chi }_1^2\stackrel{~}{\chi }_2^2\right)12\lambda \chi _1^0\stackrel{~}{\chi }_1\stackrel{~}{\chi }_2,`$
$`\alpha _2(\eta )`$ $`=`$ $`\chi _2^{0^{\prime \prime }}+m_\chi ^2(\eta )\chi _2^0+2\lambda \left[3\chi _1^0\left(\chi _2^0\right)^2\left(\chi _1^0\right)^3\right]6\lambda \chi _1^0\left(\stackrel{~}{\chi }_1^2\stackrel{~}{\chi }_2^2\right)+12\lambda \chi _2^0\stackrel{~}{\chi }_1\stackrel{~}{\chi }_2,`$
$`\alpha _{12}(\eta )`$ $`=`$ $`6\lambda \left[\left(\chi _1^0\right)^2\left(\chi _2^0\right)^2\right]+6\lambda \left(\stackrel{~}{\chi }_1^2\stackrel{~}{\chi }_2^2\right).`$ (31)
We have used the fact that correlation functions of fields evaluated at the same space-time point in the forward and backward time branches are equal. This can be seen from Eq. (18) by identifying $`(𝐱,t)`$ with $`(𝐱^{},t^{})`$. In addition, due to the spatial translational invariance, the correlation functions depend only on time and should be understood as: $`\stackrel{~}{\chi }_i^{+\mathrm{\hspace{0.17em}2}}(𝐱,\eta )=\stackrel{~}{\chi }_i^{\mathrm{\hspace{0.17em}2}}(𝐱,\eta )=\stackrel{~}{\chi }_i^2(\eta )(i=1,2)`$, $`\stackrel{~}{\chi }_1^+(𝐱,\eta )\stackrel{~}{\chi }_2^+(𝐱,\eta )=\stackrel{~}{\chi }_1^{}(𝐱,\eta )\stackrel{~}{\chi }_2^{}(𝐱,\eta )=\stackrel{~}{\chi }_1\stackrel{~}{\chi }_2(\eta )`$.
In Eq. (27), $`_0`$ is the unperturbed quadratic Lagrangian with respect to which the nonequilibrium Green functions will be defined and the interaction Lagrangian $`_{\mathrm{int}}`$ contains the linear terms in $`\stackrel{~}{\chi }_i`$ as well as the $`\stackrel{~}{\chi }_1\stackrel{~}{\chi }_2`$ terms which will be treated as perturbations. The presence of the nondiagonal $`\stackrel{~}{\chi }_1\stackrel{~}{\chi }_2`$ terms implies that these fields do not form a good basis to construct the nonequilibrium propagators in a perturbative expansion. One may therefore consider invoking the Bogoliubov transformation which is a canonical transformation in the sense that it leaves the measure of the path integral invariant. In this case, the transformation is a two-dimensional field rotation between the old and new basis fields with one parameter specified by the rotation angle. The angle for the field rotation can be determined so as to rid the nondiagonal term in the potential energy of the Lagrangian. In the presence of the time-dependent background fields, it turns out that the Bogoliubov transformation acquires a time-dependent rotation angle, which in turn gives rise to a new kind of unwanted nondiagonal terms in the kinetic energy. Thus, it seems that the Bogoliubov transformation fails to diagonalize the Lagrangian in the presence of the time-dependent background fields. To proceed, we can further approximate the term $`\stackrel{~}{\chi }_1\stackrel{~}{\chi }_2`$ by its mean value $`\stackrel{~}{\chi }_1\stackrel{~}{\chi }_2`$. In other words, we write
$$\stackrel{~}{\chi }_1\stackrel{~}{\chi }_2=\stackrel{~}{\chi }_1\stackrel{~}{\chi }_2+\left(\stackrel{~}{\chi }_1\stackrel{~}{\chi }_2\stackrel{~}{\chi }_1\stackrel{~}{\chi }_2\right),$$
(32)
and treat the terms in the parenthesis as a perturbation.
The fluctuation fields can be decomposed in terms of their Fourier modes with time-dependent mode functions as
$`\stackrel{~}{\chi }_i(𝐱,\eta )`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{\mathrm{\Omega }}}}{\displaystyle \underset{𝐤}{}}\stackrel{~}{\chi }_{i,𝐤}(\eta )e^{i𝐤𝐱}`$ (33)
$`=`$ $`{\displaystyle \frac{1}{\sqrt{\mathrm{\Omega }}}}{\displaystyle \underset{𝐤}{}}\left[a_{i,𝐤}f_{i,𝐤}(\eta )+a_{i,𝐤}^{}f_{i,𝐤}^{}(\eta )\right]e^{i𝐤𝐱},i=1,2,`$
where $`a_{i,𝐤}`$ and $`a_{i,𝐤}^{}`$ are, respectively, the annihilation and creation operators for the fluctuation field $`\stackrel{~}{\chi }_i`$ with momentum $`𝐤`$. The equations of motion for the mode functions $`f_{i,𝐤}`$ can be read off from the quadratic Lagrangian $`_0`$ in Eq. (28) as
$$\left[\frac{d^2}{d\eta ^2}+k^2+_{\chi _i}^2(\eta )\right]f_{i,𝐤}(\eta )=0,i=1,2,$$
(34)
where $`k^2=𝐤𝐤`$. The nonequilibrium propagators can also be expressed in terms of the mode functions:
$`G_{i,𝐤}^{++}(\eta ,\eta ^{})`$ $`=`$ $`G_{i,𝐤}^>(\eta ,\eta ^{})\mathrm{\Theta }(\eta \eta ^{})+G_{i,𝐤}^<(\eta ,\eta ^{})\mathrm{\Theta }(\eta ^{}\eta ),`$
$`G_{i,𝐤}^{}(\eta ,\eta ^{})`$ $`=`$ $`G_{i,𝐤}^>(\eta ,\eta ^{})\mathrm{\Theta }(\eta ^{}\eta )+G_{i,𝐤}^<(\eta ,\eta ^{})\mathrm{\Theta }(\eta \eta ^{}),`$
$`G_{i,𝐤}^+(\eta ,\eta ^{})`$ $`=`$ $`G_{i,𝐤}^<(\eta ,\eta ^{}),`$
$`G_{i,𝐤}^+(\eta ,\eta ^{})`$ $`=`$ $`G_{i,𝐤}^>(\eta ,\eta ^{}),`$
$`G_{i,𝐤}^>(\eta ,\eta ^{})`$ $`=`$ $`{\displaystyle d^3𝐱\stackrel{~}{\chi }_i(𝐱,\eta )\stackrel{~}{\chi }_i(\mathrm{𝟎},\eta ^{})e^{i𝐤𝐱}}`$
$`=`$ $`f_{i,𝐤}(\eta )f_{i,𝐤}^{}(\eta ^{}),`$
$`G_{i,𝐤}^<(\eta ,\eta ^{^{}})`$ $`=`$ $`{\displaystyle d^3𝐱\stackrel{~}{\chi }_i(\mathrm{𝟎},\eta ^{})\stackrel{~}{\chi }_i(𝐱,\eta )e^{i𝐤𝐱}}`$ (35)
$`=`$ $`f_{i,𝐤}(\eta ^{})f_{i,𝐤}^{}(\eta ),`$
where the correlations of the fluctuation fields are taken with respect to a vacuum state that is annihilated by the $`a_{i,𝐤}`$ of Eq. (33).
The equations of motion for the expectation values of the fields can be obtained from the tadpole conditions. From the condition $`\stackrel{~}{\chi }_1^+(𝐱,\eta )=0`$, the lowest-order contributions to the equation of motion are obtained from the terms linear in $`\stackrel{~}{\chi }_1^+`$ in the interaction Lagrangian (30). We find
$$𝑑\eta \left[\alpha _1(\eta )\stackrel{~}{\chi }_1^+\stackrel{~}{\chi }_1^++\alpha _1(\eta )\stackrel{~}{\chi }_1^+\stackrel{~}{\chi }_1^{}\right]=0,$$
(36)
where the spatial arguments have been suppressed. Since the correlation functions $`\stackrel{~}{\chi }_1^+\stackrel{~}{\chi }_1^+`$ and $`\stackrel{~}{\chi }_1^+\stackrel{~}{\chi }_1^{}`$ are linearly independent, the above equation leads to the equation of motion for $`\chi _1^0`$:
$$\chi _1^{0^{\prime \prime }}+m_\chi ^2(\eta )\chi _1^02\lambda \left[3\left(\chi _1^0\right)^2\chi _2^0\left(\chi _2^0\right)^3\right]6\lambda \chi _2^0\left(\stackrel{~}{\chi }_1^2\stackrel{~}{\chi }_2^2\right)12\lambda \chi _1^0\stackrel{~}{\chi }_1\stackrel{~}{\chi }_2=0,$$
(37)
where Eq. (31) has been used. It is straightforward to see that implementing the condition $`\stackrel{~}{\chi }_1^{}(𝐱,\eta )=0`$ will lead to the same equation of motion for $`\chi _1^0`$. Following the similar procedure with the conditions $`\stackrel{~}{\chi }_2^\pm (𝐱,\eta )=0`$, we find the equation of motion for $`\chi _2^0`$ to be
$$\chi _2^{0^{\prime \prime }}+m_\chi ^2(\eta )\chi _2^0+2\lambda \left[3\chi _1^0\left(\chi _2^0\right)^2\left(\chi _1^0\right)^3\right]6\lambda \chi _1^0\left(\stackrel{~}{\chi }_1^2\stackrel{~}{\chi }_2^2\right)+12\lambda \chi _2^0\stackrel{~}{\chi }_1\stackrel{~}{\chi }_2=0.$$
(38)
The quantum fluctuations $`\stackrel{~}{\chi }_1^2`$ and $`\stackrel{~}{\chi }_2^2`$ can be determined self-consistently and have the form
$$\stackrel{~}{\chi }_i^2(\eta )=\frac{d^3𝐤}{(2\pi )^3}f_{i,𝐤}(\eta )^2,$$
(39)
where the infinite volume limit has been taken. On the other hand, the term $`\stackrel{~}{\chi }_1\stackrel{~}{\chi }_2`$ vanishes at tree level and one has to include an interaction vertex from the perturbations to obtain its loop corrections. To lowest order in perturbation theory, the vertex $`6i\lambda \left[\left(\chi _1^0\right)^2\left(\chi _2^0\right)^2\right]\stackrel{~}{\chi }_1\stackrel{~}{\chi }_2`$ is included and one finds that
$`\stackrel{~}{\chi }_1\stackrel{~}{\chi }_2(\eta )`$ $`=`$ $`{\displaystyle \frac{d^3𝐤}{(2\pi )^3}\stackrel{~}{\chi }_{1,𝐤}(\eta )\stackrel{~}{\chi }_{2,𝐤}(\eta )}`$ (40)
$`=`$ $`6i\lambda {\displaystyle _{\eta _0}^\eta }d\eta ^{^{}}\{\left[\chi _1^0(\eta ^{})\right]^2\left[\chi _2^0(\eta ^{})\right]^2\}[\stackrel{~}{\chi }_{1,𝐤}^+(\eta )\stackrel{~}{\chi }_{1,𝐤}^+(\eta ^{})\stackrel{~}{\chi }_{2,𝐤}^+(\eta )\stackrel{~}{\chi }_{2,𝐤}^+(\eta ^{})`$
$`\stackrel{~}{\chi }_{1,𝐤}^+(\eta )\stackrel{~}{\chi }_{1,𝐤}^{}(\eta ^{})\stackrel{~}{\chi }_{2,𝐤}^+(\eta )\stackrel{~}{\chi }_{2,𝐤}^{}(\eta ^{})]`$
$`=`$ $`6i\lambda {\displaystyle _{\eta _0}^\eta }d\eta ^{^{}}\{\left[\chi _1^0(\eta ^{})\right]^2\left[\chi _2^0(\eta ^{})\right]^2\}[G_{1,𝐤}^>(\eta ,\eta ^{})G_{2,𝐤}^>(\eta ,\eta ^{})`$
$`G_{1,𝐤}^<(\eta ,\eta ^{})G_{2,𝐤}^<(\eta ,\eta ^{})]\mathrm{\Theta }(\eta \eta ^{}).`$
As mentioned earlier, the $`\eta _0`$ dependence arises from the perturbations being switched on at time $`\eta _0`$. We have shown above only the calculation (second equality) for the $`\stackrel{~}{\chi }_1^+\stackrel{~}{\chi }_2^+`$ correlation. However, as we have stated after Eq. (31), the final result is also applicable to $`\stackrel{~}{\chi }_1^{}\stackrel{~}{\chi }_2^{}`$. In terms of the mode functions, this correlation function becomes
$$\stackrel{~}{\chi }_1\stackrel{~}{\chi }_2(\eta )=6i\lambda _{\eta _0}^\eta d\eta ^{}\{\left[\chi _1^0(\eta ^{})\right]^2\left[\chi _2^0(\eta ^{})\right]^2\}[f_{1,𝐤}(\eta )f_{1,𝐤}^{}(\eta ^{})f_{2,𝐤}(\eta )f_{2,𝐤}^{}(\eta ^{})\mathrm{c}.\mathrm{c}.].$$
(41)
Equations (34), (37), (38), (39) and (41) form the full set of coupled equations from which we will solve for the field expectation values and the mode functions.
## V Nonequilibrium Effects and Baryon Asymmetry
The proper normalization to measure baryon asymmetry is given by the particle number density. Let us consider the initial state at time $`t_0`$ to be specified by the adiabatic modes for uncoupled harmonic oscillations defined in cosmic time under Hartree approximations. The corresponding initial particle number density can be expressed as
$`\widehat{n}(t_0)`$ $`=`$ $`{\displaystyle }{\displaystyle \frac{d^3𝐤}{(2\pi )^3}}\{{\displaystyle \frac{a^3(t_0)}{2𝒲_{1,𝐤}(t_0)}}[\dot{\mathrm{\Phi }}_{1,𝐤}(t_0)\dot{\mathrm{\Phi }}_{1,𝐤}(t_0)+𝒲_{1,𝐤}^2(t_0)\mathrm{\Phi }_{1,𝐤}(t_0)\mathrm{\Phi }_{1,𝐤}(t_0)]`$ (42)
$`+{\displaystyle \frac{a^3(t_0)}{2𝒲_{2,𝐤}(t_0)}}[\dot{\mathrm{\Phi }}_{2,𝐤}(t_0)\dot{\mathrm{\Phi }}_{2,𝐤}(t_0)+𝒲_{2,𝐤}^2(t_0)\mathrm{\Phi }_{2,𝐤}(t_0)\mathrm{\Phi }_{2,𝐤}(t_0)]1\},`$
where $`\mathrm{\Phi }=(\mathrm{\Phi }_1+i\mathrm{\Phi }_2)/\sqrt{2}`$. The instantaneous frequencies for the real and imaginary part of the complex scalar field $`\mathrm{\Phi }`$ can be gleaned from Eqs. (34) and (29) to be
$`𝒲_{1,𝐤}(t_0)`$ $`=`$ $`\left[{\displaystyle \frac{k^2}{a^2(t_0)}}+m_\mathrm{\Phi }^212\lambda a^2(t_0)\left(\chi _1^0(t_0)\chi _2^0(t_0)+\stackrel{~}{\chi }_1\stackrel{~}{\chi }_2(t_0)\right)\right]^{\frac{1}{2}},`$
$`𝒲_{2,𝐤}(t_0)`$ $`=`$ $`\left[{\displaystyle \frac{k^2}{a^2(t_0)}}+m_\mathrm{\Phi }^2+12\lambda a^2(t_0)\left(\chi _1^0(t_0)\chi _2^0(t_0)+\stackrel{~}{\chi }_1\stackrel{~}{\chi }_2(t_0)\right)\right]^{\frac{1}{2}}.`$ (43)
The expectation value of the number operator with respect to an initial vacuum state as defined after Eq. (35) evolves in time and has the following form in the Heisenberg picture:
$`\widehat{n}(t)`$ $`=`$ $`{\displaystyle }{\displaystyle \frac{d^3𝐤}{(2\pi )^3}}\{{\displaystyle \frac{a^3(t_0)}{2𝒲_{1,𝐤}(t_0)}}[\dot{\mathrm{\Phi }}_{1,𝐤}(t)\dot{\mathrm{\Phi }}_{1,𝐤}(t)+𝒲_{1,𝐤}^2(t_0)\mathrm{\Phi }_{1,𝐤}(t)\mathrm{\Phi }_{1,𝐤}(t)]`$ (44)
$`+{\displaystyle \frac{a^3(t_0)}{2𝒲_{2,𝐤}(t_0)}}[\dot{\mathrm{\Phi }}_{2,𝐤}(t)\dot{\mathrm{\Phi }}_{2,𝐤}(t)+𝒲_{2,𝐤}^2(t_0)\mathrm{\Phi }_{2,𝐤}(t)\mathrm{\Phi }_{2,𝐤}(t)]1\}.`$
In terms of the conformally rescaled fields, the expectation value of the particle number density can be written as the mean value constructed from the expectation values of the quantum fields plus its quantum fluctuations:
$`\widehat{n}(\eta )`$ $``$ $`n(\eta )=n_0(\eta )+n_\mathrm{q}(\eta ),`$
$`n_0(\eta )`$ $`=`$ $`{\displaystyle \frac{a^3(\eta _0)}{2𝒲_{1,\mathrm{𝟎}}(\eta _0)a^4(\eta )}}\left\{\left[{\displaystyle \frac{d}{d\eta }}\chi _1^0(\eta )\right]^2{\displaystyle \frac{a^{}(\eta )}{a(\eta )}}{\displaystyle \frac{d}{d\eta }}\left[\chi _1^0(\eta )\right]^2+\mathrm{\Omega }_{1,𝐤}(\eta )\left[\chi _1^0(\eta )\right]^2\right\}`$
$`+\left(\chi _1\chi _2;\mathrm{\Omega }_{1,𝐤}(\eta )\mathrm{\Omega }_{2,𝐤}(\eta )\right),`$
$`n_\mathrm{q}(\eta )`$ $`=`$ $`{\displaystyle }{\displaystyle \frac{d^3𝐤}{(2\pi )^3}}\{{\displaystyle \frac{a^3(\eta _0)}{2𝒲_{1,𝐤}(\eta _0)a^4(\eta )}}[\stackrel{~}{\chi }_{1,𝐤}^{}(\eta )\stackrel{~}{\chi }_{1,𝐤}^{}(\eta ){\displaystyle \frac{a^{}(\eta )}{a(\eta )}}\left({\displaystyle \frac{d}{d\eta }}\stackrel{~}{\chi }_{1,𝐤}(\eta )\stackrel{~}{\chi }_{1,𝐤}(\eta )\right)`$ (45)
$`+\mathrm{\Omega }_{1,𝐤}(\eta )\stackrel{~}{\chi }_{1,𝐤}(\eta )\stackrel{~}{\chi }_{1,𝐤}(\eta )]{\displaystyle \frac{1}{2}}\}+(\stackrel{~}{\chi }_1\stackrel{~}{\chi }_2;\mathrm{\Omega }_{1,𝐤}(\eta )\mathrm{\Omega }_{2,𝐤}(\eta )),`$
where
$$\mathrm{\Omega }_{i,𝐤}(\eta )=𝒲_{i,𝐤}^2(\eta _0)a^2(\eta )+\frac{a^2(\eta )}{a^2(\eta )}$$
(46)
and $`\eta _0\eta (t_0)`$ is the initial conformal time.
The correlation functions can be expressed in terms of the mode functions as in Eq. (39):
$`\stackrel{~}{\chi }_{i,𝐤}^{}(\eta )\stackrel{~}{\chi }_{i,𝐤}^{}(\eta )`$ $`=`$ $`|f_{i,𝐤}^{}(\eta )|^2,`$
$`\stackrel{~}{\chi }_{i,𝐤}(\eta )\stackrel{~}{\chi }_{i,𝐤}(\eta )`$ $`=`$ $`|f_{i,𝐤}(\eta )|^2.`$ (47)
In the weak coupling limit, and with appropriate choices for the initial expectation values of the quantum fields, we may approximate
$$𝒲_{1,𝐤}(t_0)𝒲_{2,𝐤}(t_0)𝒲_𝐤(t_0);𝒲_𝐤(t_0)=\left[\frac{k^2}{a^2(t_0)}+m_\mathrm{\Phi }^2\right]^{\frac{1}{2}}.$$
(48)
This approximation simplifies the expression for the baryon number density which is given by the current in Eq. (3): $`\widehat{n}_\mathrm{B}`$ = $`i(\mathrm{\Phi }^{}\dot{\mathrm{\Phi }}\dot{\mathrm{\Phi }}^{}\mathrm{\Phi })`$. In parallel with the discussion above for the particle number density, we find
$$\widehat{n}_\mathrm{B}(t)=\frac{a^3(t_0)}{2}\frac{d^3𝐤}{(2\pi )^3}i\left[\mathrm{\Phi }_𝐤^{}(t)\dot{\mathrm{\Phi }}_𝐤(t)+\dot{\mathrm{\Phi }}_𝐤(t)\mathrm{\Phi }_𝐤^{}(t)\dot{\mathrm{\Phi }}_𝐤^{}(t)\mathrm{\Phi }_𝐤(t)\mathrm{\Phi }_𝐤(t)\dot{\mathrm{\Phi }}_𝐤^{}(t)\right].$$
(49)
After converting to conformal time, the expectation value of the baryon number density is found to be
$`n_\mathrm{B}(\eta )\widehat{n}_\mathrm{B}(\eta )`$ $`=`$ $`{\displaystyle \frac{a^3(\eta _0)}{a^3(\eta )}}\left(\chi _2^0(\eta )\chi _1^0^{}(\eta )\chi _1^0(\eta )\chi _2^0^{}(\eta )\right)`$ (50)
$`+{\displaystyle \frac{a^3(\eta _0)}{a^3(\eta )}}{\displaystyle \frac{d^3𝐤}{(2\pi )^3}\left[\stackrel{~}{\chi }_{1,𝐤}^{}(\eta )\stackrel{~}{\chi }_{2,𝐤}(\eta )\stackrel{~}{\chi }_{1,𝐤}(\eta )\stackrel{~}{\chi }_{2,𝐤}^{}(\eta )\right]}.`$
Like the particle number density, it can be separated into a mean value plus its quantum fluctuations. Just as in the earlier computation for the term $`\stackrel{~}{\chi }_1\stackrel{~}{\chi }_2`$, the quantum fluctuations can be computed by introducing a vertex $`6i\lambda \left[(\chi _1^0)^2(\chi _2^0)^2\right]\stackrel{~}{\chi }_1\stackrel{~}{\chi }_2`$ at the lowest order, which yields
$`n_\mathrm{B}(\eta )`$ $`=`$ $`{\displaystyle \frac{a^3(\eta _0)}{a^3(\eta )}}\left(\chi _2^0(\eta )\chi _1^0^{}(\eta )\chi _1^0(\eta )\chi _2^0^{}(\eta )\right)`$ (51)
$`+6ia^3(\eta _0)\lambda {\displaystyle _{\eta _0}^\eta }d\eta ^{}\{\left[\chi _1^0(\eta ^{})\right]^2\left[\chi _2^0(\eta ^{})\right]^2\}\{[f_{1,𝐤}^{}(\eta )f_{1,𝐤}^{}(\eta ^{})f_{2,𝐤}(\eta )f_{2,𝐤}^{}(\eta ^{})\mathrm{c}.\mathrm{c}.]`$
$`[f_{1,𝐤}(\eta )f_{1,𝐤}^{}(\eta ^{})f_{2,𝐤}^{}(\eta )f_{2,𝐤}^{}(\eta ^{})\mathrm{c}.\mathrm{c}.]\}.`$
The baryon number density can be evaluated once the mode functions and field expectation values are determined by the method described in Section IV. In order to ensure the canonical quantization rules, the mode functions must satisfy the relation
$$f_{i,𝐤}^{}(\eta )f_{i,𝐤}^{}(\eta )f_{i,𝐤}(\eta )f_{i,𝐤}^{^{}}(\eta )=i,$$
(52)
which follows from the mode equations (34). With the initial states described by the adiabatic modes for uncoupled harmonic oscillations, the mode functions have the following initial values at the conformal time $`\eta _0`$:
$$f_{i,𝐤}(\eta _0)=\frac{a(\eta _0)}{\sqrt{2a^3(\eta _0)𝒲_𝐤(\eta _0)}},f_{i,𝐤}^{}(\eta _0)=\frac{a^{}(\eta _0)}{\sqrt{2a^3(\eta _0)𝒲_𝐤(\eta _0)}}i\frac{a^2(\eta _0)𝒲_𝐤(\eta _0)}{\sqrt{2a^3(\eta _0)𝒲_𝐤(\eta _0)}}.$$
(53)
As expected, these initial conditions guarantee that the baryon and particle number densities have zero quantum fluctuations initially.
Consistent with the initial conditions (5), the initial expectation values of the conformally rescaled fields are
$$\chi _1^0(\eta _0)=0,\chi _1^0^{}(\eta _0)=0;\chi _2^0(\eta _0)=\sqrt{2}a(\eta _0)\mathrm{\Phi }_0,\chi _2^0^{}(\eta _0)=\sqrt{2}a^{}(\eta _0)\mathrm{\Phi }_0.$$
(54)
Together with (53), these initial conditions guarantee that the initial baryon asymmetry is zero: $`n_\mathrm{B}(\eta _0)=0`$.
In the next section, we will present the results of our calculation that assumes a radiation-dominated epoch in which the scale factor follows $`a(\eta )=m_\mathrm{\Phi }\eta `$. We find that the results are similar in a matter-dominated universe and will not present explicit results for this case.
## VI Results
Before we describe the numerical solutions to the coupled equations, let us try to understand the issue of ultraviolet divergence associated with the loop integrals in these equations. The large-$`k`$ behavior of the mode functions can be obtained from WKB type solutions to the mode equations:
$$f_{i,𝐤}(\eta )=\frac{a(\eta _0)}{2\sqrt{2a^3(\eta _0)𝒲_𝐤(\eta _0)}}\left[(1+\gamma )𝒟_{i,𝐤}^{}(\eta )+(1\gamma )𝒟_{i,𝐤}(\eta )\right],$$
(55)
where
$$𝒟_{i,𝐤}(\eta )=e^{_{\eta _0}^\eta _{i,𝐤}(\eta ^{})𝑑\eta ^{}}$$
(56)
and
$$_{i,𝐤}(\eta )=ik\frac{i}{2k}_{\chi _i}^2(\eta )\frac{1}{4k^2}\frac{d}{d\eta }_{\chi _i}^2(\eta )+𝒪\left(\frac{1}{k^3}\right)+\mathrm{}.$$
(57)
The parameter $`\gamma `$ that appears in Eq. (55) can be determined from the initial conditions (53) on the mode functions. It has the large-$`k`$ expansion
$$\gamma =1\frac{i}{k}\left(\frac{a^{}(\eta _0)}{a(\eta _0)}\right)+𝒪\left(\frac{1}{k^3}\right)+\mathrm{},$$
(58)
which secures
$$|f_{i,𝐤}(\eta )|^2=\frac{1}{2k}\frac{1}{4k^3}\left(_{\chi _i}^2(\eta )\frac{a^{}_{}{}^{}2(\eta _0)}{a^2(\eta _0)}\right)+𝒪\left(\frac{1}{k^4}\right)+\mathrm{}.$$
(59)
These large-$`k`$ behaviors lead to quadratic and logarithmic divergence for the loop integrals as follows:
$$\stackrel{~}{\chi }_i^2(\eta )=\frac{\mathrm{\Lambda }^2}{8\pi ^2}\frac{1}{8\pi ^2}\left(_{\chi _i}^2(\eta )\frac{a^{}_{}{}^{}2(\eta _0)}{a^2(\eta _0)}\right)\mathrm{ln}\frac{\mathrm{\Lambda }}{\kappa }+\mathrm{finite},$$
(60)
where $`\mathrm{\Lambda }`$ is the ultraviolet momentum cutoff and $`\kappa `$ is the renormalization scale. Let us recall that the Affleck-Dine model under consideration is used as an effective field theory to describe the relevant low-energy physics from a supersymmetric grand unified theory. $`\mathrm{\Lambda }`$ is therefore a physical cutoff which can be set at the scale of grand unification, $`\mathrm{\Lambda }M_G`$, while $`\kappa `$ can be chosen to be the supersymmetry breaking scale, $`\kappa M_Sm_\mathrm{\Phi }`$. Since the relevant terms involved in the equations for the field expectation values are of the form $`\left(\stackrel{~}{\chi }_1^2\stackrel{~}{\chi }_2^2\right)`$, only the logarithmic dependence on the cutoff $`\mathrm{\Lambda }`$ remains. For weak couplings, such a mild cutoff dependence does not present any difficulties for our numerical study. Using Eqs. (55), (56), and (57), the quantum fluctuations $`\stackrel{~}{\chi }_1\stackrel{~}{\chi }_2`$ prove to be ultraviolet safe.
Let us turn now to the numerical results of our study. Figure 1 shows the evolution of the expectation values of the quantum fields $`\chi _1^0(\eta )`$ and $`\chi _2^0(\eta )`$ in units of $`m_\mathrm{\Phi }/\sqrt{\lambda }`$, the scale of which is set by the dimensionless parameter $`\varphi _0\sqrt{\lambda }\mathrm{\Phi }_0/m_\mathrm{\Phi }=10^2`$. According to Eqs. (37) and (38), the evolution of the field expectation values are determined by the fluctuations $`(\stackrel{~}{\chi }_1^2\stackrel{~}{\chi }_2^2)(\eta )`$ and $`\stackrel{~}{\chi }_1\stackrel{~}{\chi }_2(\eta )`$, which are displayed in Fig. 2.
For an extremely weak coupling, e.g., $`\lambda =10^{20}`$, Fig. 2(a) shows that the quantum fluctuations are very small and $`\chi _1^0(\eta )`$ and $`\chi _2^0(\eta )`$ follow essentially their classical trajectories which correspond to oscillations with time-dependent frequencies. For a larger, but still weak, coupling (e.g., $`\lambda =10^3`$), the fluctuations become more significant (see Fig. 2(b)) and the amplitudes of the oscillations of $`\chi _1^0`$ and $`\chi _2^0`$ become damped, as can be seen from Fig. 1. From Eq. (37) and the initial conditions (54), we can estimate the order of magnitude of $`\chi _1^0(\eta )`$ to be
$$\chi _1^0(\eta )\lambda \frac{\left[\chi _2^0(\eta )\right]^3}{m_\mathrm{\Phi }^2},$$
(61)
which is consistent with Fig. 1.
The time evolution of the baryon asymmetry, as measured by the ratio of the baryon number density to the particle number density, $`r(\eta )=n_\mathrm{B}(\eta )/n(\eta )`$, is shown in Fig. 3. As stated earlier, the initial baryon asymmetry is zero due to the initial conditions in Eqs. (53) and (54). For very weak couplings, the evolution of $`r`$ is determined by the classical dynamics and is depicted in Fig. 3(a). It starts from zero and decreases toward an equilibrium value $`r(\mathrm{})10\varphi _0^2`$. This saturation value can be understood as follows:
$$|r(\mathrm{})|=\frac{|n_\mathrm{B}(\mathrm{})|}{n(\mathrm{})}\frac{\chi _1^0\chi _2^0}{\left(\chi _2^0\right)^2}\left[\frac{\lambda \left(\chi _2^0\right)^4}{m_\mathrm{\Phi }^2}\right]\left[\frac{1}{\left(\chi _2^0\right)^2}\right]\frac{\lambda \mathrm{\Phi }_0^2}{m_\mathrm{\Phi }^2},$$
(62)
where the approximation for $`\chi _1^0`$ in Eq. (61) has been applied to compute the baryon number density $`n_\mathrm{B}`$, but neglected in the number density $`n`$ as compared with that of $`\chi _2^0`$. The negative baryon asymmetry should not be a concern as the sign can be fixed by the baryon number assignment for the field $`\mathrm{\Phi }`$. The alert reader will recognize that the estimate of $`r(\mathrm{})`$ above is nothing but the result quoted in Eq. (6).
As can be seen from Fig. 3(b), an interesting phenomenon occurs for larger couplings when the quantum fluctuations become large, typically of the order of the field expectation values. The growth of quantum fluctuations is a consequence of parametric amplification lee ; linde induced by the oscillating mean fields. In this case the part of the ratio $`r(\eta )`$ generated from the quantum fluctuations turns out to dilute the mean-field contribution, and therefore drives the ratio toward a much smaller equilibrium value: $`|r(\mathrm{})|10^5\lambda \mathrm{\Phi }_0^4/m_\mathrm{\Phi }^4`$. It thus appears that, under appropriate conditions, the effects of the nonequilibrium quantum fluctuations can wash out the baryon asymmetry generated from the classical dynamics. The reason for this may be understood as follows. The set of coupled equations, Eqs. (34), (37), (38), (39) and (41), which we solve to obtain the baryon asymmetry, involves fluctuation terms with nonzero baryon number such as $`(\stackrel{~}{\chi }_1^2\stackrel{~}{\chi }_2^2)`$ and $`\stackrel{~}{\chi }_1\stackrel{~}{\chi }_2`$. As such, the amount of baryon asymmetry that can be generated is limited by the Hartree approximation implemented in these equations. On the other hand, as a result of the nature of the couplings under consideration, it is possible to create a large particle number density from the fluctuation term $`(\stackrel{~}{\chi }_1^2+\stackrel{~}{\chi }_2^2)`$, which has nonzero particle number but zero baryon number under the global $`U(1)`$ symmetry. This can be seen in Fig. 4, which shows that for a sufficiently large coupling the total particle number density deviates substantially from its classical value, while the baryon number density remains close to its classical value. A similar plot for a much weaker coupling, e.g., $`\lambda =10^{20}`$, would show that both the particle number density and the baryon number density remain essentially at their respective classical values. We are thus led to the conclusion that the ratio $`r`$ can be reduced significantly through nonequilibrium quantum fluctuations with a proper choice of the couplings.
Figure 5 shows the saturation values of $`r`$ as a function of the coupling $`\lambda `$ for a wide range of values of $`\mathrm{\Phi }_0`$. Plotted are equal $`\varphi _0`$ contours, where $`\varphi _0=\sqrt{\lambda }\mathrm{\Phi }_0/m_\mathrm{\Phi }`$. For a given contour, the saturation value for small values of $`\lambda `$ is given by the classical result, $`|r(\mathrm{})|10\varphi _0^2`$, which corresponds to the horizontal part of the curve. When $`\lambda `$ becomes larger than about $`10^6\varphi _0^2`$, the nonequilibrium effects start to operate and significantly diminish the saturation ratio. For a wide range of parameters, this can result in $`r`$ saturating toward the observed value of the baryon asymmetry, as shown by the right end-points of the contours.
As an illustration, let us recall the sample values of the model parameters cited after Eq. (6), which corresponds to the left end of the $`\varphi _0=10^2`$ contour in Fig. 5. Now let us entertain the possibility that the supersymmetry breaking scale may be as high as the grand unification scale, i.e., $`M_SM_G`$. We then find $`\lambda 10^3`$ and the baryon asymmetry is reduced by the nonequilibrium dynamics to the present observed value, $`|r(\mathrm{})|10^{10}`$, corresponding to the right end of the $`\varphi _0=10^2`$ contour in Fig. 5.
## VII Conclusions
We have studied the effects of nonequilibrium dynamics in the Affleck-Dine mechanism of baryogensis. Within the context of the model (2), we find that, for very weak couplings, the ratio of the baryon to particle number density is approximately $`10\lambda \mathrm{\Phi }_0^2/m_\mathrm{\Phi }^2`$, as dictated by the classical dynamics, but for larger couplings it is driven by the nonequilibrium effects toward a smaller value given by $`|r|10^5\lambda \mathrm{\Phi }_0^4/m_\mathrm{\Phi }^4`$. Interestingly, in some cases these nonequilibrium effects can reduce the baryon asymmetry generated in the classical Affleck-Dine mechansim to the present observed value without having recourse to additional dilution processes. This encouraging result suggests that nonequilibrium dynamics can play a significant role in the evolution of the early Universe. It is therefore important to explore other possible manifestations of these nonequilibrium effects. The methodology developed and presented in this paper will be useful for such studies.
###### Acknowledgements.
This work was supported in part by the National Science Council, ROC, under grant NSC92-2112-M-001-029 (NSC93-2112-M-259-007) as well as the short-term visiting program of the Royal Society, and in part by the U. S. Department of Energy under grant DE-FG02-84ER40163. CNL would like to thank the members of the Institute of Physics at the Academia Sinica in Taipei, Taiwan, especially H.-Y. Cheng, C.-Y. Cheung, S.-P. Li, and K.-W. Ng, for their hospitality. He also thanks S.-Y. Wang for a useful discussion.
|
warning/0506/cond-mat0506373.html
|
ar5iv
|
text
|
# I Introduction
## I Introduction
Recently a considerable interest appeared towards analyzing of integrable structure behind Gaussian random matrix theories (RMT) Meh . An important development in this context was made by Kanzieper in Kanz1 , where the link between nonlinear replica $`\sigma `$-models and Painlev$`\stackrel{´}{e}`$ hierarchy of exactly solvable Toda lattice equations Toda was established. More precisely it has been shown, that the replica partition function of the Gaussian unitary ensemble (GUE), which in it’s original EA and dual (in the replica space) $`\sigma `$-model Weg ,ELK representation satisfies the Toda lattice equation (TLE), can be reduced to Painlev$`\stackrel{´}{e}`$ transcendent $`\phi _{IV}`$ FW . From here exact nonperturbative results were obtained demonstrating the ability of the method to overcome the critique of replica trick Zirn2 . In the chiral unitary ensemble (chUE), the replica trick on the basis of TLE was analyzed in Kanz1 ; SV1 .
In this paper we show how the above picture may arise and work in a unitary ensemble defined by probability distribution function $`P(H)=\mathrm{exp}[trV(H)]`$ with the confining potential
$`V(x)={\displaystyle \underset{n=0}{\overset{\mathrm{}}{}}}\mathrm{ln}[1+2q^{n+1}\mathrm{cosh}\beta \chi +q^{2n+2}],`$ (1)
that behaves as $`V(x)\mathrm{ln}^2(|x|)`$ at $`|x|\mathrm{}`$. Here $`x=\mathrm{sinh}\beta \chi /2`$ and the parameter $`q=e^\beta `$, with $`\beta >0`$.
Defined and solved using the orthogonal polynomials method in Muttalib , this unitary random matrix model has attracted considerable attention during the last decade. One reason for that is that it leads to the eigenvalue distribution which for not very large $`\beta <\pi ^2`$ is intermediate between the Wigner-Dyson Meh and the Poisson distribution and has the features typical of the critical level statistics KM ; KT .
Another reason, perhaps more important, is that the potential Eq.(1) fits very well the function $`a[\mathrm{ln}(1+bx^2)]^2`$ in a wide region of the variable $`x`$. This behavior was obtained numerically in the tight binding Anderson model with random site energies as a distribution potential of variables $`x_i`$, $`i=1,\mathrm{},N`$, connected with the conductance as $`g=_{i=1}^N(1+x_i^2)^1`$ Muttalib2 , reflecting a log-normal behavior of the conductance $`g`$ in a product of random matrices Imry . Therefore, by treating matrices $`H`$ as $`H^2=(TT^++T^+T2I)/4`$, where $`T`$ is the transfer matrix characterizing the disordered conductor, the model with the potential (1) can be considered as a qualitatively correct phenomenological model for the Anderson insulator with the parameter $`\beta `$ playing a role of the ratio of the system size to the localization length $`L/\xi `$ and $`\chi /\xi `$ playing a role of the Lyapunov exponent ($`I`$ is an identity matrix). At large $`\beta `$ the probability density $`\rho (\chi )`$ is strongly peaked Muttalib , Bogomol near positive integers exhibiting the phenomenon which is known in theory of quasi-1d Anderson localization as transmission eigenvalue crystallization Muttalib3 , PZIS , BR , Lama .
Up to now the model Eq.(1) was investigated only within the approach of orthogonal polynomials which was successful because the corresponding set of orthogonal polynomials (q-deformed Hermite polynomials) is explicitly known Ismail . Though this method gives an exact solution to the problem it does not tell us anything about the connection of this problem to exactly integrable lattice models. On the other hand, such a connection is expected by analogy with the Wigner-Dyson RMT which is solvable by the ordinary Hermite polynomials. Thus the main question we ask is as to whether or not the RMT Eq.(1) is related with exactly integrable lattice models and what precisely these models are.
As a first step we compute the replica partition function exactly using the pole structure of the distribution function $`P(H)`$. Indeed, the form of the potential allows us to integrate over poles in complex plane and obtain the new exact expression for the partition function as an infinite sum of residues. Such a result of nonperturbative nature turns out to be exactly deducible from the dual representation of the partition function in a form of the Itzikson-Zuber (IZ) type matrix integral over homogeneous factor space of pseudounitary supergroups $`\frac{SU(nM,M)}{SU(nMN,M)}`$ (Stiefel manifold) with $`M\mathrm{}`$. Thus one of the results of the paper is a calculation of this specific IZ integral by use of Duistermaat-Heckman theorem. To the best of our knowledge this integral was not calculated earlier in the literature.
One of the main results of the present paper is the establishment of the link of RMT Eq.(1) with integrable lattice models. For the original replica partition functions we infer the same Toda lattice equations as in the case of the Gaussian RMT Kanz1 . However the dual partition functions $`Z_{n,N}`$ obtained by a super-symmetric mapping onto a variant of the replica $`\sigma `$ model (color-flavor transformation Zirn ) has a more delicate relationship with the Toda lattice models. Namely, semi-infinite hierarchies of fermionic and bosonic flavors involved in this mapping manifest themselves as an extension of graded TLE first found in SV2 in the context of chiral unitary ensemble relevant for QCD.
Finally, we analyze the new representation for the two-level correlation function $`R_2(x)`$ and the probability density $`\rho (\chi )`$ and discuss the phenomenon of $`\chi `$–crystallization.
The paper is organized as follows. In the second section we make a color-flavor transformation in the model under consideration and give its sigma-model formulation defined on the homogeneous factor space of pseudo-unitary supergroups. We calculate the replica partition function in both, color and flavor spaces respectively. In the third section we infer Toda lattice equations in both spaces. In the fourth section, by use of the method of pole integration, we calculate exactly the density of states and the two point correlation function at large N. Comparison with known results follows. Section 5 contains our conclusions. The calculation of the Itzikson-Zuber matrix integral over Stiefel manifold and the derivation of TLE in flavor space are presented in two Appendices.
## II The replica partition function.
The replica partition function of the model Eq.(1), defined by $`Z_{n,N}(ϵ)=det^n(ϵH)_H`$, can be represented as an integral
$`(C_M)^N{\displaystyle 𝑑H\stackrel{n}{det}(ϵH)\underset{k=1}{\overset{M}{}}\frac{1}{det\left[(\mu _k+iH)(\mu _kiH)\right]}},`$ (2)
where $`\mu _k=\mathrm{cosh}\frac{k\beta }{2}`$, $`C_M=(_{n=1}^M4q^n)^1`$ and the limit $`M\mathrm{}`$ is understood. The main idea of the replica trick is the possibility of analytic continuation of $`Z_{n,N}`$ to non-integer values of $`n`$, which was shown to be exact in GUE Kanz1 .
Diagonalizing Hermitian $`N\times N`$ matrices $`H=Vdiag\{E_i\}V^{}`$ and taking into account the Vandermonde determinant one can calculate Eq.(2) by closing the contour of integration at infinity (which is correct for $`n<2MN`$) and using the Cauchy formula of pole integration:
$`Z_{n,N}(ϵ)=(2\pi )^N(C_M)^N{\displaystyle \underset{\{k_i\}}{}}{\displaystyle \frac{_{i=1}^N(ϵi\mu _{k_i})^n}{_{i=1}^N_{j=1}^{2MN}(\mu _{k_i}\mu _{p_j})}},`$ (3)
where the sum is taken over all possible subsets $`\{k_1,\mathrm{},k_N\}`$ of the set of indices $`k\{1,\mathrm{},M\}`$ of the poles $`\mu _k+iE_i`$ on the upper half plane and $`\{p_1,\mathrm{},p_{2MN}\}\{1,\mathrm{},M,M+1,\mathrm{}2M\}\{k_1,\mathrm{},k_N\}`$ includes the numeration of poles $`\mu _kiE_i`$ with the condition $`\mu _{M+j}=\mu _j,j=1,\mathrm{}M`$. This formula will be a starting point of the further analyzes.
### II.1 Color–flavor transformation
Now we define another, dual representation of the partition function in which the replica space and the space of matrix indices interchange. Usually (see e.g. KMez ) such a representation is obtained from the fermionic replica sigma-model derived using the Hubbard-Stratonovich transformation.
In our case the potential (1) is not Gaussian and the Hubbard-Stratonovich transformation is not useful. But nevertheless it appears that one can make a transformation of the partition function given by an integral over original $`N\times N`$ Hermitian matrices (following Zirn let us call the space where they are acting a color space) and pass to the integration over matrices, which are acting in replica space (let us call it flavor space). Similar type of transformations effective in supersymmetric $`\sigma `$-models Zirn are called color-flavor transformations.
In order to make the color-flavor transformation let us transform the determinants in the integrand of Eq. (2) into the Gaussian integral over superfields $`\mathrm{\Phi }_i^a=(\{\mathrm{\Psi }_i^\alpha \};\{\varphi _i^{(1),k}\};\{\varphi _i^{(2),p}\})`$ with $`a=\{\alpha ;k;p\}\{1\mathrm{}n;1\mathrm{}M;1\mathrm{}M\}`$, where n is the number of fermionic components $`\mathrm{\Psi }_i^\alpha `$ while M is the number of bosonic ones $`\varphi _i^{(1),k},\varphi _i^{(2),p}`$. The supersymmetric structure of the fields $`\mathrm{\Phi }_i^a`$ which can be regarded as $`i=1,\mathrm{}N`$ vectors in (n+2M) dimensional flavor superspace, is dictated by the fact that the partition function Eq.(2) is a product of a certain number of determinants and inverse determinants. Using these superfields we obtain
$`Z_{n,N}(ϵ)={\displaystyle \frac{(i^nC_M)^N}{(2\pi )^{2MN}}}{\displaystyle 𝑑H\underset{i,a}{}d\mathrm{\Phi }_i^{+a}d\mathrm{\Phi }_i^a\mathrm{exp}(\mathrm{\Phi }_i^{+a}\stackrel{~}{V}_{ij}^{ab}\mathrm{\Phi }_j^b)}.`$ (4)
The indices $`i,j=1\mathrm{}N`$ correspond to the original color space while indices $`a`$ and $`b`$ are forming a basis in the flavor space. The matrix $`\stackrel{~}{V}_{ij}^{ab}=diag\{[iϵ\delta _{ij}iH_{ij}]\delta _{\alpha \beta };[\mu _k\delta _{ij}+iH_{ij}]\delta _{kp};[\mu _k\delta _{ij}iH_{ij}]\delta _{kp}\}`$ is diagonal with $`\alpha ,\beta =1\mathrm{}n`$; $`k,p=1\mathrm{}M`$ and Kronecker $`\delta `$-s. As one can see the integral over Hermitian random matrices $`H`$ in the expression (4) simply produces a Dirac delta-function $`(2\pi )^{N(N+1)/2}_{i,j}\delta [\mathrm{\Phi }_i^{+a}g^{ab}\mathrm{\Phi }_j^b]`$ with $`g^{ab}=diag\{\delta _{ij},\delta _{kp},\delta _{kp}\}`$.
For further treatment it is convenient to make use of the following trick: we introduce a set of real numbers $`\omega _1,\mathrm{},\omega _N`$ and instead of the delta-function $`_{i,j}\delta [\mathrm{\Phi }_i^{+a}g^{ab}\mathrm{\Phi }_j^b]`$ we substitute $`lim_{\{\omega _i\}0}_{i,j}\delta [\mathrm{\Phi }_i^{+a}g^{ab}\mathrm{\Phi }_j^b\omega _i\omega _j\delta _{ij}]`$. Then after the rescaling of the superfields $`\mathrm{\Phi }_i^a\omega _i\mathrm{\Phi }_i^a`$ one obtains
$`Z_{n,N}(ϵ)`$ $`=`$ $`{\displaystyle \frac{(i^nC_M)^N}{(2\pi )^{2MNN(N+1)/2}}}\underset{\omega _i0}{lim}{\displaystyle \underset{i,j,k=1}{\overset{N}{}}}{\displaystyle \underset{a}{}}\omega _i^{2(2MnN)}{\displaystyle 𝑑\mathrm{\Phi }_k^{+a}𝑑\mathrm{\Phi }_k^a}`$ (5)
$``$ $`{\displaystyle \underset{i,j}{}}\delta [\mathrm{\Phi }_i^{+a}g^{ab}\mathrm{\Phi }_j^b\delta _{ij}]\mathrm{exp}(\omega _i^2\mathrm{\Phi }_i^{+a}g^{ab}V_{ij}^{bc}\mathrm{\Phi }_j^c),`$
with
$`V_{ij}^{ab}=\left(\begin{array}{ccc}iϵ\delta _{ij}\delta _{\alpha \beta }& 0& 0\\ 0& \mu _k\delta _{ij}\delta _{kp}& 0\\ 0& 0& \mu _k\delta _{ij}\delta _{kp}\end{array}\right)`$ (9)
and the summation is understood over repeating indices $`b`$. The argument of the new $`\delta `$\- function in Eq.(5) imposes that all possible field configurations contributing to the integral are $`N`$ normalized vectors in the $`(n+2M)`$ dimensional complex superspace with metric $`g^{ab}`$. By definition this space can be called complex Stiefel manifold and is equivalent to the homogeneous factor space $`S=\frac{SU(nM,M)}{SU(nMN,M)}`$ of pseudounitary supergroups. Therefore the formula (5) is nothing but a replica $`\sigma `$ model formulation of the model (2) in the form of graded IZ type integral over Stiefel manifold in the $`(n+2M)`$ dimensional flavor superspace:
$`Z_{n,N}(ϵ)={\displaystyle \frac{(i^nC_M)^N}{(2\pi )^{2MNN(N+1)/2}}}\underset{\omega _i0}{lim}{\displaystyle \underset{i=1}{\overset{N}{}}}\omega _i^{2(2MnN)}{\displaystyle _{U𝒮}}𝑑U\mathrm{exp}\{Str(\mathrm{\Omega }U^1VU)\},`$ (10)
where $`U^1=gU^+g`$, the matrices $`V=diag\{iϵ_1\mathrm{}iϵ_n,\mu _1\mathrm{}\mu _M,\mu _1\mathrm{}\mu _M\};`$ $`\mathrm{\Omega }=diag\{\underset{n}{\underset{}{0\mathrm{}0}},\omega _1^2\mathrm{}\omega _N^2,\underset{2MN}{\underset{}{0\mathrm{}0}}\}`$ and the limit $`ϵ_1,\mathrm{},ϵ_nϵ`$ is supposed.
The evaluation of standard (super) IZ integral over the group manifold is based on the analyze of heat kernel equation in the Hermitian matrix space and on calculation of induced metric on the subspace of diagonal matrices IZ , Guhr , Alfaro . In our case, since the tangent to the Stiefel manifold vector defined by $`dH=UdU^1GdG^1`$, with $`USU(nM,M)`$, $`GSU(nMN,M)`$, the corresponding induced metric does not coincide with the Vandermonde determinant, but with some modification.
Another elegant calculation of IZ integral over unitary supergroups has been made in Szabo by use of Duistermaat-Heckman theorem. In Fyo it has been shown, that the theorem can be applied to pseudounitary supergroups as well. In Appendix A we apply this theorem to our IZ integral over manifold $`SU(nM,M)/SU(nMN,M)`$. Leaving the mathematical details of calculations aside we present here only the answer. Let us introduce a family of maps $`\sigma :\{1,2,\mathrm{},N\}\{1,\mathrm{},M\}`$ of the indices of the original color space into the subset of indices of the positive root part of the flavor superspace. Correspondingly $`\overline{\sigma }`$ will denote the complement part of the $`\sigma `$: $`\overline{\sigma }=\{1,\mathrm{},2M\}\{\sigma (1),\mathrm{},\sigma (N)\}`$. Then the answer is (see Appendix A for details)
$`{\displaystyle _{U𝒮}}𝑑U\mathrm{exp}\{Str(\mathrm{\Omega }U^1VU)\}=(2\pi )^{2MNN(N1)/2}{\displaystyle \underset{\{\sigma \}}{}}{\displaystyle \frac{det[\mathrm{exp}\omega _k^2\mu _{\sigma (p)}]}{\mathrm{\Delta }_\sigma (\mathrm{\Omega })\mathrm{\Delta }_\sigma (V)}}`$ (11)
where $`\alpha ,\beta =\overline{1,n}`$; $`k,p=\overline{1,N}`$ are fermionic and bosonic components correspondingly, the sum is taken over all possible maps $`\sigma `$ and $`\mathrm{\Delta }_\sigma (\omega )`$, $`\mathrm{\Delta }_\sigma (V)`$ are defined by the induced metric as
$`\mathrm{\Delta }_\sigma (\nu )={\displaystyle \frac{_{i=1}^N_{p=1}^{2MN}(\nu _{\sigma (i)}\nu _{\overline{\sigma }(p)})(\nu _{\sigma (i)}\nu _{\sigma (j)})}{_{\alpha =1}^n_{i=1}^N(\nu _\alpha \nu _{\sigma (i)})}}.`$ (12)
The formulae (11) and (12) present the extension of IZ matrix integral for the case of integration over Stiefel manifold $`𝒮`$. It is clear, that the expressions (11) (12) were obtained for integer values of $`n`$. Now taking the limit $`\omega _i0`$ in (10) we will reproduce the same formula (3) for replica partition function as it was obtained in original color formulation. This demonstrates the equivalence of the color and flavor formulations of the model under consideration.
## III Toda Lattice Equations
In the present section we derive Toda lattice equations in both, color and flavor spaces respectively. The first part is devoted to the so called $`\tau `$ function, defined for a unitary ensemble with any given distribution function, and TLE for it in color space with initial conditions corresponding to the RMT Eq. (1). In the second part we show how the fermionic and bosonic flavors involved in the replica partition functions $`Z_{n,N}`$, connect them with integrable hierarchies.
### III.1 TLE in color space
It is easy to see, that the replica partition function (2) is the $`T0`$ limit of so called $`\tau `$-function $`\tau _N(T,ϵ)=𝑑H\mathrm{exp}Tr[TH+\stackrel{~}{V}(H)]`$ of the model with the potential $`\stackrel{~}{V}(H)=n\mathrm{ln}(ϵH)+V(H)`$, which can be regarded also as a matrix Fourier transform. $`N\times N`$ matrix $`T`$, being conjugate variable to energy matrix $`H`$, can be regarded as an effective time. The $`\tau `$-function can be calculated exactly as Morozov
$`\tau _N(T;ϵ)={\displaystyle \frac{1}{\mathrm{\Delta }(T)}}det[W_\delta (t_j)]_{\delta ,j=0,\mathrm{},N1},`$ (13)
where $`\mathrm{\Delta }(T)`$ is the Vandermonde determinant corresponding to the matrix $`T`$ and
$`W_\delta (t)={\displaystyle 𝑑\lambda \lambda ^\delta \mathrm{exp}\{\lambda t+\stackrel{~}{V}(\lambda )\}}`$ (14)
with $`\delta =0,\mathrm{},N1`$. In the $`t_i0`$ limit (13) gives us the replicated partition function $`Z_{n,N}(ϵ)=\tau _N(0,ϵ)`$ and $`\tau _N(t,ϵ)=\tau _N(T,ϵ)_{\{t_i\}=t}`$ can be expressed via Hankel determinant
$`\tau _N(t,ϵ)=\underset{\delta ,j=0}{\overset{N1}{det}}[_t^{j+\delta }W_0(t)].`$ (15)
Then it is clear Kanz1 ,Morozov , that $`\tau _N(t,ϵ)`$ will satisfy the TLE
$`\tau _N(t,ϵ)_t^2\tau _N(t,ϵ)(_t\tau _N(t,ϵ))^2=\tau _{N+1}(t,ϵ)\tau _{N1}(t,ϵ)`$
with the initial condition
$`\tau _1(0,ϵ)=2\pi C(M){\displaystyle \underset{k=1}{\overset{2M}{}}}[(ϵi\mu _k)^n/{\displaystyle \underset{p=1;pk}{\overset{2M}{}}}(\mu _k\mu _p)].`$ (16)
### III.2 TLE in flavor (supersymmetric) space
After some algebraic manipulations the replicated partition function (3) can be rewritten as
$`Z_{n,N}(ϵ)={\displaystyle \frac{(C_M)^N}{𝒱_{2M}(\mu _1\mathrm{}\mu _{2M})}}\underset{ϵ_\alpha ϵ}{lim}{\displaystyle \frac{1}{𝒱_n(ϵ_1\mathrm{}ϵ_n)}}`$ (17)
$`{\displaystyle \underset{\{\sigma \}}{}}(1)^{\pi (\sigma )}𝒱_{N+n}(ϵ_1\mathrm{}ϵ_n,\mu _{\sigma (1)}\mathrm{}\mu _{\sigma (N)})𝒱_{2MN}(\mu _{\overline{\sigma }(1)}\mathrm{}\mu _{\overline{\sigma }(2MN)})`$
where $`𝒱(\{\mathrm{}\})`$ is the Vandermonde determinant defined by the set of arguments $`\{\mathrm{}\}`$ and $`\pi (\sigma )`$ is the parity of permutation of $`1,\mathrm{},N,N+1,\mathrm{},M`$ to $`\sigma (1),\mathrm{},\sigma (N),\overline{\sigma }(1)\mathrm{},\overline{\sigma }(MN)`$. This form of $`Z_{n,N}(ϵ)`$ can be understood as a formula for a determinant calculated by minors
$`Z_{n,N}(ϵ)={\displaystyle \frac{(i^nC_M)^N}{𝒱_{2M}(\mu _1\mathrm{}\mu _{2M})}}\underset{ϵ_\alpha ϵ}{lim}{\displaystyle \frac{I^1det[A]}{𝒱_n(ϵ_1\mathrm{}ϵ_n)}}`$ (18)
where we have introduced the $`(n+2M)\times (n+2M)`$matrix $`A`$ as
$`A=\left(\begin{array}{ccccccc}I_1& \delta _1I_1& \mathrm{}& \delta _1^{n+N1}I_1& 0& \mathrm{}& 0\\ I_2& \delta _2I_2& \mathrm{}& \delta _2^{n+N1}I_2& 0& \mathrm{}& 0\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\\ I_n& \delta _nI_n& \mathrm{}& \delta _n^{n+N1}I_n& 0& \mathrm{}& 0\\ \overline{I}_1& \overline{\delta }_1\overline{I}_1& \mathrm{}& \overline{\delta }_1^{n+N1}\overline{I}_1& \overline{I}_1& \mathrm{}& \overline{\delta }_1^{2MN1}\overline{I}_1\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\\ \overline{I}_M& \overline{\delta }_M\overline{I}_M& \mathrm{}& \overline{\delta }_M^{n+N1}\overline{I}_M& \overline{I}_M& \mathrm{}& \overline{\delta }_M^{2MN1}\overline{I}_M\\ 0& 0& \mathrm{}& 0& \overline{I}_{M+1}& \mathrm{}& \overline{\delta }_{M+1}^{2MN1}\overline{I}_{M+1}\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\\ 0& 0& \mathrm{}& 0& \overline{I}_{2M}& \mathrm{}& \overline{\delta }_{2M}^{2MN1}\overline{I}_{2M}\end{array}\right),`$ (29)
involving the functions $`I_\alpha =\mathrm{exp}(iϵ_\alpha ),\alpha =1,\mathrm{}n;\overline{I}_r=\mathrm{exp}\mu _r,r=1,\mathrm{}2M;I=_{\alpha ,r=1}^{n,2M}I_\alpha \overline{I}_r`$ and the notations $`\delta _\alpha =ϵ_\alpha \frac{}{ϵ_\alpha };\overline{\delta }_r=\mu _r\frac{}{\mu _r}.`$ Define $`\delta ^{1+2M}=ϵ\frac{}{ϵ}+_{r=1}^{2M}(\overline{\delta }_r)`$.
Now, following the technique developed in SV2 , the modified graded Toda lattice equation for replicated partition function $`Z_{n,N}(ϵ)`$ will follow directly from its expression (18) as a determinant and two sets of Sylvester identity Syl for the matrix $`A`$ (see Appendix B for a details). As a result, the following equation is established
$`ni_ϵ\delta ^{1+2M}\mathrm{ln}[Z_{n,N}(ϵ)]`$ $`=`$
$`n{\displaystyle \frac{Z_{n+1,N}(ϵ)Z_{n1,N}(ϵ)}{Z_{n,N}^2(ϵ)}}`$ $`+`$ $`n{\displaystyle \frac{Z_{n+1,N1}(ϵ)Z_{n1,N+1}(ϵ)}{Z_{n,N}^2(ϵ)}},`$ (30)
where $`n<2MN`$, when $`Z_{n,N}`$ is finite. It has a graded form since contains derivatives over both, fermionic and bosonic parameters. One can see here, that in the replica limit $`n0`$ the right hand side (RHS) of this equation involves a factorized product of fermionic and bosonic partition functions $`Z_{n=1}`$ and $`Z_{n=1}`$ correspondingly. Moreover, the second term in RHS shows that indeed fermionic, bosonic and supersymmetric partition functions for any N belongs to the same integrable hierarchy.
In the limit $`N\mathrm{}`$ (but $`N<M`$), when $`Z_{n,N1}=Z_{n,N+1}=Z_{n,\mathrm{}}`$, the equation (III.2) simplifies to
$`ni_ϵ\delta ^{1+\mathrm{}}\mathrm{ln}[Z_{n,\mathrm{}}(ϵ)]=2n{\displaystyle \frac{Z_{n+1,\mathrm{}}(ϵ)Z_{n1,\mathrm{}}(ϵ)}{Z_{n,\mathrm{}}^2(ϵ)}},`$ (31)
where $`\delta ^{1+\mathrm{}}=ϵ\frac{}{ϵ}+_{r=1}^{\mathrm{}}(\overline{\delta }_r)`$.
Similar to (31) type of equations, but for a finite amount of mass parameters $`\mu `$ were considered earlier in SV3 , SV2 , Akemann , Mironov . The factorization properties of Toda lattice equations in the replica limit have been analyzed in SV3 .
As we see the equation (III.2) differs essentially from the known equation (31), which defines integrable Toda lattice model Toda . Since the equation (III.2) appears for the exactly calculable partition function (3) of the unitary matrix model with potential (1), one can conjecture the existence of corresponding integrable lattice chain model. It is definitely interesting to investigate the interplay between color and flavor degrees of freedom in the equation (III.2) further and analyze its consequences.
## IV Density of States, Mean Conductance and Two Point Correlation Function
In the original articles Muttalib , Ismail where the model with the potential (1) was initiated, the authors, by use of q-Hermite orthogonal polynomials, have analytically calculated the $`N\mathrm{}`$ asymptotic of the two point correlation function. For the appropriately renormalized two point kernel in the bulk of the spectrum exact expression approximately becomes (except very close to the origin)
$$K(\xi ,\eta )\frac{\beta }{2\pi }\frac{\mathrm{sin}[\pi (\xi \eta )]}{\mathrm{sinh}[(\xi \eta )\beta /2]},$$
(32)
which is translational invariant. In its $`\beta 0`$ limit $`K(\xi ,\eta )`$ reproduces the universal behavior of Wigner-Dyson RMT, while for the large $`\beta `$, as authors of Muttalib have argued, the model approaches to the uncorrelated systems with Poisson distribution.
In our approach the calculation of the density of states $`\rho (x)`$ will be reduced to Cauchy integration over poles of the distribution function in the
$`\rho (x)={\displaystyle \frac{1}{Z_{0,N}(ϵ)}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑Htr\delta (xH)P(H)={\displaystyle \frac{N(C_M)^N}{Z_{0,N}(ϵ)}}{\displaystyle \underset{i=2}{\overset{N}{}}dx_i\underset{i,j=1}{\overset{N}{}}(x_ix_j)\underset{k,j=1}{\overset{M,N}{}}\frac{1}{(\mu _k+ix_j)(\mu _kix_j)}},`$ (33)
where the normalization factor $`Z_{0,N}(ϵ)`$ is defined by the expression (3), for the case $`n=0`$. In order to analyze the large $`N`$ limit we will consider $`N=M\mathrm{}`$. In this case the calculation of the Cauchy integrals over poles in our problem essentially simplifies. From (3) one can deduce
$`Z_{0,M}(ϵ)=(2\pi )^M(C_M)^MM!{\displaystyle \underset{k,p=1}{\overset{M}{}}}{\displaystyle \frac{1}{(\mu _{k_i}+\mu _{p_j})}}.`$ (34)
Now, in order to take the Cauchy integral over poles in (33) we close the integration contours at infinity of upper half plane and notice, that because of presence of Vandermonde determinant, the integration variabes $`x_i,i=2,\mathrm{},M`$ will utilize $`M1`$ poles, all different. Only one pole will be left untouched and we will have the summation over it in the result:
$`\rho (x)={\displaystyle \frac{1}{2\pi }}{\displaystyle \underset{k=1}{\overset{M}{}}}{\displaystyle \frac{2\mu _k}{x^2+\mu _k^2}}{\displaystyle \underset{qk}{\overset{M}{}}}{\displaystyle \frac{(xi\mu _q)(\mu _k+\mu _q)}{(x+i\mu _q)(\mu _k\mu _q)}}={\displaystyle \frac{1}{\pi }}{\displaystyle \underset{k=1}{\overset{M}{}}}{\displaystyle \frac{\mu _k}{x^2+\mu _k^2}}.`$ (35)
Constituents of this formula have the following meaning. The factor $`(x^2+\mu _k^2)_{qk}^M(x+i\mu _q)`$ in the denominator comes from the distribution function of the remnant eigenvalue $`x`$, the part of which, namely $`_{qk}^M(xi\mu _q)`$, was cancelled with the similar bracket in the Vandermonde determinant. The factor $`_{qk}^M(xi\mu _q)`$ in the nominator is the remnant from the Vandermonde determinant. The factor $`_{qk}^M\frac{(\mu _k+\mu _q)}{(\mu _k\mu _q)}`$ is left after mutual cancellations between the residues of integrated poles, the Vandermonde determinant and the normalization factor $`Z_{0,N}(ϵ)`$ defined by formula (34). The second part of the formula (35) is an identity.
Hence, we have obtained a simple formula:
$`\rho (x)={\displaystyle \frac{1}{\pi }}{\displaystyle \underset{k=1}{\overset{M}{}}}Im({\displaystyle \frac{1}{xi\mu _k}}).`$ (36)
One will arrive at the same formula (36) taking a replica limit for the one point Green’s function $`G(x)=lim_{n0}n^1Z_{n,M}(x)/x`$ and calculating the DOS.
The model under consideration with potential (1) first was considered in Muttalib , in order to describe the disordered conductors connecting matrices $`H`$ with the ensemble of random transfer matrices $`T`$ as
$`H^2={\displaystyle \frac{1}{4}}(TT^++T^+T2I).`$ (37)
Therefore substituting $`x=\mathrm{sinh}[\beta \chi /2]`$ ($`e^{\beta \chi /2}`$ is the eigenvalue of $`T`$) into the $`\rho (x)`$ one will obtain an exact expression for the density of $`\chi `$:
$`\rho (\chi )={\displaystyle \frac{dx}{d\chi }}\rho (x).`$ (38)
which exhibits crystallization at the integer values $`n`$, as it is presented on Fig.1 for two values of $`\beta `$. This phenomenon for different ensembles was observed earlier in the articles Bogomol , BR , Lama .
The eigenvalues $`x_i`$ of $`H`$ and $`\chi _i`$ are connected with the conductance $`g=tr(tt^+)/2`$ ($`t`$ is the transmission matrix) as
$`g={\displaystyle \underset{i=1}{\overset{N}{}}}{\displaystyle \frac{1}{1+x_i^2}}={\displaystyle \underset{i=1}{\overset{N}{}}}{\displaystyle \frac{1}{\mathrm{cosh}[\beta \chi _i/2]^2}}`$ (39)
and by the same type of Cauchy integration over poles as in $`\rho (x)`$ one can find the average conductance
$`g={\displaystyle \underset{n=1}{\overset{M}{}}}{\displaystyle \frac{1}{1+\mu _n^2}}={\displaystyle \underset{n=1}{\overset{M}{}}}{\displaystyle \frac{1}{1+\mathrm{cosh}[n\beta /2]^2}}.`$ (40)
In the same way one can calculate the two point correlation function $`R_2[x,y]`$, again in the limit $`N=M\mathrm{}`$
$`R_2(x,y)={\displaystyle \frac{1}{Z_{0,N}(ϵ)}}{\displaystyle _{\mathrm{}}^{\mathrm{}}}𝑑Htr\delta (xH)tr\delta (yH)P(H).`$ (41)
By Cauchy integration over poles, in this case, we should now leave untouched two poles of the distribution function $`P(H)`$ (see formula (33)). Therefore, in the result we will have a sum over all possible positions $`m`$ and $`n`$ of that poles. The answer reads:
$`R_2(x,y)={\displaystyle \frac{1}{\pi ^2}}{\displaystyle \underset{m,n}{\overset{M}{}}}{\displaystyle \frac{\mu _m\mu _n}{(x^2+\mu _n^2)(x^2+\mu _m^2)(y^2+\mu _n^2)(y^2+\mu _m^2)}}`$
$`\times {\displaystyle \underset{km,n}{}}{\displaystyle \frac{(xi\mu _k)(\mu _m+\mu _k)(yi\mu _k)(\mu _n+\mu _k)}{(x+i\mu _k)(\mu _m\mu _k)(y+i\mu _k)(\mu _n\mu _k)}}.`$ (42)
Details of the calculation of $`R_2(x,y)`$ repeat the detales of the one for the density of states $`\rho (x)`$, presented above. Since here we hold two eigenvalues $`x`$ and $`y`$ untouched, the formula (IV) is similar (but not equal) to the product of two structures (35) of $`\rho (x)`$ with corresponding adjustment of the summation procedure. Unlike to the case of $`\rho (x)`$, I was not able to find a simplification of the final formula (IV), as it was done in the second part of Eq.(35).
In spite of explicit presence of the imaginary unite $`i`$ in the expression (IV) the resultant $`R_2(x,y)`$ is real. After unfolding, namely after the re-scaling $`x\xi `$, defined by $`d\xi /dx=\rho (x)`$, $`R_2(\xi ,\eta )`$ is pictured in Fig.2 by dots for $`\beta =0.3`$. Moreover, the numerical analyze of the expression (IV) after unfolding shows that $`R_2(\xi ,\eta )`$ in fact is a function of difference $`\xi \eta `$.
Numerical analyses show, that consideration of other potentials, say by taking the product in formula (2) from $`k=0`$ or $`k=2`$, after appropriate unfolding gives us the same behavior as approximately defined by formula (32).
## Conclusions
We give a sigma-model formulation of the RMT with confining potential (1). The replica partition function of the model has been calculated exactly in both, original color space and in flavor space (after a color-flavor transformation) with the same result. In both spaces we have established connections between the RMT Eq.(1) and exactly solvable lattice equations. A generalization of graded TLE is defined and solved.
We have exactly calculated the probability density and the two point correlation function of the model at $`N\mathrm{}`$ and confirmed the known results obtained earlier. The effect of crystallization of transmission eigenvalues has been observed.
## Acknowledgments
I am greatly thankful to V. E. Kravtsov for numerous stimulating discussions and critical reading of the paper. I would like also to acknowledge G. Akemann, E. Bogomolny, B. N. Narozhny, J. J. M. Verbaarschot and especially E. Kanzieper for comments.
## Appendix A. Evaluation of Supersymmetric Itzikson-Zuber Integral over $`SU(nM,M)/SU(nMN,M)`$
In this appendix we present a simple derivation of the supersymmetric extension of the Itzikson-Zuber formula over Stiefel manifold $`SU(nM,M)/SU(nMN,M)`$.
$`(X,Y;r)={\displaystyle _{USU(nM,M)/SU(nMN,M)}}𝑑U\mathrm{exp}\{rStr(XU^1YU)\}`$ (43)
In this calculation one may follow the standard technique IZ developed in Guhr , Alfaro for unitary supergroups based on the solution of the heat kernel equation. In And it was argued, that the heat kernel equation technique can be applied to pseudounitary supergroups. However we will present here the calculations based on the application of Duistermaat-Heckman theorem to integrals over supermanifolds, which states that under certain conditions the quasi-classical saddle-point approximation is exact ASw . This theorem was successfully applied to IZ integral over unitary supergroups $`U(nm)`$ in Szabo and pseudounitary supergroups $`U(nm,k)`$ in Fyo . In what follows we will assume, that this criteria are met by our integral (57) as well and evaluate the integral in the saddle-point approximation. The fact, that final result reproduces the same answer (4) for $`Z_{n,N}(ϵ)`$, as we got by simple calculation of the integral (3) by the Cauchy formula confirms, that this assumption is correct. The same answer one will obtain following the standard technique Guhr , Alfaro .
We need to find the extremum of the function
$`𝒜(U)=Str(XU^1YU),`$ (44)
where $`USU(nM,M)/SU(nMN,M)`$ and $`X`$ and $`Y`$ are diagonal matrices from the expression (10). It is easy to find the saddle-point equation
$$[X,U^1YU]=0$$
(45)
from the condition $`\delta 𝒜(U)=Str\{(XU^1YUU^1YUX)U^1\delta U\}=0`$.
Since we have $`X=diag\{\underset{n}{\underset{}{0\mathrm{}0}},x_1\mathrm{}x_N,\underset{2MN}{\underset{}{0\mathrm{}0}}\}`$ and
$`USU(nM,M)/SU(nMN,M)`$ the only solutions $`U`$ of (45) (up to irrelevant $`USU(nMN,M)`$) are those matrices, which permute the diagonal elements of the matrix $`Y`$ in three- fermionic, positive and negative metric parts of the bosonic sectors separately. In other words the set of solutions of (45) is
$`U_{ab}^{(0)}=\left(\begin{array}{ccc}\overline{}_{\alpha \beta }& 0& 0\\ 0& _{kp}& 0\\ 0& 0& _{k^{}p^{}}^{}\end{array}\right),\begin{array}{c}\alpha ,\beta =1,\mathrm{},n\\ k,p=1,\mathrm{},M\\ k^{},p^{}=1,\mathrm{},M\end{array}`$ (52)
where $`\overline{}_{\alpha \beta }`$ and $`_{kp}(_{k^{}p^{}}^{})`$ are the elements of the permutation groups $`S_n`$ and $`S_M`$ of the fermionic and bosonic sectors respectively. Notice, that the set of permutation elements $`_{ij},i,j=1,\mathrm{}N`$ coincides with the set of maps $`\sigma :\{1,\mathrm{}N\}\{1,\mathrm{}M\}`$ defined earlier.
We now set $`U=U^{(0)}e^{iL}`$ in (44), with an infinitesimal Hermitian supermatrix $`L=U^1\delta Usu(nM,M)su(nMN,M)`$ and expand the function $`𝒜(U)`$ up to the quadratic order in $`L`$. In this factor space $`L`$ has only following nonzero matrix elements $`L_{ij},i,j=1,\mathrm{}N;L_{ik},i=1,\mathrm{}N;k=1,\mathrm{}2MN`$ and $`L_{i\alpha },i=1,\mathrm{}N;\alpha =1,\mathrm{}n`$. In the second order over $`L`$, $`𝒜(U)`$ is
$`𝒜(U)`$ $`=`$ $`StrXU^{(0)}YU^{(0)}+StrXLU^{(0)}YU^{(0)}LStr{\displaystyle \frac{1}{2}}[X,U^{(0)}YU^{(0)}]L^2`$ (53)
$`=`$ $`StrXU^{(0)}YU^{(0)}+{\displaystyle \frac{1}{2}}Str[X,L][U^{(0)}YU^{(0)},L].`$
Due to Duistermaat-Heckman theorem in evaluation of IZ integral (43) we should take the sum over all extrema, therefore substituting (53) to (43) we obtain
$`(X,Y;r)`$ $`=`$ $`{\displaystyle \underset{\sigma S_{2M}}{}}\mathrm{exp}(r{\displaystyle \underset{i=1}{\overset{N}{}}}x_iy_{\sigma (i)}){\displaystyle \underset{i,j=1}{\overset{N}{}}dL_{ij}\underset{i,k=1}{\overset{N,2MN}{}}dL_{ik}\underset{i,\alpha =1}{\overset{N,n}{}}dL_{i\alpha }}`$ (54)
$`\times `$ $`\mathrm{exp}r[{\displaystyle \frac{1}{2}}{\displaystyle \underset{i,j=1}{\overset{N}{}}}L_{ij}^2(x_ix_j)(y_{\sigma (i)}y_{\sigma (j)})`$
$`+`$ $`{\displaystyle \frac{1}{2}}{\displaystyle \underset{i,k=1}{\overset{N,2MN}{}}}L_{ik}^2x_i(y_{\sigma (i)}y_{\overline{\sigma }(k)})`$
$`+`$ $`{\displaystyle \underset{i,\alpha =1}{\overset{N,n}{}}}L_{i\alpha }^2x_i(y_{\sigma (i)}\overline{y}_\alpha )],`$
where $`\overline{y}_\alpha `$ are fermionic components of the diagonal matrix $`Y`$ and $`\overline{\sigma }`$ is the complement to $`\sigma `$ permutation, defined after the formula (10). Here we essentially have used $`diag\{\underset{n}{\underset{}{0\mathrm{}0}},x_1\mathrm{}x_N,\underset{2MN}{\underset{}{0\mathrm{}0}}\}`$ form of the matrix $`X`$.
Evaluating now the Gaussian integrals in (54) over complex bosonic $`L_{ij},ij;L_{i,k},ik`$ and the complex Grassmann variables $`L_{i\alpha }`$, we arrive at
$`(X,Y;r)`$ $`=`$ $`(2\pi )^{2MNN(N1)/2}(r)^{Nn2MN+(N1)N/2}`$ (55)
$`\times `$ $`{\displaystyle \underset{\{\sigma \}}{}}{\displaystyle \frac{det[\mathrm{exp}rx_iy_{\sigma (j)}]}{\mathrm{\Delta }_\sigma (X)\mathrm{\Delta }_\sigma (Y)}}`$
with
$`\mathrm{\Delta }_\sigma (\nu )={\displaystyle \frac{_{i=1}^N_{p=1}^{2MN}(\nu _{\sigma (i)}\nu _{\overline{\sigma }(p)})(\nu _{\sigma (i)}\nu _{\sigma (j)})}{_{\alpha =1}^n_{i=1}^N(\nu _\alpha \nu _{\sigma (i)})}}\nu =X,Y.`$ (56)
## Appendix B. Graded Toda Lattice Equation
In this Appendix we show how the graded Toda lattice equation for the replica partition function $`Z_{n,N}(ϵ)`$ appears. First we would like to demonstrate, that $`Z_{n,N}(ϵ)`$ is a ratio of certain determinants. The expression (3) can be regarded as $`ϵ_\alpha ϵ;\alpha =1,\mathrm{}n`$ limit of n different fermionic eigenvalues. For the convenience and compactness of formulas we will drop from further transformations the constant factor $`(2\pi )^N(C_M)^Ni^{nN}`$ and introduce the notation $`e_\alpha =iϵ_\alpha `$ for a while. Also, instead of fixed $`\mu _1\mathrm{}\mu _M,\mu _1,\mathrm{}\mu _M`$ set of parameters of the model, we consider here general situation with different $`\mu _k,k=1,\mathrm{}2M`$. Condition $`\mu _{k+M}=\mu _k`$ can be easily imposed at the end. We have
$`Z_{n,N}(ϵ)=\underset{\{e_\alpha \}e}{lim}{\displaystyle \underset{\{\sigma \}}{}}{\displaystyle \frac{_{\alpha =1}^n_{i=1}^N(e_\alpha \mu _{\sigma (i)})}{_{i=1}^N_{j=1}^{2MN}(\mu _{\sigma (i)}\mu _{\overline{\sigma }(j)})}}`$ (57)
$`=`$ $`\underset{\{e_\alpha \}e}{lim}{\displaystyle \underset{\{\sigma \}}{}}{\displaystyle \frac{_{\alpha <\beta =1}^n(e_\alpha e_\beta )_{\alpha =1}^n_{i=1}^N(e_\alpha \mu _{\sigma (i)})}{_{\alpha <\beta =1}^n(e_\alpha e_\beta )_{i=1}^N_{j=1}^{2MN}(\mu _{\sigma (i)}\mu _{\overline{\sigma }(j)})}}`$
$`\times `$ $`{\displaystyle \frac{_{i<j=1}^N(\mu _{\sigma (i)}\mu _{\sigma (j)})_{k<p=1}^{2MN}(\mu _{\overline{\sigma }(k)}\mu _{\overline{\sigma }(p)})}{_{i<j=1}^N(\mu _{\sigma (i)}\mu _{\sigma (j)})_{k<p=1}^{2MN}(\mu _{\overline{\sigma }(k)}\mu _{\overline{\sigma }(p)})}},`$
where the nominator and denominator have been multiplied by the same expressions. Introduce now a bosonic Vandermonde determinant:
$`𝒱_s(\{x_1,\mathrm{}x_s\})={\displaystyle \underset{i<j=1}{\overset{s}{}}}(x_ix_j)=Det\left[\begin{array}{cccc}1& x_1& \mathrm{}& x_1^{s1}\\ 1& x_2& \mathrm{}& x_2^{s1}\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\\ 1& x_s& \mathrm{}& x_s^{s1}\end{array}\right]`$ (62)
The expression (57) for the partition function can be rewritten as
$`Z_{n,N}(ϵ)=\underset{\{e_\alpha \}e}{lim}{\displaystyle \underset{\{\sigma \}}{}}(1)^P{\displaystyle \frac{𝒱_{n+N}(\{\theta _1,\mathrm{}\theta _{n+N}\})𝒱_{2MN}(\{\mu _{\overline{\sigma }(1)},\mathrm{}\mu _{\overline{\sigma }(2MN)}\})}{𝒱_n(\{e_1,\mathrm{}e_n\})𝒱_{2M}(\{\mu _1,\mathrm{}\mu _{2M}\})}},`$ (63)
where $`\{\theta _1,\mathrm{}\theta _{n+N}\}=\{e_1,\mathrm{}e_n,\mu _{\sigma (1)},\mathrm{}\mu _{\sigma (N)}\}`$ and $`(1)^P`$ is the parity of permutation
$$P:\{e_1,\mathrm{}e_n,\mu _1,\mathrm{}\mu _{2M}\}\{e_1,\mathrm{}e_n,\mu _{\sigma (1)},\mathrm{}\mu _{\sigma (N)},\mu _{\overline{\sigma }(1)},\mathrm{}\mu _{\overline{\sigma }(2MN)}\}.$$
(64)
The denominator in the expression (63) does not depend on $`\sigma `$ and can be taken out of summation, while $``$ of nominators is nothing but the determinant of the matrix $`A_0`$:
$`A_0=\left(\begin{array}{ccccccccc}1& e_1& e_1^2& \mathrm{}& e_1^{n+N1}& 0& 0& \mathrm{}& 0\\ 1& e_2& e_2^2& \mathrm{}& e_2^{n+N1}& 0& 0& \mathrm{}& 0\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\\ 1& e_n& e_n^2& \mathrm{}& e_n^{n+N1}& 0& 0& \mathrm{}& 0\\ 1& \mu _1& \mu _1^2& \mathrm{}& \mu _1^{n+N1}& 1& \mu _1& \mathrm{}& \mu _1^{2MN1}\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\\ 1& \mu _M& \mu _M^2& \mathrm{}& \mu _M^{n+N1}& 1& \mu _M& \mathrm{}& \mu _M^{2MN1}\\ 0& 0& 0& \mathrm{}& 0& 1& \mu _{M+1}& \mathrm{}& \mu _{M+1}^{2MN1}\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\\ 0& 0& 0& \mathrm{}& 0& 1& \mu _{2M}& \mathrm{}& \mu _{2M}^{2MN1}\end{array}\right),`$ (75)
calculated by $`N+n`$ and $`2MN`$ dimensional minors and by use of the formula (62).
Now let us consider the operator $`\delta _x=x\frac{}{x}`$ introduced earlier and observe that
$$e^x(\delta _x)^pe^x=x^p+𝒫^{(p1)}(x)$$
(76)
where $`𝒫^{(p1)}(x)`$ is the polynomial of $`p1`$ order of $`x`$. From this it follows, that if we replace the monomials $`x^p`$ in the matrix $`A_0`$ with $`e^x(\delta _x)^pe^x`$, we will obtain the matrix $`A`$ defined in (29 ). Their determinants relate as
$`det[A_0]=det[I^1A]`$ (77)
and define the nominators in the expressions (18) and (63) for the partition function $`Z_{n,N}(ϵ)`$. The notation $`I=\mathrm{exp}(_{\alpha =1}^ne_s+_{k=1}^{2M}\mu _k)`$ is defined just after formula (18).
The next step is the consideration of the degenerate limit $`e_\alpha e;\alpha =1,\mathrm{},n`$. We put $`e_1=e`$ and consider Taylor expansions $`f(e_\alpha )=f(e)+(e_\alpha e)_ef(e)+\mathrm{}\frac{(e_\alpha e)^{n1}}{(n1)!}_e^{n1}f(e)+\mathrm{}`$ around $`e`$ of all entities in the upper-left $`n\times n`$ corner of the matrix $`A`$ in (18). It is easy to see that in the limit $`e_\alpha e;\alpha =1,\mathrm{},n`$, higher than $`n1`$ orders of the expansions will not contribute into the matrix $`A`$, while the coefficients $`(e_\alpha e)^k`$ will form a Vandermonde determinant (62), which will cancel the same $`𝒱_n(\{e_1,\mathrm{},e_n\})`$ in the denominator of $`Z_{n,N}(ϵ)`$ in the formula (63).
Therefore we obtain
$`Z_{n,N}(ϵ)={\displaystyle \frac{(2\pi )^N(C_M)^Ni^{nN}}{𝒱_{2M}(\{\mu _1,\mathrm{}\mu _{2M}\})}}I^1det[B_{n,N}],`$ (78)
with $`I=\mathrm{exp}(ne+_r\mu _r)`$ and $`B_{n,N}`$ equal to
$`\left(\begin{array}{ccccccc}I_1& \delta _1I_1& \mathrm{}& \delta _1^{n+N1}I_1& 0& \mathrm{}& 0\\ \frac{1}{1!}_eI_1& \frac{1}{1!}_e\delta _1I_1& \mathrm{}& \frac{1}{1!}_e\delta _1^{n+N1}I_1& 0& \mathrm{}& 0\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\\ \frac{1}{(n1)!}_e^{n1}I_1& \frac{1}{(n1)!}_e^{n1}\delta _1I_1& \mathrm{}& \frac{1}{(n1)!}_e^{n1}\delta _1^{n+N1}I_1& 0& \mathrm{}& 0\\ \overline{I}_1& \overline{\delta }_1\overline{I}_1& \mathrm{}& \overline{\delta }_1^{n+N1}\overline{I}_1& \overline{I}_1& \mathrm{}& \overline{\delta }_1^{2MN1}\overline{I}_1\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\\ \overline{I}_M& \overline{\delta }_M\overline{I}_M& \mathrm{}& \overline{\delta }_M^{n+N1}\overline{I}_M& \overline{I}_M& \mathrm{}& \overline{\delta }_M^{2MN1}\overline{I}_M\\ 0& 0& \mathrm{}& 0& \overline{I}_{M+1}& \mathrm{}& \overline{\delta }_{M+1}^{2MN1}\overline{I}_{M+1}\\ \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}& \mathrm{}\\ 0& 0& \mathrm{}& 0& \overline{I}_{2M}& \mathrm{}& \overline{\delta }_{2M}^{2MN1}\overline{I}_M\end{array}\right).`$ (89)
In this expression $`I_1=\mathrm{exp}(e)`$ and $`\delta _1=e_e`$.
In order to derive the Toda lattice equation (III.2) we will follow the technique presented in the article SV2 in connection with QCD partition function. The method is based on use of the Sylvester identity Syl . For the determinant of any given matrix $`D`$
$$C_{i,j}C_{p,q}C_{i,q}C_{p,j}=det[D]C_{i,j;p,q},$$
(90)
where $`C_{i,j}`$ is the cofactor of the matrix element $`ij`$:
$$C_{i,j}=\frac{det[D]}{D_{i,j}},$$
(91)
and $`C_{i,j;p,q}`$ is the double cofactor of the matrix elements $`ij`$ and $`pq`$:
$$C_{i,j;p,q}=\frac{^2det[D]}{D_{i,j}D_{p,q}}.$$
(92)
Let us consider now the $`(2M+n+1)\times (2M+n+1)`$ dimensional matrices $`B_{n+1,N}`$ and $`B_{n+1,N1}`$ and write Sylvester identities (90) for different set of indices $`i,j;p,q`$ for each of them.
For the $`B_{n+1,N}`$ we take $`i=n+1,j=n+N+1,p=n,q=n+N`$ and obtain
$`C_{n+1,n+N+1}C_{n,n+N}`$ $``$ $`C_{n,n+N+1}C_{n+1,n+N}`$ (93)
$`=`$ $`det[B_{n+1,N}]C_{n+1,n+N+1;n,n+N}.`$
From analogous to (89) structure of $`B_{n+1,N}`$ follows that
$`C_{n+1,n+N+1}=det[B_{n,N}],C_{n,n+N+1}={\displaystyle \frac{1}{n}}_edet[B_{n,N}],`$
$`C_{n+1,n+N+1;n,n+N}=det[B_{n1,N}],`$ (94)
$`C_{n,n+N}={\displaystyle \frac{1}{n}}_edet[B_{n,N}^{}],C_{n+1,n+N}=det[B_{n,N}^{}],`$
where the matrix $`B_{n,N}^{}`$ is defined by acting by the operator $`\delta ^{1+2M}=e\frac{}{e}+_{r=1}^{2M}\mu _r\frac{}{\mu _r}`$ on the $`(n+N)`$-th column of the matrix $`B_{n,N}`$.
Now consider the Sylvester identity for the matrix $`B_{n+1,N1}`$. Setting $`i=n+1,j=n+2M+1,p=n,q=n+2M,`$ we will have
$`C_{n+1,n+2M+1}^{}C_{n,n+2M}^{}`$ $``$ $`C_{n,n+2M+1}^{}C_{n+1,n+2M}^{}`$ (95)
$`=`$ $`det[B_{n+1,N1}]C_{n+1,n+2M+1;n,n+2M}^{}.`$
As in the previous case and from the structure of (89) for $`B_{n+1,N1}`$ it follows that
$`C_{n+1,n+2M+1}^{}=det[B_{n,N}],C_{n,n+2M+1}^{}={\displaystyle \frac{1}{n}}_edet[B_{n,N}],`$
$`C_{n+1,n+2M+1;n,n+N}^{}=det[B_{n1,N+1}],`$ (96)
$`C_{n,n+N}^{}={\displaystyle \frac{1}{n}}_edet[B_{n,N}^{\prime \prime }],C_{n+1,n+N}^{}=det[B_{n,N}^{\prime \prime }],`$
where the matrix $`B_{n,N}^{\prime \prime }`$ is defined by acting by the operator $`\delta ^{1+2M}`$ on the $`(n+2M)`$-th column of the matrix $`B_{n,N}`$.
It is easy to observe, that
$$det[B_{n,N}^{}]+det[B_{n,N}^{\prime \prime }]=\delta ^{1+2M}det[B_{n,N}].$$
(97)
Inserting now expressons (Appendix B. Graded Toda Lattice Equation) and (Appendix B. Graded Toda Lattice Equation) into the Sylvester identities (93) and (95) respectively, taking their sum and using (97) we get
$`det[B_{n,N}]^2{\displaystyle \frac{}{e}}\delta ^{1+2M}\mathrm{ln}[det[B_{n,N}]]`$ $`=`$ (98)
$`=ndet[B_{n+1,N}]det[B_{n1,N}]`$ $`+`$ $`ndet[B_{n+1,N1}]det[B_{n1,N+1}].`$
Finally, using the relation (78) between the partition function and the matrix $`B_{n,N}`$ we obtain the desired graded Toda lattice equation as
$`n+_e\delta ^{1+M}\mathrm{ln}[Z_{n,N}(ϵ)]`$ $`=`$
$`n{\displaystyle \frac{Z_{n+1,N}(ϵ)Z_{n1,N}(ϵ)}{Z_{n,N}^2(ϵ)}}`$ $`+`$ $`n{\displaystyle \frac{Z_{n+1,N1}(ϵ)Z_{n1,N+1}(ϵ)}{Z_{n,N}^2(ϵ)}},`$ (99)
which coincides with (III.2) after substitution $`e=iϵ`$. Notice that the Vandermonde determinants, $`𝒱_{2M}(\{\mu _1,\mathrm{}\mu _{2M}\})`$, do not contribute to the left hand side of the Toda lattice equation since $`\delta ^{1+2M}\mathrm{ln}[𝒱_{2M}(\{\mu _1,\mathrm{}\mu _{2M}\})]`$ is a number, while the contribution of $`I=\mathrm{exp}(ne+_{r=1}^{2M}\mu _r)`$ gives addition of $`n`$.
If we take $`N=M`$, then, since $`B_{n1,N+1}=0`$, the graded Toda equation (Appendix B. Graded Toda Lattice Equation) simplifies.
|
warning/0506/cs0506068.html
|
ar5iv
|
text
|
# Quantum Arthur-Merlin Games
## 1 Introduction
Interactive proof systems and Arthur-Merlin games were introduced by \[GMR89\] and \[Bab85\] (see also \[BM88\]) in order to model the notion of computationally efficient verification. In an interactive proof system, a polynomial-time verifier with a private source of uniformly generated random bits interacts with a computationally unbounded prover in an attempt to check the validity of the claim that a common input string is contained in some prespecified language. Arthur-Merlin games are similar in principle to interactive proof systems, but are somewhat more restricted—the verifier (called Arthur in this setting) no longer has a private source of randomness, but instead has only a public source of randomness that is visible to the prover (called Merlin). Because Arthur is deterministic aside from the bits produced by the random source, one may without loss of generality view that an Arthur-Merlin game is simply an interactive proof system in which the verifier’s messages to the prover consist only of uniformly generated bits from the public random source.
Although Arthur-Merlin games are more restricted than interactive proof systems in the sense just described, the two models are known to be computationally equivalent. In particular, any language having an interactive proof system in which a constant number of messages is exchanged between the prover and verifier also has an Arthur-Merlin game in which precisely two messages are exchanged, the first from Arthur to Merlin and the second from Merlin back to Arthur \[GS89, BM88\]. The complexity class consisting of all such languages is $`\mathrm{AM}`$. Also following from \[GS89\] is the fact that any language having an unrestricted (polynomial-message) interactive proof system also has a polynomial-message Arthur-Merlin game. The complexity class consisting of all such languages was initially called $`\mathrm{IP}`$, but is now known to be equal to $`\mathrm{PSPACE}`$ \[LFKN92, Sha92\].
A third complexity class arising from these models is $`\mathrm{MA}`$, which is the class consisting of all languages having an interactive proof system in which a single message is sent, from the prover to the verifier. One may view the definition of this class as a slight variation on the “guess and check” definition of $`\mathrm{NP}`$, where instead of being deterministic the checking procedure may use randomness. As the usual convention for Arthur-Merlin games is to disallow Arthur the use of the public random source except for the generation of messages, the class $`\mathrm{MA}`$ would typically be described as consisting of all languages having two-message Arthur-Merlin games in which the first message is sent from Merlin to Arthur and the second from Arthur to Merlin. However, given that the information transmitted to Merlin in the second message is irrelevant from the point of view of the game, and may instead be viewed as just a use of the random source and not as a message, it is natural to refer to such games as one-message Arthur-Merlin games.
Quantum computational variants of interactive proof systems have previously been considered in several papers, including the general multiple-message case \[Wat03, KW00, KM03, RW04, GW05\] as well as the single-message case \[AR03, JWB03, KR03, KKR04, KMY03, RS04, Vya03, Wat00\]. As for classical interactive proof systems, quantum interactive proof systems consist of two parties—a prover with unlimited computation power and a computationally bounded verifier. Now, however, the two parties may process and exchange quantum information. The complexity class consisting of all languages having quantum interactive proof systems is denoted $`\mathrm{QIP}`$, and satisfies $`\mathrm{PSPACE}\mathrm{QIP}\mathrm{EXP}`$ \[KW00\]. Here, $`\mathrm{EXP}`$ denotes the class of languages decidable by a deterministic Turing machine running in time $`2^q`$ for some polynomial $`q`$.
There are both similarities and some apparent differences in the properties of quantum and classical interactive proof systems. Perhaps the most significant difference is that any language having an unrestricted (polynomial-message) quantum interactive proof system also has a three-message quantum interactive proof system \[KW00\]. This cannot happen classically unless $`\mathrm{AM}=\mathrm{PSPACE}`$.
This paper investigates various aspects of quantum Arthur-Merlin games. In analogy to the classical case, we define quantum Arthur-Merlin games to be restricted forms of quantum interactive proof systems in which the verifier’s (Arthur’s) messages to the prover (Merlin) are uniformly generated random bits, as opposed to arbitrary messages. Consequently, Arthur is not capable of sending quantum information to Merlin at any point during a quantum Arthur-Merlin game. Similar to the classical case, quantum Arthur-Merlin games give rise to complexity classes depending on the number of messages exchanged between Arthur and Merlin. In particular, we obtain three primary complexity classes corresponding to Arthur-Merlin games with one message, two messages, and three or more messages.
In the one-message case, Merlin sends a single message to Arthur, who checks it and makes a decision to accept or reject the input. The corresponding complexity class is denoted $`\mathrm{QMA}`$, and has been considered previously in the papers cited above. In this situation Merlin’s message to Arthur may simply be viewed as a quantum witness or certificate that Arthur checks in polynomial time with a quantum computer. To our knowledge, the idea of a quantum state playing the role of a certificate in this sense was first proposed by \[Kni96\], and the idea was later studied in greater depth by \[Kit99\]. Kitaev proved various fundamental properties of $`\mathrm{QMA}`$, which are described in \[KSV02\] and \[AN02\].
One of the facts that Kitaev proved was that the completeness and soundness errors in a $`\mathrm{QMA}`$ protocol may be efficiently reduced by parallel repetition. Because quantum information cannot be copied, however, and Arthur’s verification procedure is potentially destructive to Merlin’s message, Arthur requires multiple copies of Merlin’s message for this method to work. This method therefore requires a polynomial increase in the length of Merlin’s message to Arthur in order to achieve exponentially decreasing error. In this paper, we prove that this increase in the length of Merlin’s message is not required after all—using a different error reduction method, an exponential reduction in error is possible with no increase whatsoever in the length of Merlin’s message to Arthur.
It is known that $`\mathrm{QMA}`$ is contained in the class $`\mathrm{PP}`$, which can be proved using the $`\mathrm{GapP}`$-based method of \[FR99\] together with some simple facts from matrix analysis. This fact was noted without proof in \[KW00\]. A proof of this fact was, however, given by \[Vya03\], who in fact strengthened this result to show that $`\mathrm{QMA}`$ is contained in a subclass $`\mathrm{A}_0\mathrm{PP}`$ of $`\mathrm{PP}`$. (Definitions of the classes $`\mathrm{PP}`$ and $`\mathrm{A}_0\mathrm{PP}`$ can be found in 2 of this paper.) Based on our new error reduction method, we give a simplified proof of this containment. We also use our error reduction method to prove that one-message quantum Arthur-Merlin games in which Merlin’s message has logarithmic length give no increase in power over $`\mathrm{BQP}`$.
In the two-message case, Arthur flips some number of fair coins, sends the results of those coin-flips to Merlin, and Merlin responds with some quantum state. Arthur performs a polynomial-time quantum computation on the random bits together with Merlin’s response, which determines whether Arthur accepts or rejects. The corresponding complexity class will be denoted $`\mathrm{QAM}`$. Two facts about $`\mathrm{QAM}`$ are proved in this paper. The first is the very basic fact that parallel repetition reduces error exactly as in the classical case. (This fact does not follow from known facts about quantum interactive proof systems, as parallel repetition is only known to reduce error for general quantum interactive proof systems having perfect completeness.) The second fact is that $`\mathrm{QAM}`$ is contained in $`\mathrm{BP}\mathrm{PP}`$, the class obtained by applying the $`\mathrm{BP}`$ operator to the class $`\mathrm{PP}`$.
Finally, in the three-message case, Merlin sends Arthur a message consisting of some number of qubits, Arthur flips some number of fair coins and sends the results to Merlin, and then Merlin responds with a second collection of qubits. Arthur performs a polynomial-time quantum computation on all of the qubits sent by Merlin together with the values of his own coin-flips, and decides whether to accept or reject. The corresponding complexity class will be denoted $`\mathrm{QMAM}`$. It is proved that any language having an ordinary quantum interactive proof system is contained in $`\mathrm{QMAM}`$, implying $`\mathrm{QMAM}=\mathrm{QIP}`$.
In spirit, the equality $`\mathrm{QMAM}=\mathrm{QIP}`$ resembles the theorem of \[GS89\] establishing that classical Arthur-Merlin games and interactive proof systems are equivalent in power. However, there is no similarity in the proofs of these facts. Moreover, our result is stronger than what is likely to hold classically. Specifically, we prove that any language having a quantum interactive proof system also has a three-message quantum Arthur-Merlin game in which Arthur’s only message to Merlin consists of just a single coin-flip (in order to achieve perfect completeness and soundness error exponentially close to 1/2). This is impossible classically unless interaction is useless in classical interactive proof systems; for if Arthur flips only one coin, Merlin may as well send his first message and the two possible second messages to Arthur in a single message. The reason why this strategy fails in the quantum case is that Merlin’s first and second messages may need to be entangled in order to be convincing to Arthur, but it may not possible for Merlin to simultaneously entangle his two possible second messages with the first in a way that convinces Arthur to accept. This is an example of the principle that Bennett refers to as the “monogamy of entanglement” (see, for example, \[Ter04\]); the more a given system is entangled with a second system, the less it can be entangled with a third.
### Organization of the paper
The remainder of this paper is organized as follows. We begin with 2, which discusses background information needed elsewhere in the paper, including a summary of basic notation and conventions that are used, definitions of some relevant counting complexity classes, and background on quantum computation and quantum interactive proof systems. The next three sections correspond to the three complexity classes $`\mathrm{QMA}`$, $`\mathrm{QAM}`$, and $`\mathrm{QMAM}`$, respectively; 3 discusses one-message quantum Arthur-Merlin games, 4 discusses the two-message case, and 5 discusses the case of three or more messages. The paper concludes with 6, which mentions some open problems relating to quantum Arthur-Merlin games.
## 2 Background Information
This section summarizes various background information that is needed for the remainder of the paper, including information on quantum computation, counting complexity, and quantum interactive proof systems.
We begin with some remarks about notation and other simple conventions that are followed throughout. All strings and languages in this paper will be over the alphabet $`\mathrm{\Sigma }=\{0,1\}`$. We denote by $`\mathrm{𝑝𝑜𝑙𝑦}`$ the set of all functions $`f:\backslash \{0\}`$ (where $`=\{0,1,2,\mathrm{}\}`$) for which there exists a polynomial-time deterministic Turing machine that outputs $`1^{f(n)}`$ on input $`1^n`$. For every integer $`k2`$, we fix a polynomial-time computable function that, for every choice of $`x_1,\mathrm{},x_k\mathrm{\Sigma }^{}`$, encodes the $`k`$-tuple $`(x_1,\mathrm{},x_k)`$ as a single element of $`\mathrm{\Sigma }^{}`$. These functions are assumed to satisfy the usual properties of tuple-functions, namely that they are one-to-one and polynomial-time invertible in each argument. As is typical, reference to these functions is often implicit; for instance, we write $`f(x_1,\mathrm{},x_k)`$ as shorthand for $`f((x_1,\mathrm{},x_k))`$ when $`x_1,\mathrm{},x_k\mathrm{\Sigma }^{}`$ and the domain of the function $`f`$ is understood to be $`\mathrm{\Sigma }^{}`$.
### Quantum computation
We will assume that the reader has familiarity with the mathematics of quantum information, which is discussed in the books of \[KSV02\] and \[NC00\]. The quantum complexity classes discussed in this paper are based on the quantum circuit model, with which we also assume familiarity.
All quantum circuits considered in this paper will be assumed to be composed only of Toffoli gates, Hadamard gates, and $`i`$-shift gates (which induce the mapping $`|0|0`$, $`|1i|1`$). This is a universal set of gates \[Kit97\], so there is no loss of generality in restricting our attention to this set. We assume that a reasonable encoding scheme has been fixed that allows quantum circuits to be encoded as binary strings having length at least the size of the encoded circuit and at most some fixed polynomial in the circuit’s size.
A collection $`\{A_x:x\mathrm{\Sigma }^{}\}`$ of quantum circuits is said to be generated in polynomial-time if there exists a polynomial-time deterministic Turing machine that, on input $`x\mathrm{\Sigma }^{}`$, outputs an encoding of the circuit $`A_x`$. When such a family is parameterized by tuples of strings, it is to be understood that we are implicitly referring to one of the tuple-functions discussed previously. For instance, we will consider families of the form $`\{A_{x,y}:x,y\mathrm{\Sigma }^{}\}`$ when two- and three-message quantum Arthur-Merlin games are discussed.
The notion of a polynomial-time generated family is similar to the usual notion of a polynomial-time uniform family of circuits, except that it allows the procedure generating the circuits to have access to the input $`x`$ rather than just the length of $`x`$ written in unary. In essence, the input $`x`$ may be “hard-coded” into a given circuit in a polynomial-time generated family, so that it is not necessary to assume that the input $`x`$ is input to the circuit itself. This is simply done as a matter of convenience and simplicity—all of the polynomial-time generated families of quantum circuits in this paper could be replaced by polynomial-time uniform families where the string given to the generating procedure is instead input directly into the circuit.
Let us illustrate the use of polynomial-time generated families of quantum circuits by defining $`\mathrm{BQP}`$, the class of languages recognizable in quantum polynomial time with bounded error. A language $`L`$ is in $`\mathrm{BQP}`$ if and only if there exists a polynomial-time generated family $`\{A_x\}`$ of quantum circuits such that the following conditions hold. First, it is required that there exist a function $`k\mathrm{𝑝𝑜𝑙𝑦}`$ such that each circuit $`A_x`$ act on precisely $`k(|x|)`$ qubits. (This condition is not really necessary, but will simplify further discussions.) Let $`\mathrm{\Pi }_1=|11|I_{k1}`$, where $`k`$ is shorthand for $`k(|x|)`$ and, in general, $`I_n`$ denotes the identity operator acting on $`n`$ qubits. Then it is required that
* if $`xL`$ then $`\mathrm{\Pi }_1A_x|0^k^2\frac{2}{3}`$, and
* if $`xL`$ then $`\mathrm{\Pi }_1A_x|0^k^2\frac{1}{3}`$.
In words, if the input is $`x`$, then the circuit $`A_x`$ is run on the all-zero input and the first qubit is measured in the standard basis. If the measurement result is 1, the computation is viewed as accepting, otherwise it is rejecting. The usual notion of bounded error is required.
It will sometimes be helpful when describing certain quantum Arthur-Merlin games to refer to quantum registers. These are simply collections of qubits to which we assign some name. When we refer to the reduced state of a given register, we mean the mixed state obtained by tracing out all other registers beside the one to which we are referring.
### Counting classes
Some of the results in this paper involve relations between complexity classes based on quantum Arthur-Merlin games and classes based on the notion of counting complexity. Here we briefly discuss this notion and the classes relevant to this paper; for more information about counting complexity, see \[For97\].
A function $`f:\mathrm{\Sigma }^{}`$ is an element of the function class $`\mathrm{\#}\mathrm{P}`$ if and only if there exists a polynomial-time nondeterministic Turing machine that, on each input $`x\mathrm{\Sigma }^{}`$, has precisely $`f(x)`$ accepting computation paths. For any function $`f\mathrm{\#}\mathrm{P}`$ there exists a function $`q\mathrm{𝑝𝑜𝑙𝑦}`$ such that $`f(x)2^{q(|x|)}`$ for all $`x\mathrm{\Sigma }^{}`$.
A function $`f:\mathrm{\Sigma }^{}`$ is an element of the function class $`\mathrm{FP}`$ if it is computable in polynomial time, with the understanding that the output of the function is the integer represented in binary notation by the output of the computation.
A function $`f:\mathrm{\Sigma }^{}`$ is an element of the function class $`\mathrm{GapP}`$ if and only if there exist functions $`g,h\mathrm{\#}\mathrm{P}`$ such that $`f(x)=g(x)h(x)`$ for all $`x\mathrm{\Sigma }^{}`$. The function class $`\mathrm{GapP}`$ possesses remarkable closure properties, including closure under subtraction, exponential sums, and polynomial products. In particular, if $`f\mathrm{GapP}`$ and $`q\mathrm{𝑝𝑜𝑙𝑦}`$, then the functions $`g`$ and $`h`$ defined as
$$g(x)=\underset{i=1}{\overset{2^{q(|x|)}}{}}f(x,i),h(x)=\underset{i=1}{\overset{q(|x|)}{}}f(x,i)$$
are elements of $`\mathrm{GapP}`$. (Here the integer $`i`$ is identified with the string having no leading zeroes that encodes it in binary notation.) It is not difficult to show that $`\mathrm{FP}\mathrm{GapP}`$.
The complexity class $`\mathrm{PP}`$ consists of all languages $`L\mathrm{\Sigma }^{}`$ for which there exists a function $`f\mathrm{GapP}`$ such that $`xL`$ if and only if $`f(x)>0`$ for all $`x\mathrm{\Sigma }^{}`$. The class $`\mathrm{A}_0\mathrm{PP}`$ consists of all languages $`L\mathrm{\Sigma }^{}`$ for which there exist functions $`f\mathrm{GapP}`$ and $`g\mathrm{FP}`$ satisfying
$$xLf(x)g(x),xL\mathrm{\hspace{0.33em}0}f(x)\frac{g(x)}{2},$$
for all $`x\mathrm{\Sigma }^{}`$. Finally, the complexity class $`\mathrm{BP}\mathrm{PP}`$ refers to the $`\mathrm{BP}`$ operator applied to the class $`\mathrm{PP}`$; it contains all languages $`L\mathrm{\Sigma }^{}`$ such that there exists a language $`A\mathrm{PP}`$ and a function $`q\mathrm{𝑝𝑜𝑙𝑦}`$ such that
$$\left|\{y\mathrm{\Sigma }^{q(|x|)}:(x,y)AxL\}\right|\frac{2}{3}\mathrm{\hspace{0.17em}2}^{q(|x|)}.$$
Counting complexity and quantum complexity were related by \[FR99\], who gave a simple proof that $`\mathrm{BQP}\mathrm{PP}`$ based on the closure properties of $`\mathrm{GapP}`$ functions discussed above. (The containment $`\mathrm{BQP}\mathrm{PP}`$ had been proved earlier by \[ADH97\] using a different method.) In fact, Fortnow & Rogers proved the stronger containment $`\mathrm{BQP}\mathrm{AWPP}`$, where $`\mathrm{AWPP}`$ is a subclass of $`\mathrm{PP}`$ that we will not define in this paper. As a couple of the facts we prove are based on the method of Fortnow & Rogers, it will be helpful for us to summarize this method. The quantum Turing machine model was used in the original proof, but our summary is instead based on polynomial-time generated families of quantum circuits.
Suppose that $`L\mathrm{BQP}`$, which implies the existence of a polynomial-time generated family $`\{A_x\}`$ of quantum circuits satisfying the conditions of the definition of $`\mathrm{BQP}`$ discussed previously. The goal is to construct a $`\mathrm{GapP}`$ function $`f`$ and a polynomially bounded $`\mathrm{FP}`$ function $`g`$ such that
$$\frac{f(x)}{2^{g(x)}}=0^k|A_x^{}\mathrm{\Pi }_1A_x|0^k=\mathrm{\Pi }_1A_x|0^k^2.$$
Once this is done, the $`\mathrm{GapP}`$ function $`h(x)=2f(x)2^{g(x)}`$ satisfies the required property to establish $`L\mathrm{PP}`$; namely that $`h(x)>0`$ if and only if $`xL`$.
The functions $`f`$ and $`g`$ are of course based on the circuit family $`\{A_x\}`$. For a given string $`x`$, assume that the circuit $`A_x`$ consists of gates $`G_1,\mathrm{},G_{q(|x|)}`$ for some function $`q\mathrm{𝑝𝑜𝑙𝑦}`$. Each of the gates $`G_j`$, when tensored with the identity operator on the qubits not affected by $`G_j`$, gives rise to a $`2^k\times 2^k`$ matrix whose individual entries, indexed by pairs of strings of length $`k`$, can be computed in polynomial time given $`x`$. These entries are elements of the set
$$\{0,1,i,1/\sqrt{2},1/\sqrt{2}\}$$
because we assume $`A_x`$ is composed only of Toffoli, Hadamard, and $`i`$-shift gates. Similarly, $`\mathrm{\Pi }_1`$ is a $`2^k\times 2^k`$ matrix whose entries (this time restricted to the set $`\{0,1\}`$) are also computable in polynomial time given $`x`$.
The value $`0^k|A_x^{}\mathrm{\Pi }_1A_x|0^k`$ therefore corresponds to the $`(0^k,0^k)`$ entry of the matrix product
$$G_1^{}\mathrm{}G_q^{}\mathrm{\Pi }_1G_q\mathrm{}G_1,$$
which can be expressed as an exponential sum of a polynomial product of the entries of these matrices. By letting the function $`g`$ represents the total number of Hadamard transforms in the circuit $`A_x`$, it is fairly straightforward to construct an appropriate $`\mathrm{GapP}`$ function $`f`$ based on closure properties of the class $`\mathrm{GapP}`$. Further details can be found in \[FR99\] as well as in \[Vya03\].
### Quantum interactive proofs
Here we discuss background information on quantum interactive proof systems that will be used later in the paper when it is proved that quantum Arthur-Merlin games have the same power as arbitrary quantum interactive proof systems. It will only be necessary for us to discuss the particular case of three-message quantum interactive proof systems, as any polynomial-message quantum interactive proof system can be simulated by a three-message quantum interactive proof. Moreover, such a proof system may be taken to have perfect completeness and exponentially small soundness error. These facts are proved in \[KW00\], to which the reader is referred for a more complete discussion of quantum interactive proof systems.
For a fixed input $`x`$, a three-message quantum interactive proof system operates as follows. The verifier begins with a $`k`$-qubit register $`𝖵`$ and the prover begins with two registers: an $`m`$-qubit register $`𝖬`$ and an $`l`$-qubit register $`𝖯`$. The register $`𝖵`$ corresponds to the verifier’s work-space, the register $`𝖬`$ corresponds to the message qubits that are sent back and forth between the prover and verifier, and the register $`𝖯`$ corresponds to the prover’s workspace. The register $`𝖬`$ begins in the prover’s possession because the prover sends the first message. The verifier’s work-space register $`𝖵`$ begins initialized to the state $`|0^k`$, while the prover initializes the pair $`(𝖬,𝖯)`$ to some arbitrary quantum state $`|\psi `$.
In the first message, the prover sends $`𝖬`$ to the verifier. The verifier applies some unitary transformation $`V_1`$ to the pair $`(𝖵,𝖬)`$ and returns $`𝖬`$ to the prover in the second message. The prover now applies some arbitrary unitary transformation $`U`$ to the pair $`(𝖬,𝖯)`$ and returns $`𝖬`$ to the verifier in the third and final message. Finally, the verifier applies a second unitary transformation $`V_2`$ to the pair $`(𝖵,𝖬)`$ and measures the first qubit of the resulting collection of qubits in the standard basis. The outcome 1 is interpreted as “accept” and 0 is interpreted as “reject”.
Let $`\mathrm{\Pi }_0`$, $`\mathrm{\Pi }_1`$, $`\mathrm{\Delta }_0`$, and $`\mathrm{\Delta }_1`$ be projections defined as
$$\mathrm{\Pi }_1=|11|I_{k+m1},\mathrm{\Delta }_1=|0^k0^k|I_m,\mathrm{\Pi }_0=|00|I_{k+m1},\mathrm{\Delta }_0=I_{k+m}\mathrm{\Delta }_1.$$
In other words, these are $`k+m`$ qubit projections that act on the pair of registers $`(𝖵,𝖬)`$; $`\mathrm{\Pi }_1`$ and $`\mathrm{\Pi }_0`$ are projections onto those states for which the first qubit of the register $`𝖵`$ is 1 or 0, respectively, and $`\mathrm{\Delta }_1`$ and $`\mathrm{\Delta }_0`$ are projections onto those states for which the register $`𝖵`$ contains the state $`|0^k`$ or contains a state orthogonal to $`|0^k`$, respectively.
The maximum probability with which a verifier specified by $`V_1`$ and $`V_2`$ can be made to accept is
$$(\mathrm{\Pi }_1V_2I_l)(I_kU)(V_1I_l)(|0^k|\psi )^2,$$
(1)
maximized over all choices of the state $`|\psi `$ and the unitary transformation $`U`$. The number $`l`$ is determined by the prover’s strategy, so one may maximize over this number as well. However, there is no loss of generality in assuming $`l=m+k`$; with this many work qubits, the prover may store a purification of the reduced state of the pair $`(𝖵,𝖬)`$, which is sufficient for an optimal strategy.
There is another way to characterize the maximum acceptance probability for a given verifier based on the fidelity function
$$F(\rho ,\xi )=\mathrm{tr}\sqrt{\sqrt{\rho }\xi \sqrt{\rho }}.$$
To describe this characterization we will need to define various sets of states of the pair of registers $`(𝖵,𝖬)`$. For any projection $`\mathrm{\Lambda }`$ on $`k+m`$ qubits let $`𝒮(\mathrm{\Lambda })`$ denote the set of all mixed states $`\rho `$ of $`(𝖵,𝖬)`$ that satisfy $`\rho =\mathrm{\Lambda }\rho \mathrm{\Lambda }`$, i.e., the collection of states whose support is contained in the space onto which $`\mathrm{\Lambda }`$ projects. Also let $`𝒮_𝖵(\mathrm{\Lambda })`$ denote the set of all reduced states of $`𝖵`$ that result from some state $`\rho 𝒮(\mathrm{\Lambda })`$, i.e.,
$$𝒮_𝖵(\mathrm{\Lambda })=\{\mathrm{tr}_𝖬\rho :\rho 𝒮(\mathrm{\Lambda })\},$$
where $`\mathrm{tr}_𝖬`$ denotes the partial trace over the register $`𝖬`$.
###### Proposition 2.1.
The maximum probability with which a verifier specified by $`V_1`$ and $`V_2`$ can be made to accept is
$$\mathrm{max}\left\{F(\rho ,\xi )^2:\rho 𝒮_𝖵(V_1\mathrm{\Delta }_1V_1^{}),\xi 𝒮_𝖵(V_2^{}\mathrm{\Pi }_1V_2)\right\}.$$
This proposition is essentially a restatement based on Uhlmann’s Theorem (see \[NC00\]), of the fact that the quantity 1 above represents the maximum acceptance probability of the verifier described by $`V_1`$ and $`V_2`$. This equivalence is discussed further in \[KW00\].
## 3 QMA
A $`\mathrm{QMA}`$ verification procedure $`A`$ is a family of quantum circuits $`\{A_x:x\mathrm{\Sigma }^{}\}`$ that is generated in polynomial time, together with a function $`m\mathrm{𝑝𝑜𝑙𝑦}`$. The function $`m`$ specifies the length of Merlin’s message to Arthur, and it is assumed that each circuit $`A_x`$ acts on $`m(|x|)+k(|x|)`$ qubits for some function $`k`$ specifying the number of work qubits used by the circuit. As we have done in the previous section, when the input $`x`$ has been fixed or is implicit we will generally write $`m`$ to mean $`m(|x|)`$, $`k`$ to mean $`k(|x|)`$, and so forth, in order to simplify our notation. When we want to emphasize the length of Merlin’s message, we will refer to $`A`$ as an $`m`$-qubit $`\mathrm{QMA}`$ verification procedure.
Consider the following process for a string $`x\mathrm{\Sigma }^{}`$ and a quantum state $`|\psi `$ on $`m`$ qubits:
* Run the circuit $`A_x`$ on the input state $`|\psi |0^k`$.
* Measure the first qubit of the resulting state in the standard basis, interpreting the outcome 1 as accept and the outcome 0 as reject.
The probability associated with the two possible outcomes will be referred to as $`\mathrm{Pr}[A_x\text{ accepts }|\psi ]`$ and $`\mathrm{Pr}[A_x\text{ rejects }|\psi ]`$ accordingly.
###### Definition 3.1.
The class $`\mathrm{QMA}(a,b)`$ consists of all languages $`L\mathrm{\Sigma }^{}`$ for which there exists a $`\mathrm{QMA}`$ verification procedure $`A`$ for which the following holds:
* For all $`xL`$ there exists an $`m`$ qubit quantum state $`|\psi `$ such that $`\mathrm{Pr}[A_x\text{ accepts }|\psi ]a`$.
* For all $`xL`$ and all $`m`$ qubit quantum states $`|\psi `$, $`\mathrm{Pr}[A_x\text{ accepts }|\psi ]b`$.
For any $`m\mathrm{𝑝𝑜𝑙𝑦}`$, the class $`\mathrm{QMA}_m(a,b)`$ consists of all languages $`L\mathrm{\Sigma }^{}`$ for which there exists an $`m`$-qubit $`\mathrm{QMA}`$ verification procedure that satisfies the above properties.
One may consider the cases where $`a`$ and $`b`$ are constants or functions of the input length $`n=|x|`$ in this definition. If $`a`$ and $`b`$ are functions of the input length, it is assumed that $`a(n)`$ and $`b(n)`$ can be computed deterministically in time polynomial in $`n`$. When no reference is made to the probabilities $`a`$ and $`b`$, it is assumed $`a=2/3`$ and $`b=1/3`$.
### Strong Error Reduction
It is known that $`\mathrm{QMA}`$ is robust with respect to error bounds in the following sense.
###### Theorem 3.2 (Kitaev).
Let $`a,b:[0,1]`$ and $`q\mathrm{𝑝𝑜𝑙𝑦}`$ satisfy
$$a(n)b(n)\frac{1}{q(n)}$$
for all $`n`$. Then $`\mathrm{QMA}(a,b)\mathrm{QMA}(12^r,2^r)`$ for every $`r\mathrm{𝑝𝑜𝑙𝑦}`$.
A proof of this theorem appears in Section 14.2 of \[KSV02\]. The idea of the proof is as follows. If we have a verification procedure $`A`$ with completeness and soundness probabilities given by $`a`$ and $`b`$, we construct a new verification procedure that independently runs $`A`$ on some sufficiently large number of copies of the original certificate and accepts if the number of acceptances of $`A`$ is larger than $`(a+b)/2`$. The only difficulty in proving that this construction works lies in the fact that the new certificate cannot be assumed to consist of several copies of the original certificate, but may be an arbitrary (possibly highly entangled) quantum state. Intuitively, however, entanglement cannot help Merlin to cheat; under the assumption that $`xL`$, the probability of acceptance for any particular execution of $`A`$ is bounded above by $`b`$, and this is true regardless of whether one conditions on the outcomes of any of the other executions of $`A`$. This construction requires an increase in the length of Merlin’s message to Arthur in order to reduce error.
The main result of this section is the following theorem, which states that one may decrease error without any increase in the length of Merlin’s message.
###### Theorem 3.3.
Let $`a,b:[0,1]`$ and $`q\mathrm{𝑝𝑜𝑙𝑦}`$ satisfy
$$a(n)b(n)\frac{1}{q(n)}$$
for all $`n`$. Then $`\mathrm{QMA}_m(a,b)\mathrm{QMA}_m(12^r,2^r)`$ for every $`m,r\mathrm{𝑝𝑜𝑙𝑦}`$.
###### Proof.
Assume $`L\mathrm{QMA}_m(a,b)`$, and $`A`$ is an $`m`$-qubit $`\mathrm{QMA}`$ verification procedure that witnesses this fact. We will describe a new $`m`$-qubit $`\mathrm{QMA}`$ verification procedure $`B`$ with exponentially small completeness and soundness error for the language $`L`$, which will suffice to prove the theorem.
It will simplify matters to assume hereafter that the input $`x`$ is fixed—it will be clear that the new verification procedure can be generated in polynomial-time. As the input $`x`$ is fixed, we will write $`A`$ and $`B`$ to denote $`A_x`$ and $`B_x`$, respectively.
It will be helpful to refer to the $`m`$ message qubits along with the $`k`$ work-space qubits of $`A`$ as a single $`m+k`$ qubit quantum register $`𝖱`$. Define projections acting on the vector space corresponding to $`𝖱`$ as follows:
$$\mathrm{\Pi }_1=|11|I_{m+k1},\mathrm{\Delta }_1=I_m|0^k0^k|,\mathrm{\Pi }_0=|00|I_{m+k1},\mathrm{\Delta }_0=I_{m+k}\mathrm{\Delta }_1.$$
(2)
The measurement described by $`\{\mathrm{\Pi }_0,\mathrm{\Pi }_1\}`$ is just a measurement of the first qubit of $`𝖱`$ in the computational basis; this measurement determines whether Arthur accepts or rejects after the circuit $`A`$ is applied. The measurement described by $`\{\mathrm{\Delta }_0,\mathrm{\Delta }_1\}`$ gives outcome 1 if the last $`k`$ qubits of $`𝖱`$, which correspond to Arthur’s work-space qubits, are set to their initial all-zero state, and gives outcome 0 otherwise. (These projections are similar to those in 2 except that the message qubits and Arthur’s work qubits are reversed for notational convenience.)
The procedure $`B`$ operates as follows. It assumes that initially the first $`m`$ qubits of $`𝖱`$ contain Merlin’s message $`|\psi `$ and the remaining $`k`$ qubits are set to the state $`|0^k`$.
* Set $`y_01`$ and $`i1`$.
* Repeat:
+ Apply $`A`$ to $`𝖱`$ and measure $`𝖱`$ with respect to the measurement described by $`\{\mathrm{\Pi }_0,\mathrm{\Pi }_1\}`$. Let $`y_i`$ denote the outcome, and set $`ii+1`$.
+ Apply $`A^{}`$ to $`𝖱`$ and measure $`𝖱`$ with respect to the measurement described by $`\{\mathrm{\Delta }_0,\mathrm{\Delta }_1\}`$. Let $`y_i`$ denote the outcome, and set $`ii+1`$.
Until $`iN`$, where $`N=8q^2r`$.
* For each $`i=1,\mathrm{},N`$ set
$$z_i\{\begin{array}{cc}1\hfill & \text{if }y_i=y_{i1}\hfill \\ 0\hfill & \text{if }y_iy_{i1}\text{.}\hfill \end{array}$$
Accept if $`_{i=1}^Nz_iN\frac{a+b}{2}`$ and reject otherwise.
Although the description of this procedure refers to various measurements, it is possible to simulate these measurements with unitary gates in the standard way, which allows the entire procedure to be implemented by a unitary quantum circuit. 1 illustrates a quantum circuit implementing this procedure for the case $`N=5`$.
In this figure, $`S`$ represents the computation described in the last step of $`B`$, and the last qubit rather than the first represents the output qubit to simplify the figure.
We first consider the behavior of the verification procedure $`B`$ in the situation that the state $`|\psi `$ is an eigenvector of the operator
$$Q=(I_m0^k|)A^{}\mathrm{\Pi }_1A(I_m|0^k),$$
with corresponding eigenvalue $`p`$. We have
$$p=\psi |Q|\psi =\mathrm{\Pi }_1A(|\psi |0^k)^2,$$
and thus $`p`$ is the probability that the verification procedure $`A`$ accepts $`|\psi `$. Let $`|\varphi =|\psi |0^k`$, which implies that $`|\varphi `$ is an eigenvector of $`\mathrm{\Delta }_1A^{}\mathrm{\Pi }_1A\mathrm{\Delta }_1`$, also having corresponding eigenvalue $`p`$. We will show that the verification procedure $`B`$ accepts $`|\psi `$ with probability
$$\underset{N\frac{a+b}{2}jN}{}\left(\genfrac{}{}{0pt}{}{N}{j}\right)p^j(1p)^{Nj}.$$
(3)
Using standard Chernoff-type bounds, this probability can be shown to be greater than $`12^r`$ when $`pa`$ and less than $`2^r`$ when $`pb`$, given the choice of $`N=8q^2r`$.
The fact that $`|\psi `$ is accepted with the probability given in equation 3 will follow from the fact that the procedure $`B`$ obtains each possible sequence $`(z_1,\mathrm{},z_N)`$ with probability $`p^{w(z)}(1p)^{Nw(z)}`$ for $`w(z)=_{i=1}^Nz_i`$. This is straightforward if $`p=0`$ or $`p=1`$, so assume $`0<p<1`$.
Define vectors $`|\gamma _0`$, $`|\gamma _1`$, $`|\delta _0`$, and $`|\delta _1`$ as follows:
$$|\gamma _0=\frac{\mathrm{\Pi }_0A\mathrm{\Delta }_1|\varphi }{\sqrt{1p}},|\gamma _1=\frac{\mathrm{\Pi }_1A\mathrm{\Delta }_1|\varphi }{\sqrt{p}},|\delta _0=\frac{\mathrm{\Delta }_0A^{}\mathrm{\Pi }_1|\gamma _1}{\sqrt{1p}},|\delta _1=\frac{\mathrm{\Delta }_1A^{}\mathrm{\Pi }_1|\gamma _1}{\sqrt{p}}.$$
As $`\mathrm{\Delta }_1A^{}\mathrm{\Pi }_1A\mathrm{\Delta }_1|\varphi =p|\varphi `$ and $`|\varphi `$ is a unit vector we have
$`\varphi |\mathrm{\Delta }_1A^{}\mathrm{\Pi }_1A\mathrm{\Delta }_1|\varphi `$ $`=p,`$
$`\varphi |\mathrm{\Delta }_1A^{}\mathrm{\Pi }_0A\mathrm{\Delta }_1|\varphi `$ $`=\varphi |\mathrm{\Delta }_1A^{}(I\mathrm{\Pi }_1)A\mathrm{\Delta }_1|\varphi =1p,`$
and thus $`|\gamma _0`$ and $`|\gamma _1`$ are unit vectors. Moreover, as
$$\mathrm{\Pi }_1A\mathrm{\Delta }_1A^{}\mathrm{\Pi }_1|\gamma _1=\frac{\mathrm{\Pi }_1A\mathrm{\Delta }_1\left(\mathrm{\Delta }_1A^{}\mathrm{\Pi }_1A\mathrm{\Delta }_1\right)|\varphi }{\sqrt{p}}=p|\gamma _1,$$
we have that $`|\delta _0`$ and $`|\delta _1`$ are unit vectors by similar reasoning. Note also that $`|\delta _1=|\varphi `$, which follows immediately from the fact that $`|\varphi `$ is an eigenvector of $`\mathrm{\Delta }_1A^{}\mathrm{\Pi }_1A\mathrm{\Delta }_1`$ with eigenvalue $`p`$. Based on these observations we conclude that
$$\begin{array}{cc}\hfill A|\delta _0& =\sqrt{p}|\gamma _0+\sqrt{1p}|\gamma _1\hfill \\ \hfill A|\delta _1& =\sqrt{1p}|\gamma _0+\sqrt{p}|\gamma _1.\hfill \end{array}$$
(4)
It will also be helpful to note that
$$\begin{array}{cc}\hfill A^{}|\gamma _0& =\sqrt{p}|\delta _0+\sqrt{1p}|\delta _1\hfill \\ \hfill A^{}|\gamma _1& =\sqrt{1p}|\delta _0+\sqrt{p}|\delta _1\hfill \end{array}$$
(5)
which follows from the equations 4 along with the fact that $`A`$ is unitary.
With the above equations 4 and 5 in hand, it is now possible to calculate the probability associated with each sequence of measurement outcomes. The procedure $`B`$ begins in state $`|\varphi =|\delta _1`$, and the procedure $`A`$ is performed. After the measurement described by $`\{\mathrm{\Pi }_0,\mathrm{\Pi }_1\}`$ the (renormalized) state of register $`𝖱`$ becomes $`|\gamma _0`$ or $`|\gamma _1`$ according to whether the outcome is 0 or 1, with associated probabilities $`1p`$ and $`p`$, respectively. If instead the procedure $`B`$ were to start in state $`|\delta _0`$, the renormalized states after measurement would be the same, but the probabilities would be reversed; probability $`p`$ is associated with outcome 0 and probability $`1p`$ with outcome 1. For the second step of the loop the situation is similar. If the register $`𝖱`$ is in state $`|\gamma _1`$, the transformation $`A^{}`$ is applied, and the state is measured with respect to the measurement $`\{\mathrm{\Delta }_0,\mathrm{\Delta }_1\}`$, the renormalized state after measurement will be either $`|\delta _1`$ or $`|\delta _0`$, with associated probabilities $`p`$ and $`1p`$. If instead the initial state were $`|\gamma _0`$ rather than $`|\gamma _1`$, the renormalized states after the measurement would again be the same, but the probabilities would be reversed. These transition probabilities are illustrated in 2.
In all cases we see that the probability of obtaining the same outcome as for the previous measurement is $`p`$, and the probability of the opposite outcome is $`1p`$. The probability associated with a given sequence $`z=(z_1,\mathrm{},z_N)`$ is therefore $`p^{w(z)}(1p)^{Nw(z)}`$ as claimed, as each $`z_i`$ is 1 if the measurement outcomes $`y_{i1}`$ and $`y_i`$ are equal, and is 0 otherwise. (Setting $`y_0=1`$ includes the first measurement outcome in this pattern.)
At this point we are ready to consider the completeness and soundness properties of the procedure $`B`$. Suppose first that the input $`x`$ is in $`L`$, which implies that the procedure $`A`$ can be made to accept with probability at least $`a`$. As an arbitrary state $`|\psi `$ is accepted by $`A`$ with probability $`\psi |Q|\psi `$, we therefore have $`\psi |Q|\psi a`$ for some choice of $`|\psi `$. Because $`Q`$ is positive semidefinite it is the case that $`\psi |Q|\psi `$ is bounded above by the largest eigenvalue of $`Q`$. Consequently there must exist a unit eigenvector $`|\psi `$ of $`Q`$ having associated eigenvalue $`pa`$. The procedure $`B`$ has been shown to accept such a choice of $`|\psi `$ with probability at least $`12^r`$ as required.
Now let us consider the soundness of the procedure $`B`$. If the input $`x`$ is not contained in $`L`$, then every choice for the state $`|\psi `$ causes $`A`$ to accept with probability at most $`b`$. Therefore, every eigenvalue of the operator $`Q`$ is at most $`b`$. We have shown that if $`|\psi `$ is an eigenvector of $`Q`$, then the procedure $`B`$ will accept $`|\psi `$ with probability less than $`2^r`$. Unfortunately, we may not assume that Merlin chooses $`|\psi `$ to be an eigenvector of $`Q`$. Nevertheless, the previous analysis can be extended to handle this possibility.
Specifically, let $`\{|\psi _1,\mathrm{},|\psi _{2^m}\}`$ be a complete orthonormal collection of eigenvectors of $`Q`$, with $`p_j`$ denoting the eigenvalue corresponding to $`|\psi _j`$ for $`j=1,\mathrm{},2^m`$. An arbitrary unit vector $`|\psi `$ may be written as
$$|\psi =\underset{j=1}{\overset{2^m}{}}\alpha _j|\psi _j$$
for $`\alpha _1,\mathrm{},\alpha _{2^m}`$ satisfying $`_j|\alpha _j|^2=1`$. Given such a state $`|\psi `$ as input, the procedure $`B`$ obtains each sequence $`z=(z_1,\mathrm{},z_N)`$ with probability
$$\underset{j=1}{\overset{2^m}{}}|\alpha _j|^2p_j^{w(z)}(1p_j)^{Nw(z)}$$
and so the probability of acceptance is
$$\underset{j=1}{\overset{2^m}{}}|\alpha _j|^2\underset{N\frac{a+b}{2}iN}{}\left(\genfrac{}{}{0pt}{}{N}{i}\right)p_j^i(1p_j)^{Ni}<2^r.$$
This does not follow from linearity because measurements are nonlinear. Instead, to see that it is indeed the case, one may repeat the analysis given previously in somewhat more generality. Specifically, let $`|\varphi _j=|\psi _j|0^k`$ and
$$|\gamma _{j,0}=\frac{\mathrm{\Pi }_0A\mathrm{\Delta }_1|\varphi _j}{\sqrt{1p_j}},|\gamma _{j,1}=\frac{\mathrm{\Pi }_1A\mathrm{\Delta }_1|\varphi _j}{\sqrt{p_j}},|\delta _{j,0}=\frac{\mathrm{\Delta }_0A^{}\mathrm{\Pi }_1|\gamma _{j,1}}{\sqrt{1p_j}},|\delta _{j,1}=\frac{\mathrm{\Delta }_1A^{}\mathrm{\Pi }_1|\gamma _{j,1}}{\sqrt{p_j}},$$
for each $`j=1,\mathrm{},2^m`$. As before, each of these vectors is a unit vector, $`|\delta _{j,1}=|\varphi _j`$, and
$`A|\delta _{j,0}`$ $`=\sqrt{p_j}|\gamma _{j,0}+\sqrt{1p_j}|\gamma _{j,1},`$
$`A|\delta _{j,1}`$ $`=\sqrt{1p_j}|\gamma _{j,0}+\sqrt{p_j}|\gamma _{j,1},`$
$`A^{}|\gamma _{j,0}`$ $`=\sqrt{p_j}|\delta _{j,0}+\sqrt{1p_j}|\delta _{j,1},`$
$`A^{}|\gamma _{j,1}`$ $`=\sqrt{1p_j}|\delta _{j,0}+\sqrt{p_j}|\delta _{j,1}.`$
Moreover, each of the sets $`\{|\gamma _{j,0}\}`$, $`\{|\gamma _{j,1}\}`$, $`\{|\delta _{j,0}\}`$, and $`\{|\delta _{j,1}\}`$ is an orthonormal set. Because of this fact, when $`B`$ is performed on the state $`|\psi `$, a similar pattern to the single eigenvector case arises independently for each eigenvector $`|\psi _j`$. This results in the stated probability of acceptance, which completes the proof. ∎
### Applications of strong error reduction
Two applications of 3.3 will now be discussed. The first is a simplified proof that $`\mathrm{QMA}`$ is contained in the class $`\mathrm{PP}`$.
###### Theorem 3.4.
$`\mathrm{QMA}\mathrm{PP}`$.
###### Proof.
Let $`L\mathrm{\Sigma }^{}`$ be a language in $`\mathrm{QMA}`$. By 3.3 there exists a function $`m\mathrm{𝑝𝑜𝑙𝑦}`$ such that
$$L\mathrm{QMA}_m(12^{(m+2)},2^{(m+2)}).$$
Let $`A`$ be a verification procedure that witnesses this fact. Specifically, each circuit $`A_x`$ acts on $`k+m`$ qubits, for some $`k\mathrm{𝑝𝑜𝑙𝑦}`$, and satisfies the following. If $`xL`$, then there exists an $`m`$ qubit state $`|\psi `$ such that
$$\mathrm{Pr}[A_x\text{ accepts }|\psi ]12^{m2},$$
while if $`xL`$, then
$$\mathrm{Pr}[A_x\text{ accepts }|\psi ]2^{m2}$$
for every $`m`$ qubit state $`|\psi `$.
For each $`x\mathrm{\Sigma }^{}`$, define a $`2^m\times 2^m`$ matrix $`Q_x`$ as
$$Q_x=\left(I_m0^k|\right)A_x^{}\mathrm{\Pi }_1A_x\left(I_m|0^k\right).$$
Each $`Q_x`$ is positive semidefinite, and $`\psi |Q_x|\psi =\mathrm{Pr}[A_x\text{ accepts }|\psi ]`$ for any unit vector $`|\psi `$ on $`m`$ qubits. The maximum probability with which $`A_x`$ can be made to accept is the largest eigenvalue of $`Q_x`$. Because the trace of a matrix is equal to the sum of its eigenvalues and all eigenvalues of $`Q_x`$ are nonnegative, it follows that if $`xL`$, then $`\mathrm{tr}(Q_x)12^{m2}3/4`$, while if $`xL`$, then $`\mathrm{tr}(Q_x)2^m2^{m2}1/4`$.
Now, based on a straightforward modification of the method of \[FR99\] discussed previously, we have that there exists a polynomially-bounded $`\mathrm{FP}`$ function $`g`$ and $`\mathrm{GapP}`$ functions $`f_1`$ and $`f_2`$ such that the real and imaginary parts of the entries of $`Q_x`$ are represented by $`f_1`$, $`f_2`$, and $`g`$ in the sense that
$$\mathrm{}(Q_x[i,j])=\frac{f_1(x,i,j)}{2^{g(x)}}\text{and}\mathrm{}(Q_x[i,j])=\frac{f_2(x,i,j)}{2^{g(x)}}$$
for $`0i,j<2^m`$. Define
$$h(x)=\underset{i=0}{\overset{2^m1}{}}f_1(x,i,i).$$
Because $`\mathrm{GapP}`$ functions are closed under exponential sums, we have $`h\mathrm{GapP}`$. It holds that $`h(x)=2^{g(x)}\mathrm{tr}(Q_x)`$, and therefore
$$xLh(x)\frac{3}{4}\mathrm{\hspace{0.17em}2}^{g(x)}\text{and}xLh(x)\frac{1}{4}\mathrm{\hspace{0.17em}2}^{g(x)}.$$
Because $`2^{g(x)}`$ is an FP function, it follows that $`2h(x)2^{g(x)}`$ is a GapP function that is positive if $`xL`$ and negative if $`xL`$. Thus, $`L\mathrm{PP}`$ as required. ∎
###### Remark 3.5.
A simple modification of the above proof yields $`\mathrm{QMA}\mathrm{A}_0\mathrm{PP}`$. Specifically, the $`\mathrm{GapP}`$ function $`2h`$ and the $`\mathrm{FP}`$ function $`2^{g(x)}`$ satisfy the required properties to prove $`L\mathrm{A}_0\mathrm{PP}`$, namely
$$xL\mathrm{\hspace{0.33em}2}h(x)2^{g(x)}\text{and}xL\mathrm{\hspace{0.33em}2}h(x)\frac{1}{2}\mathrm{\hspace{0.17em}2}^{g(x)}.$$
The second application concerns one-message quantum Arthur-Merlin games where Merlin sends only a logarithmic number of qubits to Arthur. Classical one-message Arthur-Merlin games with logarithmic-length messages from Merlin to Arthur are obviously equivalent in power to $`\mathrm{BPP}`$, because Arthur could simply search through all possible messages in polynomial time in lieu of interacting with Merlin. In the quantum case, however, this argument does not work, as one may construct exponentially large sets of pairwise nearly-orthogonal quantum states on a logarithmic number of qubits, such as those used in quantum fingerprinting \[BCWdW01\]. Nevertheless, logarithmic length quantum messages can be shown to be useless in the context of $`\mathrm{QMA}`$ using a different method, based on the strong error reduction property of $`\mathrm{QMA}`$ proved above.
For $`a,b:[0,1]`$ define $`\mathrm{QMA}_{\mathrm{log}}(a,b)`$ to be the class of all languages contained in $`\mathrm{QMA}_m(a,b)`$ for $`m(n)=O(\mathrm{log}n)`$, and let
$$\mathrm{QMA}_{\mathrm{log}}=\mathrm{QMA}_{\mathrm{log}}(2/3,1/3).$$
The choice of the constants 2/3 and 1/3 is arbitrary, which follows from 3.3.
###### Theorem 3.6.
$`\mathrm{QMA}_{\mathrm{log}}=\mathrm{BQP}`$.
###### Proof.
The containment $`\mathrm{BQP}\mathrm{QMA}_{\mathrm{log}}`$ is trivial, so it suffices to prove $`\mathrm{QMA}_{\mathrm{log}}\mathrm{BQP}`$. Assume $`L\mathrm{QMA}_m`$ for $`m`$ logarithmic, and assume $`A`$ is a $`\mathrm{QMA}`$ verification procedure that witnesses this fact and has completeness and soundness error less than $`2^{(m+2)}`$. Let
$$Q_x=\left(I_m0^k|\right)A_x^{}\mathrm{\Pi }_1A_x\left(I_m|0^k\right).$$
Similar to the proof of 3.4, we have
$$xL\mathrm{tr}(Q_x)3/4,xL\mathrm{tr}(Q_x)1/4.$$
We will describe a polynomial-time quantum algorithm $`B`$ that decides $`L`$ with bounded error. The algorithm $`B`$ simply constructs a totally mixed state over $`m`$ qubits and runs the verification procedure $`A`$ using this state in place of Merlin’s message. Running the verification procedure on the totally mixed state is equivalent to running the verification procedure on $`m`$ qubits initialized to some uniformly generated standard basis state, which is straightforward to simulate using Hadamard transforms and reversible computation. The totally mixed state on $`m`$ qubits corresponds to the density matrix $`2^mI_m`$, from which it follows that the probability of acceptance of $`B`$ is given by
$$\mathrm{Pr}[B\text{ accepts }x]=\mathrm{tr}\left(Q_x\mathrm{\hspace{0.17em}2}^mI_m\right)=2^m\mathrm{tr}(Q_x).$$
Given that $`m`$ is logarithmic in $`|x|`$, we have that the probabilities with which $`B`$ accepts inputs $`xL`$ and inputs $`xL`$ are bounded away from one another by the reciprocal of some polynomial. This difference can be amplified by standard methods, implying that $`L\mathrm{BQP}`$. ∎
## 4 QAM
A $`\mathrm{QAM}`$ verification procedure $`A`$ consists of a polynomial-time generated family
$$\{A_{x,y}:x\mathrm{\Sigma }^{},y\mathrm{\Sigma }^{s(|x|)}\}$$
of quantum circuits together with functions $`m,s\mathrm{𝑝𝑜𝑙𝑦}`$. As for $`\mathrm{QMA}`$ verification procedures, each circuit $`A_{x,y}`$ acts on two collections of qubits: $`m(|x|)`$ qubits sent by Merlin and $`k(|x|)`$ qubits corresponding to Arthur’s workspace. The notion of a circuit $`A_{x,y}`$ accepting a message $`|\psi `$ is defined in the same way as for $`\mathrm{QMA}`$. In the present case, the string $`y`$ corresponds to a sequence of coin-flips sent by Arthur to Merlin, on which Merlin’s message may depend.
###### Definition 4.1.
The class $`\mathrm{QAM}(a,b)`$ consists of all languages $`L\mathrm{\Sigma }^{}`$ for which there exists a $`\mathrm{QAM}`$ verification procedure $`A`$ satisfying the following conditions.
* If $`xL`$ then there exists a collection of states $`\{|\psi _y\}`$ on $`m`$ qubits such that
$$\frac{1}{2^s}\underset{y\mathrm{\Sigma }^s}{}\mathrm{Pr}[A_{x,y}\text{ accepts }|\psi _y]a.$$
* If $`xL`$ then for every collection of states $`\{|\psi _y\}`$ on $`m`$ qubits it holds that
$$\frac{1}{2^s}\underset{y\mathrm{\Sigma }^s}{}\mathrm{Pr}[A_{x,y}\text{ accepts }|\psi _y]b.$$
Similar to $`\mathrm{QMA}`$, one may consider the cases where $`a`$ and $`b`$ are constants or functions of $`n=|x|`$, and in the case that $`a`$ and $`b`$ are functions of the input length it is assumed that $`a(n)`$ and $`b(n)`$ can be computed deterministically in time polynomial in $`n`$. Also as before, let $`\mathrm{QAM}=\mathrm{QAM}(2/3,1/3)`$.
### Error reduction for QAM
The first fact about $`\mathrm{QAM}`$ that we prove is that completeness and soundness errors may be reduced by running many copies of a given game in parallel. The proof is similar in principle to the proof of Lemma 14.1 in \[KSV02\], which corresponds to our 3.2.
###### Theorem 4.2.
Let $`a,b:[0,1]`$ and $`q\mathrm{𝑝𝑜𝑙𝑦}`$ satisfy
$$a(n)b(n)\frac{1}{q(n)}$$
for all $`n`$. Then $`\mathrm{QAM}(a,b)\mathrm{QAM}(12^r,2^r)`$ for every $`r\mathrm{𝑝𝑜𝑙𝑦}`$.
###### Proof.
Let $`L\mathrm{QAM}(a,b)`$, and let $`A`$ be a $`\mathrm{QAM}`$ verification procedure witnessing this fact. We consider a new $`\mathrm{QAM}`$ verification procedure that corresponds to playing the game described by $`\{A_{x,y}\}`$ in parallel $`N`$ times. The new procedure accepts if and only if the number of acceptances of the original game is at least $`N\frac{a+b}{2}`$. Although Merlin is not required to play the repetitions independently, we will show that playing the repetitions independently in fact gives him an optimal strategy. The theorem then follows by choosing an appropriately large value of $`N`$ and applying a Chernoff-type bound.
Assume hereafter that the input $`x`$ is fixed, and define
$`Q_y^{(0)}`$ $`=(I0^k|)A_{x,y}^{}\mathrm{\Pi }_0A_{x,y}(I|0^k),`$
$`Q_y^{(1)}`$ $`=(I0^k|)A_{x,y}^{}\mathrm{\Pi }_1A_{x,y}(I|0^k)`$
for each $`y\mathrm{\Sigma }^s`$. We have $`Q_y^{(1)}=IQ_y^{(0)}`$, and consequently $`Q_y^{(0)}`$ and $`Q_y^{(1)}`$ share a complete set of orthonormal eigenvectors. Let $`\{|\psi _{y,1},\mathrm{},|\psi _{y,2^m}\}`$ be such a set, and let
$$p_{y,1}^{(z)},\mathrm{},p_{y,2^m}^{(z)}$$
be the corresponding eigenvalues for $`Q_y^{(z)}`$, $`z\{0,1\}`$. As $`Q_y^{(0)}`$ and $`Q_y^{(1)}`$ are positive semidefinite and sum to the identity, $`p_{y,i}^{(0)}`$ and $`p_{y,i}^{(1)}`$ are nonnegative real numbers with $`p_{y,i}^{(0)}+p_{y,i}^{(1)}=1`$ for each $`y`$ and $`i`$. Assume without loss of generality that the eigenvectors and eigenvalues are ordered in such a way that
$$p_{y,1}^{(1)}\mathrm{}p_{y,2^m}^{(1)}.$$
This implies that the maximum acceptance probability of $`A_{x,y}`$ is $`p_{y,1}^{(1)}`$.
Under the assumption that Arthur’s coin-flips for the $`N`$ repetitions are given by strings $`y_1,\mathrm{},y_N\mathrm{\Sigma }^s`$, if Merlin plays the repetitions independently, and optimally for each repetition, his probability of convincing Arthur to accept is
$$\underset{\stackrel{z_1,\mathrm{},z_N\mathrm{\Sigma }}{z_1+\mathrm{}+z_NN\frac{a+b}{2}}}{}p_{y_1,1}^{(z_1)}\mathrm{}p_{y_N,1}^{(z_N)}.$$
(6)
Without any assumption on Merlin’s strategy, the maximum probability with which Merlin can win $`N\frac{a+b}{2}`$ repetitions of the original game when Arthur’s coin-flips are given by $`y_1,\mathrm{},y_N`$ is equal to the largest eigenvalue of
$$\underset{\stackrel{z_1,\mathrm{},z_N\mathrm{\Sigma }}{z_1+\mathrm{}+z_NN\frac{a+b}{2}}}{}Q_{y_1}^{(z_1)}\mathrm{}Q_{y_N}^{(z_N)}.$$
(7)
Therefore, to prove the proposition it suffices to show that these quantities are equal.
All of the summands in equation 7 share the complete set of orthonormal eigenvalues given by
$$\left\{|\psi _{y_1,i_1}\mathrm{}|\psi _{y_N,i_N}:i_1,\mathrm{},i_N\{1,\mathrm{},2^m\}\right\},$$
and so this set also describes a complete set of orthonormal eigenvectors of the sum. The eigenvalue associated with $`|\psi _{y_1,i_1}\mathrm{}|\psi _{y_N,i_N}`$ is
$$\underset{\stackrel{z_1,\mathrm{},z_N\mathrm{\Sigma }}{z_1+\mathrm{}+z_NN\frac{a+b}{2}}}{}p_{y_1,i_1}^{(z_1)}\mathrm{}p_{y_N,i_N}^{(z_N)}.$$
(8)
Define $`u_1(X)=X`$, $`u_0(X)=1X`$, and let
$$f(X_1,\mathrm{},X_N)=\underset{\stackrel{z_1,\mathrm{},z_N\mathrm{\Sigma }}{z_1+\mathrm{}+z_NN\frac{a+b}{2}}}{}u_{z_1}(X_1)\mathrm{}u_{z_N}(X_N).$$
The quantity in equation 8 is equal to
$$f(p_{y_1,i_1}^{(1)},\mathrm{},p_{y_N,i_N}^{(1)}).$$
The function $`f`$ is multi-linear and nondecreasing in each variable everywhere on the unit hypercube. Thus, the maximum of the quantity in equation 8 is
$$f(p_{y_1,1}^{(1)},\mathrm{},p_{y_N,1}^{(1)}),$$
which is equal to the quantity in equation 6. This completes the proof. ∎
### An upper bound on QAM
We now observe that the upper bound
$$\mathrm{QAM}\mathrm{BP}\mathrm{PP}$$
holds. The following fact concerning the maximum probabilities of acceptance of $`A_{x,y}`$ for random $`y`$ will be used. Here we let $`\mu (A_{x,y})`$ denote the maximum probability that $`A_{x,y}`$ can be made to accept (maximized over all choices of Merlin’s message $`|\psi _y`$).
###### Proposition 4.3.
Suppose that
$$\{A_{x,y}:x\mathrm{\Sigma }^{},y\mathrm{\Sigma }^{s(|x|)}\}$$
is a $`\mathrm{QAM}`$ verification procedure for a language $`L`$ that has completeness and soundness errors bounded by 1/9. Then for any $`x\mathrm{\Sigma }^{}`$ and for $`y\mathrm{\Sigma }^s`$ chosen uniformly at random,
$`xL`$ $`\mathrm{Pr}[\mu (A_{x,y})2/3]\mathrm{\hspace{0.25em}2}/3`$
$`xL`$ $`\mathrm{Pr}[\mu (A_{x,y})1/3]\mathrm{\hspace{0.25em}2}/3.`$
###### Proof.
Suppose that $`xL`$. Let $`z(y)=1\mu (A_{x,y})`$, and let $`Z`$ be a random variable whose value is $`z(y)`$ for a uniformly chosen $`y\mathrm{\Sigma }^s`$. The assumption of the proposition implies that $`E[Z]1/9`$. By Markov’s inequality we have
$$\mathrm{Pr}[Z>1/3]\frac{E[Z]}{1/3}1/3,$$
and therefore
$$\mathrm{Pr}[\mu (A_{x,y})2/3]=\mathrm{Pr}[Z1/3]2/3.$$
The proof for $`xL`$ is similar. ∎
###### Theorem 4.4.
$`\mathrm{QAM}\mathrm{BP}\mathrm{PP}`$.
###### Proof.
Let $`L\mathrm{QAM}`$, and let
$$A=\{A_{x,y}:x\mathrm{\Sigma }^{},y\mathrm{\Sigma }^{s(|x|)}\}$$
be a $`\mathrm{QAM}`$ verification procedure for $`L`$ with completeness and soundness errors bounded by 1/9. Such a procedure exists by 4.2. By a straightforward modification of the proof of 3.4, one may conclude that there exists a language $`K\mathrm{PP}`$ such that
$`\mu (A_{x,y})2/3`$ $`(x,y)K,`$
$`\mu (A_{x,y})1/3`$ $`(x,y)K.`$
It is possible that $`\mu (A_{x,y})(1/3,2/3)`$ for some values of $`y`$, but in this case no requirement is made on whether or not $`(x,y)K`$. The theorem now follows from 4.3. ∎
## 5 QMAM
A $`\mathrm{QMAM}`$ verification procedure $`A`$ consists of a polynomial-time generated family
$$\{A_{x,y}:x\mathrm{\Sigma }^{},y\mathrm{\Sigma }^{s(|x|)}\}$$
of quantum circuits, together with functions $`m_1,m_2,s\mathrm{𝑝𝑜𝑙𝑦}`$. The functions $`m_1`$ and $`m_2`$ specify the number of qubits in Merlin’s first and second messages to Arthur, while $`s`$ specifies the number of random bits Arthur sends to Merlin. Each circuit $`A_{x,y}`$ acts on $`m_1(|x|)+m_2(|x|)+k(|x|)`$ qubits, where as before $`k(|x|)`$ denotes the number of qubits corresponding to Arthur’s workspace.
In the $`\mathrm{QMAM}`$ case, it becomes necessary to discuss possible actions that Merlin may perform rather than just discussing states that he may send. This is because Merlin’s strategy could involve preparing some quantum state, sending part of that state to Arthur on the first message, and transforming the part of that state he did not send to Arthur (after receiving Arthur’s coin-flips) in order to produce his second message.
###### Definition 5.1.
A language $`L\mathrm{\Sigma }^{}`$ is in $`\mathrm{QMAM}(a,b)`$ if there exists a $`\mathrm{QMAM}`$ verification procedure $`A`$ such that the following conditions are satisfied.
* If $`xL`$ then for some $`l`$ there exists a quantum state $`|\psi `$ on $`m_1+m_2+l`$ qubits and a collection of unitary operators $`\{U_y:y\mathrm{\Sigma }^s\}`$ acting on $`m_2+l`$ qubits such that
$$\frac{1}{2^s}\underset{y\mathrm{\Sigma }^s}{}\mathrm{Pr}[A_{x,y}\text{ accepts }(I_{m_1}U_y)|\psi ]a.$$
* If $`xL`$ then for every $`l`$, every quantum state $`|\psi `$ on $`m_1+m_2+l`$ qubits, and every collection of unitary operators $`\{U_y:y\mathrm{\Sigma }^s\}`$ acting on $`m_2+l`$ qubits,
$$\frac{1}{2^s}\underset{y\mathrm{\Sigma }^s}{}\mathrm{Pr}[A_{x,y}\text{ accepts }(I_{m_1}U_y)|\psi ]b.$$
The same assumptions regarding $`a`$ and $`b`$ apply in this case as in the $`\mathrm{QMA}`$ and $`\mathrm{QAM}`$ cases.
In the above definition, the circuit $`A_{x,y}`$ is acting on $`m_1+m_2`$ qubits sent by Merlin in addition to Arthur’s $`k`$ workspace qubits, while $`(I_{m_1}U_y)|\psi `$ is a state on $`m_1+m_2+l`$ qubits. It is to be understood that the last $`l`$ qubits of $`(I_{m_1}U_y)|\psi `$ remain in Merlin’s possession, so $`A_{x,y}`$ is effectively tensored with the identity acting on these qubits.
### Equivalence of QMAM and QIP
We now prove $`\mathrm{QMAM}=\mathrm{QIP}`$. Because quantum Arthur-Merlin games are a restricted form of quantum interactive proof systems, $`\mathrm{QMAM}\mathrm{QIP}`$ is obvious. To prove the opposite containment, we will require the following lemmas. The first lemma is a corollary of Uhlmann’s Theorem (see \[NC00\]).
###### Lemma 5.2.
Suppose the pair of registers $`(𝖵,𝖬)`$ is in a mixed state for which the reduced state of $`𝖵`$ is $`\sigma `$. If the pair $`(𝖵,𝖬)`$ is measured with respect to a binary valued measurement described by orthogonal projections $`\{\mathrm{\Lambda }_0,\mathrm{\Lambda }_1\}`$, then the probability of obtaining the outcome 1 is at most $`F(\sigma ,\rho )^2`$ for some $`\rho 𝒮_𝖵(\mathrm{\Lambda }_1)`$.
The second lemma is a simple property of the fidelity function.
###### Lemma 5.3 (\[NS02, SR02\]).
For any choice of density matrices $`\rho `$, $`\xi `$, and $`\sigma `$, we have
$$F(\rho ,\sigma )^2+F(\sigma ,\xi )^21+F(\rho ,\xi ).$$
###### Theorem 5.4.
Let $`L\mathrm{QIP}`$ and let $`r\mathrm{𝑝𝑜𝑙𝑦}`$. Then $`L`$ has a three message quantum Arthur-Merlin game with completeness error 0 and soundness error at most $`1/2+2^r`$. Moreover, in this quantum Arthur-Merlin game, Arthur’s message consists of a single coin-flip.
###### Proof.
Let $`L\mathrm{QIP}`$, which implies that $`L`$ has a three-message quantum interactive proof system with completeness error 0 and soundness error $`\epsilon (n)=2^{2r(n)}`$ on inputs of length $`n`$.
Consider a $`\mathrm{QMAM}`$ verification procedure $`A`$ that corresponds to the following actions for Arthur. (It will be assumed that the input $`x`$ is fixed, and it will be clear that the family of quantum circuits corresponding to this verification procedure can be generated in polynomial-time given that the same is true of the verifier being simulated.)
* Receive register $`𝖵`$ from Merlin.
* Flip a fair coin and send the result to Merlin.
* Receive register $`𝖬`$ from Merlin. If the coin flipped in step 2 was heads, apply $`V_2`$ to $`(𝖵,𝖬)`$ and accept if the first qubit of $`𝖵`$ (i.e., the output qubit of the quantum interactive proof system) is 1, otherwise reject. If the coin in step 2 was tails, apply $`V_1^{}`$ to $`(𝖵,𝖬)`$ and accept if all qubits of $`𝖵`$ are set to 0, otherwise reject.
Suppose first that $`xL`$, so that some prover, whose actions are described by a state $`|\psi `$ and a unitary operator $`U`$ can convince $`V`$ to accept with certainty. Then Merlin can convince Arthur to accept with certainty as follows:
* Prepare state $`|0^k`$ in register $`𝖵`$ and state $`|\psi `$ in registers $`(𝖬,𝖯)`$. Apply $`V_1`$ to registers $`(𝖵,𝖬)`$, and send $`𝖵`$ to Arthur.
* If Arthur flips heads, apply $`U`$ to $`(𝖬,𝖯)`$ and send $`𝖬`$ to Arthur. If Arthur flips tails, send $`𝖬`$ to Arthur without applying $`U`$.
Now assume $`xL`$, so that no prover can convince $`V`$ to accept with probability exceeding $`\epsilon `$. Suppose that the reduced density matrix of register $`𝖵`$ sent by Merlin is $`\sigma `$. By 5.2 and 5.3, the probability that Arthur can be made to accept is at most
$$\frac{1}{2}F(\rho ,\sigma )^2+\frac{1}{2}F(\xi ,\sigma )^2\frac{1}{2}+\frac{1}{2}F(\rho ,\xi )$$
maximized over $`\rho 𝒮_𝖵(V_1\mathrm{\Delta }_1V_1^{})`$ and $`\xi 𝒮_𝖵(V_2^{}\mathrm{\Pi }_1V_2)`$. By 2.1 this probability is at most
$$\frac{1}{2}+\frac{\sqrt{\epsilon }}{2}\frac{1}{2}+2^{r(|x|)},$$
which completes the proof. ∎
###### Corollary 5.5.
For any function $`r\mathrm{𝑝𝑜𝑙𝑦}`$ we have $`\mathrm{QIP}\mathrm{QMAM}(1,1/2+2^r)`$.
### Error reduction for QMAM
Now, suppose that we have a $`\mathrm{QMAM}`$ protocol for a language $`L`$ with perfect completeness and soundness error $`b`$, and we repeat the protocol $`N`$ times in parallel, accepting if and only if all $`N`$ of the repetitions accept. It is clear that this resulting protocol has perfect completeness, because Merlin can play optimally for each parallel repetition independently and achieve an acceptance probability of 1 for any $`xL`$. In the case that $`xL`$, Merlin can gain no advantage whatsoever over playing the repetitions independently, and so the soundness error decreases to $`b^N`$ as we would hope. This follows from the fact that the same holds for arbitrary three-message quantum interactive proof systems \[KW00\], of which three-message quantum Arthur-Merlin games are a restricted type. This implies the following corollary.
###### Corollary 5.6.
For any function $`r\mathrm{𝑝𝑜𝑙𝑦}`$ we have $`\mathrm{QIP}=\mathrm{QMAM}(1,2^r)`$.
### More than three messages
Finally, we note that one may define quantum Arthur-Merlin games having any polynomial number of messages in a similar way to three-message quantum Arthur-Merlin games. Such games are easily seen to be equivalent in power to three-message quantum Arthur-Merlin games. Specifically, polynomial-message quantum Arthur-Merlin games will be special cases of quantum interactive proof systems, and can therefore be parallelized to three-message interactive proofs and simulated by three-message quantum Arthur-Merlin games as previously described.
## 6 Open questions
Many interesting questions about quantum Arthur-Merlin games remain unanswered, including the following questions.
* Are there interesting examples of problems in $`\mathrm{QMA}`$ or $`\mathrm{QAM}`$ that are not known to be in $`\mathrm{AM}`$? A similar question may be asked for $`\mathrm{QMAM}`$ vs. $`\mathrm{PSPACE}`$.
* The question of whether there exists an oracle relative to which $`\mathrm{BQP}`$ is outside of the polynomial-time hierarchy appears to be a difficult problem. In fact it is currently not even known if there is an oracle relative to which $`\mathrm{BQP}\mathrm{AM}`$. Is there an oracle relative to which $`\mathrm{QMA}`$ or $`\mathrm{QAM}`$ is not contained in $`\mathrm{AM}`$? If so, what about $`\mathrm{QMA}`$ or $`\mathrm{QAM}`$ versus $`\mathrm{PH}`$? Such results might shed some light on the problem of $`\mathrm{BQP}`$ versus the polynomial-time hierarchy.
* \[NW94\] proved $`\mathrm{almost}`$-$`\mathrm{NP}=\mathrm{AM}`$. Is it the case that $`\mathrm{almost}`$-$`\mathrm{QMA}=\mathrm{QAM}`$?
### Acknowledgements
Thanks to Dorit Aharonov, Oded Regev, and Umesh Vazirani for their comments on error reduction for $`\mathrm{QMA}`$, Ashwin Nayak for helpful references, and Alexei Kitaev for discussions about quantum proof systems. This research was supported by Canada’s NSERC, the Canadian Institute for Advanced Research (CIAR), and the Canada Research Chairs program.
|
warning/0506/hep-ph0506147.html
|
ar5iv
|
text
|
# Pions emerging from an Arbitrarily Disoriented Chiral Condensate.
Relativistic heavy ion collisions either with fixed target or with head on collisions are providing results which suggest that the chiral phase transition does take place with temperatures crossing 170Mev. The comparatively large sizes of the heavy ions involved in the collisions along with their energy/nucleon, provides a large interaction region for the quarks and gluons. This region is where the Quark Gluon Plasma (QGP) forms. The picture we therefore have is that of a collision region where a QGP is formed in a highly non-equilibrium situation with chiral symmetry being restored. This plasma then expands and cools, causing the restored chiral symmetry to be broken once again. In a recent paper , we have given a quantum field theoretical model providing multiplicities of pions which would be observed if the QGP undergoes a rapid quench in its expansion or if the expansion is adiabatic. It was also shown how the rapid quench scenario, which is appropriate for the formation of the disoriented chiral condensate (DCC) has a enhancement in the multiplicities of the observed pions. Thus the enhancement signatures in the pion multiplicities are signals for the formation of the DCC also. In our analysis, we had assumed the expansion of the QGP to be isotropic for simplicity. However, recent results from RHIC suggest that there may be anisotropy built in from the nature of the collisions. Not all collisions will be head on with zero impact parameter. There will be instances where there will be non-zero impact parameter causing the impact region to be elliptical and the corresponding QGP to have an elliptical flow. From the nature of our quantum field theoretical model, borrowing from the studies of scalar fields on expanding universes, such as the Friedmann-Robertson-Walker universe, it is easy to extend the study to include anisotropy such as that caused by an elliptic flow. We shall report these results elsewhere, in a separate communication . Here, we shall be examining another aspect of the restoration of chiral symmetry. The particular form of our model allows for many metastable states through which the DCC can form. In particular, we find that mixing between various erstwhile pion states in the collision region can produce different signals in the multiplicities of the final state pions.
We start with an $`O(4)`$ quartet of scalar fields in order that we can construct the dynamics of quenched pions in the formation of a disoriented chiral condensate. The background field is the classical disoriented vacuum and the Hamiltonian of the quantum fluctuations around the background field is derived from the $`O(4)`$ sigma model keeping one-loop quantum corrections (quadratic order in the fluctuations). The resulting dynamical Hamiltonian for the pion and sigma fields fields in terms of the neutral , charged pion and sigma creation and annihilation operators ($`a,a^{},c,c^{},b`$ $`b^{}`$, and $`d,d^{}`$) is:
$`H`$ $`=`$ $`{\displaystyle }{\displaystyle \frac{d^3\underset{¯}{k}}{(2\pi )^3}}{\displaystyle \frac{1}{2}}\{{\displaystyle \frac{\omega _\pi }{a^3}}(a_k^{}a_k+a_ka_k^{})+{\displaystyle \frac{\omega _\pi }{2a^3}}({\displaystyle \frac{\mathrm{\Omega }_\pi ^2}{\omega _\pi ^2}}1)(a_k^{}a_k+a_ka_k^{}+a_ka_k+a_k^{}a_k^{})`$
$`+`$ $`{\displaystyle \frac{\omega _\mathrm{\Sigma }}{a^3}}(d_k^{}d_k+d_kd_k^{})+{\displaystyle \frac{\omega _\mathrm{\Sigma }}{2a^3}}({\displaystyle \frac{\mathrm{\Omega }_\mathrm{\Sigma }^2}{\omega _\mathrm{\Sigma }^2}}1)(d_k^{}d_k+d_kd_k^{}+d_kd_k+d_k^{}d_k^{})\}`$
$`+`$ $`{\displaystyle \frac{\omega _\pi }{a^3}}(b_k^{}b_k+c_kc_k^{})+{\displaystyle \frac{\omega _\pi }{2a^3}}({\displaystyle \frac{\mathrm{\Omega }_{\pi _\pm }^2}{\omega _\pi ^2}}1)(b_k^{}b_k+c_kc_k^{}+b_kc_k+c_k^{}b_k^{})`$
$`+`$ $`{\displaystyle \frac{\lambda a^3v^2cos^2(\rho )sin^2(\theta )}{4\omega _\pi }}(b_kb_k+b_kc_k^{}+c_k^{}b_k+c_kc_k+c_k^{}c_k^{}+c_kb_k^{}+b_k^{}c_k+b_k^{}b_k^{}))`$
$`+`$ $`{\displaystyle \frac{\lambda a^3v^2cos(\rho )sin(\rho )sin^2(\theta )}{2\omega _\pi }}\left(b_ka_k+b_ka_k^{}+c_k^{}a_k+c_ka_k+c_k^{}a_k^{}+c_ka_k^{}+b_k^{}a_k+b_k^{}a_k^{}\right)`$
$`+`$ $`{\displaystyle \frac{\lambda a^3v^2sin(\rho )sin(\theta )cos(\theta )}{\sqrt{\omega _\pi \omega _\mathrm{\Sigma }}}}\left(d_ka_k+d_ka_k^{}+d_k^{}a_k+d_k^{}a_k^{}\right)`$
$`+`$ $`{\displaystyle \frac{\lambda a^3v^2cos(\rho )sin(\theta )cos(\theta )}{\sqrt{\omega _\pi \omega _\mathrm{\Sigma }}}}(b_kd_k+b_kd_k^{}+c_k^{}d_k+c_kd_k+c_k^{}d_k^{}+c_kd_k^{}+b_k^{}d_k+b_k^{}d_k^{})\}`$ (2)
where
$$\frac{\omega _\pi ^2(k)}{a^6}\frac{\omega _{\pi _0}^2(k)}{a^6}=\frac{\omega _{\pi _\pm }^2(k)}{a^6}=(m_\pi ^2+\frac{\underset{¯}{k}^2}{a^2}):\frac{\omega _\mathrm{\Sigma }^2(k)}{a^6}=(m_\mathrm{\Sigma }^2+\frac{\underset{¯}{k}^2}{a^2})$$
(3)
and
$`{\displaystyle \frac{\mathrm{\Omega }_{\pi _0}^2\omega _{\pi _0}^2}{a^6}}`$ $`=`$ $`\lambda [(<\mathrm{\Phi }^2>v^2)+2v_3^2]`$
$`{\displaystyle \frac{\mathrm{\Omega }_{\pi \pm }^2\omega _{\pi _\pm }^2}{a^6}}`$ $`=`$ $`\lambda [(<\mathrm{\Phi }^2>v^2)+2v_+v_{}]`$
$`{\displaystyle \frac{\mathrm{\Omega }_\mathrm{\Sigma }^2\omega _\mathrm{\Sigma }^2}{a^6}}`$ $`=`$ $`\lambda [(<\mathrm{\Phi }^2>v^2)+2\sigma ^2]`$ (4)
The background field in the disoriented phase is given by :
$$\left(\begin{array}{c}v_+\\ v_{}\\ v_3=v\\ \sigma \end{array}\right)<\mathrm{\Phi }>,$$
(5)
In keeping with the disorientation of the vacuum on a three sphere we parametrize the background field through three angles:
$$<\mathrm{\Phi }>=\left(\begin{array}{c}vCos(\rho )Sin(\theta )Sin(\alpha )\\ vCos(\rho )Sin(\theta )Cos(\alpha )\\ vSin(\rho )Sin(\theta )\\ vCos(\theta )\end{array}\right).$$
(6)
The dynamical evolution of a system governed by this Hamiltonian is considered from a non equilibrium state state of restored symmetry $`<\mathrm{\Phi }>=0`$ at high temperatures to the equilibrium state of broken symmetry $`<\mathrm{\Phi }>=v`$ due to the subsequent expansion and cooling of the QGP. The time evolution can be modeled through a time dependence of the parameter $`<\mathrm{\Phi }>(t)`$, which depends on the way in which the system relaxes to equilibrium. In addition there is a time dependence of the expansion factor $`a(t)`$ which allows for the rate of the expansion of the plasma. The following scenarios are possible:
A. A sudden quench from a state of restored symmetry to a state of broken symmetry in a rapidly cooling expanding plasma, the configuration of the field lags behind the expansion of the plasma.If $`\tau `$ is the time the state spends in the symmetry restored phase, a quench can be modeled by assuming that for $`0`$ the vacuum expectation value $`<\mathrm{\Phi }^2>=0`$ and for $`t>\frac{\tau }{2}`$ the vacuum relaxes to its value $`<\mathrm{\Phi }^2>=<v^2>`$. In such a case the dynamical evolution of $`<\mathrm{\Phi }>`$ could be modeled by $`<\mathrm{\Phi }>(t)=v\mathrm{\Theta }[(t\tau )]`$ In addition in a realistic scenario the expansion coefficient $`a(t)`$ must also be a function of time and has been modeled appropriately in ref . For a sudden quench scenario the Hamiltonian given in equation 1 diagonalizes to
$$H=\frac{d^3\underset{¯}{k}}{(2\pi )^3}\frac{1}{2a^3}\{\mathrm{\Omega }_\pi (k,t)\{(A_k^{}A_k+\frac{1}{2})+(C_k^{}C_k+B_k^{}B_k+1)\}+\mathrm{\Omega }_\mathrm{\Sigma }(k,t)(D_k^{}D_k+\frac{1}{2})\}.$$
(7)
by a unitary matrix given by a squeezed transformation
$$U(r,t)=e^{{\scriptscriptstyle {\scriptscriptstyle \frac{d^3\underset{¯}{k}}{(2\pi )^3}}r(k,t)\{(a_k^{}a_k^{}a_ka_k)+(d_k^{}d_k^{}d_kd_k)+(c_kb_k+b_kc_k)(c_k^{}b_k^{}+b_k^{}c_k^{})\}}}$$
(8)
where $`r_k`$ is the squeezing parameter related to the physical variables $`\mathrm{\Omega }_\pi (k,t)`$ and $`\omega _\pi (k)`$ through
$$Tanh(2r_k)=\frac{(\frac{\mathrm{\Omega }_k(t)}{\omega _\pi })^21}{(\frac{\mathrm{\Omega }_k(t)}{\omega _\pi })^2+1}$$
(9)
The quench consists of a rapid change of frequency of the time dependent harmonic oscillator from $`\mathrm{\Omega }_\pi (k,t))`$ to $`\omega _\pi (k)`$ though the time evolution of $`<\mathrm{\Phi }>(t)`$ and $`a(t)`$. It can be shown that for a sudden frequency jump, such as that of a quench, there is substantial squeezing, resulting in an enhancement of low momentum modes signalling a DCC. The details of the evolution of the Hamiltonian of a system undergoing a sudden quench (enhanced squeezing) also show up in the difference in multiplicity distributions of the charged and neutral pions illustrated in Figure 1.
B. Another non-equilibrium situation that can arise is one where the system can go through a metastable disordered vacuum given by eqn and then relax by quantum fluctuations to an equilibrium configuration. Here $`\theta `$ and $`\rho `$ measure the degree of disorientation of the condensate in isospin space ( for charge conservation $`\alpha =\frac{\pi }{4}`$) . The disorientation can be in the neutral sector $`\rho =\frac{\pi }{2}`$ in which case the angle $`\theta `$ mixes the neutral pion and sigma field resulting in oscillations and enhancement of neutral pions over the charged pions, we will call this case 1 . The disorientation can be in the charged sector $`\theta =\frac{\pi }{2}`$, where the angle $`\rho `$ mixes the charged pions and the neutral pions resulting in charged oscillations and enhancement of charged pions over the neutral pions we will call this case 2. As an illustration of these effects we examine case 1 briefly. All the details of case 1 and 2 will be given in an expanded later communication . For case 1 the examination of the Hamiltonian ( obtained by putting $`\rho =\frac{\pi }{2}`$ ) shows a non-zero mixing term coming from the $`\pi _0`$-$`\mathrm{\Sigma }`$ sector. The misalignment of the vacuum through an angle $`\theta `$ induces a mixing of the two fields. The mixed fields are
$$\left(\begin{array}{c}A_{\theta (k)}\\ D_{\theta (k)}\end{array}\right)=\left(\begin{array}{c}Cos(\theta )Sin(\theta )\\ Sin(\theta )Cos(\theta )\end{array}\right)\left(\begin{array}{c}a_k\\ d_k\end{array}\right)$$
(10)
The diagonalization procedure for this case involves two squeezing transformations of the mixed fields
$`F_k`$ $`=`$ $`\mu A_{\theta (k)}+\nu A_{\theta (k)}^{}`$
$`G_k`$ $`=`$ $`\rho D_{\theta (k)}+\sigma D_{\theta (k)}^{}.`$ (11)
The diagonalized form is
$$H_{\pi /2}=\frac{d^3\underset{¯}{k}}{(2\pi )^3}\{\frac{\mathrm{\Omega }_{\pi _\pm }}{a^3}(C_k^{}C_k+B_k^{}B_k+1)+\frac{\sqrt{\mathrm{\Omega }_\pi \mathrm{\Omega }_\mathrm{\Sigma }}}{4a^3}\left((F_k^{}F_k+\frac{1}{2})+(G_k^{}G_k+\frac{1}{2})\right)\}$$
(12)
During time evolution again we have a time dependent frequency for the state of mixed neutral pions and sigma which changes from $`\frac{\sqrt{\mathrm{\Omega }_\pi (k,t)\mathrm{\Omega }_\mathrm{\Sigma }(k,t)}}{4a^3}`$ to$`\frac{\sqrt{\omega _\pi (k,t)\omega _\mathrm{\Sigma }(k,t)}}{4a^3}`$ this time through the time evolution of $`\theta (t)`$ from the disoriented value to 0 and $`a(t)`$. There is again an enhancement of neutral pions as a result of the DCC formation and furthermore a mixing in the neutral sector . In addition there are oscillations in the number of neutral pions as the disorientation $`\theta `$ changes with time shown in Figure 2.
A case of particular interest , in view of latest preliminary experimental results is case 2 i.e. $`\theta =\frac{\pi }{2}`$ because there is a mixing of charged and neutral fields due to the disorientation. This gives rise to interesting charge-neutral fluctuations which can be measured in a DCC detection experiment. A full discussion of the detailed theory is given in .
To conclude we have modeled the evolution of the disoriented chiral condensate through both a sudden quench and with a transition through a metastable state with arbitrary disorientation and have shown that the total multiplicity distributions of charged and neutral pions functions are dramatic characteristic signals for the DCC and are related directly to the way in which the DCC forms. These are unambiguous, therefore they must be examined thoroughly in searches for the DCC . We are encouraged by the first preliminary data of references and , from the analysis of the WA98 experiment at the CERN SPS in which some events show an excess of photons (neutral pion excess) within the overlap region of charged and photon multiplicity detectors, the preliminary results given in ref are shown in figure 3.
In view of the results presented in this section, we hope that the future will bring more exotic events with charge excess and neutral and charged multiplicity oscillations.
References
* B. Bambah and C. Mukku, Phys. Rev. D. 70 (3) 0340001 (2004).
B.A. Bambah and C.Mukku, Annals of Physics 314, 34,(2004)
* B. A. Bambah , C. Mukku and K.V.S Shiv Chaitanya Dynamics of the Dcc with Arbitrary Disorientation in Isospin Space( In preparation).
* T. Nayak ,Pramana 57 ,285,(2001).
* M.M. Aggarwal et al., e-print archive:nucl-ex/0012004.
Madan M Aggarwal, Pramana 60, No.5, 987 , (2003)
|
warning/0506/hep-lat0506016.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
In lattice simulations of QCD on a cubic volume ($`V=L^3`$) with periodic boundary conditions on the fields, the components of hadronic momenta $`p_i`$ are quantized in integer multiples of $`2\pi /L`$. For currently available lattices this implies that the lowest non-zero momentum is large, typically $`500\mathrm{MeV}`$ or so, and there are large gaps between neighbouring momenta. This limits the phenomenological reach of simulations, particularly for momentum dependent quantities such as the form-factors of weak semileptonic decays of hadrons. In ref. Bedaque proposed the use of twisted boundary conditions<sup>1</sup><sup>1</sup>1See the references cited in and for earlier related ideas. for the quark fields $`\psi `$
$$\psi (x_i+L)=e^{i\theta _i}\psi (x_i).$$
(1)
Twisted boundary conditions allow for simulations with arbitrary components of hadronic momenta. For example, the momentum of a meson composed of a quark with flavour $`1`$ satisfying boundary conditions with a twisting angle $`\stackrel{}{\theta }_1=(\theta _{11},\theta _{12},\theta _{13})`$ and an antiquark of flavour $`2`$ with angle $`\stackrel{}{\theta }_2`$ is
$$\stackrel{}{p}=\frac{2\pi }{L}\stackrel{}{n}\frac{\stackrel{}{\theta }_1\stackrel{}{\theta }_2}{L},$$
(2)
where $`\stackrel{}{n}`$ is a vector of integers.
The practical difficulty in using twisted boundary conditions in lattice simulations with dynamical quarks is that it requires the generation of a new set of gauge field configurations for every choice of twisting angle(s). In refs. it was shown that for many physical quantities one can use partially twisted boundary conditions, i.e. impose twisted boundary conditions for the valence quarks but periodic boundary conditions for the sea quarks, thus eliminating the need for new simulations for every choice of momentum and making the technique practicable. The physical quantities for which partially twisted boundary conditions can be applied include those with at most a single hadron in the initial and final states (and possibly even in intermediate states), for which the finite-volume effects decrease exponentially with the volume. For these processes the finite-volume effects depend on the twisting angle(s) but remain exponentially small.
For some processes with energies above a two-body threshold, such as $`K\pi \pi `$ decays with the two-pions in an isospin zero state, the finite-volume effects decrease only as powers of the volume and must be subtracted for acceptable precision to be reached. We are not able to perform these subtractions if partially twisted boundary conditions are used. Here we will only consider processes for which such a problem does not arise.
In this letter we confirm the theoretical results of ref. in a numerical study of partially twisted boundary conditions for dynamical, non-perturbatively improved Wilson fermions. In particular we find that:
* The energies of $`\pi `$ and $`\rho `$-mesons (with masses below the two-pion threshold) satisfy the expected dispersion relation
$$E_{\pi ,\rho }^2=m_{\pi ,\rho }^2+\left(\stackrel{}{p}_{\mathrm{lat}}\frac{\stackrel{}{\theta }_1\stackrel{}{\theta }_2}{L}\right)^2,$$
(3)
where $`\stackrel{}{\theta }_1`$ and $`\stackrel{}{\theta }_2`$ are the twisting angles of the two valence quarks and $`\stackrel{}{p}_{\mathrm{lat}}=(2\pi /L)\stackrel{}{n}`$ is the contribution to the meson’s momentum introduced by the Fourier transform of the correlation function.
This study extends the one in ref. where the dispersion relation for pseudo scalar mesons with twisted boundary conditions in the quenched approximation was found to be consistent with expectations.
* The values of the leptonic decay constants of $`\pi `$ and $`\rho `$ mesons and of the matrix element $`0|P|\pi `$ of the pseudo scalar density $`P`$ are independent of the twisting angles as expected.
A further reassuring result of our study is that twisted boundary conditions do not introduce additional noise in the data. As we increase the meson’s momentum by suitably varying the angles $`\stackrel{}{\theta }_{1,2}`$, the statistical errors on meson masses and matrix elements increase smoothly. However, when comparing results obtained with twisted and periodic boundary conditions with similar momenta (i.e. momenta close to $`2\pi /L`$ or $`\sqrt{2}(2\pi /L)`$) the errors are found to be comparable.
The plan of the remainder of this letter is as follows. In the next section we present the details of our computation, the parameters of the simulation (including the choice of twisting angles) and a description of the analysis. We present our results in sec. 3 and conclusions in sec. 4.
## 2 Details of the Simulation and Analysis
We study meson observables on sets of gauge configurations which were generated with two degenerate flavours of sea quarks using non-perturbatively improved Wilson fermions and the plaquette gauge action on the torus with periodic boundary conditions ($`\beta =5.2`$, $`a0.1\mathrm{fm}`$, $`c_{\mathrm{SW}}=2.0171`$, $`(L/a)^3\times T/a=16^3\times 32`$). We used the ensembles of field configurations which were studied in detail in and took over the suggested separation of measurements by $`40`$ trajectories in our analysis. The simulated quark masses are summarized in table 1. Propagators and correlators were calculated using the FermiQCD libraries . We stress that the aim of the present study is to investigate the consistency and effectiveness of using partially twisted boundary conditions at fixed values of the quark mass. We do not attempt to perform a chiral extrapolation.
For each flavour of valence quark we impose the boundary conditions in eq. (1) for a variety of twisting angles $`\stackrel{}{\theta }=(\theta _1,\theta _2,\theta _3)`$. When evaluating the corresponding propagators we make use of the change of quark field variables
$$\psi (x)=e^{i\frac{\stackrel{}{\theta }\stackrel{}{x}}{L}}\stackrel{~}{\psi }(x),$$
(4)
where $`\stackrel{~}{\psi }(x)`$ satisfies periodic boundary conditions. The phase factor cancels in all terms of the lattice fermion action except for the spatial hopping terms which now become (for $`i=1,2,3`$)
$$\overline{\stackrel{~}{\psi }}(x)\left[e^{i\frac{a\theta _i}{L}}U_i(x)(1\gamma _i)\stackrel{~}{\psi }(x+\widehat{i})+e^{i\frac{a\theta _i}{L}}U_i^{}(x\widehat{i})(1+\gamma _i)\stackrel{~}{\psi }(x\widehat{i})\right].$$
(5)
In practice therefore, the partially twisted quark propagator can be computed by inverting the standard improved Wilson-Dirac operator in a gauge field background where the link variables {$`U_i(x)`$} have been replaced by {$`e^{i\frac{a\theta _i}{L}}U_i(x)`$}.
The physical observables which we study in this letter are the energies and leptonic decay constants of the pseudo scalar and vector mesons and the matrix element of the pseudo scalar density. In order to determine these, we compute the following correlation functions:
$`C_{A_0P}(t,\stackrel{}{p})`$ $`=`$ $`{\displaystyle \underset{\stackrel{}{x}}{}}e^{i\stackrel{}{p}_{\mathrm{lat}}\stackrel{}{x}}0|A_0^I(\stackrel{}{x},t)P^{}(0)|0,`$ (6)
$`C_{PP}(t,\stackrel{}{p})`$ $`=`$ $`{\displaystyle \underset{\stackrel{}{x}}{}}e^{i\stackrel{}{p}_{\mathrm{lat}}\stackrel{}{x}}0|P(\stackrel{}{x},t)P^{}(0)|0,`$ (7)
$`C_{A_0A_0}(t,\stackrel{}{p})`$ $`=`$ $`{\displaystyle \underset{\stackrel{}{x}}{}}e^{i\stackrel{}{p}_{\mathrm{lat}}\stackrel{}{x}}0|A_0^I(\stackrel{}{x},t)(A_0^I(0))^{}|0,`$ (8)
$`C_{V_iV_i}(t,\stackrel{}{p})`$ $`=`$ $`{\displaystyle \underset{\stackrel{}{x}}{}}e^{i\stackrel{}{p}_{\mathrm{lat}}\stackrel{}{x}}0|V_i^I(\stackrel{}{x},t)(V_i^I(0))^{}|0\text{(no sum on }i\text{)},`$ (9)
where $`P(x)`$ is the pseudo scalar density
$$P(x)=\overline{\psi }_2(x)\gamma _5\psi _1(x)$$
(10)
for quarks of flavour $`1`$ and $`2`$ (with twisting angles $`\stackrel{}{\theta }_1`$ and $`\stackrel{}{\theta }_2`$), and $`V_\mu ^I(x)`$ and $`A_\mu ^I(x)`$ are the improved vector and axial-vector currents
$`V_\mu ^I(x)`$ $`=`$ $`\overline{\psi }_2(x)\gamma _\mu \psi _1(x)+ac_V(g_0){\displaystyle \frac{1}{2}}(_\nu ^{}+_\nu )\overline{\psi }_2(x)\sigma _{\mu \nu }\psi _1(x)`$
$`A_\mu ^I(x)`$ $`=`$ $`\overline{\psi }_2(x)\gamma _\mu \gamma _5\psi _1(x)+ac_A(g_0){\displaystyle \frac{1}{2}}(_\mu ^{}+_\mu )P(x).`$
Here, $`_\mu `$ and $`_\mu ^{}`$ are the forward and backward derivatives and $`c_V(g_0)`$ and $`c_A(g_0)`$ are improvement coefficients which we take from and respectively. Since we are primarily interested in the effects of twisted boundary conditions we do not attempt to compute the renormalization constants of $`P`$, $`V_\mu `$ and $`A_\mu `$, nor do we implement improvement factors of the form $`1+b(g_0)m_qa`$, where $`m_q`$ is the mass of the quark. The inclusion of these factors would of course be necessary if we were attempting to determine the physical leptonic decay constants. However, they are overall factors for each choice of quark mass and are independent of the twisting angles, while it is precisely the dependence on these angles which is the object of our study.
The momentum, $`\stackrel{}{p}`$, of the meson is given by
$$\stackrel{}{p}=\stackrel{}{p}_{\mathrm{lat}}\frac{\stackrel{}{\theta }_1\stackrel{}{\theta }_2}{L},$$
(11)
where $`p_{\mathrm{lat}}=(2\pi /L)\stackrel{}{n}`$ and $`\stackrel{}{n}`$ is a vector of integers.
At large values of $`t`$ the time dependences of (6)–(9) approach:
$`C_{A_0P}(t,\stackrel{}{p})`$ $``$ $`{\displaystyle \frac{1}{E_\pi }}Z_PM_0(\stackrel{}{p})e^{E_\pi T/2}\mathrm{sinh}((tT/2)E_\pi ),`$ (12)
$`C_{PP}(t,\stackrel{}{p})`$ $``$ $`{\displaystyle \frac{1}{E_\pi }}Z_P^2e^{E_\pi T/2}\mathrm{cosh}((tT/2)E_\pi ),`$ (13)
$`C_{A_0A_0}(t,\stackrel{}{p})`$ $``$ $`{\displaystyle \frac{1}{E_\pi }}M_0^2(\stackrel{}{p})e^{E_\pi T/2}\mathrm{cosh}((tT/2)E_\pi ),`$ (14)
$`C_{V_iV_i}(t,\stackrel{}{p})`$ $``$ $`{\displaystyle \frac{1}{E_\rho }}N_i^2(\stackrel{}{p})e^{E_\rho T/2}\mathrm{cosh}((tT/2)E_\rho )(i=1,2,3),`$ (15)
where, for each choice of quark masses, we have denoted the lightest pseudo scalar and vector mesons by $`\pi `$ and $`\rho `$ respectively and $`E_\pi `$ and $`E_\rho `$ are the corresponding energies which we expect to satisfy the dispersion relations in eq. (3). The notation for the matrix elements is as follows:
$`Z_P`$ $`=`$ $`0|P(0)|\pi (\stackrel{}{p}),`$ (16)
$`M_0(\stackrel{}{p})`$ $`=`$ $`0|A_0(0)|\pi (\stackrel{}{p})=f_\pi E_\pi ,`$ (17)
$`N_i^2(\stackrel{}{p})`$ $`=`$ $`{\displaystyle \underset{\lambda }{}}|\mathrm{\hspace{0.17em}0}|V_i(0)|\rho (\stackrel{}{p},\lambda )|^2=f_\rho ^2m_\rho ^2\left(1+{\displaystyle \frac{p_i^2}{m_\rho ^2}}\right)`$ (18)
where the index $`\lambda `$ labels the $`\rho `$-meson’s polarization state.
In this letter we study the validity of the dispersion relation in eq. (3) and the independence of $`f_\pi `$, $`f_\rho `$ and $`Z_P`$ of the momentum. We evaluate the quark propagators for four values of the twisting angle $`\stackrel{}{\theta }`$:
$$\stackrel{}{\theta }=\stackrel{}{0},(2,0,0),(0,\pi ,0)\text{and}(3,3,3).$$
(19)
For each value of $`\kappa _{\mathrm{val}}`$, quark and antiquark propagators with all possible pairs $`\stackrel{}{\theta }_1`$ and $`\stackrel{}{\theta }_2`$ were combined to construct correlation functions for mesons with a variety of momenta. Moreover we also combined them with Fourier momenta $`\stackrel{}{p}_{\mathrm{lat}}=(0,\pm \mathrm{\hspace{0.17em}2}\pi /L,0)`$ to increase the range of momenta which can be reached. When presenting our results in the following section, we include for comparison results without twisting ($`\stackrel{}{\theta }_1=\stackrel{}{\theta }_2=0`$), obtained by averaging over the $`12`$ equivalent momenta with $`|\stackrel{}{p}_{\mathrm{lat}}|=\sqrt{2}\times 2\pi /L`$ and those obtained by averaging over the eight equivalent momenta with $`|\stackrel{}{p}_{\mathrm{lat}}|=\sqrt{3}\times 2\pi /L`$. Of course this averaging reduces the statistical errors and this should be borne in mind when comparing the errors at these untwisted momenta with those at momenta with $`\stackrel{}{\theta }_1\stackrel{}{\theta }_2\stackrel{}{0}`$ for which such averaging is not possible.
Applying the jackknife procedure to the data for the correlation functions in eqs. (6)–(8), we have extracted all observables in the pseudo scalar channel from a combined non-linear $`\chi ^2`$ fit to the functional form suggested by (12)–(14), (16) and (17). The fit-ranges were chosen to yield compatible results under variation of the range by at least one unit in $`t/a`$. We applied the same procedure in the vector channel, combining the data for the correlation function (9) for $`i=1,2`$ and $`3`$ in one fit using the expressions in eqs. (15) and (18).
## 3 Results
The series of plots in figures 1 and 3 show our data as a function of $`(\stackrel{}{p}L)^2`$ in the range $`|\stackrel{}{p}L|[0,\sqrt{3}\times 2\pi ]`$. Fig. 1 contains the results for the energies as a function of momentum and fig. 3 those for the decay constants and $`Z_P`$. To ease orientation, the positions of the discrete Fourier momenta $`|\stackrel{}{p}_{\mathrm{lat}}L|=0`$, $`2\pi `$, $`\sqrt{2}\times 2\pi `$ and $`\sqrt{3}\times 2\pi `$ are indicated by dashed vertical lines. We emphasize that it is only at these values of momenta that one can obtain results using periodic boundary conditions. In fig. 2 we zoom into the region $`|\stackrel{}{p}L|2\pi `$ for the dispersion relations. In this region we would expect lattice artefacts to be small and the use of twisted boundary conditions to be particularly useful.
In each plot, the (blue) triangles correspond to points in which the correlation function was evaluated with $`\stackrel{}{p}_{\mathrm{lat}}=\stackrel{}{0}`$, but with all possible pairs of $`\stackrel{}{\theta }_1`$ and $`\stackrel{}{\theta }_2`$ from the set in (19). The (red) diamonds and (green) squares represent the results obtained with $`\stackrel{}{p}_{\mathrm{lat}}=(0,2\pi /L,0)`$ and $`\stackrel{}{p}_{\mathrm{lat}}=(0,2\pi /L,0)`$ respectively, combined with all possible pairs of $`\stackrel{}{\theta }_1`$ and $`\stackrel{}{\theta }_2`$. The four points with $`\stackrel{}{\theta }_1=\stackrel{}{\theta }_2=\stackrel{}{0}`$ with $`|\stackrel{}{p}_{\mathrm{lat}}|=0,2\pi /L,\sqrt{2}\times 2\pi /L`$ and $`\sqrt{3}\times 2\pi /L`$ are denoted by (black) circles.
For the discussion of our results it is convenient to rewrite the dispersion relation in eq. (3) in the form
$$(aE_{\pi /\rho })^2=(am_{\pi /\rho })^2+\mathrm{\Delta }^2(\stackrel{}{p}L)^2$$
(20)
where $`\mathrm{\Delta }^2=(a/L)^2=0.0039`$. The dispersion relation (20) is displayed as the dashed line in the plots of fig. 1. In the first row of table 2 we present the $`\chi ^2/`$d.o.f of the comparison of our data to eq. (20) over the range $`0|\stackrel{}{p}|^2L^2(2\pi )^2`$ using the values of the meson masses obtained from fits at zero momenta. In the third row of the table we present the values of $`\mathrm{\Delta }^2`$ obtained by fitting the lattice data to the functional form in eq. (20) over the same range in momentum, but allowing $`\mathrm{\Delta }^2`$ to be a parameter of the fit. We note that our values for the ratios $`m_\pi /m_\rho `$ agree with those found earlier on the same configurations in .
As the momentum of the meson grows so do the expected discretization effects in the dispersion relation. For example, a free scalar particle with a wave-function $`\varphi (x)`$ satisfying the (Minkowski-Space) Klein-Gordon equation $`(\mathrm{}+m^2)\varphi (x)=0`$, with a generic discretized second derivative defined by $`^2f(x)/x^2=(f(x+a)+f(xa)2f(x))/a^2`$, satisfies the following lattice dispersion relation:
$$\mathrm{sinh}^2\left(\frac{aE_{\pi /\rho }}{2}\right)=\mathrm{sinh}^2\left(\frac{am}{2}\right)+\underset{i}{}\mathrm{sin}^2\left(\mathrm{\Delta }\frac{p_iL}{2}\right).$$
(21)
To indicate the possible size of lattice artefacts we plot the dispersion relation of eq. (21) as the solid curve in fig. 1. We stress, however, that in an interacting theory the discretization errors will in general be different from those in eq. (21). Indeed there seems to be no evidence from our data that eq. (21) is a particularly good representation of the lattice artefacts (see the second row of table 2, where we show the $`\chi ^2/`$d.o.f. from a comparison of our data with (21)). At small momenta the solid curve merges of course with the dashed line representing the continuum dispersion relation.
We conclude from the results for the dispersion relation plotted in figs. 1 and 2 that the use of partially twisted boundary conditions is beautifully consistent with expectations, particularly at low momenta where the lattice artefacts are small<sup>2</sup><sup>2</sup>2The different data points at $`|\stackrel{}{p}L|=0`$ and $`2\pi `$ correspond to $`\stackrel{}{\theta }_1=\stackrel{}{\theta }_2`$ but with different choices of $`\stackrel{}{\theta }_1`$ and $`\stackrel{}{\theta }_2`$.. An important further observation is that there is no evidence in our data that the introduction of twisted boundary conditions increases significantly the statistical or systematic uncertainties. This is illustrated in the second line of figure 1, which shows the relative error in the pion energy
$$\delta _{E_\pi }\frac{\delta E_\pi }{E_\pi }$$
(22)
as $`|\stackrel{}{p}L|`$ is varied. The error $`\delta E_\pi `$ is the jackknife error, including the statistical error and the systematic uncertainty stemming from the improvement constants in the improved quark currents. The plot shows that the errors increase smoothly as the momentum increases, and that no appreciable additional noise is introduced by partial twisting <sup>3</sup><sup>3</sup>3For $`\kappa =0.13550`$ at $`|\stackrel{}{p}L|=2\pi /L`$ we observe a fluctuation in the effective mass from one of the gauge configurations and with our choice of the position of the source (this fluctuation has also been observed by our colleagues in the UKQCD collaboration ). The fluctuation is particularly noticeable when twisting both quarks by $`\stackrel{}{\theta }=(3,3,3)`$ and this is the reason for the larger jackknife error at this particular momentum and twist combination (see figure 1).. We observe the same behaviour for all analyzed quantities.
In fig. 3 we plot our results for the decay constants $`f_\pi `$ and $`f_\rho `$ and for $`Z_P`$. The values for $`af_\pi `$ agree with the ones obtained in at $`|\stackrel{}{p}L|=0`$. Again we see that the results are completely consistent with theoretical expectations, being independent of the twisting angles and Fourier momenta.
## 4 Conclusions
We have investigated the use of partially twisted boundary conditions in evaluating the energies of pseudo scalar and vector mesons and their leptonic decay constants. The results are very encouraging; it does appear that the method allows the evaluation of physical quantities with any momentum. Moreover the use of these boundary conditions does not appear to increase the errors in any appreciable way. It will be important to monitor whether this continues to be true as the quark masses are decreased. Once the quark masses are such that two-pion intermediate states contribute significantly to the $`\rho `$-meson’s correlation function the finite-volume effects will no longer fall exponentially with the volume, but only as powers. For the pion observables studied in this paper this is not the case.
Partially twisted boundary conditions will be particularly useful for evaluating momentum-dependent physical quantities. One important application is to the determination of the form-factors of semileptonic weak decays of heavy ($`D`$ and $`B`$) mesons to light mesons. For these processes, with conventional periodic boundary conditions, the initial and final state hadrons are restricted to have momenta $`(2\pi /L)\stackrel{}{n}`$ where $`\stackrel{}{n}`$ is a vector of integers. In order to avoid lattice artefacts the possible values of $`|\stackrel{}{n}|`$ are frequently limited to $`0`$, $`1`$, and perhaps $`\sqrt{2}`$. Thus, for any particular choice of quark masses, the number of values of the momentum transfer, $`q^2`$, or the light meson energy, $`E`$, is also very limited. Moreover, chiral extrapolations are conveniently performed at fixed $`q^2`$ or fixed $`E`$ (and heavy quark extrapolations at fixed $`E`$), but $`q^2`$ and $`E`$ vary with both the momentum and quark masses. Ansätze for the form factors, such as the Becirevic-Kaidalov model, are used to interpolate and extrapolate simulation data to sets of common $`q^2`$ or $`E`$ values before the extrapolations are performed. Using twisted boundary conditions would enable the form factors to be evaluated directly at these common values, removing the need for the intermediate form-factor fit.
For some other physical quantities, such as the moments of hadronic deep inelastic structure functions or light-cone distribution amplitudes, it may be helpful to use twisted boundary conditions even though it is not strictly necessary. The corresponding matrix elements are proportional to factors of $`p_i`$, where $`\stackrel{}{p}`$ is the momentum of the hadron, so that the correlation functions must be computed with $`\stackrel{}{p}0`$. The use of twisted boundary conditions allows $`|\stackrel{}{p}|`$ to be decreased and hence the lattice artefacts to be reduced. Moreover by varying $`\stackrel{}{p}`$ one can verify that the leading twist component has been extracted correctly.
Following the successful conclusion of this exploratory numerical study of the implementation of partially twisted boundary conditions we now look forward to applying them in lattice computations of a wide variety of phenomenologically important quantities.
Acknowledgements We warmly thank Daragh Byrne, Steve Downing and Craig McNeile for their support with QCDgrid and Massimo di Pierro for help using FermiQCD. We thank the Iridis parallel computing team at the University of Southampton, in particular Oz Parchment and Ivan Wolton, for their assistance. We also acknowledge Alan Irving, Craig McNeile and Chris Michael for correspondence on the gauge field configurations. This work was supported by PPARC grants PPA/G/S/2002/00467 and PPA/G/O/2002/00468.
|
warning/0506/astro-ph0506376.html
|
ar5iv
|
text
|
# The Remnants of Intergalactic Supernovae
## 1. Introduction
Studies of nearby clusters have revealed a population of intergalactic stars. Already noticed by Zwicky (1951) as excess starlight between the galaxies in the core of the Coma cluster, this diffuse stellar emission has been confirmed and quantified in recent deep images of Coma (Gregg & West 1998; Trentham & Mobasher 1998; Feldmeier et al. 2002), in other nearby clusters (Calcáneo-Roldán et al. 2000), and in stacked images of redshift $`z0.25`$ clusters from the Sloan Digital Sky Survey (Zibetti et al. 2005). Intergalactic red giant stars have been detected in the Virgo Cluster (Ferguson, Tanvir, & von Hippel 1998; Durrell et al. 2002), and intergalactic planetary nebulae have been found in Virgo and Fornax (Arnaboldi et al. 1996; Theuns & Warren 1997; Mendez et al. 1997; Ciardullo et al. 1998, 2002; Feldmeier, Ciardullo, & Jacoby 1998). Some 10–20% of the stars in galaxy clusters are in the intergalactic component. This population is believed to have been stripped off the cluster galaxies through tidal disruption by other galaxies and by the cluster potential as a whole (Dubinski, Mihos, & Hernquist 1996; Moore et al. 1996; Korchagin, Tsuchiya, & Miyama 2001).
In the course of a survey for supernovae (SNe) in rich galaxy clusters at redshifts $`0.08<z<0.2`$, Gal-Yam et al. (2003) recently discovered two type-Ia SNe at the redshifts of their respective clusters, but spatially and and kinematically distinct from any galaxy in the cluster. The two events, constituting 2/7 of the cluster SNe found in the survey, had no detectable host galaxy, even in deep images taken by the Keck 10 m telescope. Gal-Yam et al. argued, based on the galaxy luminosity function of clusters, that dwarf-galaxies below the detection limit, and contributing only $`10^3`$ of the cluster stellar luminosity, could not plausibly be the hosts of the two SNe. Accounting for the relative detection efficiencies of events within and outside galaxies, Gal-Yam et al. estimated that $`21_{14}^{+18}`$ percent of the SN Ia parent stellar population in clusters is intergalactic. This fraction is consistent with the intergalactic stellar fraction found by other tracers.
Quantifying the properties of the intergalactic stellar population via its different tracers is important for understanding galaxy interactions and evolution in dense environments. SNe are particularly useful because, as opposed to other tracers, they can be seen out to clusters at large lookback times, and can thus reveal the history of the galaxy evolution process.
Supernovae are brief optical events. Supernova remnants (SNRs), however, exist for thousands of years and are detectable over a wide range of wavelengths. The intergalactic SN population in clusters could hence potentially be detected and characterized by means of the SNRs it leaves behind. In this paper, we predict the observational signatures of SNRs in the unusual intracluster medium (ICM) environment. We show that the SN ejecta is visible via its interaction with the surrounding circumstellar medium (CSM), if the latter exists, over a timescale of $`10^2`$$`10^3`$ years. In galactic environments, after traversing the CSM, the ejecta interacts with a galaxy’s interstellar medium (ISM) and the SNR remains visible for $`10^4`$ yr, or more. However, in cluster environments, once the ejecta traverses the CSM and enters into the hot and tenuous intracluster medium (ICM), the SNR emission fades. We will thus argue that the emission from the remnants of intergalactic type-Ia SNe is detectable only if they are surrounded by a dense CSM (implying progenitors with a giant-star companion that had a high mass loss rate; see review by Branch et al. 1995), and if the remnants are not much older than a thousand years. A CSM has been detected in only one type-Ia SN, 2002ic, where hydrogen lines have appeared in the late-time spectra of the explosion (Hamuy et al. 2003). However, it is unclear if this single event is representative of the physical conditions of most SNe-Ia which (by definition) have no signs of hydrogen in their spectra. Thus, detection of intergalactic SN-Ia remnants could demonstrate the existence of a CSM around type-Ia SN progenitors.
In §2 we use some simple physical arguments to derive the main observed features of SNRs and their dependence on progenitor properties and on the peculiar ICM environment. In §3, we review the characteristics of the unresolved intergalactic emission-line objects recently discovered in nearby clusters and groups, and discuss whether some or all of these objects could be intergalactic SNRs, rather than H II regions. In §4, we examine whether radio observations could detect the intergalactic SNR population. We summarize our results in §5.
## 2. The appearance of a SNR in the ICM
### 2.1. Wind dynamics
We will construct a simple physical model for SNRs that reproduces their main observed features, and will allow prediction of the properties of intergalactic SNRs. For a detailed treatment of the physics of SNRs, see, e.g., McCray & Wang(1996), or Truelove & McKee (1999).
We consider a type-Ia SN explosion due to a white dwarf that exceeds the Chandrasekhar mass through accretion of wind material from a giant star companion. The giant is assumed to have a mass loss rate of $`\dot{M}=10^6\dot{M}_6M_{}\mathrm{yr}^1`$, which creates a CSM with which the SN ejecta eventually collides. Only later does the ejecta reach the external medium, be it the ISM in the case of normal galactic SNe, or the ICM for the intracluster SNe under discussion.
First, we consider the case where the binary star system is at rest relative to an ambient ICM with a pressure $`p_{\mathrm{ICM}}`$. In this case, the entire mass lost by the giant star accumulates in a shell around the binary. This case applies when the speed of the binary system relative to the ICM, $`v_{}`$, is much smaller than the giant star’s wind speed, $`v_w=10^{1.5}v_{w,1.5}\mathrm{km}\mathrm{s}^1`$.
The outgoing wind acts as a piston and generates a forward shock in the external medium (e.g., Parker 1963). Since the wind velocity is typically much smaller than the sound speed of the surrounding ICM, the forward shock is weak. The interaction with the ICM sends a reverse shock back into the cold wind material. The shell of material trapped between the forward and reverse shocks includes wind material separated from shocked ICM material by a contact discontinuity. If this shell is moving forward with a velocity $`v`$, then the reverse shock moves backward relative to the unperturbed wind at a speed of,
$$v_s=\frac{(\gamma +1)}{2}\left(v_wv\right),$$
(1)
where $`\gamma =5/3`$ is the adiabatic index of a monoatomic gas. The pressure in this shell satisfies
$$p_s=\frac{2}{(\gamma +1)}\rho _wv_s^2p_{\mathrm{ICM}},$$
(2)
implying a reverse shock speed relative to the ICM:
$$v_{\mathrm{rs}}=(v_wv_s)=v_w\left(1\sqrt{\frac{(\gamma +1)p_{\mathrm{ICM}}}{2\rho _wv_w^2}}\right).$$
(3)
Here, $`\rho _wR^\eta `$ is the density of the wind material, with $`\eta =2`$ if the star had a constant mass loss rate. For such a density profile, the reverse shock turns into a standing shock in the observer’s frame at the radius $`R=R_{\mathrm{rs}}`$ where $`v_{\mathrm{rs}}=0`$, namely $`v_w=v_s`$. Setting $`v_{\mathrm{rs}}=0`$ in equation (3), we have
$`R_{\mathrm{rs}}`$ $`=`$ $`\left({\displaystyle \frac{\dot{M}v_w}{2\pi (\gamma +1)p_{\mathrm{ICM}}}}\right)^{1/2}`$ (4)
$`=0.6\times 10^{18}\dot{M}_6^{1/2}v_{w,1.5}^{1/2}n_3^{1/2}T_1^{1/2}\mathrm{cm},`$
where $`\rho _wn_wm_p=\dot{M}/4\pi R^2v_w`$ is the mass density in the wind (with $`m_p`$ being the mean particle mass), $`T_{\mathrm{ICM}}=10T_1\mathrm{keV}`$ is the ICM temperature, and $`n_{\mathrm{ICM}}=10^3n_3\mathrm{cm}^3`$ is the electron density in the ICM (with $`p_{\mathrm{ICM}}2n_{\mathrm{ICM}}kT_{\mathrm{ICM}}`$). At this point the wind velocity is reduced to
$$v=\left(\frac{\gamma 1}{\gamma +1}\right)v_w=\frac{1}{4}v_w.$$
(5)
The density profile beyond $`R_{\mathrm{rs}}`$ can be derived by the following argument. Once a wind fluid element crosses the reverse shock, its entropy is conserved and its pressure is approximately time independent and equal to the ICM pressure. This implies that the density of the fluid is independent of time and radius at $`R>R_{\mathrm{rs}}`$, $`\rho _w(R>R_{\mathrm{rs}})=4\dot{M}/4\pi R_{\mathrm{rs}}^2v_w=2(\gamma +1)p_{\mathrm{ICM}}/v_w^2`$,
$$\rho _w(R>R_{\mathrm{rs}})=9.2m_p\frac{n_3T_1}{v_{w,1.5}^2}\mathrm{cm}^3.$$
(6)
The radius independent density implies a deceleration of the wind velocity beyond $`R_{\mathrm{rs}}`$, $`v_w(R>R_{\mathrm{rs}})R^2`$. The outer radius of the wind is determined by mass conservation, $`4\pi R_{\mathrm{max}}^3\rho _w(R>R_{\mathrm{rs}})/3=\dot{M}\tau `$,
$$R_{\mathrm{max}}=\left(\frac{3}{4}R_{\mathrm{rs}}^2v_w\tau \right)^{1/3}=3.0\times 10^{18}\left(\frac{\dot{M}_6v_{w,1.5}^2\tau _6}{n_3T_1}\right)^{1/3}\mathrm{cm}.$$
(7)
Here, $`\tau =10^6\tau _6\mathrm{yr}`$ is the duration of the period over which the wind is active before the SN explodes. For typical parameters, most of the wind material is trapped between this radius and the stationary radius of the reverse shock, $`R_{\mathrm{rs}}`$.
In the ICM of massive clusters, an intergalactic stellar binary will typically move at a high speed of $`v_{\mathrm{star}}10^3\mathrm{km}\mathrm{s}^1`$ relative to the ICM. Under these circumstances, the wind remnant will be stripped by the ram pressure of the ICM in its rest frame. We define an effective ”ram-pressure” speed, $`v_{}=\sqrt{v_{\mathrm{star}}^2+v_{\mathrm{th}}^2}=10^3v_{,3}\mathrm{km}\mathrm{s}^1`$, where $`v_{\mathrm{th}}^2=p_{\mathrm{ICM}}/\rho _{\mathrm{ICM}}`$ is the thermal speed of the ICM gas. The wind remnant (or equivalently, the CSM) will have the same structure as before, except that the ICM temperature, $`T`$, is increased by a factor $`(v_{}/v_{\mathrm{th}})^2`$ to $`T_{}=(1+v_{\mathrm{star}}^2/v_{\mathrm{th}}^2)T`$. This leads to modification of the stalling radius of the reverse shock, $`R_{\mathrm{rs}}`$, of $`R_{\mathrm{max}}`$, and of $`\rho _w(R>R_{\mathrm{rs}})`$. A bow-shock structure will form, with these modified parameters describing its forward region. In the other directions, there will be larger radii and lower densities, but only by factors of order unity.
We will assume that the CSM gas is neutral, both in front and behind the reverse shock, based of the following arguments. First, we consider the effect of the reverse shock on the CSM gas that has passed through it. Since the pressure of this gas approximately equals the ICM pressure (Eq. 2), then (Eq. 6) implies that the post-reverse-shock CSM temperature is
$$T_{\mathrm{CSM}}10^4Kv_{w,1.5}^2.$$
(8)
Thus, the relatively mild reverse shock may ionize the gas at $`R>R_{\mathrm{rs}}`$, but will not heat it excessively. If the gas gets ionized, its recombination time will be
$$t_{\mathrm{rec}}1.5\times 10^4\mathrm{yr}v_{w,1.5}^2n_3^1T_1^1,$$
(9)
much shorter than the $`10^6`$ yr duration of the wind phase. Thus, by the time the SN explodes, the CSM will have become neutral again.
Second, the radiation from the ICM also cannot photoionize the CSM. The photoionization cross section of neutral hydrogen declines with increasing frequency as $`\nu ^3`$ above the ionization threshold, while the optically thin thermal bremsstrahlung flux from the ICM, per unit frequency, is independent of frequency for $`h\nu kT`$. Thus, only photons with energy within a factor of two above 13.6 eV contribute to ionizations. For the fiducial ICM parameters, $`n_3=1`$ and $`T_1=1`$, and assuming a cluster core radius of $`300`$ kpc, the bremsstrahlung impinging on the CSM surface has a photon flux of $`4\times 10^3\mathrm{cm}^2\mathrm{s}^1`$ between 13.6 eV, and 27.2 eV (e.g., Rybicki & Lightman 1979). Multiplying by the ionization cross-section at 13.6 eV, $`6.3\times 10^{18}\mathrm{cm}^2`$, the ionization rate is $`2\times 10^{14}\mathrm{s}^1`$ per hydrogen atom. For the fiducial CSM density of $`10\mathrm{cm}^3`$ (Eq. 6), the recombination rate is $`2\times 10^{12}\mathrm{s}^1`$, 100 times higher than the photoionization rate, and therefore the CSM gas will remain mostly neutral. Finally, the ICM gas particles cannot collisionally heat and ionize the CSM because magnetic fields keep their mean free path small, thus suppressing heat conduction. This situation is actually observed in the case of “cold fronts”, clouds of relatively cool gas that maintain their lower temperature as they plunge through an X-ray cluster (e.g., Vikhlinin & Markevitch 2002). The magnetic fields also provide surface tension that prevents shearing of the cold gas clous via the Kelvin-Helmholtz instability.
### 2.2. Ejecta dynamics and line emission
Once the SN explodes, the fast SN ejecta expands and drives a strong shock into the CSM. By definition, the ejecta of type Ia explosions have a low abundance of hydrogen. The H$`\alpha `$ emission from their remnants originates from the neutral hydrogen intercepted by the ejecta’s forward shock in the surrounding medium. The underlying physics of H$`\alpha `$ emission at the young stage (when the shock is called a “non-radiative” shock, and the SNR is called “Balmer dominated”) is well understood (e.g., Raymond 1991). Of order one tenth of all hydrogen atoms entering the shock produce an H$`\alpha `$ photon before they get ionized, and roughly another tenth of the atoms do it by charge exchange with the hot electrons behind the shock. The first component generates a narrow line (since the emission occurs before the atoms get kicked by the shock), and the second results in a broad H$`\alpha `$ component with a velocity width of order the shock speed.
The total H$`\alpha `$ luminosity can be calculated from the radius of the forward shock, $`R_{\mathrm{fs}}`$,
$$L_{H\alpha }=4\pi R_{\mathrm{fs}}^2\frac{dR_{\mathrm{fs}}}{dt}\left(\frac{\rho _{\mathrm{CSM}}}{m_p}\right)\times 0.2\times h\nu _{H\alpha },$$
(10)
where $`h\nu _{H\alpha }=3\times 10^{12}`$ erg is the energy of an H$`\alpha `$ photon. The CSM mass density at radii $`R<R_{\mathrm{rs}}`$ is given by $`\rho _{\mathrm{CSM}}=\dot{M}/4\pi R^2v_w`$, yielding
$$L_{H\alpha }=\frac{\dot{R_{\mathrm{fs}}}}{v_w}\frac{\dot{M}}{m_p}0.2h\nu _{H\alpha }.$$
(11)
The forward shock velocity may be estimated as follows. Ignoring, at first, the interaction of the SN ejecta with the CSM, the time dependent ejecta velocity profile is given, after significant expansion of the ejecta, by $`v=R/t`$. This velocity profile describes “free expansion” (expansion into vacuum of a pressureless fluid), in which the velocity of each fluid element is independent of time. The density profile of such a flow is self-similar, $`\rho (R,t)g(R/t)t^3=g(v)t^3`$. Numerical models of SN-Ia explosions yield $`g(v)=\mathrm{exp}(v/v_{\mathrm{ej}})`$ (e.g., Höflich & Khokhlov 1996; Dwarkadas & Chevalier 1998). The characteristic ejecta speed, $`v_{\mathrm{ej}}`$, defines the ratio between the ejecta kinetic energy, $`E`$, and mass, $`M`$,
$$\frac{E}{M}=\frac{1}{2}\frac{𝑑RR^2\rho v^2}{𝑑RR^2\rho }=\frac{𝑑vv^4g(v)}{2𝑑vv^2g(v)}.$$
(12)
For $`g(v)=\mathrm{exp}(v/v_{\mathrm{ej}})`$ we have
$$v_{\mathrm{ej}}=\left(\frac{E}{6M_{\mathrm{ej}}}\right)^{1/2}=10^{3.5}\left(\frac{E_{51}}{M_{\mathrm{ej},0}}\right)^{1/2}\mathrm{km}\mathrm{s}^1,$$
(13)
where $`E=10^{51}E_{51}`$ erg and $`M_{\mathrm{ej}}=10^0M_{\mathrm{ej},0}M_{}`$ (see, e.g., Truelove & McKee 1999, for a summary of the values of these parameters in historical SNRs). The fraction of the ejecta mass which has a velocity in excess of $`v`$ is determined by the algebraic relation
$`{\displaystyle \frac{M(>v)}{M_{\mathrm{ej}}}}`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle _{v/v_{\mathrm{ej}}}}𝑑xx^2e^x`$ (14)
$`=`$ $`\left[1+\left(1+{\displaystyle \frac{1}{2}}{\displaystyle \frac{v}{v_{\mathrm{ej}}}}\right){\displaystyle \frac{v}{v_{\mathrm{ej}}}}\right]e^{v/v_{\mathrm{ej}}}.`$
The part of the ejecta with velocity $`>v`$ starts decelerating once it has propagated to a radius out to which the CSM mass is comparable to $`M(>v)`$. As long as $`R_{\mathrm{fs}}<R_{\mathrm{rs}}`$, the time at which this deceleration occurs is given by $`t=M(>v)v_w/\dot{M}v`$. Thus, the velocity of the “piston” driving the shock into the surrounding CSM is given, as a function of time, by
$$t=\frac{M_{\mathrm{ej}}v_w}{\dot{M}v_{\mathrm{ej}}}\left[\frac{v_{\mathrm{ej}}}{v}+\left(1+\frac{1}{2}\frac{v}{v_{\mathrm{ej}}}\right)\right]e^{v/v_{\mathrm{ej}}}.$$
(15)
Defining a characteristic time for deceleration,
$$t_{\mathrm{ej}}\frac{M_{\mathrm{ej}}v_w}{\dot{M}v_{\mathrm{ej}}}=1.1\times 10^4\frac{M_{\mathrm{ej},0}^{3/2}v_{w,1.5}}{\dot{M}_6E_{51}^{1/2}}\mathrm{yr},$$
(16)
the solution of equation (15) for $`t<t_{\mathrm{ej}}`$ is
$$v1.1\mathrm{ln}(2t_{\mathrm{ej}}/t)v_{\mathrm{ej}}.$$
(17)
Equations (15) and (16) hold for $`R_{\mathrm{fs}}<R_{\mathrm{rs}}`$. The shock first reaches this radius at $`t=t_{\mathrm{rs}}`$, given by
$$t_{\mathrm{rs}}\mathrm{ln}\frac{2t_{\mathrm{ej}}}{t_{\mathrm{rs}}}\frac{R_{\mathrm{rs}}}{1.1v_{\mathrm{ej}}}=58\left(\frac{\dot{M}_6v_{w,1.5}M_{\mathrm{ej},0}}{n_3T_1E_{51}}\right)^{1/2}\mathrm{yr}.$$
(18)
Since the forward shock is strong, its speed is simply related to the piston speed by $`(dR_{\mathrm{fs}}/dt)=\frac{\gamma +1}{2}v=\frac{4}{3}v`$, and consequently
$$L_{H\alpha }=3.0\times 10^{33}\frac{\dot{M}_6E_{51}^{1/2}}{v_{w,1.5}M_{\mathrm{ej},0}^{1/2}}\mathrm{ln}\frac{2t_{\mathrm{ej}}}{t}\mathrm{erg}\mathrm{s}^1$$
(19)
for $`t<t_{\mathrm{rs}}`$.
For $`t>t_{\mathrm{rs}}`$, $`R_{\mathrm{fs}}>R_{\mathrm{rs}}`$, the piston velocity and time are related by $`M(>v)=4\pi (vt)^3\rho _w(R>R_{\mathrm{rs}})/3=(4/3)\dot{M}(vt)^3/R_{\mathrm{rs}}^2v_w`$, i.e. by
$$\left(\frac{t}{t_{\mathrm{ej}}^{}}\right)^3=\left[\left(\frac{v_{\mathrm{ej}}}{v}\right)^3+\left(1+\frac{1}{2}\frac{v}{v_{\mathrm{ej}}}\right)\left(\frac{v_{\mathrm{ej}}}{v}\right)^2\right]e^{v/v_{\mathrm{ej}}},$$
(20)
with
$$t_{\mathrm{ej}}^{}\left(\frac{3M_{\mathrm{ej}}v_wR_{\mathrm{rs}}^2}{4\dot{M}v_{\mathrm{ej}}^3}\right)^{1/3}=360\left(\frac{M_{\mathrm{ej},0}^{5/2}v_{w,1.5}^2}{E_{51}^{3/2}n_3T_1}\right)^{1/3}\mathrm{yr}.$$
(21)
At $`tt_{\mathrm{ej}}^{}`$ we have $`vv_{\mathrm{ej}}`$. Here, we neglect the logarithmic corrections to $`v`$ implied by equation (20), since for a uniform $`\rho _{\mathrm{CSM}}`$ the time dependence of $`L_{H\alpha }`$, $`L_{H\alpha }R_{\mathrm{fs}}^2v\rho _{\mathrm{CSM}}vt^2`$, and the logarithmic evolution of $`v`$ is not important. We therefore obtain
$$L_{H\alpha }=2.8\times 10^{35}M_{\mathrm{ej},0}^{1/6}E_{51}^{1/2}\left(\frac{n_3T_1}{v_{w,1.5}^2}\right)^{1/3}\left(\frac{t}{t_{\mathrm{ej}}^{}}\right)^\beta \mathrm{erg}\mathrm{s}^1$$
(22)
where $`\beta =2`$ for $`R_{\mathrm{rs}}/v_{\mathrm{ej}}<t<t_{\mathrm{ej}}^{}`$ and $`\beta =1/5`$ for $`t_{\mathrm{ej}}^{}<t`$. At $`t_{\mathrm{ej}}^{}<t`$, equation (20) (which implies $`vv_{\mathrm{ej}}(t_{\mathrm{ej}}^{}/t)`$) no longer holds, and the shock approaches the self-similar Sedov-von Neumann-Taylor regime with $`R_{\mathrm{fs}}t^{2/5}`$.
Equation (22) holds up to the time where the ejecta traverses the CSM and enters the ICM. For $`v_{\mathrm{ej}}t_{\mathrm{ej}}^{}>R_{\mathrm{max}}`$ the ejecta suffers little deceleration prior to crossing the CSM, and hence the crossing time is given by $`t_{\mathrm{CSM}}R_{\mathrm{max}}/v_{\mathrm{ej}}`$. For $`v_{\mathrm{ej}}t_{\mathrm{ej}}^{}<R_{\mathrm{max}}`$, the ejecta enters the self-similar deceleration phase before crossing the CSM and $`t_{\mathrm{CSM}}(R_{\mathrm{max}}/v_{\mathrm{ej}})(R_{\mathrm{max}}/v_{\mathrm{ej}}t_{\mathrm{ej}}^{})^{3/2}`$. Since $`v_{\mathrm{ej}}t_{\mathrm{ej}}^{}/R_{\mathrm{max}}=(M_{\mathrm{ej}}/\dot{M}\tau )^{1/3}`$, we find
$`t_{\mathrm{CSM}}`$ $``$ $`{\displaystyle \frac{R_{\mathrm{max}}}{v_{\mathrm{ej}}}}\mathrm{max}[1,\left({\displaystyle \frac{\dot{M}\tau }{M_{\mathrm{ej}}}}\right)^{1/2}]`$ (23)
$`=`$ $`360\left({\displaystyle \frac{\dot{M}_6v_{w,1.5}^2\tau _6}{n_3T_1}}\right)^{1/3}\left({\displaystyle \frac{E_{51}}{M_{\mathrm{ej},0}}}\right)^{1/2}`$
$`\times `$ $`\mathrm{max}[1,\left({\displaystyle \frac{\dot{M}\tau }{M_{\mathrm{ej}}}}\right)^{1/2}]\mathrm{yr}.`$
This time may change by a factor of order unity, depending on the binary speed $`v_{\mathrm{star}}`$, and one could average over a Maxwellian distribution of possible stellar velocities to examine the statistics of SN events with different durations. Depending on the giant’s wind speed, the duration of mass loss, and the ejecta velocity, the time to traverse the CSM may vary in the range $`10^2`$$`10^3`$ years.
As long as the ejecta is interacting with the CSM, an intergalactic SNR is thus similar to a galactic SNR at this stage of its development. We can therefore test the applicability of our simple model by comparing its predictions to the observed properties of young ($`1000`$ yr) type-Ia SNRs in the Milky Way and in nearby galaxies. The remnants of the historical SNe of 1006 and of 1572 (Tycho) belong to this class. Observational estimates of sizes and luminosities for Galactic SNRs are complicated by uncertainties in distance and extinction, but in recent years this problem is being overcome by the detection of SNRs in nearby galaxies. In the Large Magellanic Cloud (LMC), the typical H$`\alpha `$ luminosities of young, Balmer-dominated, SNRs are of order $`10^{3435}`$ erg s<sup>-1</sup> (Tuohey et al. 1982; Smith et al. 1991). This is comparable to our estimate in equation (22). Note that the typical pressures, $`nT`$, of the ICM and the ISM are similar, and hence in view of the weak, 1/3 power, dependence on pressure, we expect similar luminosities at this stage from galactic SNRs and intergalactic SNRs exploding into a CSM.
In older (a few thousand to 20,000 years old) Galactic and nearby SNRs, the radii are typically 10-15 pc, and H$`\alpha `$ luminosities are in the range of $`10^{3537}`$ erg s<sup>-1</sup> (e.g., Blair & Long 2004; Williams et al. 2004). However, at this stage, the shocks, now called “radiative shocks”, have slowed down considerably. Cooling is dominated by hydrogen recombination and by collisionally excited lines of low-ionization metals, especially in regions where the shock encounters denser clumps with shorter recombination times. In fact, optical surveys often distinguish SNRs from H II regions based on an emission-line ratio criterion of \[S II\]$`\lambda \lambda 6717,6731`$/H$`\alpha >0.4`$, but of course, this will select against young, often Balmer-dominated, SNRs. It is possible that intergalactic SNRs can approach this stage of greater Balmer line luminosity and a metal-line-cooling spectrum in regions of higher pressure in the ICM (e.g., near the cluster center), or if the pre-explosion wind included clumps of dense material. More likely, however, the forward shock will reach $`R_{\mathrm{max}}`$ and enter the ICM when it is still fast and non-radiative. Since the ICM is fully ionized, the H$`\alpha `$ emission is expected to diminish as soon as the ejecta’s forward shock traverses the CSM. This is in contrast to SNRs in the ISM of galaxies, where the forward shock continues to intercept neutral atoms even beyond the CSM.
### 2.3. Continuum emission
The collisionless forward shock of the SN remnant is expected to accelerate electrons to relativistic energies by the Fermi mechanism (e.g., Blandford & Eichler 1987). In the presence of intracluster magnetic fields, the accelerated electrons are expected to emit non-thermal radio photons via the synchrotron process (see Chevalier & Raymond 1978, and references therein). Let us first consider the shock driven into the ICM, at the late SNR stage. In analogy with SNRs in galaxies, we assume that a fraction $`\xi _e`$ of the post-shock thermal energy is given to the relativistic electrons. For a strong shock, the accelerated electron number is expected to be distributed with Lorentz factor $`\gamma `$ as $`dN_e/d\gamma \gamma ^2`$ up to a maximum Lorentz factor, $`\gamma _{\mathrm{max}}`$. The adiabatic compression of the magnetic field in X-ray clusters is expected to produce a magnetic field strength of $`B10B_5\mu \mathrm{G}`$ in the gas behind the forward shock front. The accelerated electrons will emit synchrotron radiation at a frequency,
$$\nu =\gamma ^2\left(\frac{eB}{2\pi m_ec}\right)=0.3B_5\gamma _4^2\mathrm{GHz},$$
(24)
where $`\gamma _4=(\gamma /10^4)`$. The synchrotron cooling time of the electrons emitting at a frequency $`\nu `$ is $`t_{\mathrm{syn}}2.5\times 10^7\mathrm{yr}B_5^2\gamma _4^1`$. The synchrotron emission spans up to $`2\mathrm{ln}\gamma _{\mathrm{max}}`$ decades in frequency, and the electrons carry an equal amount of energy per logarithmic Lorentz factor interval. Assuming that the ejecta does not decelerate significantly during the preceding CSM crossing, which holds for $`\dot{M}\tau <M_{\mathrm{ej}}`$, the synchrotron luminosity is therefore given by
$$\nu L_\nu \left(\frac{\xi _e}{2\mathrm{ln}\gamma _{\mathrm{max}}}\right)\times \left(3\times \frac{1}{6}v_{\mathrm{ej}}^2\times \frac{4\pi }{3}R^3\rho _{\mathrm{ICM}}\right)t_{\mathrm{syn}}^1.$$
(25)
Substituting $`\gamma _{\mathrm{max}}10^{10}`$ and $`\xi _e0.05`$, based on SNR observations (e.g. Dyer et al. 2001, Ellison et al. 2001; for a discussion see Keshet et al. 2003), we find that during the passage of the ejecta through the surrounding ICM, but before the ejecta starts to decelerate, the radio luminosity is given by
$$\nu L_\nu 6\times 10^{32}\left(\frac{\nu }{1\mathrm{GHz}}\right)^{1/2}n_3v_{\mathrm{ej},3.5}^5t_4^3\mathrm{erg}\mathrm{s}^1,$$
(26)
where $`t_4=(t/10^4\mathrm{yr})`$. Equation (26) holds until the ejecta begins to decelerate, at
$$t_{\mathrm{dec}}10^4\left(\frac{M_{\mathrm{ej},0}}{n_3}\right)^{1/3}v_{\mathrm{ej},3.5}^1\mathrm{yr}.$$
(27)
The luminosities given in equation (26) are consistent with those measured in galactic radio SNRs – $`10^{3234}\mathrm{erg}\mathrm{s}^1`$ (e.g., Payne et al. 2004; Warren & Hughes 2004) – for which $`n_3101000`$, and $`t_40.11`$.
The continuum emission (accompanied by a sub-dominant inverse-Compton component of upscattered microwave background photons), extends to high frequencies, following equation (26), up to optical frequencies, where the cooling time of electrons becomes comparable to the dynamical time. At higher frequencies the luminosity is frequency independent, $`\nu L_\nu \nu ^0`$. The peak luminosity is achieved at the deceleration time, $`t=t_{\mathrm{dec}}`$, when the flux and spectrum are given by
$`\nu L_\nu `$ $``$ $`1.3\times 10^{36}M_{\mathrm{ej},0.6}^{2/3}v_{\mathrm{ej},3.5}^3n_3^{1/3}`$ (28)
$`\times `$ $`\left({\displaystyle \frac{\nu /10^{15}\mathrm{Hz}}{n_3^{2/3}v_{\mathrm{ej},3.5}^2M_{\mathrm{ej},0.6}^{2/3}}}\right)^\alpha \mathrm{erg}\mathrm{s}^1,`$
where $`\alpha =1/2`$ for $`\nu <10^{15}n_3^{2/3}v_{\mathrm{ej},3.5}^2M_{\mathrm{ej},0.6}^{2/3}\mathrm{Hz}`$ and $`\alpha =0`$ otherwise. This continuum emission can be searched for at optical and X-ray wavelengths.
Looking back now to the expansion of the ejecta through the CSM, the emission of synchrotron radiation from the forward collisionless shock driven into the wind depends on the fraction $`\xi _B`$ of post shock thermal energy carried by the magnetic field. Radio supernovae are commonly modelled assuming a near-equipartition magnetic field, $`\xi _B0.1`$ (e.g., Weiler et al. 1998). Under this assumption, the synchrotron emission during the CSM crossing phase is given by
$`\nu L_\nu 6`$ $`\times `$ $`10^{33}\mathrm{ergs}^1\left({\displaystyle \frac{\nu }{1\mathrm{GHz}}}\right)^{1/2}\xi _{B,1}^{3/4}\left({\displaystyle \frac{n_3T_1}{v_{\mathrm{w},1.5}^2}}\right)^{7/4}`$ (29)
$`\times `$ $`v_{\mathrm{ej},3.5}^{13/2}t_2^3\times \{\begin{array}{cc}1,\hfill & t>t_{\mathrm{rs}};\hfill \\ \left(\frac{R_{\mathrm{rs}}/v_{\mathrm{ej}}}{2t}\right)^{7/2}\left[\mathrm{ln}\left(\frac{2t_{\mathrm{ej}}}{t}\right)\right]^3,\hfill & t<t_{\mathrm{rs}}\hfill \end{array},`$
where $`\xi _B=0.1\xi _{B,1}`$ and $`t=10^2t_2`$ yr. The logarithmic factor cubed is due to the fact that part of the ejecta moves at a velocity $`v`$ faster than $`v_{ej}`$. Although this factor formally diverges at small $`t`$, $`v/v_{ej}`$ should not exceed a factor of 10, since the simulations for various models do not show shells moving at $`v>10v_{ej}`$ (e.g., Höflich & Khokhlov 1996; Dwarkadas & Chevalier 1998). Since $`v/v_{ej}\mathrm{ln}(2t_{\mathrm{ej}}/2)10`$ is obtained for $`t1`$ yr, equation (29) is not applicable at earlier times. For the fiducial parameters, the 1 GHz luminosity $`1`$ year after the explosion is $`\nu L_\nu 1.4\times 10^{35}`$ erg s<sup>-1</sup>. It should be kept in mind that, at this stage, the luminosity depends sensitively on the velocity distribution in the ejecta – if the maximum velocity is, say, $`5v_{ej}`$, then the luminosity will be be 10 times smaller.
Observationally, prompt or very early (few years) radio emission from type-Ia SNe has not been detected. Sramek & Weiler (1990; see Boffi & Branch 1995) have presented upper limits on the radio fluxes for several events at distances of $`20`$ Mpc, which translate to luminosity limits of $`\nu L_\nu 10^{35}`$ erg s<sup>-1</sup>. Three events were observed $`2`$ months after the explosion, but one event, SN1981B, was observed near optical maximum. A $`3\sigma `$ upper limit of 1mJy at 5 GHz was also obtained a week before optical maximum by Eck et al. (1995) for SN1986G in the nearby (4.2 Mpc; Tonry et al. 2001) galaxy NGC 5128 (Cen A), corresponding to $`\nu L_\nu <1.1\times 10^{35}`$ erg s<sup>-1</sup>. However 1986G was a peculiar and underluminous SN-Ia. These upper limits are comparable to the radio luminosities that we predict at an age $`1`$ yr. Given the freedom in input parameters to equation (29), the dependence of its range of applicability on the speed of the fastest ejecta, and the small number of observed cases, this does not yet pose a serious discrepancy. However, tighter upper limits on the radio flux from additional SNe-Ia would constitute independent evidence for the absence of a CSM. Similar conclusions were reached by Boffi & Branch (1995).
## 3. Have Optical Intergalactic SNRs Been Detected?
In the course of narrow-band imaging of the Virgo cluster in search of intracluster planetary nebulae, Gerhard et al. (2002) have recently discovered a detached, spatially unresolved, emission-line object, $`17`$ kpc in projection from the spiral galaxy NGC 4388. Optical spectroscopy of the object revealed line ratios, including \[S II\]$`\lambda \lambda 6717,6731`$/H$`\alpha =0.1`$, which are characteristic of H II regions. Based on this, Gerhard et al. (2002) classified this object as an intergalactic H II region, in which the line emission is powered by one or two O stars. The massive stars are presumably members of a young cluster that was formed in situ in the ICM as a result of a recent collision between galaxies. The H$`\alpha `$ equivalent width constrains the age of the star cluster to $`3`$ Myr, too short for the stars to have formed in NGC 4388 and to have traversed the large distance, although the similar radial velocities of the galaxy and the emission-line region do suggest that the two are associated. The physical mechanism that could have led to intergalactic star formation is unclear, but perhaps the galaxy collision caused stripping and compression of gas along tidal tails, which then fragmented into stars when the gas was already unbound from its original galaxy. Gerhard et al. (2002) found a total of 17 candidate objects of this type in their narrow-band data, with H$`\alpha `$+\[N II\] luminosities of order $`10^{37}`$ erg s<sup>-1</sup> (for an assumed Virgo distance of 17 Mpc; since the exact distances are unknown, the luminosities could be lower or higher by an order of magnitude).
Several, possibly related, objects have also been found by Ryan-Weber et al. (2004) in a narrow-band imaging survey around nearby galaxies. Again, the emission-line objects are at large projected distances from their associated galaxies, but at similar velocities. Two of these candidates have been spectroscopically confirmed to be low-redshift objects based on the detection of both H$`\alpha `$ and \[OIII\]$`\lambda 5007`$ (the signal-to-noise ratio was too low for detection of additional lines). One object is 33 kpc from the S0 galaxy NGC 1533 in the Doradus Group (distance 21 Mpc), and another is 19 kpc from the galaxies NGC 833 and NGC 835 in the compact group HCG 16 (distance 53 Mpc). The H$`\alpha `$ luminosities, and the deduced ages and numbers of ionizing stars are of the same order of magnitude as found by Gerhard et al. (2002) in Virgo.
While the intergalactic star-formation option is a possibility, we point out that, within current observational constraints, the emission line objects could also be the remnants of the intracluster type-Ia SN population discovered by Gal-Yam et al. (2003), the observational signatures of which we have estimated above. The angular scale occupied by a remnant at the distance of the Virgo cluster, $`d=10d_1`$Mpc, is
$$\theta \frac{R_{\mathrm{max}}}{d}=0\stackrel{}{\mathrm{.}}02\left(\frac{R_{18.5}}{d_1}\right),$$
(30)
i.e., unresolved by ground-based optical telescopes, but resolvable with the Hubble Space Telescope (HST) if $`R_{18.5}3`$. The H$`\alpha `$ luminosities of the putative intergalactic H II regions are similar to those of the more luminous SNRs in nearby galaxies (e.g., Blair & Long 2004). Nearby SNRs of this luminosity are at least several thousand years old, and are in their slow, radiative-shock, phase, with strong low-ionization metal lines, in addition to Balmer lines. We estimated above that, under typical conditions, intergalactic SNRs will not reach this phase. Furthermore, the object found by Gerhard et al. (2003) has detectable but relatively weak, metal lines. However, considering the distance uncertainties (which lead to an order of magnitude uncertainty in the luminosity), stretching the parameters in equation (22), and allowing for line emission from dense clumps in the giant’s wind, it cannot be excluded that this object is an intergalactic SNR. In the case of the objects studied by Ryan-Weber et al. (2004), the spectra have too low a signal-to-noise ratio to detect the metal lines, let alone perform the usual diagnostic tests distinguishing H II regions from SNRs.
In terms of numbers, Gerhard et al. (2003) estimate that there are $`10^3`$ similar emission-line objects in Virgo, but of order one-half, or even more, of these sources are likely background objects at high redshift (see Ryan-Weber et al. 2004). The measured type-Ia SN rate, per unit $`B`$-band stellar luminosity, in clusters and in elliptical galaxies (see summary in Gal-Yam et al. 2002) is $`R=0.2\pm 0.1h_{70}^2`$ SNu, where 1SNu$`=`$SN century$`{}_{}{}^{1}(10^{10}L_{B,})_{}^{1}`$, and $`h_{70}`$ is the Hubble parameter in units of 70 km s<sup>-1</sup> Mpc<sup>-1</sup>. The expected number of intergalactic SNRs in Virgo is therefore
$$N_{\mathrm{SNR}}RfLt150R_{0.2}f_{0.2}L_{12.5}t_3,$$
(31)
where $`R_{0.2}=R/(0.2\mathrm{SNu})`$, $`f=0.2f_{0.2}`$ is the intergalactic stellar fraction, $`L=3\times 10^{12}L_{B,}L_{12.5}`$ is the total stellar $`B`$-band luminosity of Virgo (Sandage et al. 1985; Trentham & Hodgkin 2002), and $`t=10^3\mathrm{yr}t_3`$ is the time during which a SNR emits H$`\alpha `$ via non-radiative shocks that encounter the CSM. The number of emission-line objects is therefore also consistent, to an order of magnitude, with the SNR option, if the outer CSM radius is large enough to keep the SNR bright for about one thousand years.
Thus, if the intergalactic emission-line objects are SNRs, high-angular-resolution imaging should reveal resolved, shell-like morphologies with diameters of order 3 pc. If these objects are the remnants of the intergalactic type-Ia SN population, this would establish the fact that SNe-Ia have a CSM at the time of explosion, with far-reaching implications for progenitor models.
We note that, based on the rate parameters above, there should be an intergalactic SN-Ia in Virgo about once per decade. However, such SNe may be missed because surveys for nearby SNe (e.g., Li et al. 2003) monitor individual galaxies, rather than the entire cluster. A candidate intergalactic type-Ia SN that went off in Virgo is SN1980I (Smith et al. 1981), which occurred in between three elliptical/S0 galaxies, but was separated in projection by $`50`$ kpc from each. Twenty-five years after the explosion, the SNR should be about halfway to its CSM reverse-shock crossing stage, with an H$`\alpha `$ luminosity of $`10^{34}`$ erg s<sup>-1</sup>, i.e., an H$`\alpha `$ flux of $`3\times 10^{19}`$ erg s<sup>-1</sup> cm<sup>-2</sup>. Detection of such a weak line will be possible with the next generation of large ($`30`$ m) telescopes.
## 4. Intergalactic SNRs in the Radio
At the distance of the Virgo cluster of $`10\mathrm{Mpc}`$, the peak radio luminosities of intergalactic SNRs, lasting of order $`10^4t_4`$ yr, translate to fluxes of $`0.1`$mJy at $`\nu 1`$GHz. Considering the intergalactic SN rates above (Eq. 31), there should be of order $`10^3t_4`$ radio SNRs visible above this flux limit in Virgo, or about 10 SNRs per square degree. This is much smaller than the surface density of the background radio source population. A deep VLA survey of the Hubble Deep Field by Richards (2000) showed that the mean density of background radio sources brighter than 0.1 mJy is about 3000 per square degree. He found that about one half of the background sources were spatially extended, above the $`2^{\prime \prime }`$ resolution limit. Richards et al. (1998) found that the majority of the radio sources can be associated with luminous galaxies at redshifts $`z0.11`$, with mean $`R`$-band optical AB magnitudes of $`22`$, and generally brighter than 24.5 mag. Star-forming disk galaxies and AGNs (generally in early-type galaxies) both contribute to the source counts at these fluxes, although the relative contribution of each class is unclear.
Intergalactic SNRs in Virgo would have an angular extent of order $`1^{\prime \prime }`$, and would not be associated with any background galaxy. Thus, by surveying one square degree in Virgo, to $`0.1`$ mJy in the radio and to 24.5 mag in the optical, one could exclude most of the $`3000`$ background radio sources, based on their large sizes, their association with distant galaxies, or both. Using follow-up radio observations, with higher angular resolution, of the remaining candidates, one could find of order 10 radio SNRs of intracluster SNe, based on their characteristic morphologies and sizes. A detection of this population, and measurements of its properties (SNR sizes, luminosities, spectra), would provide constraints on the intergalactic SN rate and on SN-Ia progenitors.
Finally, from equation (29), the 1 GHz continuum flux from the remnant of SN1980I should be $`0.25`$ mJy, which is detectable, although for some parameters the flux could be lower by an order of magnitude. With a radius of about 1 milliarcsecond, the remnant should be unresolved.
## 5. Conclusions
We have estimated the properties of the remnants of type-Ia SNe that explode in the ICM. We have shown that, if such SNe explode into a CSM, presumably produced by the wind from the donor giant star from which the white dwarf accreted mass, then intergalactic SNRs are no different from galactic SNRs, for an age of up to a thousand years. This conclusion is mostly unaffected by the large speed with which the SNR may travel through the ICM. Galactic and intergalactic supernovae are distinct after they have expanded beyond the CSM radius. Since CSM and galaxy-ISM densities are comparable, galactic SNRs remain luminous emission-line sources in the ISM stage. Indeed, for any particular galactic SNR, it is difficult to determine observationally whether the ejecta is interacting with the CSM or the ISM, or even whether there ever was a CSM.
Intergalactic SNRs, in contrast, are optically luminous only during the CSM stage, if there is one. Once the ejecta reaches the ICM, the SNR quickly fades in the optical band. As a result, intergalactic SNRs can provide a unique test for the existence of a CSM around type-Ia SNe. If intracluster SNRs are found in the optical band, the corollary would be that type-Ia SNe have a CSM, supporting the progenitor model of accretion from a giant star companion. Alternatively, a non-detection of these SNRs would imply the absence of a CSM. This would point to accretion from a main-sequence or sub-giant companion, or to the “double-degenerate” progenitor model, involving the merger of a white-dwarf pair following the loss of orbital energy to gravitational radiation.
We have also shown that irrespective of the existence of a CSM, old intracluster SNRs are detectable in the radio band. Although identifying them among the more numerous background radio populations is challenging, one could take advantage of the low optical luminosity of SNRs in the ICM stage. In contrast with most of the background radio sources, which are associated with galaxies, the old intergalactic radio SNRs will have no associated optical source. Thus, a radio search could establish the existence of the old intracluster SNR population, and an optical search would provide valuable information on type-Ia SN physics.
Finally, we have speculated that several examples of compact intergalactic emission-line objects, recently discovered in Virgo (Gerhard et al. 2002) and in two galaxy groups (Ryan-Weber et al. 2004), could be intergalactic SNRs, rather than H II regions, as postulated before. We have argued that the luminosities and numbers of the newly discovered objects, although somewhat on the high side, are still roughly consistent with our estimates for SNRs. Consistency is possible given the simplifications inherent in our model, the allowed range of input parameters, and the uncertainties in the parameters describing the observed objects. High resolution imaging with HST could potentially discriminate among the two options, based on the different morphologies of SNRs and H II regions. Deep pointed observations of the site of SN1980I, a possible intergalactic SN-Ia in Virgo, could test our predictions for the early-stage development of intergalactic SNRs.
We thank John Raymond and Jacco Vink for enlightening discussions on radiative processes in supernova remnants. This work was supported in part by NASA grant NAG 5-13292, and by NSF grants AST-0071019, AST-0204514 (for A.L.). A. L. is grateful for the kind hospitality of the Einstein Minerva Center at the Weizmann Institute of Science, where this work started.
|
warning/0506/math0506294.html
|
ar5iv
|
text
|
# 1 Preliminaries
## 1 Preliminaries
Our notations is standard. If $`n`$ is a natural number, $`\pi `$ a set of primes, then by $`\pi (n)`$ we denote the set of all prime divisors of $`n`$, by $`n_\pi `$ we denote the maximal divisor $`t`$ of $`n`$ such that $`\pi (t)\pi `$. Note that for a finite group $`G`$, $`\pi (G)=\pi (|G|)`$ by definition.
The adjacency criterion of two prime divisors for alternating groups is obvious and can be given as follows.
###### Proposition 1.1.
Let $`G=Alt_n`$ be an alternating group of degree $`n`$.
* Let $`r,s\pi (G)`$ be odd primes. Then $`r,s`$ are non-adjacent if and only if $`r+s>n`$.
* Let $`r\pi (G)`$ be an odd prime. Then $`2,r`$ are non-adjacent if and only if $`r+4>n`$.
The information on the adjacency of vertices in a prime graph for every sporadic group and Tits group $`{}_{}{}^{2}F_{4}^{}(2)^{}`$ can be extracted from or . Thus, we need to consider only simple groups of Lie type.
For groups of Lie type and linear algebraic groups our notations agrees with that in and respectively. Denote by $`G_{sc}`$ a universal group of Lie type. Then every factor group $`G_{sc}/Z`$, where $`ZZ(G_{sc})`$, we call a group of Lie type. Almost in all cases $`G_{sc}/Z(G_{sc})`$ is simple and we say that $`G_{ad}=G_{sc}/Z(G_{sc})`$ is of adjoint type. Some groups of Lie type over small fields are not simple. In Table 1 we assemble the information of \[7, Theorems 11.1.2 and 14.4.1\] and give all such exceptions. Sometimes we shall use notations $`A_n^\epsilon (q),D_n^\epsilon (q)`$, and $`E_6^\epsilon (q)`$, where $`\epsilon \{+,\}`$, and $`A_n^+(q)=A_n(q)`$, $`A_n^{}(q)={}_{}{}^{2}A_{n}^{}(q)`$, $`D_n^+(q)=D_n(q)`$, $`D_n^{}(q)={}_{}{}^{2}D_{n}^{}(q)`$, $`E_6^+(q)=E_6(q)`$, $`E_6^{}(q)={}_{}{}^{2}E_{6}^{}(q)`$.
If $`G`$ is isomorphic to $`{}_{}{}^{2}A_{n}^{}(q)`$, $`{}_{}{}^{2}D_{n}^{}(q)`$, or $`{}_{}{}^{2}E_{6}^{}(q)`$ we say that $`G`$ is defined over $`GF(q^2)`$, if $`G{}_{}{}^{3}D_{4}^{}(q)`$ we say that $`G`$ is defined over $`GF(q^3)`$ and we say that $`G`$ is defined over $`GF(q)`$ for other finite groups of Lie type. The field $`GF(q)`$ in all cases is called the base field of $`G`$. If $`G`$ is a universal finite group of Lie type with the base field $`GF(q)`$, then there are a natural number $`N`$($`=|\mathrm{\Phi }^+|`$ in most cases) and a polynomial $`f(t)[t]`$ such that $`|G|=f(q)q^N`$ and $`(q,f(q))=1`$ (see \[7, Theorems 9.4.10 and 14.3.1\]). This polynomial we denote by $`f_G(t)`$. If $`G`$ is not universal then there is a universal group $`K`$ with $`G=K/Z`$, where $`Z=Z(K)`$, and we define $`f_G(t)=f_K(t)`$.
Assume that $`\overline{G}`$ is a connected simple linear algebraic group defined over an algebraically closed field of a positive characteristic $`p`$. Let $`\sigma `$ be an endomorphism of $`\overline{G}`$ such that $`\overline{G}_\sigma =C_{\overline{G}}(\sigma )`$ is a finite set. Then $`\sigma `$ is said to be a Frobenius map and $`G=O^p^{}(\overline{G}_\sigma )`$ is appeared to be a finite group of Lie type. Moreover all finite groups of Lie type both split and twisted can be derived in this way. Below we assume that for every finite group $`G`$ of Lie type we fix (in some way) a linear algebraic group $`\overline{G}`$ and a Frobenius map $`\sigma `$ such that $`G=O^p^{}(\overline{G}_\sigma )`$. If $`\overline{G}`$ is simply connected, then $`G=\overline{G}_\sigma =O^p^{}(\overline{G}_\sigma )`$; and if $`\overline{G}`$ is of adjoint type, then $`\overline{G}_\sigma =\widehat{G}`$ is the group of inner-diagonal automorphisms of $`G`$ (see \[9, § 12\]).
If $`\overline{R}`$ is a $`\sigma `$-stable reductive subgroup of $`\overline{G}`$, then $`\overline{R}_\sigma G=\overline{R}G`$ is called a reductive subgroup of $`G`$. If $`\overline{R}`$ is of maximal rank, then $`\overline{R}_\sigma G`$ is also said to be of maximal rank. Note that if $`\overline{R}`$ is $`\sigma `$-stable reductive of maximal rank, then $`\overline{R}=\overline{T}\overline{G}_1\mathrm{}\overline{G}_k`$, where $`\overline{T}`$ is some $`\sigma `$-stable maximal torus of $`\overline{R}`$ and $`\overline{G}_1,\mathrm{},\overline{G}_k`$ are subsystem subgroups of $`\overline{G}`$. Furthermore $`\overline{G}_1\mathrm{}\overline{G}_k=[\overline{R},\overline{R}]`$. It is known that $`\overline{R}_\sigma =\overline{T}_\sigma G_1\mathrm{}G_m`$, subgroups $`G_1,\mathrm{},G_m`$ we call subsystem subgroups of $`G`$. In general $`mk`$ and for all $`i`$ the base field of $`G_i`$ is equal to $`GF(q^{\alpha _i})`$, where $`\alpha _i1`$. There is a nice algorithm due to Borel and de Siebental determining all subsystem subgroups of $`\overline{G}`$ (see and also ). One has to consider the extended Dynkin diagram of $`\overline{G}`$ and remove any number of nodes. Connected components of the remaining graph are Dynkin diagrams of subsystem subgroups of $`\overline{G}`$ and Dynkin diagrams of all subsystem subgroups can be derived in this way.
If $`\overline{T}`$ is a $`\sigma `$-stable torus of $`\overline{G}`$ then $`T=\overline{T}G=\overline{T}_\sigma G`$ is called a torus of $`G`$. If $`\overline{T}`$ is maximal, then $`T`$ is a maximal torus of $`G`$. If $`G`$ is neither a Suzuki group, nor a Ree group, then for every maximal torus $`T`$ we have that $`|\overline{T}_\sigma |=g(q)`$, where $`GF(q)`$ is the base field of $`G`$, $`g(t)`$ is a polynomial of degree $`n`$ dividing $`f_G(t)`$ and $`n`$ is the rank of $`\overline{G}`$. For more details see \[12, Chapter 1\].
###### Table 1.
Non simple groups of Lie type
In Lemmas 1.2 and 1.3 we assemble the information about maximal tori in finite simple groups of Lie type.
###### Lemma 1.2.
(see \[13, Propositions 7–10\] and ) Let $`\overline{G}`$ be a connected simple classical algebraic group of adjoint type and let $`G=O^p^{}(\overline{G}_\sigma )`$ be the finite simple classical group.
* Every maximal torus $`T`$ of $`G=A_{n1}^\epsilon (q)`$ has the order
$$\frac{1}{(n,q(\epsilon 1))(q(\epsilon 1))}(q^{n_1}(\epsilon 1)^{n_1})(q^{n_2}(\epsilon 1)^{n_2})\mathrm{}(q^{n_k}(\epsilon 1)^{n_k})$$
for appropriate partition $`n_1+n_2+\mathrm{}+n_k=n`$ of $`n`$. Moreover, for every partition there exists a torus of corresponding order.
* Every maximal torus $`T`$ of $`G`$, where $`G=B_n(q)`$ or $`G=C_n(q)`$, has the order
$$\frac{1}{(2,q1)}(q^{n_1}1)(q^{n_2}1)\mathrm{}(q^{n_k}1)(q^{l_1}+1)(q^{l_2}+1)\mathrm{}(q^{l_m}+1)$$
for appropriate partition $`n_1+n_2+\mathrm{}+n_k+l_1+l_2+\mathrm{}+l_m=n`$ of $`n`$. Moreover, for every partition there exists a torus of corresponding order.
* Every maximal torus $`T`$ of $`G=D_n^\epsilon (q)`$ has the order
$$\frac{1}{(4,q^n\epsilon 1)}(q^{n_1}1)(q^{n_2}1)\mathrm{}(q^{n_k}1)(q^{l_1}+1)(q^{l_2}+1)\mathrm{}(q^{l_m}+1)$$
for appropriate partition $`n_1+n_2+\mathrm{}+n_k+l_1+l_2+\mathrm{}+l_m=n`$ of $`n`$, where $`m`$ is even if $`\epsilon =+`$ and $`m`$ is odd if $`\epsilon =`$. Moreover, for every partition there exists a torus of corresponding order.
###### Lemma 1.3.
(see and ) Let $`\overline{G}`$ be a connected simple exceptional algebraic group of adjoint type and let $`G=O^p^{}(\overline{G}_\sigma )`$ be the finite simple exceptional group of Lie type.
* Every maximal torus $`T`$ of $`G=G_2(q)`$ has one of the following orders:
$$(q\pm 1)^2,q^21,q^2\pm q+1.$$
Moreover, for every number given above there exists a torus of corresponding order.
* Every maximal torus $`T`$ of $`G=F_4(q)`$ has one of the following orders:
$$(q\pm 1)^4,(q\pm 1)^2(q^2\pm 1),(q^2\pm 1)^2,(q\pm 1)(q^3\pm 1),q^4\pm 1,(q^2\pm q+1)^2,q^4q^2+1.$$
Here and below symbol $`\pm `$ means that we can choose either “$`+`$” or “$``$” independently for all multiplies, i. e., $`(q\pm 1)^2(q^2\pm 1)`$ is equal to either $`(q1)^2(q^21)`$, or $`(q+1)^2(q^21)`$, or $`(q1)^2(q^2+1)`$, or $`(q+1)^2(q^2+1)`$. Moreover, for every number given above there exists a torus of corresponding order.
* For every maximal torus $`T`$ of $`G=E_6^\epsilon (q)`$, the number $`(3,q\epsilon 1)|T|`$ is equal to one of the following:
$`(q\epsilon 1)^k(q+\epsilon 1)^{6k}\text{}2k6;(q^k(\epsilon 1)^k)(q^{6k}(\epsilon 1)^{6k})\text{}1k5;`$
$`(q^k(\epsilon 1)^k)(q\epsilon 1)^{6k}\text{}3k6;(q^3\epsilon 1)(q^21)(q\pm 1);(q^5\epsilon 1)(q+\epsilon 1);`$
$`(q^3+\epsilon 1)(q^2\pm 1)(q\epsilon 1);(q^4+1)(q^21);(q^2+1)^2(q\epsilon 1)^2;(q^2+\epsilon q+1)^3;`$
$`(q^2+\epsilon q+1)^2(q^21);(q^41)(q+\epsilon 1)^2;(q^3+\epsilon 1)(q^2+\epsilon q+1)(q+\epsilon 1);`$
$`(q^4q^2+1)(q^2+\epsilon q+1);q^6+\epsilon q^3+1;(q^2+\epsilon q+1)(q^2\epsilon q+1)^2.`$ Moreover, for every number $`n`$ given above there exists a torus $`T`$ with $`(3,q\epsilon 1)|T|=n`$.
* For every maximal torus $`T`$ of $`G=E_7(q)`$, the number $`m=(2,q1)|T|`$ is equal to one of the following: $`(q+1)^{n_1}(q1)^{n_2},`$ $`n_1+n_2=7;`$ $`(q^2+1)^{n_1}(q+1)^{n_2}(q1)^{n_3},`$ $`1n_12,`$ $`2n_1+n_2+n_3=7,`$ and $`m(q^2+1)(q\pm 1)^5;`$ $`(q^3+1)^{n_1}(q^31)^{n_2}(q^2+1)^{n_3}(q+1)^{n_4}(q1)^{n_5},`$ $`1n_1+n_22,`$ $`3n_1+3n_2+2n_3+n_4+n_5=7,`$ and $`m(q^3+ϵ1)(qϵ1)^4,`$ $`m(q^3\pm 1)(q^2+1)^2,`$ $`m(q^3+ϵ1)(q^2+1)(q+ϵ1)^2;`$ $`(q^4+1)(q^2\pm 1)(q\pm 1);`$ $`(q^5\pm 1)(q^21);`$ $`(q^5+ϵ1)(q+ϵ1)^2;`$ $`q^7\pm 1;`$ $`(qϵ1)(q^2+ϵq+1)^3;(q^5ϵ1)(q^2+ϵq+1);(q^3\pm 1)(q^4q^2+1);(qϵ1)(q^6+ϵq^3+1);`$ $`(q^3ϵ1)(q^2ϵq+1)^2,`$ where $`ϵ=\pm `$. Moreover, for every number $`m`$ given above there exists a torus $`T`$ with $`(2,q1)|T|=m`$. $`(2,q1)|T|=n`$.
* Every maximal torus $`T`$ of $`G=E_8(q)`$ has one of the following orders: $`(q+1)^{n_1}(q1)^{n_2},`$ $`n_1+n_2=8;`$ $`(q^2+1)^{n_1}(q+1)^{n_2}(q1)^{n_3},`$ $`1n_14,`$ $`2n_1+n_2+n_3=8,`$ and $`|T|(q^2+1)^3(q\pm 1)^2,`$ $`|T|(q^2+1)(q\pm 1)^6;`$ $`(q^3+1)^{n_1}(q^31)^{n_2}(q^2+1)^{n_3}(q+1)^{n_4}(q1)^{n_5},`$ $`1n_1+n_22,`$ $`3n_1+3n_2+2n_3+n_4+n_5=8,`$ and $`|T|(q^3\pm 1)^2(q^2+1),`$ $`|T|(q^3+ϵ1)(qϵ1)^5,`$ $`|T|(q^3+ϵ1)(q^2+1)(q+ϵ1)^3,`$ $`|T|(q^3+ϵ1)(q^2+1)^2(qϵ1);`$ $`q^81;`$ $`(q^4+1)^2;`$ $`(q^4+1)(q^2\pm 1)(q\pm 1)^2;`$ $`(q^4+1)(q^21)^2;`$ $`(q^4+1)(q^3+ϵ1)(qϵ1);`$ $`(q^5+ϵ1)(q+ϵ1)^3;`$ $`(q^5\pm 1)(q+ϵ1)^2(qϵ1);`$ $`(q^5+ϵ1)(q^2+1)(qϵ1);`$ $`(q^5+ϵ1)(q^3+ϵ1);`$ $`(q^6+1)(q^2\pm 1);`$ $`(q^7\pm 1)(q\pm 1);`$ $`(qϵ1)(q^2+ϵq+1)^3(q\pm 1);`$ $`(q^5ϵ1)(q^2+ϵq+1)(q+ϵ1);`$ $`(q^3\pm 1)(q^4q^2+1)(q\pm 1);`$ $`(qϵ1)(q^6+ϵq^3+1)(q\pm 1);`$ $`(q^3ϵ1)(q^2ϵq+1)^2(q\pm 1);`$ $`q^8q^4+1;`$ $`q^8+q^7q^5q^4q^3+q+1;`$ $`q^8q^6+q^4q^2+1;`$ $`(q^4q^2+1)^2;`$ $`(q^6+ϵq^3+1)(q^2+ϵq+1);`$ $`q^8q^7+q^5q^4+q^3q+1;`$ $`(q^4+ϵq^3+q^2+ϵq+1)^2;`$ $`(q^4q^2+1)(q^2\pm q+1)^2;`$ $`(q^2q+1)^2(q^2+q+1)^2;`$ $`(q^2\pm q+1)^4,`$ where $`ϵ=\pm `$. Moreover, for every number given above there exists a torus of corresponding order.
* Every maximal torus $`T`$ of $`G={}_{}{}^{3}D_{4}^{}(q)`$ has one of the following orders:
$$(q^3\pm 1)(q\pm 1);(q^2\pm q+1)^2;q^4q^2+1.$$
Moreover, for every number given above there exists a torus of corresponding order.
* Every maximal torus $`T`$ of $`G={}_{}{}^{2}B_{2}^{}(2^{2n+1})`$ has one of the following orders:
$$q1;q\pm \sqrt{2q}+1,$$
where $`q=2^{2n+1}`$. Moreover, for every number given above there exists a torus of corresponding order.
* Every maximal torus $`T`$ of $`G={}_{}{}^{2}G_{2}^{}(3^{2n+1})`$ has one of the following orders:
$$q\pm 1;q\pm \sqrt{3q}+1,$$
where $`q=3^{2n+1}`$. Moreover, for every number given above there exists a torus of corresponding order.
* Every maximal torus $`T`$ of $`G={}_{}{}^{2}F_{4}^{}(2^{2n+1})`$ with $`n1`$ has one of the following orders: $`q^2+ϵq\sqrt{2q}+q+ϵ\sqrt{2q}+1;`$ $`q^2ϵq\sqrt{2q}+ϵ\sqrt{2q}1;`$ $`q^2q+1;`$ $`(q\pm \sqrt{2q}+1)^2;`$ $`(q1)(q\pm \sqrt{2q}+1);`$ $`(q\pm 1)^2;q^2\pm 1;`$ where $`q=2^{2n+1}`$ and $`ϵ=\pm `$. Moreover, for every number given above there exists a torus of corresponding order.
If $`q`$ is a natural number, $`r`$ is an odd prime and $`(r,q)=1`$, then by $`e(r,q)`$ we denote the minimal natural number $`n`$ with $`q^n1(\mathrm{mod}r)`$. If $`q`$ is odd, let $`e(2,q)=1`$ if $`q1(\mathrm{mod}4)`$ and $`e(2,q)=2`$ if $`q1(\mathrm{mod}4)`$.
The main technical tools in Section 6 is the following statement.
###### Lemma 1.4.
(Corollary to Zsigmondy’s theorem ) Let $`q`$ be a natural number greater than $`1`$. For every natural number $`m`$ there exists a prime $`r`$ with $`e(r,q)=m`$ but for the cases $`q=2`$ and $`m=1`$, $`q=3`$ and $`m=1`$, and $`q=2`$ and $`m=6`$.
The prime $`r`$ with $`e(r,q)=n`$ is said to be a primitive prime divisor of $`q^n1`$. By Zsigmondy theorem it exists excepting the cases indicated above. If $`q`$ is fixed, we denote by $`r_n`$ some primitive prime divisor of $`q^n1`$ (obviously, $`q^n1`$ can have more than one such divisor). Note that according to our definition every prime divisor of $`q1`$ is a primitive prime divisor of $`q1`$ with sole exception: $`2`$ is not a primitive prime divisor of $`q1`$ if $`e(2,q)=2`$. In the last case $`2`$ is a primitive prime divisor of $`q^21`$.
In view \[7, Theorems 9.4.10 and 14.3.1\] the order of any finite simple group of Lie type $`G`$ of rank $`n`$ over the field $`GF(q)`$ of characteristic $`p`$ is given by
$$|G|=\frac{1}{d}q^N(q^{m_1}\pm 1)\mathrm{}(q^{m_n}\pm 1).$$
It follows that any prime divisor $`r`$ of $`|G|`$ distinct from the characteristic $`p`$ is a primitive divisor of $`q^m1`$ for some natural $`m`$. Thus, the Zsigmondy Theorem allows us to “find” prime divisors of $`|G|`$. Moreover, if $`G`$ is neither a Suzuki group nor a Ree group, Lemmas 1.2 and 1.3 imply that for a fixed $`m`$ every two primitive prime divisors of $`q^m1`$ are adjacent in $`GK(G)`$.
For Suzuki and Ree groups we use following
###### Lemma 1.5.
Let $`n`$ be a natural number.
1. Let $`m_1(B,n)=2^{2n+1}1`$,
$`m_2(B,n)=2^{2n+1}2^{n+1}+1`$,
$`m_3(B,n)=2^{2n+1}+2^{n+1}+1`$.
Then $`(m_i(B,n),m_j(B,n))=1`$ if $`ij`$.
2. Let $`m_1(G,n)=3^{2n+1}1`$,
$`m_2(G,n)=3^{2n+1}+1`$,
$`m_3(G,n)=3^{2n+1}3^{n+1}+1`$,
$`m_4(G,n)=3^{2n+1}+3^{n+1}+1`$.
Then $`(m_1(G,n),m_2(G,n))=2`$ and $`(m_i(G,n),m_j(G,n))=1`$ otherwise.
3. Let $`m_1(F,n)=2^{2n+1}1`$,
$`m_2(F,n)=2^{2n+1}+1`$,
$`m_3(F,n)=2^{4n+2}+1`$,
$`m_4(F,n)=2^{4n+2}2^{2n+1}+1`$,
$`m_5(F,n)=2^{4n+2}2^{3n+2}+2^{2n+1}2^{n+1}+1`$,
$`m_6(F,n)=2^{4n+2}+2^{3n+2}+2^{2n+1}+2^{n+1}+1`$.
Then $`(m_2(F,n),m_4(F,n))=3`$ and $`(m_i(F,n),m_j(F,n))=1`$ otherwise.
###### Proof.
It easy to check by the direct computation. ∎
By Lemma 1.3 every distinct from the characteristic prime divisor $`s`$ of order of the Suzuki group $`{}_{}{}^{2}B_{2}^{}(2^{2n+1})`$ divides one of the numbers $`m_i(B,n)`$ defined in Lemma 1.5. The same is true for Ree groups $`{}_{}{}^{2}G_{2}^{}(3^{2n+1})`$, $`{}_{}{}^{2}F_{2}^{}(2^{2n+1})`$ and all prime divisors of the numbers $`m_i(G,n)`$, $`m_i(F,n)`$ respectively. Thus, Lemma 1.5 allow us to find prime divisors of orders of Suzuki and Ree groups. Moreover, Lemma 1.3 implies that for a fixed $`k`$ every two prime divisors of $`m_k(B,n)`$ are adjacent in $`GK({}_{}{}^{2}B_{2}^{}(2^{2n+1}))`$. The same is also true for Ree groups and all prime divisors of $`m_k(G,n)`$ and $`m_k(F,n)`$.
## 2 Adjacent odd primes
In this section we consider whether two odd primes distinct from the characteristic are adjacent in the Gruenberg — Kegel graph of a finite group of Lie type.
###### Proposition 2.1.
Let $`G=A_{n1}(q)`$ be a finite simple group of Lie type over a field of characteristic $`p`$. Let $`r,s`$ be odd primes and $`r,s\pi (G)\{p\}`$. Denote $`k=e(r,q)`$, $`l=e(s,q)`$ and suppose that $`2kl`$. Then $`r`$ and $`s`$ are non-adjacent if and only if $`k+l>n`$ and $`k`$ does not divide $`l`$.
###### Proof.
Note first that, for any odd prime $`cp`$:
$$c\text{ divides }q^x1\text{ if and only if }e(c,q)\text{ divides }x.$$
(1)
Indeed, by definition, $`c`$ divides $`q^{e(c,q)}1`$ and does not divide $`q^y1`$ for all $`y<e(c,q)`$, i. e., $`e(c,q)`$ is the order of $`q`$ in the multiplicative group $`GF(c)^{}`$ of the finite field $`GF(c)`$. So if $`c`$ divides $`q^z1`$, then $`q^z=1`$ in $`GF(c)^{}`$, hence $`e(c,q)`$ divides $`z`$. Now assume that $`e(c,q)`$ divides $`z`$. Then $`q^z1=(q^{e(c,q)}1)f(q)`$ for some $`f(t)[t]`$, hence $`c`$ divides $`q^z1`$.
Assume that $`k+ln`$. Consider a maximal torus $`T`$ of $`G`$ of order
$$\frac{1}{(n,q1)(q1)}(q^k1)(q^l1)(q1)^{nkl}.$$
The torus $`T`$ is an abelian subgroup of $`G`$ and $`r,s\pi (T)`$. It follows that $`T`$ contains an element of order $`rs`$, hence $`r,s`$ are adjacent. If $`k`$ divides $`l`$ then both $`r`$ and $`s`$ divide $`q^l1`$, therefore a maximal torus of order $`\frac{1}{(n,q1)}(q^l1)(q1)^{nl1}`$ contains an element of order $`rs`$.
Assume now that $`k+l>n`$, $`k`$ does not divide $`l`$, and assume that $`gG`$ is an element of order $`rs`$. Then $`(|g|,p)=1`$ hence $`g`$ is semisimple. Therefore there exists a maximal torus $`T`$ such that $`gT`$. By Lemma 1.2 the order of $`T`$ is equal to
$$\frac{1}{(n,q1)(q1)}(q^{n_1}1)(q^{n_2}1)\mathrm{}(q^{n_x}1)$$
for appropriate partition $`n_1+n_2+\mathrm{}+n_x=n`$ of $`n`$. Since $`r,s`$ are prime, there exist $`n_i,n_j`$ such that $`r`$ divides $`q^{n_i}1`$ and $`s`$ divides $`q^{n_j}1`$. In view of (1), it follows that $`n_i=ak`$, $`n_j=bl`$ for some $`a,b1`$. Moreover, since $`k+l>n`$ and $`kl`$, we have that $`b=1`$. Indeed, otherwise $`n_jl+lk+l>n`$, a contradiction with $`n_1+\mathrm{}+n_x=n`$. Since $`k`$ does not divide $`l`$ it follows that $`n_in_j`$. Hence $`n_1+n_2+\mathrm{}+n_xn_i+n_j=ak+l>n`$; a contradiction. ∎
###### Proposition 2.2.
Let $`G={}_{}{}^{2}A_{n1}^{}(q)`$ be a finite simple group of Lie type over a field of characteristic $`p`$. Define a function
$$\nu (m)=\{\begin{array}{cc}\hfill m& \text{, if }m0(\mathrm{mod}4),\hfill \\ \hfill \frac{m}{2}& \text{, if }m2(\mathrm{mod}4),\hfill \\ \hfill 2m& \text{, if }m1(\mathrm{mod}2).\hfill \end{array}$$
Let $`r,s`$ be odd primes and $`r,s\pi (G)\{p\}`$. Denote $`k=e(r,q)`$, $`l=e(s,q)`$ and suppose that $`2\nu (k)\nu (l)`$. Then $`r`$ and $`s`$ are non-adjacent if and only if $`\nu (k)+\nu (l)>n`$ and $`\nu (k)`$ does not divide $`\nu (l)`$.
###### Proof.
Note first that, for any odd prime $`cp`$:
$$c\text{ divides }q^x(1)^x\text{ if and only if }\nu (e(c,q))\text{ divides }x.$$
(2)
Assume first that $`e(c,q)`$ is odd. If $`c`$ divides $`q^z(1)^z`$ then $`c`$ divides $`q^{2z}1`$, hence, by (1), $`e(c,q)`$ divides $`2z`$. Since $`e(c,q)`$ is odd it follows that $`e(c,q)`$ divides $`z`$ and therefore (again by (1)) $`c`$ divides $`q^z1`$. Since $`c`$ is odd we have that $`q^z+1`$ is not divisible by $`c`$, so $`q^z(1)^z=q^z1`$, i. e., $`z`$ is even. But $`e(c,q)`$ is odd hence $`2e(c,q)=\nu (e(c,q))`$ divides $`z`$. Now assume that $`\nu (e(c,q))`$ divides $`z`$. Then $`z`$ is even, hence $`q^z(1)^z=q^z1`$ and $`c`$ divides $`q^z(1)^z`$ by (1). So (2) is true in this case.
Assume that $`e(c,q)2(\mathrm{mod}4)`$. If $`c`$ divides $`q^z(1)^z`$ then $`c`$ divides $`q^{2z}1`$, hence $`e(c,q)`$ divides $`2z`$. But $`\nu (e(c,q))=\frac{e(c,q)}{2}`$, therefore $`\nu (e(c,q))`$ divides $`z`$. If $`\nu (e(c,q))`$ divides $`z`$, and $`z`$ is odd, then $`q^z(1)^z=q^z+1`$. We have that $`e(c,q)`$ divides $`2z`$, hence $`c`$ divides $`q^{2z}1`$. Now $`q^{2z}1=(q^z1)(q^z+1)`$. Since $`z`$ is odd, $`e(c,q)`$ does not divide $`z`$, therefore, by (1), $`c`$ does not divide $`q^z1`$, hence $`c`$ divides $`q^z+1=q^z(1)^z`$. If $`\nu (e(c,q))`$ divides $`z`$, and $`z`$, is even then $`2\nu (e(c,q))=e(c,q)`$ divides $`z`$, hence, by (1), $`c`$ divides $`q^z(1)^z=q^z1`$. Thus (2) is true in this case as well.
At the end assume that $`e(c,q)0(\mathrm{mod}4)`$. If $`c`$ divides $`q^z(1)^z`$ then, as above, $`c`$ divides $`q^{2z}1`$, hence $`e(c,q)`$ divides $`2z`$ and $`\frac{e(c,q)}{2}`$ divides $`z`$. But $`\frac{e(c,q)}{2}`$ is even, hence, $`z`$ is even and $`q^z(1)^z=q^z1`$. It follows that $`e(c,q)=\nu (e(c,q))`$ divides $`z`$. If $`\nu (e(c,q))=e(c,q)`$ divides $`z`$ then $`z`$ is even and, by (1), $`c`$ divides $`q^z(1)^z=q^z1`$.
Assume that $`\nu (k)+\nu (l)n`$. Consider a maximal torus $`T`$ of $`G`$ of order
$$\frac{1}{(n,q+1)(q+1)}(q^{\nu (k)}(1)^{\nu (k)})(q^{\nu (l)}(1)^{\nu (l)})(q+1)^{n\nu (k)\nu (l)}.$$
The torus $`T`$ is an abelian subgroup of $`G`$ and $`r,s\pi (T)`$. It follows that $`T`$ contains an element of order $`rs`$, hence $`r,s`$ are adjacent. If $`\nu (k)`$ divides $`\nu (l)`$ then both $`r`$ and $`s`$ divide $`q^{\nu (l)}(1)^{\nu (l)}`$, therefore a maximal torus of order $`\frac{1}{(n,q+1)}(q^{\nu (l)}(1)^{\nu (l)})(q+1)^{n\nu (l)1}`$ contains an element of order $`rs`$.
Assume now that $`\nu (k)+\nu (l)>n`$, $`\nu (k)`$ does not divide $`\nu (l)`$, and assume that $`g{}_{}{}^{2}A_{n1}^{}(q)`$ is an element of order $`rs`$. Then $`(|g|,p)=1`$ hence $`g`$ is semisimple. Therefore there exists a maximal torus $`T`$ such that $`gT`$. Using (2) and Lemma 1.2 we obtain a contradiction like in the proof of Proposition 2.1. ∎
###### Proposition 2.3.
Let $`G`$ be one of simple groups of Lie type, $`B_n(q)`$ or $`C_n(q)`$, over a field of characteristic $`p`$. Define
$$\eta (m)=\{\begin{array}{cc}m& \text{ if }m\text{ is odd},\\ \frac{m}{2}& \text{ otherwise}.\end{array}$$
Let $`r,s`$ be odd primes with $`r,s\pi (G)\{p\}`$. Put $`k=e(r,q)`$ and $`l=e(s,q)`$, and suppose that $`1\eta (k)\eta (l)`$. Then $`r`$ and $`s`$ are non-adjacent if and only if $`\eta (k)+\eta (l)>n`$, and $`k`$, $`l`$ satisfy to (3):
$$\frac{l}{k}\text{ }\text{is not an odd natural number}$$
(3)
Note that (3) is true in the following cases:
1. Both $`k`$ and $`l`$ are even and either $`\eta (k)`$ does not divide $`\eta (l)`$ or $`\frac{\eta (l)}{\eta (k)}`$ is even. In this case $`q^{\eta (k)}+(1)^k=q^{\eta (k)}+1`$ does not divide $`q^{\eta (l)}+(1)^l=q^{\eta (l)}+1`$.
2. Both $`k`$ and $`l`$ are odd and $`\eta (k)`$ does not divide $`\eta (l)`$. In this case $`q^{\eta (k)}+(1)^k=q^k1`$ does not divide $`q^{\eta (l)}+(1)^l=q^l1`$.
3. $`k`$ is odd and $`l`$ is even. In this case $`q^{\eta (k)}+(1)^k=q^k1`$ does not divide $`q^{\eta (l)}+(1)^l=q^{\eta (l)}+1`$.
4. $`k`$ is even, $`l`$ is odd and either $`\eta (k)`$ does not divide $`\eta (l)`$ or $`\frac{\eta (l)}{\eta (k)}`$ is either even, or is equal to 1. In this case $`q^{\eta (k)}+(1)^k=q^{\eta (k)}+1`$ does not divide $`q^{\eta (l)}+(1)^l=q^l1`$.
This means that $`\eta (k)`$, $`\eta (l)`$ satisfy (3) if and only if $`q^{\eta (k)}+(1)^k`$ does not divide $`q^{\eta (l)}+(1)^l`$.
###### Proof.
First we prove that, for any odd prime $`cp`$:
$$\text{if }c\text{ divides }q^x\pm 1\text{ then }\eta (e(c,q))\text{ divides }x.$$
(4)
Assume first that $`e(c,q)`$ is odd and $`c`$ divides $`q^z\pm 1`$. By (1) it follows that $`e(c,q)`$ divides $`2z`$. But $`e(c,q)`$ is odd hence $`e(c,q)=\eta (e(c,q))`$ divides $`z`$. Assume now that $`e(c,q)`$ is even and $`c`$ divides $`q^z\pm 1`$. It follows that $`c`$ divides $`q^{2z}1`$, hence $`e(c,q)`$ divides $`2z`$ and $`\frac{e(c,q)}{2}=\eta (e(c,q))`$ divides $`z`$.
Now if $`\eta (k)+\eta (l)n`$ we may consider a maximal torus $`T`$ of order $`\frac{1}{(2,q1)}(q^{\eta (k)}+(1)^k)(q^{\eta (l)}+(1)^l)(q1)^{n\eta (k)\eta (l)}`$. We have that $`T`$ is an abelian group and $`r,s\pi (T)`$. Hence $`T`$ contains an element of order $`rs`$. If $`\eta (k)`$, $`\eta (l)`$ does not satisfy (3), then $`q^{\eta (k)}+(1)^k`$ divides $`q^{\eta (l)}+(1)^l`$, therefore both $`r`$ and $`s`$ divide $`q^{\eta (l)}+(1)^l`$. Thus a maximal torus of order $`\frac{1}{(2,q1)}(q^{\eta (l)}+(1)^l)(q1)^{n\eta (l)}`$ contains an element of order $`rs`$.
Assume that $`\eta (k)+\eta (l)>n`$, $`\eta (k)`$ and $`\eta (l)`$ satisfy (3), and assume that there exists an element $`gG`$ of order $`rs`$. Since $`(|g|,p)=1`$, it follows that $`g`$ is semisimple. So there exists a maximal torus $`T`$ containing $`g`$. In view of Lemma 1.2, the order $`|T|`$ is equal to
$$\frac{1}{(2,q1)}(q^{n_1}\pm 1)(q^{n_2}\pm 1)\mathrm{}(q^{n_x}\pm 1)$$
for appropriate partition $`n_1+n_2+\mathrm{}+n_x=n`$ of $`n`$. Using (4) we obtain a contradiction like in Proposition 2.1. ∎
###### Proposition 2.4.
et $`G=D_n^\epsilon (q)`$ be a finite simple group of Lie type over a field of characteristic $`p`$, and let the function $`\eta (m)`$ be defined as in Proposition 2.3. Suppose $`r,s`$ are odd primes and $`r,s\pi (D_n^\epsilon (q))\{p\}`$. Put $`k=e(r,q)`$, $`l=e(s,q)`$, and $`1\eta (k)\eta (l)`$. Then $`r`$ and $`s`$ are non-adjacent if and only if $`2\eta (k)+2\eta (l)>2n(1\epsilon (1)^{k+l})`$, $`k`$ and $`l`$ satisfy (3), and, if $`\epsilon =+`$, then the chain of equalities:
$$n=l=2\eta (l)=2\eta (k)=2k$$
(5)
is not true.
###### Proof.
Using (4) and Lemma 1.2 we prove the proposition as above. ∎
###### Proposition 2.5.
Let $`G`$ be a finite simple exceptional group of Lie type over a field of characteristic $`p`$, suppose that $`r,s`$ are odd primes, and assume that $`r,s\pi (G)\{p\}`$, $`k=e(r,q)`$, $`l=e(s,q)`$, and $`1kl`$. Then $`r`$ and $`s`$ are non-adjacent if and only if $`kl`$ and one of the following holds:
* $`G=G_2(q)`$ and either $`r3`$ and $`l\{3,6\}`$ or $`r=3`$ and $`l=93k`$.
* $`G=F_4(q)`$ and either $`l\{8,12\}`$, or $`l=6`$ and $`k\{3,4\}`$, or $`l=4`$ and $`k=3`$.
* $`G=E_6(q)`$ and either $`l=4`$ and $`k=3`$, or $`l=5`$ and $`k3`$, or $`l=6`$ and $`k=5`$, or $`l=8`$, $`k3`$, or $`l=8`$, $`r=3`$, and $`(q1)_3=3`$, or $`l=9`$, or $`l=12`$ and $`k3`$.
* $`G={}_{}{}^{2}E_{6}^{}(q)`$ and either $`l=6`$ and $`k=4`$, or $`l=8`$, $`k3`$, or $`l=8`$, $`r=3`$,and $`(q+1)_3=3`$, or $`l=10`$ and $`k3`$, or $`l=12`$ and $`k6`$, or $`l=18`$.
* $`G=E_7(q)`$ and either $`l=5`$ and $`k=4`$, or $`l=6`$ and $`k=5`$, or $`l\{14,18\}`$ and $`k2`$, or $`l\{7,9\}`$ and $`k2`$, or $`l=8`$ and $`k3,k4`$, or $`l=10`$ and $`k3,k6`$, or $`l=12`$ and $`k4,k6`$.
* $`G=E_8(q)`$ and either $`l=6`$ and $`k=5`$, or $`l\{7,14\}`$ and $`k3`$, or $`l=9`$ and $`k4`$, or $`l\{8,12\}`$ and $`k5,k6`$, or $`l=10`$ and $`k3,k4,6`$, or $`l=18`$ and $`k1,2,6`$, or $`l=20`$ and $`rk20`$, or $`l\{15,24,30\}`$.
* $`G={}_{}{}^{3}D_{4}^{}(q)`$ and either $`l=6`$ and $`k=3`$, or $`l=12`$.
###### Proof.
As for classical groups of Lie type, prime divisors $`r,s\pi (G)`$ satisfying the conditions of the proposition are adjacent if and only if $`rs`$ divides the order of some maximal torus of $`G`$. Thus, using Lemma 1.3 instead of Lemma 1.2 we prove the proposition as above. ∎
###### Proposition 2.6.
Let $`G`$ be a finite simple Suzuki or Ree group over a field of characteristic $`p`$, let $`r,s`$ be odd primes $`r,s\pi (G)\{p\}`$. Then $`r,s`$ are non-adjacent if and only if one of the following holds:
* $`G={}_{}{}^{2}B_{2}^{}(2^{2n+1})`$, $`r`$ divides $`m_k(B,n)`$, $`s`$ divides $`m_l(B,n)`$ and $`kl`$.
* $`G={}_{}{}^{2}G_{2}^{}(3^{2n+1})`$, $`r`$ divides $`m_k(G,n)`$, $`s`$ divides $`m_l(G,n)`$ and $`kl`$.
* $`G={}_{}{}^{2}F_{4}^{}(2^{2n+1})`$, $`r`$ divides $`m_k(F,n)`$, $`s`$ divides $`m_l(F,n)`$, $`kl`$, and if $`\{k,l\}=\{1,3\}`$, then $`r3s`$.
Numbers $`m_i(B,n)`$, $`m_i(G,n)`$, and $`m_i(F,n)`$ are defined in Lemma 1.5.
###### Proof.
We use Lemma 1.3, Lemma 1.5 and arguments as in the previous propositions of the section. ∎
## 3 Adjacency with the characteristic
In this section we consider whether a prime $`r`$ and the characteristic $`p`$ of the base field of a finite group of Lie type are adjacent.
###### Proposition 3.1.
Let $`G=O^p^{}(\overline{G}_\sigma )`$ be a finite simple classical group of Lie type defined over a field of characteristic $`p`$. Let $`r\pi (G)`$ and $`rp`$. Then $`r`$ and $`p`$ are non-adjacent if and only if one of the following holds:
* $`G=A_{n1}(q)`$, $`r`$ is odd, and $`e(r,q)>n2`$.
* $`G={}_{}{}^{2}A_{n1}^{}(q)`$, $`r`$ is odd, and $`\nu (e(r,q))>n2`$. The function $`\nu (m)`$ is defined in Proposition 2.2.
* $`G=C_n(q)`$, $`\eta (e(r,q))>n1`$. The function $`\eta (m)`$ is defined in Proposition 2.3.
* $`G=B_n(q)`$, $`\eta (e(r,q))>n1`$. The function $`\eta (m)`$ is defined in Proposition 2.3.
* $`G=D_n^\epsilon (q)`$, $`\eta (e(r,q))>n2`$. The function $`\eta (m)`$ is defined in Proposition 2.3.
* $`G=A_1(q)`$, $`r=2`$.
* $`G=A_2^\epsilon (q)`$, $`r=3`$ and $`(q\epsilon 1)_3=3`$.
###### Proof.
It is evident that $`2`$ and $`p`$ are adjacent in all classical simple groups except $`A_1(q)`$. Hence we may assume that $`r`$ is odd.
First we outline the general idea of the proof. We shall use as a technical tool in our proof, so we shall keep notations of . In order to prove that $`r`$ and $`p`$ are adjacent we find a connected reductive subgroup $`R`$ of maximal rank in $`G`$ such that $`R=T(G_1G_2)`$, where $`T`$ is a maximal torus, both $`G_1`$ and $`G_2`$ are non-trivial groups of Lie type such that $`r\pi (G_1)`$. Then $`G_1`$ contains an element of order $`r`$ and it centralizes $`G_2`$. Since $`G_2`$ is non-trivial, it contains an element of order $`p`$, so $`GR`$ contains an element of order $`rp`$.
In order to prove that $`r`$ and $`p`$ are non-adjacent we shall consider arbitrary element $`g`$ of order $`r`$ and its connected centralizer $`GC_{\overline{G}}(g)^0`$. Recall that $`C_{\overline{G}}(g)^0=\overline{S}\overline{L}`$, where $`\overline{S}=Z(C_{\overline{G}}(g)^0)`$ is a central torus and $`\overline{L}`$ is a semisimple part. Clearly $`g\overline{S}G\overline{S}_\sigma `$, hence $`r`$ divides $`|\overline{S}_\sigma |`$. This condition would imply that $`\overline{L}_\sigma `$ is trivial, hence $`C_{\overline{G}}(g)^0`$ does not contain unipotent elements. But every unipotent element of $`C_{\overline{G}}(g)`$ is contained in $`C_{\overline{G}}(g)^0N_{\overline{G}}(\overline{T})^0=\overline{T}`$. So none unipotent element of $`G`$ centralizes $`g`$. Now consider all classical groups case by case.
$`A_1(q)`$. It is known that if $`g`$ is an element of order $`rp`$, then $`(|C_{A_1(q)}(g)|,p)=1`$ (see, \[13, Proposition 7\]). So $`r,p`$ are non-adjacent for every $`r\pi (A_1(q))\{p\}`$.
$`A_2(q)`$. By using \[13, Proposition 7\] we obtain that only prime divisors of $`\frac{q1}{(3,q1)}`$ are adjacent to $`p`$ and the proposition is true in this case.
$`A_{n1}(q)`$, and $`n4`$. In our case $`T`$ is a Cartan subgroup, $`G_1=A_{n3}(q)`$ and $`G_2=A_1(q)`$. The existence of such a subgroup can be obtained by using \[13, Proposition 7\]. Thus every $`r`$ with $`e(r,q)n2`$ divides $`|G_1|`$, hence is adjacent to $`p`$. Now let $`e(r,q)=n1`$ and let $`g`$ be an element of order $`r`$. In view of (1) we obtain that $`q^{n1}1`$ must divide $`|\overline{S}_\sigma |(q1)`$. It follows from \[13, Proposition 7\] that $`GC_{\overline{G}}(g)^0`$ is a maximal torus of order $`\frac{1}{(n,q1)}(q^{n1}1)`$. Hence $`|\overline{L}_\sigma |=1`$ and $`r,p`$ are non-adjacent. If $`e(r,q)=n`$ and $`g`$ is an element of order $`r`$, then by (1), $`q^n1`$ must divide $`|\overline{S}_\sigma |(q1)`$. Again by using \[13, Proposition 7\] we obtain that $`GC_{\overline{G}}(g)^0`$ is a maximal torus of order $`\frac{1}{(n,q1)}\frac{q^n1}{q1}`$. So $`|\overline{L}_\sigma |=1`$ in this case and every $`r`$ with $`e(r,q)>n2`$ is non-adjacent to $`p`$.
$`{}_{}{}^{2}A_{2}^{}(q)`$. By using \[13, Proposition 8\] we obtain that only prime divisors of $`\frac{q+1}{(3,q+1)}`$ are adjacent to $`p`$ and the proposition is true in this case.
$`{}_{}{}^{2}A_{n1}^{}(q)`$ and $`n4`$. Again $`T`$ is a Cartan subgroup, $`G_1={}_{}{}^{2}A_{n3}^{}(q)`$ and $`G_2=A_1(q)`$. The existence of such a subgroup can be obtained by using \[13, Proposition 8\]. Thus every $`r`$ with $`\nu (e(r,q))n2`$ divides $`|G_1|`$, hence is adjacent to $`p`$. Now let $`\nu (e(r,q))=n1`$ and $`g`$ is an element of order $`r`$. Then in view of (2) we obtain that $`q^{n1}(1)^{n1}`$ divides $`|\overline{S}|(q+1)`$. It follows from \[13, Proposition 8\] that $`GC_{\overline{G}}(g)^0`$ is a maximal torus of order $`\frac{1}{(n,q+1)}(q^{n1}(1)^{n1})`$. Hence $`|\overline{L}_\sigma |=1`$ and $`r,p`$ are non-adjacent. If $`\nu (e(r,q))=n`$ and $`g`$ is an element of order $`r`$, then by (2), $`q^n(1)^n`$ divides $`|\overline{S}_\sigma |(q+1)`$. By using \[13, Proposition 8\] we obtain that $`GC_{\overline{G}}(g)^0`$ is a maximal torus of order $`\frac{1}{(n,q+1)}\frac{q^n(1)^n}{q+1}`$. Therefore $`|\overline{L}_\sigma |=1`$ and every $`r`$ with $`\nu (e(r,q))>n2`$ is non-adjacent to $`p`$.
$`C_n(q)`$. Take for $`R`$ torus $`T`$ as a Cartan subgroup, $`G_1=C_{n1}(q)`$, $`G_2=A_1(q)`$. Such a subgroup $`R`$ exists in view of \[13, Propositions 9 and 12\]. Again every $`r`$ with $`\eta (e(r,q))n1`$ divides $`|G_1|`$, hence is adjacent to $`p`$. If $`\eta (e(r,q))=n`$ and $`g`$ is an element of order $`r`$, then (4) implies that either $`q^n1`$, or $`q^n+1`$ divides $`|\overline{S}_\sigma |`$. From \[13, Propositions 9 and 12\] we obtain that $`GC_{\overline{G}}(g)^0`$ is a maximal torus of order $`\frac{1}{(2,q1)}(q^n\pm 1)`$ respectively, hence $`|\overline{L}_\sigma |=1`$ and $`r,p`$ are non-adjacent.
$`B_n(q)`$. Since $`B_2(q)C_2(q)`$ and $`B_n(2^t)C_n(2^t)`$, we may assume that $`p`$ is odd and $`n3`$. We can take $`T`$ to be a Cartan subgroup, $`G_1=B_{n2}(q)`$, and $`G_2=D_2(q)`$. Such a subgroup exists in view of \[13, Proposition 11\]. Since every prime $`r`$ with $`\eta (e(r,q))n2`$ divides the order of $`G_1`$, we obtain that $`r`$ and $`p`$ are adjacent if $`\eta (e(r,q))n2`$. If $`\eta (e(r,q))=n1`$, then, again by \[13, Proposition 11\], there exists a reductive subgroup $`R`$ such that $`|\overline{S}_\sigma |=q^{\eta (e(r,q))}+(1)^{e(r,q)}`$ and $`\overline{L}_\sigma =B_1(q)A_1(q)`$. Hence $`r,p`$ are adjacent in $`G`$. Assume now that $`\eta (e(r,q))=n`$ and that $`g`$ is an element of order $`r`$ of $`G`$. Then the order $`|\overline{S}_\sigma |`$ is given in \[13, Proposition 11\] and is equal to $`_i(q^{n_i}\pm 1)`$, where $`_in_in`$. By using (4) we obtain that $`\eta (e(r,q))=n`$ divides $`n_i`$ for some $`i`$. Hence either $`q^n1`$, or $`q^n+1`$ divides $`|S_\sigma |`$. In view of \[13, Proposition 11\] this implies that $`|\overline{L}_\sigma |=1`$, hence $`r,p`$ are non-adjacent.
$`D_n^\epsilon (q)`$. We can take $`T`$ to be a Cartan subgroup, $`G_1=D_{n2}^\epsilon (q)`$, $`G_2=A_1(q)`$. The existence of such a subgroup $`R`$ can be obtained by using \[13, Proposition 10\]. So $`r,p`$ are adjacent for all $`r`$ with $`\eta (e(r,q))n2`$, except one case: $`\eta (e(r,q))=n2`$ and $`r`$ divides $`q^{n2}+\epsilon 1`$. In this last case we obtain that there exists a reductive subgroup $`R`$ such that $`|\overline{S}_\sigma |=q^{n2}+\epsilon 1`$ and $`\overline{L}_\sigma D_2^\epsilon (q)`$. Hence in this exceptional case $`r,p`$ are adjacent also. Now if $`\eta (e(r,q))n1`$ and $`g`$ is an element of $`G`$ of order $`r`$, then as above we obtain that $`q^{\eta (e(r,q))}+(1)^{e(r,q)}`$ divides $`|\overline{S}_\sigma |`$ and \[13, Proposition 10\] implies that $`\overline{L}_\sigma `$ is trivial. Therefore $`r,p`$ are non-adjacent in this case. ∎
###### Proposition 3.2.
Let $`G`$ be a finite simple exceptional group of Lie type over a field of characteristic $`p`$. Let $`r\pi (G)`$, $`k=e(r,q)`$, and $`rp`$. Then $`r,p`$ are non-adjacent if and only if one of the following holds:
* $`G=G_2(q)`$, $`k\{3,6\}`$.
* $`G=F_4(q)`$, $`k\{8,12\}`$.
* $`G=E_6(q)`$, $`k\{8,9,12\}`$.
* $`G={}_{}{}^{2}E_{6}^{}(q)`$, $`k\{8,12,18\}`$.
* $`G=E_7(q)`$, $`k\{7,9,14,18\}`$.
* $`G=E_8(q)`$, $`k\{15,20,24,30\}`$.
* $`G={}_{}{}^{3}D_{4}^{}(q)`$, $`k=12`$.
###### Proof.
All statements are obtained by using information about conjugacy classes and centralizers of semisimple elements given in and . ∎
###### Proposition 3.3.
Let $`G`$ be a finite simple Suzuki or Ree group over a field of characteristic $`p`$, let $`r\pi (G)\{p\}`$. Then $`r,p`$ are non-adjacent if and only if one of the following holds:
* $`G={}_{}{}^{2}B_{2}^{}(2^{2n+1})`$, $`r`$ divides $`m_k(B,n)`$.
* $`G={}_{}{}^{2}G_{2}^{}(3^{2n+1})`$, $`r`$ divides $`m_k(G,n)`$ and $`r2`$.
* $`G={}_{}{}^{2}F_{4}^{}(2^{2n+1})`$, $`r`$ divides $`m_k(F,n)`$, $`r3`$, and $`k>2`$.
Numbers $`m_i(B,n)`$, $`m_i(G,n)`$, and $`m_i(F,n)`$ are defined in Lemma 1.5.
###### Proof.
All statements are obtained by using information about conjugacy classes and centralizers of semisimple elements given in . ∎
## 4 Adjacency of 2 and an odd prime $`r`$
In this section we consider whether $`2`$ and an odd prime $`r`$ are adjacent if both of them are not equal to the characteristic $`p`$ of the base field. We start from groups $`A_n^\epsilon (q)`$. Recall that for those groups we did not consider adjacency criterion for prime divisors of $`q\epsilon 1`$ in Section 2. It is natural to consider such a criterion together with a criterion for 2, hence we give it in the following two propositions.
###### Proposition 4.1.
Let $`G=A_{n1}(q)`$ be a finite simple group of Lie type. Let $`r`$ be a prime divisor of $`q1`$ and $`s`$ be an odd prime distinct from the characteristic. Denote $`k=e(s,q)`$. Then $`s`$ and $`r`$ are non-adjacent if and only if one of the following holds:
* $`k=n`$, $`n_r(q1)_r`$, and if $`n_r=(q1)_r`$, then $`2<(q1)_r`$.
* $`k=n1`$ and $`(q1)_rn_r`$.
###### Proof.
First we prove the following statement
$$\begin{array}{c}\left(\frac{q^n1}{q1}\right)_r=n_r\text{, if }(q1)_rn_r\text{ and }(q1)_r>2,\hfill \\ \hfill \left(\frac{q^n1}{q1}\right)_2>2\text{, if }(q1)_2=n_2=2.\end{array}$$
(6)
Assume that $`r`$ is odd. Then $`n=r^kl`$, where $`(l,r)=1`$ and $`q=r^km+1`$. Now $`q^n1=(q^{r^k}1)(q^{nr^k}+q^{n2r^k}+\mathrm{}+q^{r^k}+1)`$. The second multiplier is the sum of $`l`$ numbers of type $`q^i`$ and $`q^i1(\mathrm{mod}r)`$ for all $`i`$. Since $`(l,r)=1`$ it follows that the second multiplier is coprime to $`r`$. Thus $`\left(\frac{q^n1}{q1}\right)_r=\left(\frac{q^{r^k}1}{q1}\right)_r`$. Now remember that $`q=r^km+1`$, hence
$$\frac{q^{r^k}1}{q1}=(r^km)^{r^k1}+r^k(r^km)^{r^k2}+\mathrm{}+\frac{1}{2}r^k(r^k1)(r^km)+r^k.$$
Since $`r^{k+1}`$ divides all summands, except the last, and $`r^{k+1}`$ does not divide $`r^k`$ we obtain that $`r^k`$ divides $`\left(\frac{q^{r^k}1}{q1}\right)_r`$, but $`r^{k+1}`$ does not divide $`\left(\frac{q^{r^k}1}{q1}\right)_r`$. Therefore $`\left(\frac{q^{r^k}1}{q1}\right)_r=n_r`$ in this case.
Assume now that $`r=2`$, $`n=2^kl`$, where $`l`$ is odd, and $`q=2^km+1`$. Then
$$\left(\frac{q^n1}{q1}\right)_2=\left(\frac{q^l1}{q1}\right)_2(q^l+1)(q^{2l}+1)\mathrm{}(q^{2^{k1}l}+1).$$
Since $`l`$ is odd we have that $`\left(\frac{q^l1}{q1}\right)_2=1`$. If $`(q1)_2>2`$ we have that $`(q^i+1)_2=2`$ for all $`i`$. Therefore
$$\left(\frac{q^n1}{q1}\right)_2=(q^l+1)_2(q^{2l}+1)_2\mathrm{}(q^{2^{k1}l}+1)_2=2^k=n_2.$$
The second case $`(q1)_2=n_2=2`$ is evident.
Assume that $`kn2`$. By Lemma 1.2 it follows that there exists a maximal torus $`T`$ of $`G`$ of order $`\frac{1}{(n,q1)}(q^k1)(q1)^{nk1}`$. We have that $`T`$ is an abelian subgroup of $`G`$, and both $`r,s`$ divide $`|T|`$. Therefore $`T`$ contains an element of order $`rs`$.
Assume that $`k=n`$. By (1) and Lemma 1.2 every element of order $`s`$ is contained in a maximal torus $`T`$ of order $`\frac{1}{(n,q1)}\frac{q^n1}{q1}`$. In view of (6) $`r`$ does not divide $`|T|`$ if and only if condition 1 of the proposition holds.
Assume now that $`k=n1`$. By (1) and Lemma 1.2 every element of order $`s`$ is contained in a maximal torus $`T`$ of order $`\frac{1}{(n,q1)}(q^{n1}1)`$. We have that $`(q^{n1}1)=(q1)(q^{n2}+q^{n3}+\mathrm{}+q+1)`$ and $`(q^{n2}+q^{n3}+\mathrm{}+q+1)_r=1`$. Therefore $`r`$ does not divide $`|T|`$ if and only if $`(q1)_rn_r`$ and we obtain condition 2 of the proposition. ∎
###### Proposition 4.2.
Let $`G={}_{}{}^{2}A_{n1}^{}(q)`$ be a finite simple group of Lie type. Let $`r`$ be a prime divisor of $`q+1`$ and $`s`$ be an odd prime distinct from the characteristic. Denote $`k=e(s,q)`$. Then $`s`$ and $`r`$ are non-adjacent if and only if one of the following holds:
* $`\nu (k)=n`$, $`n_r(q+1)_r`$, and if $`n_r=(q+1)_r`$, then $`2<(q+1)_r`$.
* $`\nu (k)=n1`$ and $`(q+1)_rn_r`$.
The function $`\nu (m)`$ is defined in Proposition 2.2.
###### Proof.
Like in the previous proposition first we show that
$$\begin{array}{c}\left(\frac{q^n(1)^n}{q+1}\right)_r=n_r\text{, if }(q+1)_rn_r\text{ and }(q+1)_r>2,\hfill \\ \hfill \left(\frac{q^n(1)^n}{q+1}\right)_2>2\text{, if }(q+1)_2=n_2=2.\end{array}$$
(7)
Assume that $`r`$ is odd. Then $`n=r^kl`$, where $`(l,r)=1`$ and $`q=r^km1`$. Now $`q^n(1)^n=(q^{r^k}+1)(q^{nr^k}q^{n2r^k}+\mathrm{}+(1)^{l1}q^{r^k}+(1)^l)`$. The second multiplier is the sum of $`l`$ numbers of type $`(1)^tq^{n(t+1)r^k}`$ and $`q^{n(t+1)r^k}(1)^{n+t1}(\mathrm{mod}r)`$ for all $`t`$. Hence the second multiplier is equivalent $`(1)^{n1}l`$ modulo $`r`$. Since $`(l,r)=1`$ it follows that the second multiplier is coprime to $`r`$. Thus $`\left(\frac{q^n(1)^n}{q+1}\right)_r=\left(\frac{q^{r^k}+1}{q+1}\right)_r`$. Now remember that $`q=r^km1`$, hence
$$\frac{q^{r^k}+1}{q+1}=(r^km)^{r^k1}r^k(r^km)^{r^k2}+\mathrm{}+(1)^{k2}\frac{1}{2}r^k(r^k1)(r^km)+(1)^{k1}r^k.$$
Since $`r^{k+1}`$ divides all summands, except the last, and $`r^{k+1}`$ does not divide $`r^k`$ we obtain that $`r^k`$ divides $`\left(\frac{q^{r^k}+1}{q+1}\right)_r`$, but $`r^{k+1}`$ does not divide $`\left(\frac{q^{r^k}+1}{q+1}\right)_r`$. Therefore $`\left(\frac{q^{r^k}+1}{q+1}\right)_r=n_r`$ in this case.
Assume now that $`r=2`$, $`n=2^kl`$, where $`l`$ is odd, and $`q=2^km1`$. Then
$$\left(\frac{q^n(1)^n}{q+1}\right)_2=\left(\frac{q^l+1}{q+1}\right)_2(q^l1)(q^{2l}+1)\mathrm{}(q^{2^{k1}l}+1).$$
Since $`l`$ is odd we have that $`\left(\frac{q^l+1}{q+1}\right)_2=1`$. If $`(q+1)_2>2`$ we have that $`(q^{2i}+1)_2=2`$ for all $`i1`$ and $`(q^l1)_2=2`$. Therefore
$$\left(\frac{q^n(1)^n}{q+1}\right)_2=(q^l1)_2(q^{2l}+1)_2\mathrm{}(q^{2^{k1}l}+1)_2=2^k=n_2.$$
The second case $`(q+1)_2=n_2=2`$ is evident.
Assume that $`\nu (k)n2`$. By Lemma 1.2 it follows that there exists a maximal torus $`T`$ of $`G`$ of order $`\frac{1}{(n,q+1)}(q^{\eta (k)}(1)^{\eta (k)})(q+1)^{nk1}`$. We have that $`T`$ is an abelian subgroup of $`G`$ and both $`r,s`$ divide $`|T|`$. Therefore $`T`$ contains an element of order $`rs`$.
Assume that $`\nu (k)=n`$. By (2) and Lemma 1.2 every element of order $`s`$ is contained in a maximal torus $`T`$ of order $`\frac{1}{(n,q+1)}\frac{q^n(1)^n}{q+1}`$. In view of (7) $`r`$ does not divide $`|T|`$ if and only if condition 1 of the proposition holds.
Assume now that $`\nu (k)=n1`$. By (2) and Lemma 1.2 every element of order $`s`$ is contained in a maximal torus $`T`$ of order $`\frac{1}{(n,q+1)}(q^{n1}(1)^{n1})`$. We have that $`(q^{n1}(1)^{n1})=(q+1)(q^{n2}q^{n3}+\mathrm{}+(1)^{n3}q+(1)^{n2})`$ and $`(q^{n2}q^{n3}+\mathrm{}+(1)^{n3}q+(1)^{n2})_r=1`$. Therefore $`r`$ does not divide $`|T|`$ if and only if $`(q+1)_rn_r`$ and we obtain condition 2 of the proposition. ∎
###### Proposition 4.3.
Let $`G=B_n(q)`$ or $`G=C_n(q)`$ be a finite simple group of Lie type over a field of odd characteristic $`p`$. Let $`r`$ be an odd prime divisor of $`|G|`$, $`rp`$, and $`k=e(r,q)`$. Then $`2,r`$ are non-adjacent if and only if $`\eta (k)=n`$ and one of the following holds:
* $`n`$ is odd and $`k=(3e(2,q))n`$.
* $`n`$ is even and $`k=2n`$.
The function $`\eta (m)`$ is defined in Proposition 2.3.
###### Proof.
If $`\eta (k)n1`$, then by Lemma 1.2 there exists a maximal torus $`T`$ of order $`\frac{1}{2}(q^{\eta (k)}+(1)^k)(q1)^{n\eta (k)}`$. We have that $`T`$ is an abelian subgroup of $`G`$ and $`2,r`$ both divide $`|T|`$. Hence $`T`$ contains an element of order $`2r`$ and $`2,r`$ are adjacent.
Thus we may assume that $`\eta (k)=n`$. In view of (4) and Lemma 1.2 we have that every element $`g`$ of order $`r`$ is contained in a maximal torus $`T`$ of order $`\frac{1}{2}(q^n+(1)^k)`$. Therefore $`2,r`$ are non-adjacent if and only if 2 does not divide $`|T|`$ and the proposition follows. ∎
###### Proposition 4.4.
Let $`G=D_n^\epsilon (q)`$ be a finite simple group of Lie type over a field of odd characteristic $`p`$. Let $`r`$ be an odd prime divisor of $`|G|`$, $`rp`$, and $`k=e(r,q)`$. Then $`2`$ and $`r`$ are non-adjacent if and only if one of the following holds:
* $`\eta (k)=n`$ and $`(4,q^n\epsilon 1)=(q^n\epsilon 1)_2`$.
* $`\eta (k)=k=n1`$, $`n`$ is even, $`\epsilon =+`$, and $`e(2,q)=2`$.
* $`\eta (k)=\frac{k}{2}=n1`$, $`\epsilon =+`$ and $`e(2,q)=1`$.
* $`\eta (k)=\frac{k}{2}=n1`$, $`n`$ is odd, $`\epsilon =`$, and $`e(2,q)=2`$.
The function $`\eta (m)`$ is defined in Proposition 2.3.
###### Proof.
First note that if $`\eta (k)n2`$, then by Lemma 1.2 there exists a maximal torus $`T`$ of order $`\frac{1}{(4,q^n\epsilon 1)}(q^{\eta (k)}+(1)^k)(q+(\epsilon 1)^k)(q1)^{n\eta (k)1}`$. We have that $`2,r`$ both divide $`|T|`$, hence $`2,r`$ are adjacent.
Now consider the case $`G={}_{}{}^{2}D_{n}^{}(q)`$ (i. e. $`\epsilon =`$). If $`\eta (k)=n1`$ then, by (4) and Lemma 1.2, we have that every element of order $`r`$ is contained in a maximal torus $`T`$ of order $`\frac{1}{(4,q^n+1)}(q^{n1}+(1)^k)(q(1)^k)`$. Then 2 does not divide $`|T|`$ if and only if condition 4 of the proposition holds. If $`\eta (k)=n`$, then, by (4) and Lemma 1.2, every element of order $`r`$ is contained in a maximal torus $`T`$ of order $`\frac{1}{(4,q^n+1)}(q^n+1)`$ (in particular, $`k=2\eta (k)`$). Then 2 does not divide $`|T|`$ if and only if $`(4,q^n+1)=(q^n+1)_2`$ and we obtain the proposition for twisted groups.
Assume now that $`G=D_n(q)`$. If $`\eta (k)=n1`$, then, by (4) and Lemma 1.2 we obtain that every element of order $`r`$ is contained in a maximal torus $`T`$ of order $`\frac{1}{(4,q^n1)}(q^{n1}+(1)^k)(q+(1)^k)`$ and $`k=n1`$ if $`k`$ is odd. Then 2 does not divide $`|T|`$ if and only if condition 2 or condition 3 of the proposition hold. If $`\eta (k)=n`$, then $`n`$ is odd and, by (4) and Lemma 1.2, every element of order $`r`$ is contained in a maximal torus $`T`$ of order $`\frac{1}{(4,q^n1)}(q^n1)`$. Clearly, $`2`$ does not divide $`|T|`$ if and only if $`(4,q^n1)=(q^n1)_2`$ and the proposition follows. ∎
###### Proposition 4.5.
Let $`G`$ be a finite simple exceptional group of Lie type over a field of odd characteristic $`p`$. Let $`r`$ be an odd prime divisor of $`|G|`$, $`rp`$, and $`k=e(r,q)`$. Then $`2`$ and $`r`$ are non-adjacent if and only if one of the following holds:
* $`G=G_2(q)`$, $`k\{3,6\}`$.
* $`G=F_4(q)`$, $`k=12`$.
* $`G=E_6(q)`$, $`k\{9,12\}`$.
* $`G={}_{}{}^{2}E_{6}^{}(q)`$, $`k\{12,18\}`$.
* $`G=E_7(q)`$, and either $`k\{7,9\}`$ and $`e(2,q)=2`$, or $`k\{14,18\}`$ and $`e(2,q)=1`$.
* $`G=E_8(q)`$, $`k\{15,20,24,30\}`$.
* $`G={}_{}{}^{3}D_{4}^{}(q)`$, $`k=12`$.
* $`G={}_{}{}^{2}G_{2}^{}(3^{2n+1})`$, and $`r`$ divides $`m_3(G,n)`$ or $`m_4(G,n)`$, where $`m_i(G,n)`$ are defined in Lemma 1.5.
###### Proof.
The proposition is immediate from Lemmas 1.3 and 1.5. ∎
## 5 On connection between the properties of the prime graph $`GK(G)`$ and the structure of $`G`$
Let $`GK(G)`$ be a prime graph of a finite group $`G`$. Obviously, the spectrum $`\omega (G)`$ uniquely determines the structure of $`GK(G)`$. Denote by $`s(G)`$ the number of connected components of $`GK(G)`$ and by $`\pi _i(G)`$, $`i=1,\mathrm{},s(G)`$, the $`i`$th connected component of $`GK(G)`$. If $`G`$ has even order then put $`2\pi _1(G)`$. Denote by $`\omega _i(G)`$ the set of numbers $`n\omega (G)`$ such that each prime divisor of $`n`$ belongs to $`\pi _i(G)`$.
Gruenberg and Kegel obtained the following description of finite groups with disconnected prime graph.
###### Theorem 5.1.
(Gruenberg — Kegel Theorem) (see ) If $`G`$ is a finite group with disconnected graph $`GK(G)`$, then one of the following statements holds:
(a) $`s(G)=2`$ and $`G`$ is a Frobenius group;
(b) $`s(G)=2`$ and $`G`$ is a $`2`$-Frobenius group, i.e., $`G=ABC`$, where $`A`$, $`AB`$ are normal subgroups of $`G`$; $`AB`$, $`BC`$ are Frobenius groups with cores $`A`$, $`B`$ and complements $`B`$, $`C`$ respectively;
(c) there exists a nonabelian simple group $`S`$ such that $`S\overline{G}=G/K\mathrm{Aut}(S)`$, where $`K`$ is a maximal normal soluble subgroup of $`G`$. Furthermore, $`K`$ and $`\overline{G}/S`$ are $`\pi _1(G)`$-subgroups, the graph $`GK(S)`$ is disconnected, $`s(S)s(G)`$, and for every $`i`$, $`2is(G)`$, there is $`j`$, $`2js(S)`$, such that $`\omega _i(G)=\omega _j(S)`$.
Together with the classification of finite simple groups with disconnected prime graph obtained by Williams and Kondrat’ev (see and ) the Gruenberg — Kegel theorem implied a series of important corollaries (see, for example, \[1, Theorems 3–6\] and \[2, Theorems 2–3\]). In last years this theorem is used for the proving of recognizability of finite groups by spectrum (for details see and ).
The proof of the Gruenberg — Kegel Theorem substantially used the fact that in a group $`G`$ (if its order is even) there exists an element of odd prime order disconnected in $`GK(G)`$ with a prime $`2`$. It turns out that disconnectedness of $`GK(G)`$ could be successfully replaced in most cases by a weaker condition for the prime $`2`$ to be nonadjacent to at least one odd prime.
Denote by $`t(G)`$ the maximal number of prime dividers of the order of $`G`$ pairwise non-adjacent in $`GK(G)`$. In other words, $`t(G)`$ is a maximal number of vertices in independent sets of $`GK(G)`$ (a set of vertices is called independent if its elements are pairwise non-adjoint). In graph theory this number is usually called an independence number of the graph. By analogy we denote by $`t(r,G)`$ the maximal number of vertices in independent sets of $`GK(G)`$ containing the prime $`r`$. We call this number an $`r`$-independence number.
###### Theorem 5.2.
(see ) Let $`G`$ be a finite group satisfying two conditions:
(a) there exist three primes in $`\pi (G)`$ whose are pairwise non-adjacent in $`GK(G)`$, i. e., $`t(G)3`$;
(b) there exists an odd prime in $`\pi (G)`$ which is non-adjacent to prime $`2`$ in $`GK(G)`$, i. e., $`t(2,G)2`$.
Then there exists a finite nonabelian simple group $`S`$ such that $`S\overline{G}=G/K\mathrm{Aut}(S)`$ for maximal normal soluble subgroup $`K`$ of $`G`$. Furthermore, $`t(S)t(G)1`$ and one of the following statements holds:
(1) $`SAlt_7`$ or $`A_1(q)`$ for some odd $`q`$ and $`t(S)=t(2,S)=3`$.
(2) For every prime $`r\pi (G)`$ non-adjacent in $`GK(G)`$ to $`2`$ the Sylow $`r`$-subgroup of $`G`$ is isomorphic to the Sylow $`r`$-subgroup of $`S`$. In particular, $`t(2,S)t(2,G)`$.
Note that the condition (a) of Theorem 5.2 easily implies the insolubility of group $`G`$ and so, by the Feit — Thompson Odd Theorem, implies that $`G`$ is of even order. Moreover, the condition (a) can be replaced by the weaker condition of the insolubility of $`G`$ (see \[4, Propositions 2–3\]). Together with \[18, Theorem 2\] Theorem 5.2 implies the following statement.
###### Theorem 5.3.
(see \[4, Proposition 4\]) Let $`L`$ be a finite simple group with $`t(2,L)2`$ non-isomorphic to $`A_2(3)`$, $`{}_{}{}^{2}A_{3}^{}(3)`$, $`C_2(3)`$ and $`Alt_{10}`$. Let $`G`$ be a finite group with $`\omega (G)=\omega (L)`$. Then for a group $`G`$ the conclusion of Theorem 5.2 holds true. In particular, $`G`$ has a unique nonabelian composition factor.
As a matter of fact, these recent results (Theorem 5.2 and Theorem 5.3) gave the motivation for the present work. In Section 6 using results of Sections 14 we calculate the values of the independence numbers and the 2-independence numbers for all finite nonabelian simple groups. Furthermore, for every finite nonabelian simple group of Lie type over the field of characteristic $`p`$ we also determine $`p`$-independence number $`t(p,G)`$.
## 6 Independence numbers
For a finite group $`G`$ denote by $`\rho (G)`$ (by $`\rho (r,G)`$) some independence set in $`GK(G)`$ (containing $`r`$) with maximal number of vertices. Thus, $`|\rho (G)|=t(G)`$ and $`|\rho (r,G)|=t(r,G)`$. For a given finite nonabelian simple group we point out some $`\rho (G)`$ and $`\rho (2,G)`$ (obviously, they may be not uniquely defined) and so determine $`t(G)`$ and $`t(2,G)`$.
###### Proposition 6.1.
Let $`G`$ be a sporadic group. Then $`\rho (G)`$, $`t(G)`$, $`\rho (2,G)`$ and $`t(2,G)`$ are listed in Table 2. Furthermore, $`\rho (2,G)`$ is uniquely determined.
###### Proof.
The results are easy to obtain using or . ∎
Now we deal with simple alternating groups.
###### Proposition 6.2.
Let $`G=Alt_n`$ be an alternating group of degree $`n5`$. Let $`s_n^{}`$ be the largest prime which does not exceed $`n/2`$ and $`s_n^{\prime \prime }`$ be the smallest prime greater than $`n/2`$. Denote by $`\tau (n)`$ the set $`\{s|s\text{ is a prime, }n/2<sn\}`$ and by $`\tau (2,n)`$ the set $`\{s|s\text{ is a prime, }n3sn\}`$. Then $`\rho (G)`$, $`t(G)`$, $`\rho (2,G)`$ and $`t(2,G)`$ are listed in Table 3. Furthermore, $`\rho (2,G)`$ is uniquely determined.
###### Proof.
The result is immediate corollary of Proposition 1.1. ∎
Now we consider groups of Lie type. Since the problem here is more complicated we divide the process of its solution into several natural steps. The main tools will be a Zsigmondy Theorem, Lemma 1.5 and our results from Sections 2-4. Recall that for a given natural number $`q`$ we denote by $`r_n`$ some primitive prime divisor of $`q^n1`$ (if it exists). Note that a primitive prime divisor $`r_{2m}`$ of $`q^{2m}1`$ divides $`q^m+1`$ and does not divide $`q^k+1`$ for every natural $`k<m`$. Since arguments in even characteristic of the field of definition are suitable for every characteristic $`p`$ we start with determination of $`t(p,G)`$.
###### Proposition 6.3.
Let $`G`$ be a finite simple group of Lie type over a field of characteristic $`p`$ and order $`q`$. Then $`\rho (p,G)`$ and $`t(p,G)`$ are listed in Table 4 for classical groups and Table 5 for exceptional groups.
###### Proof.
(1) $`G=A_{n1}^\epsilon (q)`$.
Let $`G=A_1(q)`$. Since $`A_1(2)`$ and $`A_1(3)`$ are not simple we may assume that $`q>3`$. By Zsigmondy Theorem, if $`q2^t1`$, then there exist primitive prime divisors $`r_1`$ and $`r_2`$ of $`q1`$ and $`q+1`$. They are non-adjacent in $`GK(G)`$. By Proposition 3.1, they are non-adjacent to $`p`$. Since $`q>3`$, for any $`q=2^t1`$ there exists odd primitive prime divisor $`r_1`$ of $`q1`$ which is not adjacent to $`2`$ in $`GK(G)`$. Since in this case $`e(2,q)=2`$ and $`q+1=2^t`$, the prime $`2`$ is a unique primitive prime divisor of $`q^21`$. Hence $`\rho (p,G)=\{p,r_1,r_2\}`$ in this case too.
Suppose that $`n>2`$. In order to consider groups $`A_{n1}(q)`$ and $`{}_{}{}^{2}A_{n1}^{}(q)`$ together we define new function:
$$\nu _\epsilon (m)=\{\begin{array}{cc}\hfill m& \text{, if either }\epsilon =+,\text{ or }\epsilon =\text{ and }m0(\mathrm{mod}4),\hfill \\ \hfill \frac{m}{2}& \text{, if }\epsilon =\text{ and }m2(\mathrm{mod}4),\hfill \\ \hfill 2m& \text{, if }\epsilon =\text{ and }m1(\mathrm{mod}2).\hfill \end{array}$$
Obviously, that $`\nu _\epsilon (m)`$ is the identity function if $`\epsilon =+`$, and $`\nu _\epsilon (m)=\nu (m)`$ if $`\epsilon =`$. It is easy to check that $`\nu _\epsilon `$ is a bijection on $``$, thus $`\nu _\epsilon ^1`$ is well defined.
Let $`n=3`$ and $`G=A_2^\epsilon (q)`$. Then, by Proposition 3.1, it follows that $`r_{\nu _\epsilon ^1(2)}`$ and $`r_{\nu _\epsilon ^1(3)}`$ are non-adjacent to $`p`$. Moreover, if $`(q\epsilon 1)_3=3`$, then, by Proposition 3.1, we have that 3 is non-adjacent to $`p`$ and, by Propositions 4.1 and 4.2, in this case 3 is non-adjacent to $`r_{\nu _\epsilon ^1(2)}`$ and $`r_{\nu _\epsilon ^1(3)}`$. By Zsigmondy Theorem $`r_{\nu _\epsilon ^1(3)}`$ exists for all $`q`$, except $`\epsilon =`$ and $`q=2`$. But $`{}_{}{}^{2}A_{2}^{}(2)`$ is not simple and we do not consider this group. Therefore $`r_{\nu _\epsilon ^1(3)}`$ exists for all simple groups in this case. Again by Zsigmondy Theorem, $`r_{\nu _{\epsilon ^1(2)}}`$ exists if and only if $`q+\epsilon 1`$ is not a power of $`2`$, i.e., it is not a Mersenne prime in case $`\epsilon =+`$, and it is not a Fermat prime or $`9`$ in case $`\epsilon =`$. Thus, we have
$$\rho (p,G)=\{\begin{array}{cc}\hfill \{p,3,r_{\nu _\epsilon ^1(2)},r_{\nu _\epsilon ^1(3)}\},\text{ if }(q\epsilon 1)_3=3\text{ and }q+\epsilon 2^k;& \\ \hfill \{p,r_{\nu _\epsilon ^1(2)},r_{\nu _\epsilon ^1(3)}\},\text{ if }(q\epsilon 1)_33\text{ and }q+\epsilon 2^k;& \\ \hfill \{p,3,r_{\nu _\epsilon ^1(3)}\},\text{ if }(q\epsilon 1)_3=3\text{ and }q+\epsilon =2^k;& \\ \hfill \{p,r_{\nu _\epsilon ^1(3)}\},\text{ if }(q\epsilon 1)_33\text{ and }q+\epsilon =2^k.& \end{array}$$
Note that we avoid to use the functions $`e(r,q)`$ and $`\nu _\epsilon `$ in the tables in Section 8.
Let $`G=A_4(2),A_5(2)\text{ or }{}_{}{}^{2}A_{3}^{}(2)`$. Since there are no primitive prime divisors of $`2^61`$, by Proposition 3.1 we have that $`\rho (2,A_5(2))=\{2,31\}`$, $`\rho (2,A_6(2))=\{2,127\}`$, and $`\rho (2,{}_{}{}^{2}A_{3}^{}(2))=\{2,5\}.`$
In all other cases by Zsigmondy Theorem there exist primitive prime divisors $`r_{\nu _\epsilon ^1(n1)}`$ and $`r_{\nu _\epsilon ^1(n)}`$. By Lemma 1.2 we have $`r_{\nu _\epsilon ^1(n1)},r_{\nu _\epsilon ^1(n)}\pi (G)`$. By Propositions 2.1 and 2.2 they are non-adjacent in $`GK(G)`$. Now Proposition 3.1 yields that for $`G=A_{n1}^\epsilon (q)`$ we have $`\rho (p,G)=\{p,r_{\nu _\epsilon ^1(n1)},r_{\nu _\epsilon ^1(n)}\}`$.
(2) $`G=C_n(q)`$ or $`B_n(q)`$.
In view of Proposition 2.3, Proposition 3.1, and Proposition 4.3, we have that the prime graphs of $`C_n(q)`$ and $`B_n(q)`$ coincide. So we consider these groups together and, for brevity, use the symbol $`C_n(q)`$ in both cases.
Let $`G=C_3(2)`$. Since there are no primes $`r`$ with $`e(r,2)=6`$, only $`7`$ as the primitive prime divisor of $`2^31`$ is not adjacent to $`2`$.
Let $`G=C_n(q)`$, $`n2`$ and $`(n,q)(3,2)`$. If $`n`$ is even, by Proposition 3.1 only primitive prime divisors of $`q^n+1`$ are non-adjacent to $`p`$. Thus, in this case $`\rho (G)=\{p,r_{2n}\}`$. If $`n`$ is odd, the Propositions 2.3 and 3.1 yield $`\rho (G)=\{p,r_n,r_{2n}\}`$.
(3) $`G=D_n^\epsilon (q)`$.
With respect to well-known isomorphisms of groups of small Lie rank, we may suppose that $`n4`$. Let $`n=4`$ and $`q=2`$. Since there are no primitive prime divisors of $`2^61`$ we have that only $`7`$ as a primitive prime divisor of $`2^31`$ is non-adjacent to $`2`$ in case $`G=D_4(2)`$ and only $`7`$ and $`17`$ as primitive divisors of $`2^31`$ and $`2^81`$ are non-adjacent to $`2`$ in case $`G={}_{}{}^{2}D_{4}^{}(2)`$. All others possibilities could be easily described in the direct accordance to Proposition 2.4 and Proposition 3.1. The results of such consideration one can see in Table 4.
(4) The result for exceptional groups distinct from Suzuki and Ree groups can be obtained by using Proposition 2.5, Proposition 3.2 and the Zsigmondy Theorem.
(5) $`G`$ is a finite simple Suzuki or Ree group.
Let $`G={}_{}{}^{2}B_{2}^{}(2^{2m+1})`$, and $`s_i`$ be a prime divisor of $`m_i(B,n)`$, where $`i=1,2,3`$ (see Lemma 1.5). By Proposition 2.6 primes $`s_i`$ and $`s_j`$ are adjacent if and only if $`i=j`$. On the other hand, every $`s_i`$, where $`i=1,2,3`$, is non-adjacent to $`p=2`$. Thus, $`\rho (p,G)=\{p,s_1,s_2,s_3\}`$ and $`t(p,G)=4`$ in this case. The same arguments one can apply to Ree groups and prime divisors of $`m_i(G,n)`$ and $`m_i(F,n)`$. ∎
In general, a primitive prime divisor $`r_m`$ of $`q^m1`$ could be chosen by several ways. Thus, the set $`\rho (p,G)`$ could be not uniquely determined for a finite simple group $`G`$ of Lie type. However, it turns out that the values of $`e(r,q)`$ for all primitive prime divisors $`r`$ from $`\rho (p,G)`$ are invariants for a given group $`G`$ of Lie type.
###### Proposition 6.4.
Let $`G`$ be a finite simple group of Lie type over a field of characteristic $`p`$ and order $`q`$. Assume that $`G`$ is not isomorphic to $`{}_{}{}^{2}B_{2}^{}(q)`$, $`{}_{}{}^{2}G_{2}^{}(q)`$, $`{}_{}{}^{2}F_{4}^{}(2)^{}`$, and $`{}_{}{}^{2}F_{4}^{}(q)`$. Let $`\rho (p,G)=\{p,s_1,s_2,\mathrm{},s_m\}`$ be an independence set in $`GK(G)`$ containing $`p`$ with maximal number of vertices and $`k_i=e(s_i,q)`$. Then the set $`\{k_1,\mathrm{},k_m\}`$ is uniquely determined.
###### Proof.
It follows from the results of Sections 2-4. ∎
If we change primitive prime divisors on divisors of numbers defined in Lemma 1.5, we obtain a similar statement for Suzuki and Ree groups.
###### Proposition 6.5.
Let $`G`$ be a finite simple Suzuki or Ree group over a field of characteristic $`p`$. Let $`\rho (p,G)=\{p,s_1,\mathrm{},s_k\}`$.
* If $`G={}_{}{}^{2}B_{2}^{}(2^{2n+1})`$, then, up to reodering, $`s_i`$ divides $`m_i(B,n)`$. In particular, $`k=3`$.
* If $`G={}_{}{}^{2}G_{2}^{}(3^{2n+1})`$, then, all of $`s_i`$ are odd and, up to reodering, $`s_i`$ divides $`m_i(G,n)`$. In particular, $`k=4`$.
* If $`G={}_{}{}^{2}F_{4}^{}(2^{2n+1})`$, then, up to reodering, $`s_i`$ divides $`m_{i+2}(F,n)`$ and $`s_13`$. In particular, $`k=3`$.
Numbers $`m_i(B,n)`$, $`m_i(G,n)`$, and $`m_i(F,n)`$ are defined in Lemma 1.5.
Now we determine $`t(2,G)`$. Obviously, for a group of Lie type over a field of even characteristic $`p`$-independence number and $`2`$-independence number coincide. Thus, we may assume that $`G`$ is defined over the field of odd characteristic.
###### Proposition 6.6.
Let $`G`$ be a finite simple group of Lie type over the field of odd characteristic $`p`$ and order $`q`$. Then $`\rho (2,G)`$ and $`t(2,G)`$ are listed in Table 6 for classical groups and in Table 7 for exceptional groups.
###### Proof.
(1) $`G=A_{n1}^\epsilon (q)`$.
Let $`G=A_1(q)`$ and $`q>3`$. Since $`q`$ is odd, every prime divisor $`rp`$ of $`|G|`$ divides $`(q1)/2`$ or $`(q+1)/2`$. If $`e(2,q)=1`$, then $`2`$ is non-adjacent to some prime divisor $`r_2`$ of $`(q+1)/2`$ and $`\tau (G)=\{2,p,r_2\}`$. If $`e(2,q)=2`$, then $`2`$ is non-adjacent to some divisor $`r_1`$ of $`(q1)/2`$ and $`\tau (G)=\{2,p,r_1\}`$.
Let $`G=A_{n1}^\epsilon (q)`$ and $`n3`$. If $`(q\epsilon 1)_2<n_2`$, then by Propositions 4.1 and 4.2 only primitive prime divisor $`r_{\nu _\epsilon ^1(n1)}`$ is non-adjacent to $`2`$. Note that in this case $`n_24`$, since for $`n_22`$ the inequality $`(q\epsilon 1)_2<n_2`$ is impossible. By Zsigmondy Theorem primitive prime divisor $`r_{\nu _\epsilon ^1(n1)}`$ always exists. Thus, $`\rho (2,G)=\{2,r_{\nu _\epsilon ^1(n1)}\}`$.
If $`(q\epsilon 1)_2>n_2`$ or $`(q\epsilon 1)_2=n_2=2`$, then by Propositions 4.1 and 4.2 every primitive prime divisor $`r_{\nu _\epsilon ^1(n)}`$ is non-adjacent to $`2`$. Therefore, $`\rho (2,G)=\{2,r_{\nu _\epsilon ^1(n)}\}`$.
At last, let $`(q\epsilon 1)_2=n_2>2`$. By Propositions 4.1 and 4.2 only primitive prime divisors $`r_{\nu _\epsilon ^1(n1)}`$ and $`r_{\nu _\epsilon ^1(n)}`$ are non-adjacent to $`2`$. On the other hand, by Propositions 2.1 and 2.2, primes $`r_{\nu _\epsilon ^1(n1)}`$ and $`r_{\nu _\epsilon ^1(n)}`$ are non-adjacent. Thus, $`\rho (2,G)=\{2,r_{\nu _\epsilon ^1(n1)},r_{\nu _\epsilon ^1(n)}\}`$.
(2) $`G=C_n(q)`$ or $`B_n(q)`$.
The results from the table are directly following from Proposition 4.3.
(3) $`G=D_n^\epsilon (q)`$.
The results again are the direct corollary of Proposition 4.4. Note that the equality $`t(2,G)=2`$ is true for most groups of type $`D_n`$ over fields of odd order. Exceptions are following: $`n`$ is odd and $`q5(\mathrm{mod}8)`$ for $`G=D_n(q)`$, and $`q3(\mathrm{mod}8)`$ for $`G={}_{}{}^{2}D_{n}^{}(q)`$.
(4) Now to complete the proof of the proposition one can use Proposition 4.5 and Lemma 1.5 for Suzuki and Ree groups, and the Zsigmondy Theorem for other exceptional groups. ∎
Now we consider the uniqueness of $`\rho (2,G)`$. The situation here is very similar to the situation with $`\rho (p,G)`$.
###### Proposition 6.7.
Let $`G`$ be a finite simple group of Lie type over a field of odd characteristic $`p`$ and order $`q`$. Assume that $`G`$ is not isomorphic to $`A_1(q)`$, and $`{}_{}{}^{2}G_{2}^{}(q)`$. Let $`\rho (2,G)=\{2,s_1,s_2,\mathrm{},s_m\}`$ be an independence set in $`GK(G)`$ containing $`2`$ with maximal number of vertices and $`k_i=e(s_i,q)`$. Then the set $`\{k_1,\mathrm{},k_m\}`$ is uniquely determined.
###### Proof.
It follows from the results of Sections 24. ∎
###### Proposition 6.8.
Let $`G`$ be either $`A_1(q)`$ with $`q`$ odd, or $`{}_{}{}^{2}G_{2}^{}(3^{2n+1})`$ and let $`\rho (2,G)=\{2,s_1,s_2\}`$.
* If $`G=A_1(q)`$, $`q`$ odd, then, up to renumbering, $`s_1=p`$ and $`e(s_2,q)=3e(2,q)`$.
* If $`G={}_{}{}^{2}G_{2}^{}(3^{2n+1})`$, then, up to renumbering, $`s_i=m_{i+2}(G,n)`$.
Numbers $`m_i(G,n)`$ are defined in Lemma 1.5.
###### Proof.
It follows from the results of Sections 24. ∎
Our last task is to determine for every finite simple group $`G`$ of Lie type some independent set $`\rho (G)`$ in $`GK(G)`$ with maximal number of vertices.
###### Proposition 6.9.
Let $`G`$ be a finite simple group of Lie type over the field of characteristic $`p`$ and order $`q`$. Then $`\rho (G)`$ and $`t(G)`$ are listed in Table 8 for classical groups and in Table 9 for exceptional groups.
###### Proof.
(1) $`G=A_{n1}^\epsilon (q)`$.
If $`G=A_1(q)`$ or $`G=A_2^\epsilon (q)`$, then arguing as in proof of Proposition 6.3 we obtain $`\rho (G)=\rho (p,G)`$.
Suppose $`n=4`$. In view of Proposition 6.3 and Table 4 we have that $`t(p,G)=3`$ in all cases, except $`{}_{}{}^{2}A_{3}^{}(2)`$. By Proposition 6.6 and Table 6 we obtain that $`t(2,G)3`$. By Propositions 4.1 and 4.2 it follows that $`t(r_{\nu _\epsilon ^1(1)},G)3`$. Furthermore, there exist at most three other primitive divisors $`r_{\nu _\epsilon ^1(2)}`$, $`r_{\nu _\epsilon ^1(3)}`$, and $`r_{\nu _\epsilon ^1(4)}=r_4`$. So $`t(G)=t(p,G)=3`$ except the case: $`t({}_{}{}^{2}A_{3}^{}(2))=t(2,{}_{}{}^{2}A_{3}^{}(2))=2`$.
Let $`G=A_{n1}^\epsilon (q)`$, $`n5`$ and $`q2`$. By Zsigmondy Theorem there exist primitive prime divisors $`r_{\nu _\epsilon ^1(k)}`$ for every $`k>2`$. Denote by $`m`$ the number $`[n/2]`$, i.e., the integral part of $`n/2`$. By Propositions 2.1 and 2.2 the set
$$\rho =\{r_{\nu _\epsilon ^1(m+1)},r_{\nu _\epsilon ^1(m+2)},\mathrm{},r_{\nu _\epsilon ^1(n)}\}$$
is an independent set in $`GK(G)`$. On the other hand, by Propositions 2.1, 2.2, 3.1, 4.1 and 4.2, every two prime divisors from $`\pi \left(q_{i=1}^m(q^i(\epsilon 1)^i)\right)`$ are adjacent in $`GK(G)`$. Furthermore, each of them is adjacent to at least one number from $`\rho `$. Since every prime $`s\pi (G)\left(\rho \pi \left(q_{i=1}^m\left(q^i(\epsilon 1)^i\right)\right)\right)`$ is of the form $`r_i`$ for some $`i>m`$, it follows that every independent set in $`GK(G)`$ with prime divisor from $`\pi \left(q_{i=1}^m(q^i1)\right)`$ contains at most $`|\rho |`$ vertices. Thus, $`\rho (G)=\rho `$ and $`t(G)=|\rho |=[(n+1)/2]`$.
Let $`q=2`$. Since $`\nu _+^1(6)=\nu _{}^1(3)=6`$, the results of previous paragraph hold true for $`A_{n1}(2)`$ with $`n12`$ and $`{}_{}{}^{2}A_{n1}^{}(2)`$ with $`n6`$. If $`n=5`$ and $`\epsilon =`$, then for $`{}_{}{}^{2}A_{4}^{}(2)`$ one can check that $`\rho (G)=\rho (2,G)=\{2,5,11\}`$ and $`t(G)=3`$. So we can suppose that $`G=A_{n1}(2)`$. If $`n=5,6`$ then $`\rho (G)=\{r_3,r_4,r_5\}=\{5,7,31\}`$ and $`t(G)=3`$. If $`7n11`$ then we need to eliminate divisors of $`2^61`$ from $`\rho (G)`$ and at rest argue like in previous paragraph. In this case $`\rho (G)=\{r_i|i6,[n/2]<in\}`$ and $`t(G)=[(n1)/2]`$.
(2) $`G=C_n(q)`$ or $`B_n(q)`$. Recall that $`GK(C_n(q))=GK(B_n(q))`$ and we consider these groups together.
$`G=C_2(q)`$, $`q>2`$. Since every two prime divisors of $`q(q^21)`$ are adjacent, we obtain $`\rho (G)=\rho (p,G)=\{p,r_4\}`$.
Let $`n3`$ if $`q>2`$, and $`n7`$ if $`q=2`$. Define the set $`\rho `$ as follows:
$$\rho =\{r_{2i}|[n+1/2]<in\}\{r_i|[n/2]<in,i1(\mathrm{mod}2)\}.$$
Using results of Sections 14 and arguments as in proof for groups $`A_{n1}^\epsilon (q)`$ we obtain that $`\rho (G)=\rho `$ and as easy to verify $`t(G)=[(3n+5)/4]`$.
If $`q=2`$ and $`3n5`$, all arguments stand the same but we have to eliminate the divisors of type $`r_6`$ from $`\rho `$. Thus, in this case $`t(G)=[(3n+1)/4]`$. At last, if $`(n,q)=(6,2)`$ we have to eliminate a divisor of type $`r_6`$ from $`\rho `$, but instead it we can add a primitive prime divisor $`7`$ of $`2^31`$. Thus, $`t(G)=[(3n+5)/4]`$ as in the common situation.
(3) $`G=D_n^\epsilon (q)`$.
We argue like in two previous parts of proof. Using Proposition 2.4 we obtain a set $`\rho (G)`$ in common situation and then consider some exceptions arising with respect to exceptions in the Zsigmondy Theorem.
(4) We obtain results for exceptional groups by using Propositions 2.5 and 2.6, Tables 5 and 6, and the Zsigmondy Theorem and Lemma 1.5. ∎
## 7 Applications
Here we apply our results in a spirit of Section 5. First of all our investigations show that the condition $`t(2,G)>1`$ is realized for an extremely wide class of finite simple groups.
###### Theorem 7.1.
Let $`G`$ be a finite nonabelian simple group with $`t(2,G)=1`$, then $`G`$ is an alternating group $`Alt_n`$ with $`\tau (2,n)=\{s|s\text{ is a prime, }n3sn\}=\mathrm{}`$.
###### Proof.
See Propositions 6.16.9 and corresponding tables in Section 8. ∎
Thus, we may apply results of Theorems 5.2 and 5.3 as follows.
###### Corollary 7.2.
Let $`L`$ be a finite nonabelian simple group distinct from $`A_2(3)`$, $`{}_{}{}^{2}A_{3}^{}(3)`$, $`C_2(3)`$, $`Alt_{10}`$ and $`Alt_n`$ with $`\tau (2,n)=\mathrm{}`$. Let $`G`$ be a finite group with $`\omega (G)=\omega (L)`$. Then there exists a finite nonabelian simple group $`S`$ such that $`S\overline{G}=G/K\mathrm{Aut}(S)`$ for maximal normal soluble subgroup $`K`$ of $`G`$. Furthermore, $`t(S)t(G)1`$ and one of the following statements holds:
(1) $`SAlt_7`$ or $`A_1(q)`$ for some odd $`q`$ and $`t(S)=t(2,S)=3`$.
(2) For every prime $`r\pi (G)`$ non-adjacent in $`GK(G)`$ to $`2`$ the Sylow $`r`$-subgroup of $`G`$ is isomorphic to the Sylow $`r`$-subgroup of $`S`$. In particular, $`t(2,S)t(2,G)`$.
###### Proof.
It is direct corollary of Theorem 5.3 and Theorem 7.1. ∎
If $`G`$ is one of the groups $`A_2(3)`$, $`{}_{}{}^{2}A_{3}^{}(3)`$, $`C_2(3)`$, $`Alt_{10}`$, then $`t(2,G)2`$ and we also have
###### Corollary 7.3.
Let $`L`$ be a finite simple group distinct from $`Alt_n`$ with $`\tau (2,n)=\mathrm{}`$. Let $`G`$ be a finite group with $`\omega (G)=\omega (L)`$. Then $`G`$ has at most one nonabelian composition factor.
For an arbitrary subset $`\omega `$ of the set $``$ of natural numbers denote by $`h(\omega )`$ the number of pairwise nonisomorphic finite groups $`G`$ such that $`\omega (G)=\omega `$. We say that for a finite group $`G`$ the recognition problem is solved if we know the value of $`h(\omega (G))`$ (for brevity, $`h(G)`$). In particular, the group $`G`$ is said to be recognizable by spectrum (briefly, recognizable) if $`h(G)=1`$. In last twenty years the recognition problem is solved for many finite nonabelian simple and almost simple groups (for the review of results in this field see ). But most of those groups have the disconnected prime graph, since the Gruenberg — Kegel Theorem and classification of finite simple groups with disconnected prime graph obtained by Williams and Kondrat’ev were an important part of proof. The main theorem of (Theorem 5.2 here) and the results of the present work give a possibility to deal with a groups whose prime graph is connected. A finite nonabelian simple group $`L`$ is said to be quasirecognizable by spectrum if every finite group with the same spectrum has exactly one nonabelian composition factor $`S`$ and $`S`$ is isomorphic to $`L`$. So the investigations of quasirecognizability is an important step in the determination, whether the given group is recognizable by spectrum. In the first author gave a sketch of a proof for the following statement.
###### Theorem 7.4.
(\[4, Proposition 5\]) Let $`L={}_{}{}^{2}D_{n}^{}(q)`$, $`q=2^k`$, $`k,n`$ are natural numbers, $`n`$ is even and $`n16`$. Then $`L`$ is quasirecognizable by spectrum.
Actually, this result was proven modulo the results of Section 6 of the present article. Thus, now it is completely proven.
Now we would like to emphasize one statement that we have mentioned before.
###### Proposition 7.5.
Let $`G=B_n(q)`$ and $`H=C_n(q)`$. Then the prime graphs $`GK(G)`$ and $`GK(H)`$ coincide.
###### Proof.
It follows from the results of Sections 24. ∎
At last, we mention one recent result on prime graphs of finite groups. Recall that a set of vertices of a graph is said to be a clique if all vertices from this set are pairwise adjacent. In \[18, Theorem 1\] Lucido and Moghaddamfar describe all finite nonabelian simple groups whose prime graph connected components are cliques. We check their list of such groups using results of the present article. Unfortunately, there are some mistakes in the list. As a matter of fact, for groups $`G=A_2^\epsilon (q)`$, where $`q=2^k\epsilon 1`$ and $`(q\epsilon 1)_3=3`$, we have $`3,p\pi _1(G)`$ and $`3p\omega (G)`$. So the component $`\pi _1(G)`$ is not a clique for those groups contrary to the statement of Theorem 1 in . Below in Corollary 7.6 we give a revised list of such groups.
###### Corollary 7.6.
Let $`G`$ be a finite nonabelian simple group and all connected components of its prime graph $`GK(G)`$ are cliques. Then $`G`$ is one of the following groups:
* Sporadic groups $`M_{11}`$, $`M_{22}`$, $`J_1`$, $`J_2`$, $`J_3`$, $`HiS`$.
* Alternating groups $`Alt_n`$, where $`n=5,6,7,9,12,13`$.
* Groups of Lie type $`A_1(q)`$, where $`q>3`$; $`A_2(4)`$; $`A_2(q)`$, where $`(q1)_33`$, $`q+1=2^k`$; $`{}_{}{}^{2}A_{3}^{}(3)`$; $`{}_{}{}^{2}A_{5}^{}(2)`$; $`{}_{}{}^{2}A_{2}^{}(q)`$, where $`(q+1)_33`$, $`q1=2^k`$; $`C_3(2)`$, $`C_2(q)`$, where $`q>2`$; $`D_4(2)`$; $`{}_{}{}^{3}D_{4}^{}(2)`$; $`{}_{}{}^{2}B_{2}^{}(q)`$, where $`q=2^{2k+1}`$; $`G_2(q)`$, where $`q=3^k`$.
###### Proof.
All connected component of prime graph $`GK(G)`$ of a finite group $`G`$ are cliques if and only if the number $`s(G)`$ of the components is equal to the independence number $`t(G)`$ of $`GK(G)`$. For every finite nonabelian simple group the values of $`s(G)`$ and $`t(G)`$ are now known. Thus, using Tables 2a–2c in for the values of $`s(G)`$ and Tables 29 in Section 8 of the present article for $`t(G)`$, we obtain the result of the corollary. ∎
## 8 Resulting Tables
In the tables below $`n,k`$ are assumed to be naturals. By $`[x]`$ we denote the integral part of $`x`$. For a finite group $`G`$ we denote by $`\rho (G)`$ (by $`\rho (r,G)`$) some independence set in $`GK(G)`$ (containing $`r`$) with maximal number of vertices and put $`t(G)=|\rho (G)|`$, $`t(r,G)=|\rho (r,G)|`$. In Table 3 by $`\tau (n)`$ we denote the set $`\{s|s\text{ is a prime, }n/2<sn\}`$ and by $`\tau (2,n)`$ we denote the set $`\{s|s\text{ is a prime, }n3sn\}`$. We denote $`s_n^{}`$ to be the largest prime which does not exceed $`n/2`$ and $`s_n^{\prime \prime }`$ to be the smallest prime greater than $`n/2`$. In Tables 49 we assume $`G`$ to be a finite nonabelian simple group of Lie type over a field of characteristic $`p`$ and order $`q`$. By $`r_m`$ we define the primitive prime divisor of $`q^m1`$. If $`p`$ is odd then we say that $`2`$ is a primitive prime divisor of $`q1`$ if $`q1(\mathrm{mod}4)`$ and that $`2`$ is a primitive prime divisor of $`q^21`$ if $`q1(\mathrm{mod}4)`$.
###### Table 2.
Sporadic groups
| $`G`$ | $`t\left(G\right)`$ | $`\rho \left(G\right)`$ | $`t(2,G)`$ | $`\rho (2,G)`$ |
| --- | --- | --- | --- | --- |
| $`M_{11}`$ | $`3`$ | $`\{3,5,11\}`$ | $`3`$ | $`\{2,5,11\}`$ |
| $`M_{12}`$ | $`3`$ | $`\{3,5,11\}`$ | $`2`$ | $`\{2,11\}`$ |
| $`M_{22}`$ | $`4`$ | $`\{3,5,7,11\}`$ | $`4`$ | $`\{2,5,7,11\}`$ |
| $`M_{23}`$ | $`4`$ | $`\{3,7,11,23\}`$ | $`4`$ | $`\{2,5,11,23\}`$ |
| $`M_{24}`$ | $`4`$ | $`\{5,7,11,23\}`$ | $`3`$ | $`\{2,11,23\}`$ |
| $`J_1`$ | $`4`$ | $`\{5,7,11,19\}`$ | $`4`$ | $`\{2,7,11,19\}`$ |
| $`J_2`$ | $`2`$ | $`\{5,7\}`$ | $`2`$ | $`\{2,7\}`$ |
| $`J_3`$ | $`3`$ | $`\{5,17,19\}`$ | $`3`$ | $`\{2,17,19\}`$ |
| $`J_4`$ | $`7`$ | $`\{7,11,23,29,31,37,43\}`$ | $`6`$ | $`\{2,23,29,31,37,43\}`$ |
| $`\mathrm{Ru}`$ | $`4`$ | $`\{5,7,13,29\}`$ | $`2`$ | $`\{2,29\}`$ |
| $`\mathrm{He}`$ | $`3`$ | $`\{5,7,17\}`$ | $`2`$ | $`\{2,17\}`$ |
| $`\mathrm{McL}`$ | $`3`$ | $`\{5,7,11\}`$ | $`2`$ | $`\{2,11\}`$ |
| $`\mathrm{HN}`$ | $`3`$ | $`\{7,11,19\}`$ | $`2`$ | $`\{2,19\}`$ |
| $`\mathrm{HiS}`$ | $`3`$ | $`\{5,7,11\}`$ | $`3`$ | $`\{2,7,11\}`$ |
| $`\mathrm{Suz}`$ | $`4`$ | $`\{5,7,11,13\}`$ | $`3`$ | $`\{2,11,13\}`$ |
| $`\mathrm{Co}_1`$ | $`4`$ | $`\{7,11,13,23\}`$ | $`2`$ | $`\{2,23\}`$ |
| $`\mathrm{Co}_2`$ | $`4`$ | $`\{5,7,11,23\}`$ | $`3`$ | $`\{2,11,23\}`$ |
| $`\mathrm{Co}_3`$ | $`4`$ | $`\{5,7,11,23\}`$ | $`2`$ | $`\{2,23\}`$ |
| $`\mathrm{Fi}_{22}`$ | $`4`$ | $`\{5,7,11,13\}`$ | $`2`$ | $`\{2,13\}`$ |
| $`\mathrm{Fi}_{23}`$ | $`5`$ | $`\{7,11,13,17,23\}`$ | $`3`$ | $`\{2,17,23\}`$ |
| $`\mathrm{Fi}_{24}^{}`$ | $`6`$ | $`\{7,11,13,17,23,29\}`$ | $`4`$ | $`\{2,17,23,29\}`$ |
| $`\mathrm{O}^{}\mathrm{N}`$ | $`5`$ | $`\{5,7,11,19,31\}`$ | $`4`$ | $`\{2,11,19,31\}`$ |
| $`\mathrm{LyS}`$ | $`6`$ | $`\{5,7,11,31,37,67\}`$ | $`4`$ | $`\{2,31,37,67\}`$ |
| $`F_1`$ | $`11`$ | $`\{11,13,17,19,23,29,31,41,47,59,71\}`$ | $`5`$ | $`\{2,29,41,59,71\}`$ |
| $`F_2`$ | $`8`$ | $`\{7,11,13,17,19,23,31,47\}`$ | $`3`$ | $`\{2,31,47\}`$ |
| $`F_3`$ | $`5`$ | $`\{5,7,13,19,31\}`$ | $`4`$ | $`\{2,13,19,31\}`$ |
###### Table 3.
Simple alternating groups
| $`G`$ | Conditions | $`t\left(G\right)`$ | $`\rho \left(G\right)`$ | $`t(2,G)`$ | $`\rho (2,G)`$ |
| --- | --- | --- | --- | --- | --- |
| $`Alt_n`$ | $`n=5,6`$ | $`3`$ | $`\{2,3,5\}`$ | $`3`$ | $`\{2,3,5\}`$ |
| | $`n=8`$ | $`3`$ | $`\{2,5,7\}`$ | $`3`$ | $`\{2,5,7\}`$ |
| | $`n7`$, $`s_n^{}+s_n^{\prime \prime }>n`$ | $`\left|\tau \left(n\right)\right|+1`$ | $`\tau \left(n\right)\left\{s_n^{}\right\}`$ | $`\left|\tau (2,n)\right|+1`$ | $`\tau (2,n)\left\{2\right\}`$ |
| | $`n9`$, $`s_n^{}+s_n^{\prime \prime }n`$ | $`\left|\tau \left(n\right)\right|`$ | $`\tau \left(n\right)`$ | $`\left|\tau (2,n)\right|+1`$ | $`\tau (2,n)\left\{2\right\}`$ |
###### Table 4.
$`p`$-independence numbers for finite simple classical groups
| $`G`$ | Conditions | $`t(p,G)`$ | $`\rho (p,G)`$ |
| --- | --- | --- | --- |
| $`A_{n1}\left(q\right)`$ | $`n=2`$, $`q>3`$ | $`3`$ | $`\{p,r_1,r_2\}`$ |
| | $`n=3`$, $`\left(q1\right)_3=3`$, and $`q+12^k`$ | $`4`$ | $`\left\{p,3,r_22,r_3\right\}`$ |
| | $`n=3`$, $`\left(q1\right)_33`$, and $`q+12^k`$ | $`3`$ | $`\left\{p,r_22,r_3\right\}`$ |
| | $`n=3`$, $`\left(q1\right)_3=3\text{ and }q+1=2^k`$ | $`3`$ | $`\{p,3,r_3\}`$ |
| | $`n=3`$, $`\left(q1\right)_33\text{ and }q+1=2^k`$ | $`2`$ | $`\{p,r_3\}`$ |
| | $`n=6`$, $`q=2`$ | $`2`$ | $`\{2,31\}`$ |
| | $`n=7`$, $`q=2`$ | $`2`$ | $`\{2,127\}`$ |
| | $`n>3`$ $`(n,q)(6,2),(7,2)`$ | $`3`$ | $`\{p,r_{n1},r_n\}`$ |
| $`{}_{}{}^{2}A_{n1}^{}\left(q\right)`$ | $`n=3`$, $`\left(q+1\right)_3=3`$, and $`q12^k`$ | $`4`$ | $`\left\{p,3,r_12,r_6\right\}`$ |
| | $`n=3`$, $`\left(q+1\right)_33`$, and $`q12^k`$ | $`3`$ | $`\left\{p,r_12,r_6\right\}`$ |
| | $`n=3`$, $`\left(q+1\right)_3=3\text{ and }q1=2^k`$ | $`3`$ | $`\{p,3,r_6\}`$ |
| | $`n=3`$, $`\left(q+1\right)_33\text{ and }q1=2^k`$ | $`2`$ | $`\{p,r_6\}`$ |
| | $`n=4`$, $`q=2`$ | $`2`$ | $`\{2,5\}`$ |
| | $`n0\left(\mathrm{mod}4\right)`$, $`(n,q)(4,2)`$ | $`3`$ | $`\{p,r_{2n2},r_n\}`$ |
| | $`n1\left(\mathrm{mod}4\right)`$ | $`3`$ | $`\{p,r_{n1},r_{2n}\}`$ |
| | $`n2\left(\mathrm{mod}4\right)`$, $`n2`$ | $`3`$ | $`\{p,r_{2n2},r_{n/2}\}`$ |
| | $`n3\left(\mathrm{mod}4\right)`$, $`n3`$ | $`3`$ | $`\{p,r_{\left(n1\right)/2},r_{2n}\}`$ |
| $`B_n\left(q\right)`$ or | $`n=3`$, $`q=2`$ | $`2`$ | $`\{2,7\}`$ |
| $`C_n\left(q\right)`$ | $`n`$ is even | $`2`$ | $`\{p,r_{2n}\}`$ |
| | $`n>1`$ is odd, $`(n,q)(3,2)`$ | $`3`$ | $`\{p,r_n,r_{2n}\}`$ |
| $`D_n\left(q\right)`$ | $`n=4`$, $`q=2`$ | $`2`$ | $`\{2,7\}`$ |
| | $`n0\left(\mathrm{mod}2\right)`$, $`n4`$, $`(n,q)(4,2)`$ | $`3`$ | $`\{p,r_{n1},r_{2n2}\}`$ |
| | $`n1\left(\mathrm{mod}2\right)`$, $`n>4`$ | $`3`$ | $`\{p,r_n,r_{2n2}\}`$ |
| $`{}_{}{}^{2}D_{n}^{}\left(q\right)`$ | $`n=4`$, $`q=2`$ | $`3`$ | $`\{2,7,17\}`$ |
| | $`n0\left(\mathrm{mod}2\right)`$, $`n4`$, $`(n,q)(4,2)`$ | $`4`$ | $`\{p,r_{n1},r_{2n2},r_{2n}\}`$ |
| | $`n1\left(\mathrm{mod}2\right)`$, $`n>4`$ | $`3`$ | $`\{p,r_{2n2},r_{2n}\}`$ |
###### Table 5.
$`p`$-independence numbers for finite simple exceptional groups of Lie type
| $`G`$ | Conditions | $`t(p,G)`$ | $`\rho (p,G)`$ |
| --- | --- | --- | --- |
| $`G_2\left(q\right)`$ | $`q>2`$ | 3 | $`\{p,r_3,r_6\}`$ |
| $`F_4\left(q\right)`$ | none | 3 | $`\{p,r_8,r_{12}\}`$ |
| $`E_6\left(q\right)`$ | none | 4 | $`\{p,r_8,r_9,r_{12}\}`$ |
| $`{}_{}{}^{2}E_{6}^{}\left(q\right)`$ | none | 4 | $`\{p,r_8,r_{12},r_{18}\}`$ |
| $`E_7\left(q\right)`$ | none | 5 | $`\{p,r_7,r_9,r_{14},r_{18}\}`$ |
| $`E_8\left(q\right)`$ | none | 5 | $`\{p,r_{15},r_{20},r_{24},r_{30}\}`$ |
| $`{}_{}{}^{3}D_{4}^{}\left(q\right)`$ | none | 2 | $`\{p,r_{12}\}`$ |
| $`{}_{}{}^{2}B_{2}^{}\left(2^{2n+1}\right)`$ | $`n1`$ | 4 | $`\{2,s_1,s_2,s_3\}`$, where |
| | | | $`s_1`$ divides $`2^{2n+1}1`$, |
| | | | $`s_2`$ divides $`2^{2n+1}2^{n+1}+1`$ |
| | | | $`s_3`$ divides $`2^{2n+1}+2^{n+1}+1`$ |
| $`{}_{}{}^{2}G_{2}^{}\left(3^{2n+1}\right)`$ | $`n1`$ | 5 | $`\{3,s_1,s_2,s_3,s_4\}`$, where |
| | | | $`s_12`$ divides $`3^{2n+1}1`$ |
| | | | $`s_22`$ divides $`3^{2n+1}+1`$ |
| | | | $`s_3`$ divides $`3^{2n+1}3^{n+1}+1`$ |
| | | | $`s_4`$ divides $`3^{2n+1}+3^{n+1}+1`$ |
| $`{}_{}{}^{2}F_{4}^{}\left(2^{2n+1}\right)`$ | $`n1`$ | 4 | $`\{2,s_1,s_2,s_3\}`$, where |
| | | | $`s_13`$ and divides $`2^{4n+2}2^{2n+1}+1`$ |
| | | | $`s_2`$ divides $`2^{4n+2}+2^{3n+2}+2^{2n+1}+2^{n+1}+1`$ |
| | | | $`s_3`$ divides $`2^{4n+2}2^{3n+2}+2^{2n+1}2^{n+1}+1`$ |
| $`{}_{}{}^{2}F_{4}^{}\left(2\right)^{}`$ | none | 2 | $`\{2,13\}`$ |
###### Table 6.
2-independence numbers for finite simple classical groups of characteristic $`p2`$
| $`G`$ | Conditions | $`t(2,G)`$ | $`\rho (2,G)`$ |
| --- | --- | --- | --- |
| $`A_{n1}\left(q\right)`$ | $`n=2`$, $`q1\left(\mathrm{mod}4\right)`$ | $`3`$ | $`\{2,r_2,p\}`$ |
| | $`n=2`$, $`q3\left(\mathrm{mod}4\right)`$, $`q3`$ | $`3`$ | $`\{2,r_1,p\}`$ |
| | $`n3`$ and $`n_2<\left(q1\right)_2`$ | $`2`$ | $`\{2,r_n\}`$ |
| | $`n3`$ and either $`n_2>\left(q1\right)_2`$, | $`2`$ | $`\{2,r_{n1}\}`$ |
| | or $`n_2=\left(q1\right)_2=2`$ | | |
| | $`2<n_2=\left(q1\right)_2`$ | $`3`$ | $`\{2,r_{n1},r_n\}`$ |
| $`{}_{}{}^{2}A_{n1}^{}\left(q\right)`$ | $`n_2>\left(q+1\right)_2`$ | $`2`$ | $`\{2,r_{2n2}\}`$ |
| | $`n_2=1`$ | $`2`$ | $`\{2,r_{2n}\}`$ |
| | $`2<n_2<\left(q+1\right)_2`$ | $`2`$ | $`\{2,r_n\}`$ |
| | $`n3`$, $`2=n_2\left(q+1\right)_2`$ | $`2`$ | $`\{2,r_{n/2}\}`$ |
| | $`2<n_2=\left(q+1\right)_2`$ | $`3`$ | $`\{2,r_{2n2},r_n\}`$ |
| $`B_n\left(q\right)`$ or | $`n>1`$ is odd and $`\left(q1\right)_2=2`$ | $`2`$ | $`\{2,r_n\}`$ |
| $`C_n\left(q\right)`$ | $`n`$ is even or $`\left(q1\right)_2>2`$ | $`2`$ | $`\{2,r_{2n}\}`$ |
| $`D_n\left(q\right)`$ | $`n0\left(\mathrm{mod}2\right)`$, $`n4`$, $`q3\left(\mathrm{mod}4\right)`$ | $`2`$ | $`\{2,r_{n1}\}`$ |
| | $`n0\left(\mathrm{mod}2\right)`$, $`n4`$, $`q1\left(\mathrm{mod}4\right)`$ | $`2`$ | $`\{2,r_{2n2}\}`$ |
| | $`n1\left(\mathrm{mod}2\right)`$, $`n>4`$, $`q3\left(\mathrm{mod}4\right)`$ | $`2`$ | $`\{2,r_n\}`$ |
| | $`n1\left(\mathrm{mod}2\right)`$, $`n>4`$, $`q1\left(\mathrm{mod}8\right)`$ | $`2`$ | $`\{2,r_{2n2}\}`$ |
| | $`n1\left(\mathrm{mod}2\right)`$, $`n>4`$, $`q5\left(\mathrm{mod}8\right)`$ | $`3`$ | $`\{2,r_n,r_{2n2}\}`$ |
| $`{}_{}{}^{2}D_{n}^{}\left(q\right)`$ | $`n0\left(\mathrm{mod}2\right)`$, $`n4`$ | $`2`$ | $`\{2,r_{2n}\}`$ |
| | $`n1\left(\mathrm{mod}2\right)`$, $`n>4`$, $`q1\left(\mathrm{mod}4\right)`$ | $`2`$ | $`\{2,r_{2n}\}`$ |
| | $`n1\left(\mathrm{mod}2\right)`$, $`n>4`$, $`q7\left(\mathrm{mod}8\right)`$ | $`2`$ | $`\{2,r_{2n2}\}`$ |
| | $`n1\left(\mathrm{mod}2\right)`$, $`n>4`$, $`q3\left(\mathrm{mod}8\right)`$ | $`3`$ | $`\{2,r_{2n2},r_{2n}\}`$ |
###### Table 7.
2-independence numbers for finite simple exceptional groups of Lie type of characteristic $`p2`$
| $`G`$ | Conditions | $`t(2,G)`$ | $`\rho (2,G)`$ |
| --- | --- | --- | --- |
| $`G_2\left(q\right)`$ | none | 3 | $`\{2,r_3,r_6\}`$ |
| $`F_4\left(q\right)`$ | none | 2 | $`\{2,r_{12}\}`$ |
| $`E_6\left(q\right)`$ | none | 3 | $`\{2,r_9,r_{12}\}`$ |
| $`{}_{}{}^{2}E_{6}^{}\left(q\right)`$ | none | 3 | $`\{2,r_{12},r_{18}\}`$ |
| $`E_7\left(q\right)`$ | $`q1\left(\mathrm{mod}4\right)`$ | 3 | $`\{2,r_{14},r_{18}\}`$ |
| | $`q3\left(\mathrm{mod}4\right)`$ | 3 | $`\{2,r_7,r_9\}`$ |
| $`E_8\left(q\right)`$ | none | 5 | $`\{2,r_{15},r_{20},r_{24},r_{30}\}`$ |
| $`{}_{}{}^{3}D_{4}^{}\left(q\right)`$ | none | 2 | $`\{2,r_{12}\}`$ |
| $`{}_{}{}^{2}G_{2}^{}\left(3^{2n+1}\right)`$ | none | 3 | $`\{2,s_1,s_2\}`$, where |
| | | | $`s_1`$ divides $`3^{2n+1}3^{n+1}+1`$, |
| | | | $`s_2`$ divides $`3^{2n+1}+3^{n+1}+1`$. |
###### Table 8.
Independence numbers for finite simple classical groups
| $`G`$ | Conditions | $`t\left(G\right)`$ | $`\rho \left(G\right)`$ |
| --- | --- | --- | --- |
| $`A_{n1}\left(q\right)`$ | $`n=2`$, $`q>3`$ | $`3`$ | $`\{p,r_1,r_2\}`$ |
| | $`n=3`$, $`\left(q1\right)_3=3\text{ and }q+12^k`$ | $`4`$ | $`\{p,3,r_2,r_3\}`$ |
| | $`n=3`$, $`\left(q1\right)_33\text{ and }q+12^k`$ | $`3`$ | $`\{p,r_2,r_3\}`$ |
| | $`n=3`$, $`\left(q1\right)_3=3\text{ and }q+1=2^k`$ | $`3`$ | $`\{p,3,r_3\}`$ |
| | $`n=3`$, $`\left(q1\right)_33\text{ and }q+1=2^k`$ | $`2`$ | $`\{p,r_3\}`$ |
| | $`n=4`$ | $`3`$ | $`\{p,r_{n1},r_n\}`$ |
| | $`n=5,6`$, $`q=2`$ | $`3`$ | $`\{5,7,31\}`$ |
| | $`7n11`$, $`q=2`$ | $`\left[\frac{n1}{2}\right]`$ | $`\left\{r_i\right|i6,\left[\frac{n}{2}\right]<in\}`$ |
| | $`n5`$ and $`q>2`$ or $`n12`$ and $`q=2`$ | $`\left[\frac{n+1}{2}\right]`$ | $`\left\{r_i\right|\left[\frac{n}{2}\right]<in\}`$ |
| $`{}_{}{}^{2}A_{n1}^{}\left(q\right)`$ | $`n=3,q2,\left(q+1\right)_3=3,\text{ and }q12^k`$ | $`4`$ | $`\{p,3,r_1,r_6\}`$ |
| | $`n=3`$, $`\left(q+1\right)_33\text{ and }q12^k`$ | $`3`$ | $`\{p,r_1,r_6\}`$ |
| | $`n=3`$, $`\left(q+1\right)_3=3\text{ and }q1=2^k`$ | $`3`$ | $`\{p,3,r_6\}`$ |
| | $`n=3`$, $`\left(q+1\right)_33\text{ and }q1=2^k`$ | $`2`$ | $`\{p,r_6\}`$ |
| | $`n=4`$, $`q=2`$ | $`2`$ | $`\{2,5\}`$ |
| | $`n=4`$, $`q>2`$ | $`3`$ | $`\{p,r_4,r_6\}`$ |
| | $`n=5`$, $`q=2`$ | $`3`$ | $`\{2,5,11\}`$ |
| | $`n5`$ and $`(n,q)(5,2)`$ | $`\left[\frac{n+1}{2}\right]`$ | $`\left\{r_{i/2}\right|\left[\frac{n}{2}\right]<in,`$ |
| | | | $`i2\left(\mathrm{mod}4\right)\}`$ |
| | | | $`\left\{r_{2i}\right|\left[\frac{n}{2}\right]<in,`$ |
| | | | $`i1\left(\mathrm{mod}2\right)\}`$ |
| | | | $`\left\{r_i\right|\left[\frac{n}{2}\right]<in,`$ |
| | | | $`i0\left(\mathrm{mod}4\right)\}`$ |
| $`B_n\left(q\right)`$ or | $`n=2`$, $`q>2`$ | $`2`$ | $`\{p,r_4\}`$ |
| $`C_n\left(q\right)`$ | $`n=3`$ and $`q=2`$ | $`2`$ | $`\{5,7\}`$ |
| | $`n=4`$ and $`q=2`$ | $`3`$ | $`\{5,7,17\}`$ |
| | $`n=5`$ and $`q=2`$ | $`4`$ | $`\{7,11,17,31\}`$ |
| | $`n=6`$ and $`q=2`$ | $`5`$ | $`\{7,11,13,17,31\}`$ |
| | $`n>2`$, $`(n,q)(3,2),(4,2),(5,2),(6,2)`$ | $`\left[\frac{3n+5}{4}\right]`$ | $`\left\{r_{2i}\right|\left[\frac{n+1}{2}\right]in\}`$ |
| | | | $`\left\{r_i\right|\left[\frac{n}{2}\right]<in,`$ |
| | | | $`i1\left(\mathrm{mod}2\right)\}`$ |
| $`D_n\left(q\right)`$ | $`n=4`$ and $`q=2`$ | $`2`$ | $`\{5,7\}`$ |
| | $`n=5`$ and $`q=2`$ | $`4`$ | $`\{5,7,17,31\}`$ |
| | $`n=6`$ and $`q=2`$ | $`4`$ | $`\{7,11,17,31\}`$ |
| | $`n4`$, $`n3\left(mod4\right)`$, | $`\left[\frac{3n+1}{4}\right]`$ | $`\{r_{2i}\left[\frac{n+1}{2}\right]i<n\}`$ |
| | $`(n,q)(4,2),(5,2),(6,2)`$ | | $`\{r_i\left[\frac{n}{2}\right]<in,`$ |
| | | | $`i1\left(mod2\right)\}`$ |
| | $`n3\left(mod4\right)`$ | $`\frac{3n+3}{4}`$ | $`\{r_{2i}\left[\frac{n+1}{2}\right]i<n\}`$ |
| | | | $`\{r_i\left[\frac{n}{2}\right]in,`$ |
| $`{}_{}{}^{2}D_{n}^{}\left(q\right)`$ | $`n=4`$ and $`q=2`$ | $`3`$ | $`\{5,7,17\}`$ |
| | $`n=5`$ and $`q=2`$ | $`3`$ | $`\{7,11,17\}`$ |
| | $`n=6`$ and $`q=2`$ | $`5`$ | $`\{7,11,13,17,31\}`$ |
| | $`n=7`$ and $`q=2`$ | $`5`$ | $`\{11,13,17,31,43\}`$ |
| | $`n4`$, $`n1\left(mod4\right),`$ | $`\left[\frac{3n+4}{4}\right]`$ | $`\{r_{2i}\left[\frac{n}{2}\right]in\}`$ |
| | $`(n,q)(4,2),(6,2),(7,2)`$ | | $`\{r_i\left[\frac{n}{2}\right]<i<n,`$ |
| | | | $`i1\left(mod2\right)\}`$ |
| | $`n>4`$, $`n1\left(\mathrm{mod}4\right)`$, $`(n,q)(5,2)`$ | $`\left[\frac{3n+4}{4}\right]`$ | $`\left\{r_{2i}\right|\left[\frac{n}{2}\right]<in\}`$ |
| | | | $`\left\{r_i\right|\left[\frac{n}{2}\right]<i<n,`$ |
| | | | $`i1\left(\mathrm{mod}2\right)\}`$ |
###### Table 9.
Independence numbers for finite simple exceptional groups of Lie type
| $`G`$ | Conditions | $`t\left(G\right)`$ | $`\rho \left(G\right)`$ |
| --- | --- | --- | --- |
| $`G_2\left(q\right)`$ | $`q>2`$ | 3 | $`\{p,r_3,r_6\}`$ |
| $`F_4\left(q\right)`$ | $`q=2`$ | 4 | $`\{5,7,13,17\}`$ |
| | $`q>2`$ | 5 | $`\{r_3,r_4,r_6,r_8,r_{12}\}`$ |
| $`E_6\left(q\right)`$ | | 5 | $`\{r_4,r_5,r_8,r_9,r_{12}\}`$ |
| $`{}_{}{}^{2}E_{6}^{}\left(q\right)`$ | | 5 | $`\{r_4,r_8,r_{10},r_{12},r_{18}\}`$ |
| $`E_7\left(q\right)`$ | | 8 | $`\{r_5,r_7,r_8,r_9,r_{10},r_{12},r_{14},r_{18}\}`$ |
| $`E_8\left(q\right)`$ | | 12 | $`\{r_5,r_7,r_8,r_9,r_{10},r_{12},r_{14},r_{15},r_{18},r_{20},r_{24},r_{30}\}`$ |
| $`{}_{}{}^{3}D_{4}^{}\left(q\right)`$ | $`q=2`$ | 2 | $`\{2,13\}`$ |
| | $`q>2`$ | 3 | $`\{r_3,r_6,r_{12}\}`$ |
| $`{}_{}{}^{2}B_{2}^{}\left(2^{2n+1}\right)`$ | $`n1`$ | 4 | $`\{2,s_1,s_2,s_3\}`$, where |
| | | | $`s_1`$ divides $`2^{2n+1}1`$, |
| | | | $`s_2`$ divides $`2^{2n+1}2^{n+1}+1`$ |
| | | | $`s_3`$ divides $`2^{2n+1}+2^{n+1}+1`$ |
| $`{}_{}{}^{2}G_{2}^{}\left(3^{2n+1}\right)`$ | $`n1`$ | 5 | $`\{3,s_1,s_2,s_3,s_4\}`$, where |
| | | | $`s_12`$ divides $`3^{2n+1}1`$ |
| | | | $`s_22`$ divides $`3^{2n+1}+1`$ |
| | | | $`s_3`$ divides $`3^{2n+1}3^{n+1}+1`$ |
| | | | $`s_4`$ divides $`3^{2n+1}+3^{n+1}+1`$ |
| $`{}_{}{}^{2}F_{4}^{}\left(2^{2n+1}\right)`$ | $`n2`$, | $`5`$ | $`\{s_1,s_2,s_3,s_4,s_5\}`$, where |
| | | | $`s_1`$ and divides $`2^{2n+1}+1`$, |
| | | | $`s_2`$ divides $`2^{4n+2}+1`$, |
| | | | $`s_33`$ and divides $`2^{4n+2}2^{2n+1}+1`$, |
| | | | $`s_4`$ divides $`2^{4n+2}2^{3n+2}+2^{2n+1}2^{n+1}+1`$, |
| | | | $`s_5`$ divides $`2^{4n+2}+2^{3n+2}+2^{2n+1}+2^{n+1}+1`$, |
| $`{}_{}{}^{2}F_{4}^{}\left(2\right)^{}`$ | none | $`3`$ | $`\{3,5,13\}`$ |
| $`{}_{}{}^{2}F_{4}^{}\left(8\right)`$ | none | 4 | $`\{7,19,37,109\}`$ |
|
warning/0506/astro-ph0506289.html
|
ar5iv
|
text
|
# Extragalactic Globular Clusters: Old Spectroscopic Ages and New Views on Their Formation
## 1 Introduction
Bimodal color distributions were first discovered in extragalactic globular cluster (GC) systems in the early nineties (Ostrov, Geisler, & Forte 1993; Zepf & Ashman 1993) and are now known to be commonplace among luminous galaxies (Larsen et al. 2001; Kundu & Whitmore 2001). It was realized early on that this finding had far-reaching implications for our understanding of the formation histories of GC systems. Since GC formation is thought to accompany all major star forming episodes in a galaxy’s history (Schweizer 2000), an understanding of the origin of GC bimodality implicitly constrains the formation histories of their host galaxies.
Several scenarios have been developed to explain these GC subpopulations, which are generally termed metal-poor (MP) and metal-rich (MR). The major merger model (Schweizer 1987; Ashman & Zepf 1992) predicts the formation of MR GCs out of enriched gas during spiral–spiral mergers that form elliptical galaxies. The MP GCs are associated with the progenitor spirals. Forbes, Brodie, & Grillmair (1997) pointed out that bimodality could also arise in a multiphase in situ scenario: the MP GCs are formed in gaseous clumps within the early protogalactic potential, while the MR GCs are formed after a dormant period in a second phase of star formation. It is during this second phase that the bulk of the galaxy starlight is thought to be formed. Finally, Côté et al. (1998, 2000, 2002) suggested an accretion scenario in which a massive seed galaxy forms MR GCs in situ, while MP GCs are acquired via the accretion of dwarf galaxies and their accompanying low-metallicity GC systems. GC formation has also recently been simulated with semi-analytic models (SAMs; Beasley et al. 2002) and N-body+gasdynamic codes (Kravtsov & Gnedin 2005). The Beasley et al. model, which followed GC systems to the present day, found it necessary to truncate MP GC formation at $`z>5`$ to produce bimodality. Candidates for this truncation process include feedback in low-mass halos (Forbes et al. 1997) and cosmic reionization (Santos 2003).
Some of the scenarios mentioned above were developed before the clear success of $`\mathrm{\Lambda }`$CDM cosmology and the hierarchical merging paradigm. However, they are relevant for understanding structure formation. These scenarios describe the behavior of baryons within dark matter halos, which is dominated by processes that are relatively unconstrained at present. All of the scenarios predict old ages for MP GCs, but not necessarily for MR GCs. In the merger model, the mergers (and hence the MR GCs) will have a range of ages. In the multiphase scenario both GC subpopulations are expected to be old ($`10`$ Gyr) with the MR GCs slightly younger than the MP GCs. Some variation with environment and galaxy mass may be expected. In the accretion scenario, both subpopulations are $``$ coeval and very old. Of course, reality is likely to be some hybrid of the these alternatives, and in a very real sense the distinctions between them blur if the processes occur at high enough redshift. Nonetheless, it is important to determine whether one (or more) of these ideas represents a dominant mechanism. Thus, the goal of age-dating GC subpopulations and, in particular, detecting or ruling out age differences between MR and MP GCs, will set constraints on the processes of galaxy formation and limits on the epochs of galaxy assembly.
The goal of this work is to determine relative ages of GC subpopulations and set limits on their absolute ages (modulo modeling and observational uncertainties). Our data are among the best spectra currently available for extragalactic GCs, and it is unlikely that significantly tighter age constraints for GCs will be forthcoming using Lick index diagnostics (Trager et al. 1998 and references therein) in the era of $`810`$ m class telescopes. Since the Lick system is not optimized for studying stellar systems with low velocity dispersion, this current short study is intended as a summary of our spectroscopic work in age-dating GCs and as a starting point for our future investigations in which new, higher resolution (3 Å) index definitions will be employed.
## 2 The Sample and Method
Not all galaxies for which our group has previously obtained spectra are included in this study. Here we require all Lick indices (see details below) from H$`\delta _F`$ to Fe5335 to be measurable, and that at least two GCs per subpopulation have $`H\beta `$ and H$`\delta _F`$ errors $`<0.5`$ Å. This left a sample of eight galaxies, which are listed in Table 1. The sample is dominated by massive early-type galaxies in group and cluster environments, but includes two dwarfs.
The GC spectra were obtained at Keck with LRIS (Oke et al. 1995) over a variety of runs, from Dec. 1995 to Apr. 2004. Spectra were taken in multislit mode (except for the GCs of the Fornax dwarf spheroidal), with resolution ranging from 3.3–8 Å. The spectral coverage was typically $`40005500`$ Å, but varied with slit placement and from run to run, sometimes going further to the red or blue.
We refer the reader to the respective papers noted in Table 1 for details on the reduction of each data set; however, a short summary follows. GC spectra extracted from individual exposures (typically 30 min in length) were wavelength calibrated using arc lamps and then co-added. Small zero-point shifts to the wavelength calibration were made using bright night sky lines, and a standard flux calibration was carried out using one or two standards from Massey et al. (1988).
Lick indices were measured using the definitions in Trager et al. (1998) and Worthey & Ottaviani (1997). The spectra were smoothed to the wavelength-dependent resolution (8–11 Å) of the Lick system before the measurement of indices. An unfortunate aspect of the present study is that we cannot fully calibrate our index measurements onto the Lick system. The number and type of Lick standard stars varied from run to run, rendering consistent corrections impossible. We note that many of our previous studies of GCs in individual galaxies (e.g., Brodie et al. 2005; Strader, Brodie, & Forbes 2004a) have found such corrections unnecessary: they are generally smaller than the error bars on individual index measurements, and there appear to be no large systematic effects. However, our lack of Lick corrections could still result in small offsets when comparing to Lick index predictions of simple stellar population (SSP) models.
For this reason, we have decided to concentrate on a *differential* analysis between our data and integrated spectra of Galactic GCs. These Galactic data are from Gregg (1994) and Puzia et al. (2002b). In the former case we measured indices directly from the digital spectra, as described above. In the latter case we “de-corrected” the published indices using Lick offsets given in the same paper. The resulting datasets should be amenable to consistent comparisons.
Even with our best individual spectra, estimating accurate ages and metallicities is difficult. An error in H$`\beta `$ of only $`0.3`$ Å corresponds to an age error of $`3`$–4 Gyr for a 13 Gyr old GC. Thus, to facilitate comparison to the Galactic GCs, we constructed *mean* MP and MR (and in NGC 4365 and NGC 1407, intermediate-metallicity) subpopulations for all galaxies, except for the two dwarfs Fornax and VCC 1087. Fornax has only MP GCs; VCC 1087 may have a small number of MR GCs, but they were sparsely sampled in our spectroscopy and do not meet the criteria outlined above (Beasley et al. 2005). Photometric studies of the remainder of our sample galaxies show clear evidence for subpopulations, so this division is justified (see individual papers for details). GCs were sorted into subpopulations primarily on the basis of this photometry (with typical separation at $`VI=1.05`$, $`BI=1.85`$, or $`gz=1.1`$). These criteria were augmented by principle components analysis (PCA)-based metallicities when ambiguous (Strader & Brodie 2004). The number of GCs per subpopulation are listed in Table 1. Lick indices for these subpopulations are weighted means of the individual GC indices with $`2\sigma `$ rejection. The variance is calculated as the sum of the weights and is equivalent to a weighted standard variance of the mean. The mean subpopulation indices and standard errors of the mean are given in Tables 2 and 3. The only GCs excluded outright were two suspected young ($`1`$–2 Gyr) metal-rich GCs in the dynamically young elliptical NGC 3610 (Strader et al. 2003a; Strader et al. 2004a), since our focus is on deriving the properties of the “old” subpopulations. For each subpopulation with at least three GCs, weighted standard deviations of the indices are given in Table 4, and represent the combined effects of observational error and intrinsic differences among the GCs within each subpopulation.
The majority of our GC samples are biased. First there is the necessary luminosity bias: nearly all of our GCs have $`V<23`$. Adequate S/N cannot be achieved for GC fainter than this, even in $`810`$ hours at Keck. This apparent magnitude corresponds to $`M_V8`$ at the distance of Virgo, 0.5 mag brighter than the turnover of the GC luminosity function (LF). However, neither metallicity or age correlates strongly with GC mass in the Milky Way, suggesting that this introduces minimal bias in derived properties. More importantly, the GCs were not selected uniformly in color. This makes it impossible to construct “true” composite subpopulations that accurately reflect the underlying metallicities. However, we *can* constrain mean ages, since younger GC subpopulations with a standard log-normal LF will be, on average, brighter than their old counterparts, even taking into account their probable higher metallicities. Thus magnitude-selected samples will give lower limits to mean GC subpopulation ages, unless the LF of the younger GCs is biased to low masses, either at birth or because of differing GC destruction.
## 3 Ages
The traditional method of estimating ages and metallicities using Lick indices involves constructing index-index plots. \[MgFe\], the harmonic mean of Mg$`b`$ and (Fe5270+Fe5335)/2, is often used as a metallicity indicator largely insensitive to variations in \[$`\alpha `$/Fe\]. However, the use of \[MgFe\] does not take advantage of the metallicity information contained in the remainder of the Lick indices. We have developed a PCA-based method for estimating GC metallicities using 11 Lick indices (Strader & Brodie 2004), and choose to use these more robust metallicities instead of the standard \[MgFe\] (or the slightly different \[MgFe\]$`\mathrm{}`$, in which $`<`$Fe$`>`$ is replaced by ($`0.72\times `$ Fe5270 \+ $`0.28\times `$ Fe5335); Thomas, Maraston, & Bender 2003).
Puzia et al. (2004) have argued that the best Lick Balmer index for estimating GC ages is H$`\gamma _F`$. However, the blue pseudocontinuum bandpass of H$`\gamma _F`$ encompasses most of the CH absorption of the G band centered at $`4300`$ Å. This makes the index extremely sensitive to changes in the carbon abundance (and overall metallicity, though its sensitivity decreases toward high metallicities). Thus, we have chosen not to use either of the Lick H$`\gamma `$ indices in this study. As noted above, we are currently engaged in a redefinition of the Lick system that is optimized for GCs.
Instead, we will use the average of H$`\beta `$ and $`H\delta _F`$ as our age-sensitive Balmer indices (in fact, using H$`\gamma _F`$ in the average as well gives consistent results). In Figure 1 we plot PCA \[m/H\] vs. (H$`\beta `$+$`H\delta _F`$)/2. The formal random errors for PCA \[m/H\] are negligible, but we assign a systematic calibration error of 0.12 dex to all points (see Strader & Brodie 2004). Our GC subpopulations fall at or below the envelope defined by the Galactic GCs, suggesting that they are no younger than Galactic GCs at all metallicities. 14 Gyr model lines for \[$`Z`$/H\] = $`2.25`$ to 0 and \[$`\alpha `$/Fe\] = 0 and +0.3 from Thomas, Maraston, & Korn (2004, TMK) are superposed. These include a blue horizontal branch below \[$`Z`$/H\] = $`1.35`$. The models are plotted only to show that our index combinations are relatively insensitive to \[$`\alpha `$/Fe\] variations; we cannot make any direct model comparisons since our data are not on the Lick system. Indeed, ages derived using PCA metallicities are even less sensitive to \[$`\alpha `$/Fe\] than those derived from \[MgFe\]$`\mathrm{}`$. Thus, the use of PCA metallicities is preferable when deriving SSP ages for GCs, and this is likely to be true even for more metal-rich stellar systems, up to the current PCA calibration limit of solar metallicity. The MR Galactic GC with high Balmer line strength is NGC 6553; its location is due not to youth or warm horizontal branch stars but probably poor sampling in the crowded bulge (Puzia et al. 2002b).
Let us consider the scatter among the mean subpopulations in Figure 1 more quantitatively. The relation between \[m/H\] and Balmer line strength appears to be linear to solar metallicity; a least-squares fit to all points with \[m/H\] $`<0`$ gives: (H$`\beta `$+$`H\delta _F`$)/2 = $`1.221`$ \[m/H\] + 0.632, with a residual standard error of 0.17. The mean Balmer error for these subpopulations is 0.16. Thus, the extragalactic points are entirely consistent with expected scatter in the mean values around a single line. This need not necessarily be a single isochrone (as there could be a shift to younger ages with increasing metallicity), but it does indicate that there is not a large age spread from galaxy to galaxy at fixed metallicity.
What can we conclude about the absolute ages of our GC subpopulations? Our results are directly tied to knowledge of the ages of MP and MR GCs in the Milky Way (MW). Unfortunately, there is considerable uncertainty in this subject as well. Using $`\alpha `$-enhanced isochrones with diffusion, Gratton et al. (2003) find that the oldest MP GCs have formal ages of 13.4 $`\pm 0.8\pm 0.7`$ Gyr (random and systematic errors). Krauss & Chaboyer (2002) used a Monte Carlo technique to perform isochrone fits to simulated metal-poor color-magnitude diagrams with varying physical parameters, and found best-fit ages of 12.6 Gyr with $`1\sigma `$ errors of $`1`$ Gyr. A recent revision of the <sup>14</sup>N($`p,\gamma `$)<sup>15</sup>O reaction rate raises these ages by 0.7–1.0 Gyr (Imbriani et al. 2004); they are indistinguishable from the age of the universe within the errors. Similar ages are found through integrated spectroscopy (Proctor, Forbes, & Beasley 2004).
Few MR GCs have robust (determined through direct main-sequence fitting) age determinations. Even for the best-studied of these GCs, 47 Tuc, there are still significant discrepancies, depending on the choice of subdwarfs and colors used for fitting. 47 Tuc could be coeval with MP GCs, or $`23`$ Gyr younger at 11 Gyr (Gratton et al. 2003, Grundahl, Stetson, & Andersen 2002; or $`12`$ Gyr with the updated <sup>14</sup>N($`p,\gamma `$)<sup>15</sup>O rate). The main source of error is the distance modulus, and a convincing resolution of the issue may require direct parallax measurements from SIM or GAIA. In addition, though the bulk of the MR GCs are probably associated with the bulge (Minniti 1995; Côté 1999; Forbes, Larsen, & Brodie 2001), a subset may belong to the thick disk—including 47 Tuc. Such thick disk clusters, if they exist, probably have few counterparts in early-type galaxies. Secure turnoff ages for more MR GCs will be needed before a more certain estimate of their mean properties can be made, but for the purposes of this paper we assume a minimum age of 10 Gyr for MR GCs.
In conclusion, the GC subpopulations in our sample all appear to have mean ages $`10`$ Gyr. However, relative *mean* subpopulation ages more accurate than $`12`$ Gyr ($`1\sigma `$) cannot be determined at present, and age errors on individual clusters are $`34`$ Gyr.
## 4 Discussion
The old GC ages derived in this paper are consistent with an emerging observational and theoretical consensus that very massive galaxies, especially in dense environments, formed the bulk of their stars at high redshift (McCarthy et al. 2004; Chen & Marzke 2004; Somerville 2004; Nagamine et al. 2001). There is no sharp cutoff for the classification of a galaxy as “very massive”, but a number often adopted is 3$`L^{}`$ = $`2\times 10^{11}M_{}`$, which corresponds to $`M_V21.5`$ for an old metal-rich stellar population. Most spectroscopic studies of GC systems beyond the Local Group have focused on such galaxies, since they generally have the largest GC systems and thus are easiest to study. Our findings in this paper, coupled with the enormous number of photometric studies of GCs in such galaxies over the last several decades, are entirely consistent with the formation at high redshift of most GCs in massive galaxies in dense environments. These ages are consistent with previous spectroscopic studies of GC systems in individual galaxies by our group and by other workers (e.g., Kuntschner et al. 2002; Cohen, Blakeslee, & Ryzhov 1998).
Old ages for the majority of GCs in galaxies of all types might seem, on the face of it, to conflict with the discovery of young “proto-GC” candidates in gas-rich mergers (e.g., NGC 1275, Holtzman et al. 1992; NGC 7252, Whitmore et al. 1993; the Antennae, Whitmore & Schweizer 1995; NGC 3921, Schweizer et al. 1996) and intermediate-age ($`1`$–5 Gyr) GCs in some merger remnants. These latter systems have been targeted because they have already suffered the bulk of their dynamical evolution and can thus be more readily compared to old GC systems. There are only two galaxies for which intermediate-age GCs have been spectroscopically confirmed: NGC 1316 (Goudfrooij et al. 2001) and NGC 3610 (Strader et al. 2003a), though the existence of such GCs has been suggested on the basis of lower S/N spectra in NGC 5128 (Peng, Ford, & Freeman 2004) and NGC 1399 (Forbes et al. 2001). NGC 1316 appears to have a significant population of intermediate-age clusters, as revealed by variations of the GCLF with galactocentric radius (Goudfrooij et al. 2004). However, these clusters make up only a minor component of the total GC system in NGC 3610 (Strader et al. 2004a). The 1–2 Gyr merger remnant NGC 1052 shows no evidence for intermediate-age GCs (Pierce et al. 2005). On the basis of optical/near-IR photometry, it has been suggested that intermediate-age subpopulations exist in several other galaxies (e.g., NGC 4365, Puzia et al. 2002a; NGC 5846, Hempel et al. 2003). Recent high S/N, follow-up spectroscopy of prime intermediate-age candidates in NGC 4365 found only old GCs (Brodie et al. 2005), showing the need for high S/N spectroscopic confirmation. The intermediate-age merger remnants studied thus far are a mixed bag; at least some remnants have smaller populations of intermediate-age GCs than might be expected if the major merger model is correct.
In addition to GC ages, a successful formation scenario must also reproduce the GC metallicity–galaxy mass relations for *both* MP and MR GCs. The slopes of these relations are now well-determined (Larsen et al. 2001, Strader, Brodie, & Forbes 2004b), and can offer important quantitative constraints on GC formation models, although they have not yet been used in this respect. The relations are also useful for defining qualitative scenarios for the formation of massive ellipticals, as discussed below.
The in situ scenario (Forbes, Brodie & Grillmair 1997) is sometimes considered together with the SAM of Beasley et al. (2002) under the rubric “hierarchical”. In some sense all modern scenarios must be hierarchical, given the strength of evidence for the assembly of present-day DM halos through hierarchical merging in a $`\mathrm{\Lambda }`$CDM cosmology. Indeed, the accretion models of Côté et al. (1998, 2000, 2002) were explicitly inspired by the convergence toward hierarchical galaxy assembly. While this specific scenario is difficult to reconcile with recent observations (since the mean metallicities of GCs in dwarf galaxies are lower than those of MP GCs in massive galaxies; Strader et al. 2004b), the hierarchical nature of structure formation implies that all models are accretion models to a certain extent.
### 4.1 Metal-Poor Globular Clusters
A key problem with the accretion model (and a point not considered in Strader et al. 2004b) is that local dwarfs cannot be directly compared to their counterparts at high redshift (see, e.g., Santos 2003 as it relates to MP GCs, and Robertson et al. 2005 on the formation of the Galactic halo). Dwarf galaxies sitting “on top” of a large overdensity destined to become a massive galaxy will collapse before those on the periphery. If MP GC formation was terminated by reionization, the dwarfs that collapsed first might have produced MP GCs of a higher metallicity than those dwarfs that collapsed later.
This could happen in a variety of ways: the dwarfs collapsing first could accrete a larger quantity of pristine gas (thereby enhancing star formation), have more time to process the gas they accreted, or capture more enriched, outflowing gas from the higher density of nearby starforming halos. These more centrally concentrated halos would generally be accreted into their parent halo, with their GCs forming its MP subpopulation, while some of the outlying halos would survive to be present-day dwarfs. This scenario could qualitatively reproduce the MP GC-metallicity–galaxy mass relation, since larger overdensities collapse first and thus would have MP GCs of higher metallicity. Since the MP GCs are accreted from low-mass halos that are enriched “in place” within a DM overdensity, our picture can be seen as a synthesis of the accretion and in situ approaches. We have begun work to test this hypothesis using SAM+N-body halo merger trees and simple chemical evolution models. Recent numerical simulations of the assembly of Milky Way-sized halos (Moore et al. 2005, priv. communication) indicate that the radial distribution of MP Galactic GCs they predict is consistent with our scenario.
We also note here the continuing efforts to place GC formation in a cosmological context. Given the lack of observational evidence for DM halos around GCs (Moore 1996; Odenkirchen et al. 2003), a popular view among cosmologists is that MP GCs formed with halos that were later stripped away (e.g., Cen 2001; Bromm & Clarke 2002; Scannapieco et al. 2004) or are still present but difficult to detect (Mashchenko & Sills 2005). However, any model which requires different formation mechanisms for MP and MR GCs must contend with the strong similarities between the two subpopulations, including sizes and the shapes of the mass and metallicity distributions (see also Harris 2003).
What can we infer about the masses of halos in which MP GCs formed? A number of Galactic satellites with total masses $`10^6`$$`10^7M_{}`$ lack GCs entirely (Forbes et al. 2000), while the Fornax and Sagittarius dSphs, with masses $`10^8M_{}`$ (Walcher et al. 2003; Law et al. 2005), have at least five GCs each. The dSph DDO 78 in the M81 group is slightly less massive and has one GC (Sharina et al. 2003). Grebel et al. (2000) find a candidate GC of the M31 satellite And I ($`M_V12`$; $`M_{tot}`$ a few $`\times 10^7`$ for an $`M/L`$ similar to Fornax), but at a projected distance of $`40`$ kpc from M31 the GC could be a contaminant. These results might suggest that MP GCs formed in halos with a minimum mass of $`10^7`$$`10^8M_{}`$. This “intermediate” regime could be better probed with further study of And I (and the similar mass satellite And II).
### 4.2 Metal-rich Globular Clusters and Formation Efficiency
Despite a variety of evidence suggesting that massive, compact star clusters form in most (possibly all) major episodes of star formation (e.g., Schweizer 2000), and that a subset of these clusters may survive to become GCs, the relation between GCs and field stars is poorly understood. For lack of data, arguments for the universality of GC formation efficiency (e.g., McLaughlin 1999) have generally not been based on the separation of GC and field star subpopulations. In the MW, modern interpretations associate MP and MR GCs with the halo ($`1.5\times 10^9M_{}`$; Carney, Latham, & Laird 1990) and bulge ($`10^{10}M_{}`$), respectively. There are $`100`$ MP and 50 MR GCs in the MW. The subpopulations have mass-normalized specific frequencies ($`T`$ = the number of GCs per $`10^9`$ $`M_{}`$; Zepf & Ashman 1993) of 67 and 5, respectively. The “universal” GC formation efficiency of McLaughlin corresponds roughly to $`T8`$ (assuming all gas in a system has been converted to stars), and is rather similar to the MR value. However, the MP value is more than a factor of 10 higher than the MR one.
A strong link between MR GCs and field stars is predicted in all the GC formation scenarios discussed to date and is consistent with observations of several extragalactic GC systems (e.g., M31: Jablonka et al. 2000; NGC 5128: Harris et al. 1999, Harris & Harris 2002; NGC 1399: Forte, Faifer, & Geisler 2005) and the nearly-constant “bulge specific frequency” in elliptical galaxies and many spiral bulges (Forbes, Brodie & Larsen 2001; see Goudfrooij et al. 2003 for a slightly different view). In none of these galaxies have strong constraints been placed on the existence of a population of metal-poor “halo” field stars; existing surveys in NGC 5128 and M31 (though see Guhathakurta et al. 2005) have found few metal-poor stars, suggesting that the MP $`T`$ value may generically be higher than the MR. In our scenario, truncation by reionization offers a natural explanation for these high $`T`$ values of MP GC systems.
The intimate connection between MR GCs and the underlying galaxy starlight has implications for the present study, as the old ages of MR GCs in our sample of galaxies suggests old ages for the bulk of their field stars. However, with our small samples of GCs, we cannot constrain at present whether they were formed in a short-lived “monolithic” dissipational event nor whether a small age spread ($`<2`$ Gyr) might exist.
Both in situ and accretion scenarios predict old ages for red GCs (but not the formation mechanism). The classical major merger picture (Ashman & Zepf 1992), however, would predict younger ages, unless the main epoch of major mergers is pushed to higher redshift (say $`z>2`$); these could be visible as SCUBA sources (Smail et al. 2002). Indeed, massive disks have recently been found at these redshifts (e.g., Labbé et al. 2003; Stockton, Canalizo, & Maihara 2004). However, whether such disks had sufficient gas or were present in the requisite numbers to account for the formation of most massive ellipticals is still an open question. Our point here is not that gas-rich starbursts were uncommon in the early universe, but that these starbursts may have been driven by stochastic, ongoing merging of objects over a range of mass scales rather than “simple” major mergers of fully-formed massive disks. Other issues with the major merger scenario, such as specific frequencies, relative numbers of MP and MR GCs, and metallicities of the MR GCs, have been discussed in detail elsewhere (e.g., Ashman & Zepf 1998; Harris 2001).
It is clear that much larger samples of high S/N GC spectra will be needed to investigate the details of MR GC formation. Because of downsizing, the most massive galaxies should have smaller age differences among their MR GCs but larger numbers of bright GCs to probe these differences; less massive galaxies have fewer GCs but could have more age structure. Due to the steep central concentration of GC systems, multi-object slit spectrographs, even if highly multiplexing, will continue to be limited in their ability to observe large numbers of GCs. Fiber instruments on large telescopes or natural-seeing IFUs may be better suited to this topic.
## 5 Summary & Conclusions
We have undertaken a meta-analysis of high signal-to-noise Keck spectra of GCs in eight galaxies, ranging from dwarfs to massive ellipticals. We have established the best constraints on GC ages that can realistically be achieved using current methods and technology. Ages have been inferred from a direct comparison to Galactic GCs to avoid any dependence on stellar evolutionary synthesis models. We conclude that both the MP and MR subpopulations in these galaxies are no younger than their Galactic counterparts ($`10`$ Gyr) and, within the uncertainties ($`2`$ Gyr), both subpopulations are coeval. These results indicate that the vast majority of GCs, at least in dense environments, were formed at high redshift ($`z>2`$). Given the expectation that GC formation has accompanied all major star forming episodes in a galaxy’s history, the old ages of GCs implies an old formation epoch for the stars in their host galaxies. This is consistent with results emerging from large photometric surveys of galaxies, which are increasingly revealing massive galaxies already in place at similar redshifts.
We have synthesized a new GC formation scenario which contains elements of the in situ and accretion models that have been invoked in recent years to explain the existence of GC subpopulations. The scenario suggests that MP GCs formed in low-mass halos in the early universe, with their formation truncated by reionization. MR GCs formed along with the bulk of the field stars in massive early-type galaxies and spiral bulges. This scenario qualitatively reproduces the MP and MR GC metallicity-galaxy mass relations and can explain the relative formation efficiencies of MP and MR GCs in massive galaxies.
We thank Graeme Smith for useful comments and discussion. We acknowledge support by the National Science Foundation through Grant AST-0206139 and a Graduate Research Fellowship (J.S.). A. J. C. acknowledges financial support from a UCM Fundación del Amo Fellowship. This research has made use of the NASA/IPAC Extragalactic Database (NED) which is operated by the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration. The data presented herein were obtained at the W.M. Keck Observatory, which is operated as a scientific partnership among the California Institute of Technology, the University of California and the National Aeronautics and Space Administration. The Observatory was made possible by the generous financial support of the W.M. Keck Foundation. The authors wish to recognize and acknowledge the very significant cultural role and reverence that the summit of Mauna Kea has always had within the indigenous Hawaiian community. We are most fortunate to have the opportunity to conduct observations from this mountain.
|
warning/0506/astro-ph0506369.html
|
ar5iv
|
text
|
# Jet Formation in Black Hole Accretion Systems II: Numerical Models
## 1. Introduction
General relativistic magnetohydrodynamics (GRMHD) is the black hole mass-invariant (nonradiative) physics commonly used to describe black hole accretion systems. Such systems often exhibit jets. However, the observed jet properties, such as collimation and speed, are not uniform between systems. In McKinney (2005b), our goal was to determine the unifying, or minimum number of, pieces of physics that would explain jet formation associated with gamma-ray bursts (GRBs), active galactic nuclei (AGN), and x-ray binaries. Similar ideas have been explored by other authors (Ghisellini & Celotti, 2002; Ghisellini, 2003; Meier, 2003). The goal of McKinney (2005b) was to understand jet formation and explain the origin of the energy, composition, collimation, and Lorentz factor.
This paper implements some of the theoretical ideas of McKinney (2005b) into a GRMHD numerical model. In particular, we study the collapsar model with pair creation and an effective model for Fick diffusion of neutrons that contaminate the jet with an electron-proton plasma. The generic model parameters also allow the numerical model to be interpreted as a model for AGN systems such as M87. We also comment on the applicability of the model to x-ray binary systems.
§ 2 summarizes the proposed unified model to explain jet formation in all black hole accretion systems, where more details are provided in McKinney (2005b).
§ 3 presents the results of a fiducial numerical model corresponding to the collapsar model and describes the jet structure and formation. The relevance of the presented model to AGN and x-ray binaries is discussed. The jet is found to have some piece-wise self-similar features, and curve fits are given for use by other modellers. Acceleration of the GRB jet is found to occur over a large range in radii rather than occurring close to the black hole. The introductory estimates of the Lorentz factor in McKinney (2005b) are refined based upon these numerical models. The jet characteristic structure, such as the formation of a double transonic (transfast) flow, is discussed.
§ 4 discusses the results and their possible implications. The results are compared to similar investigations, and the limitations of the models presented are discussed.
§ 5 summarizes the key points.
In appendix A, we give the GRMHD equations of motion solved when accounting for pair creation. See also McKinney (2005b) for relevant discussions. These are the equations numerically solved that give the results discussed in section 3. These are straightforward and how one handles the details of the pair creation ends up not making any qualitative (or much quantitative) difference in the results, hence why the material is in the appendix. That is, the energetics of the Poynting-dominated jet are dominated by the electromagnetic field. Appendix B summarizes the well-known characteristics of the GRMHD equations and other surfaces of physical interest used in the discussion in section 3.
## 2. GRMHD Pair Injection Model of Jet Formation
In McKinney (2005b), we showed that one can identify a small subset of physics that can explain the jet energy, composition, collimation, and Lorentz factor for all black hole accretion systems.
One key idea of that paper is that the terminal Lorentz factor is determined by the toroidal magnetic energy per unit pair mass density energy near the location where pairs can escape to infinity (beyond the so-called “stagnation surface”). For GRBs, neutron diffusion is crucial to explain (and limit) the Lorentz factor and explain the baryon-pollution or baryon-contamination problem. For AGN and x-ray binaries, since a negligible number of baryons cross the field lines, pair-loading from radiative annihilation is crucial to determine the Lorentz factor of the Poynting-dominated jet since this determines the rest-mass flux and density in the jet.
Figure 1 shows the basic picture for GRB systems, while figure 2 shows the basic picture for AGN and x-ray binary systems. An accreting, spinning black hole creates a magnetically dominated funnel region around the polar axis. The rotating black hole drives a Poynting flux into the funnel region, where the Poynting flux is associated with the coiling of poloidal magnetic field lines into toroidal magnetic field lines. The ideal MHD approximation holds very well, and so only neutral particles such as photons, neutrinos, and neutrons can cross the field lines and load the Poynting-dominated jet that emerges.
The accretion disk emits neutrinos in a GRB model ($`\gamma `$-ray and many soft photons for AGN and x-ray binaries) that annihilate and pair-load the funnel region within some “injection region.” For GRB systems, neutrons Fick-diffuse across the field lines and collisionally decay into an electron-proton plasma.
Many pairs (any type) are swallowed by the black hole, but some escape if beyond some “stagnation surface,” where the time-averaged poloidal velocity is zero and positive beyond. Pairs beyond the stagnation surface are then accelerated by the Poynting flux in a self-consistently generated collimated outflow. In the electromagnetic (EM) jet, the acceleration process corresponds to a gradual uncoiling of the magnetic field and a release of the stored magnetic energy that originated from the spin energy of the black hole.
One key result of this paper is that the release of magnetic energy need not be gradual once the toroidal field dominates the poloidal field, in which case pinch (and perhaps kink) instabilities can occur and lead to a nonlinear coupling (e.g. a shock) that converts Poynting flux into enthalpy flux (Eichler, 1993; Begelman, 1998). In the proposed GRB model, this conversion reaches equipartition and the jet becomes a “magnetic fireball,” where the toroidal field instabilities drive large variations in the jet Lorentz factor and jet luminosity.
In McKinney (2005b), we showed that in AGN systems, nonthermal synchrotron from shock-accelerated electrons and some thermal synchrotron emission releases the shock energy until the synchrotron cooling times are longer than the jet propagation time. For AGN, jet acceleration is negligible beyond the extended shock zone, as suggested for blazars beyond the “blazar zone” (Sikora et al., 2005). In x-ray binary systems, the shock is not as hot and also unlike in the AGN (at least those like M87) case the jet can be optically thick. Thus these x-ray binary systems self-absorbed synchrotron emit if they survive Compton drag.
For all these systems, at large radii patches of energy flux and variations in the Lorentz factor develop due to toroidal instabilities. These patches in the jet could drive internal shocks and at large radii they drive external shocks with the surrounding medium. The EM jet is also surrounded by a mildly relativistic matter coronal outflow/jet/wind, which is a material extension of the corona surrounding the disk. This Poynting-baryon, coronal outflow collimates the outer edge of the Poynting-dominated jet, which otherwise internally collimates by hoop stresses. The luminosity of the Poynting-baryon jet is determined, like the Poynting-dominated jet, by the mass accretion rate, disk thickness, and black hole spin.
In this paper, this unified model is studied numerically using axisymmetric, nonradiative, GRMHD simulations to study the self-consistent process of jet formation from black hole accretion systems. These simulations extend the work of McKinney & Gammie (2004) by including pair creation (and an effective neutron diffusion for GRB-type systems) to self-consistently treat the creation of jet matter, investigating a larger dynamic range in radius, and presenting a more detailed analysis of the Poynting-dominated jet structure.
### 2.1. Units and Notation
The units in this paper have $`GM=c=1`$, which sets the scale of length ($`r_gGM/c^2`$) and time ($`t_gGM/c^3`$). The mass scale is determined by setting the (model-dependent) observed (or inferred for GRB-type systems) mass accretion rate ($`\dot{M}_0`$\[$`gs^1`$\]) equal to the accretion rate through the black hole horizon as measured in a simulation. So the mass is scaled by the mass accretion rate ($`\dot{M}_0`$) at the horizon ($`r=r_Hr_L(1+\sqrt{1j^2})`$), such that $`\rho _{0,disk}\dot{M}_0[r=r_H]t_g/r_g^3`$ and the mass scale is then just $`m\rho _{0,disk}r_g^3=\dot{M}_0[r=r_H]t_g`$. Unless explicitly stated, the magnetic field strength is given in Heaviside-Lorentz units, where the Gaussian unit value is obtained by multiplying the Heaviside-Lorentz value by $`\sqrt{4\pi }`$.
The value of $`\rho _{0,disk}`$ can be determined for different systems. For example, a collapsar model with $`\dot{M}=0.1\mathrm{M}_{}s^1`$ and $`M3\mathrm{M}_{}`$, then $`\rho _{0,disk}3.4\times 10^{10}\mathrm{g}\mathrm{cm}^3`$. M87 has a mass accretion rate of $`\dot{M}10^2\mathrm{M}_{}\mathrm{yr}^1`$ and a black hole mass of $`M3\times 10^9\mathrm{M}_{}`$ (Ho, 1999; Reynolds et al., 1996) giving $`\rho _{0,disk}10^{16}\mathrm{g}\mathrm{cm}^3`$. GRS 1915+105 has a mass accretion rate of $`\dot{M}7\times 10^7\mathrm{M}_{}\mathrm{yr}^1`$ (Mirabel & Rodriguez, 1994; Mirabel & Rodríguez, 1999; Fender & Belloni, 2004) with a mass of $`M14\mathrm{M}_{}`$ (Greiner et al., 2001), but see Kaiser et al. (2004). This gives $`\rho _{0,disk}3\times 10^4\mathrm{g}\mathrm{cm}^3`$. This disk density scales many of the results of the paper. The disk height ($`H`$) to radius ($`R`$) ratio is written as $`H/R`$.
The GRMHD notation follows MTW. For example, the 4-velocity components are $`u^\mu `$ (contravariant) or $`u_\mu `$ (covariant). For a black hole with angular momentum $`J=jGM^2/c`$, $`j=a/M`$ is the dimensionless Kerr parameter with $`1j1`$. The rest-mass density is given as $`\rho _0`$, internal energy density as $`u`$, magnetic field 3-vector as $`B^i{}_{}{}^{^{}}F_{}^{it}`$, where $`{}_{}{}^{^{}}F`$ is the dual of the Faraday tensor. The contravariant metric components are $`g^{\mu \nu }`$ and covariant components are $`g_{\mu \nu }`$, where beyond the frame-dragging of the black hole the metric in Boyer-Lindquist coordinates is approximately diagonal such that an orthonormal contravariant component is $`u^{\widehat{\mu }}=\sqrt{g_{\mu \mu }}u^\mu `$ for any spatial component of $`u^\mu `$. The comoving energy density is $`b^2/2`$, where $`b^\mu `$ is the comoving magnetic field. See McKinney & Gammie (2004) for details.
The Lorentz factor of the jet can be measured either as the current time-dependent value, or, using information about the GRMHD system of equations, one can estimate the Lorentz factor at large radii from fluid quantities at small radii. The Lorentz factor as measured by a static observer at infinity is
$$\mathrm{\Gamma }u^{\widehat{t}}=u^t\sqrt{g_{tt}}$$
(1)
in Boyer-Lindquist coordinates, where no static observers exist inside the ergosphere. This is as opposed to $`Wu^t\sqrt{1/g^{tt}}`$, which is the Lorentz factor as measured by the normal observer as used by most numerical relativists.
In McKinney (2005b) we show that the Lorentz factor and $`\varphi `$-velocity at large distances are
$`\mathrm{\Gamma }_{\mathrm{}}`$ $`=`$ $`E=hu_t+\mathrm{\Phi }\mathrm{\Omega }_FB_\varphi `$ (2)
$`u_{\mathrm{}}^{\widehat{\varphi }}`$ $`=`$ $`L=hu_\varphi +\mathrm{\Phi }B_\varphi ,`$ (3)
where $`h=(\rho _0+u_g+p)/\rho _0`$ is the specific enthalpy, $`\mathrm{\Phi }`$ is the conserved magnetic flux per unit rest-mass flux, $`\mathrm{\Omega }_F`$ is the conserved field rotation frequency, $`B_\varphi `$ is the covariant toroidal magnetic field, and $`u_{\mathrm{}}^{\widehat{\varphi }}=u_{\varphi ,\mathrm{}}`$. Here $`E`$ and $`L`$ simply represent the conserved energy and angular momentum flux per unit rest-mass flux. Notice the matter and electromagnetic pieces are separable, such that
$`\mathrm{\Gamma }_{\mathrm{}}`$ $`=`$ $`\mathrm{\Gamma }_{\mathrm{}}^{(MA)}+\mathrm{\Gamma }_{\mathrm{}}^{(EM)}`$ (4)
$`u_{\varphi ,\mathrm{}}`$ $`=`$ $`u_{\varphi ,\mathrm{}}^{(MA)}+u_{\varphi ,\mathrm{}}^{(MA)},`$ (5)
See McKinney (2005b) for more details.
## 3. Numerical Experiments
This section presents the results of GRMHD numerical models with pair creation to self-consistently describe the Poynting-dominated and Poynting-baryon jet formation process. Our numerical scheme is HARM (Gammie et al., 2003a), a conservative, shock-capturing scheme for evolving the equations of general relativistic MHD. Compared to the original HARM, the inversion of conserved quantities to primitive variables is performed by using a new faster and more robust two-dimensional non-linear solver (Noble et al., 2005). The new HARM also uses a parabolic interpolation scheme (Colella, 1984) rather than a linear interpolation scheme. The new HARM also uses an optimal TVD third order Runge-Kutta time stepping (Shu, 1997) rather than the mid-point method. For the problems under consideration, the parabolic interpolation and third order time stepping method reduce the truncation error significantly, including magnetically dominated regions with $`b^2/\rho _01`$.
Notice that no explicit reconnection model is included. However, HARM checks the effective resistivity by measuring the rest-mass flux across field lines. For the boundary between the coronal wind and the Poynting-dominated jet, the total mass flux across the field line is negligible compared to the rest-mass flux along the field line and the pair creation rate. An unresolved model would load field lines with rest-mass that crosses field lines and not properly represent any physical resistivity (McKinney, 2004).
### 3.1. Computational Domain
The computational domain is axisymmetric, with a grid that typically extends from $`r_{in}=0.98r_H`$, where $`r_H`$ is the horizon, to $`r_{out}=10^4r_g`$, and from $`\theta =0`$ to $`\theta =\pi `$. Full 3D calculations with this dynamic range are not possible with today’s computers. Our numerical integrations are carried out on a uniform grid in so-called “modified KS” (MKS) coordinates: $`x_0,x_1,x_2,x_3`$, where $`x_0=t[\mathrm{KS}]`$, $`x_3=\varphi [\mathrm{KS}]`$. The radial coordinate is chosen to be
$$r=R_0+e^{x_1^{n_r}},$$
(6)
where $`R_0`$ is chosen to concentrate the grid zones toward the event horizon (as $`R_0`$ is increased from $`0`$ to $`r_H`$) and $`n_r`$ is controls the enhancement of inner to outer radial regions. For studies where the disk and jet interaction is of primary interest, $`R_0=0`$ and $`n_r=1`$ are chosen. For studies where in addition the far-field jet is of interest, $`R_0=3`$ and $`n_r=10`$ are chosen. The $`\theta `$ coordinate is chosen to be
$$\theta =\pi x_2+\frac{1}{2}(1h(r))\mathrm{sin}(2\pi x_2),$$
(7)
where $`h(r)`$ is used to concentrate grid zones toward the equator (as $`h`$ is decreased from $`1`$ to $`0`$) or pole (as $`h`$ is increased from $`1`$ to $`2`$). The jet at large radii is resolved together with the disk at small radii using
$$h(r)=2Q_j(r/r_{0j})^{n_jg_j}$$
(8)
with the parameters of $`Q_j=1.31.8`$, $`r_{0j}=2.8`$, $`n_j=0.3`$, $`r_{1j}=20`$, $`r_{2j}=80`$, and $`g_j=g_j(r)=\frac{1}{2}+\frac{1}{\pi }atan(\frac{rr_{2j}}{r_{1j}})`$ is used to control the transition between inner and outer radial regions. To perform convergence tests, $`Q_j`$ and the overall radial and $`\theta `$ resolution are varied. An alternative to this fixed refinement of the jet and disk is an adaptive refinement (see, e.g., Zhang & MacFadyen 2005).
### 3.2. Initial Conditions and Problem Setup
All the experiments evolve a weakly magnetized torus around a Kerr black hole in axisymmetry. The focus of our numerical investigation is to study a high resolution model of the collapsar model. Other simulations were performed, and the fiducial model’s relevance to AGN and x-ray binary systems is discussed at the end of the section. A generic model is used so the results mostly can be applied to any black hole system. A black hole spin of $`j=0.9375`$ is chosen, but this produces similar results to models with $`0.5j0.97`$, which includes most black hole accretion systems.
The initial conditions consist of an equilibrium torus which is a “donut” of plasma with a black hole at the center (Fishbone & Moncrief, 1976; Abramowicz, Jaroszinski, & Sikora, 1978). The donut is supported against gravity by centrifugal and pressure forces. The solutions of Fishbone & Moncrief (1976) corresponding to $`u^tu_\varphi =const.`$ are used. The initial inner edge of the torus is set at $`r_{edge}=6`$. The equation of state is a gamma-law ideal gas with $`\gamma =4/3`$, but other $`\gamma `$ lead to similar results (McKinney & Gammie, 2004). This donut is a solution to the axisymmetric stationary equations, but the donut is unstable to global nonaxisymmetric modes (Papaloizou & Pringle, 1983). However, when the donut is embedded in a weak magnetic field, the magnetorotational instability (MRI) dominates those hydrodynamic modes. Small perturbations are introduced in the velocity field to seed the instability. The models have $`u^tu_\varphi =4.281`$, the pressure maximum is located at $`r_{max}=12`$, the inner edge at $`(r,\theta )=(6,\pi /2)`$, and the outer edge at $`(r,\theta )=(42,\pi /2)`$.
This is consistent with the accretion disk that is expected to form in the collapsar model (MacFadyen & Woosley, 1999). A GRB in the collapsar model is formed after the collapse of a massive star as a black hole with $`M_{BH}3M_{}`$ and spin $`j0.750.94`$ accretes at $`\dot{M}_00.1M_{}/s`$ from a torus of matter within $`r<200\mathrm{km}`$ that has a height ($`H`$) to radius ($`R`$) ratio (“thickness”) of $`H/R0.20.4`$. MacFadyen & Woosley (1999) find that an energetic outflow develops in the polar region where $`\rho 10^7\mathrm{g}\mathrm{cm}^3`$ near the horizon, $`\rho 10^5\mathrm{g}\mathrm{cm}^3`$ at $`r200\mathrm{km}`$, and $`\rho 10^{34}\mathrm{g}\mathrm{cm}^3`$ at $`r2000\mathrm{km}`$ (see model 14A in MacFadyen & Woosley 1999). In the parlance of collapsar models, our $`u^tu_\varphi `$ model corresponds to a $`j_{16}j/(10^{16}\mathrm{cm}^2\mathrm{s}^1)`$ of $`j_{16}5`$, within the range suggested by MacFadyen & Woosley (1999) to produce a GRB. Also, the $`H/R0.20.4`$ as suggested that forms according to neutrino emission models (Kohri & Mineshige, 2002; Kohri et al., 2005).
For radiatively inefficient AGN and x-ray binaries this is simply a reasonable approximate model for the inner-radial accretion flow. While a slightly thicker disk ($`H/R0.9`$) may be more appropriate for radiatively inefficient flows, this is not expected to significantly affect these results. A study of the effect of disk thickness, especially thin disks, is left for future work.
The orbital period at the pressure maximum $`2\pi (a+\left(r_{max}/r_g\right)^{3/2})t_g267t_g`$, as measured by an observer at infinity. The model is run for $`\mathrm{\Delta }t=1.4\times 10^4t_g`$, which is about $`52`$ orbital periods at the pressure maximum and about $`1150`$ orbital periods at the black hole horizon. For the collapsar model this is only $`0.2`$ seconds, and at the spin chosen would correspond to the late phase of accretion when the Poynting flux reaches its largest magnitude. For the AGN M87 this corresponds to $`7\mathrm{years}`$. For the x-ray binary GRS 1915+105 this corresponds to $`1\mathrm{second}`$.
Notice that since the model is axisymmetric, disk turbulence is not sustained after about $`t3000t_g`$ ; that is, the anti-dynamo theorem prevails (Cowling, 1934). However, while this affects the disk accretion, this does not affect the evolution of the Poynting-dominated jet. That is, from the time of turbulent accretion to “laminar” accretion, the funnel region is mostly unchanged. Indeed, the far-field jet that has already formed is causally disconnected from the region where the accretion disk would still be turbulent.
For the collapsar model, at the quasi-steady state mass accretion rate of $`0.1M_{}/s`$ the disk will last for $`t0.42s`$. Thus, this model requires a more extended disk or a fresh supply of plasma into the disk from the surrounding stellar envelope through an “accretion shock” to generate a long duration GRBs (MacFadyen & Woosley, 1999). However, since any such system modelled is in quasi-steady state, the results here do not strongly depend on the mass of the disk as long as the disk is not too massive, which would require taking into account the self-gravity of the disk.
The numerical resolution of these models is $`512\times 256`$ compared to $`456^2`$ in McKinney & Gammie (2004). However, due to the enhanced $`\theta `$ grid, the resolution in the far-field jet region is $`10`$ times larger. Also, with the use of a parabolic interpolation scheme, the overall resolution is additionally enhanced. Compared to our previous model this gives us an effective $`\theta `$ resolution of $`9000`$.
As with McKinney & Gammie (2004), into the initial torus is put a purely poloidal magnetic field. The field can be described using a vector potential with a single nonzero component $`A_\varphi \mathrm{MAX}(\rho _0/\rho _{0,max}0.2,0)`$ The field is therefore restricted to regions with $`\rho _0/\rho _{0,max}>0.2`$. The field is normalized so that the minimum ratio of gas to magnetic pressure is $`100`$. The equilibrium is therefore only weakly perturbed by the magnetic field. It is, however, no longer stable (Balbus & Hawley, 1991; Gammie, 2004). Hirose et al. (2004); McKinney & Gammie (2004) show that no initial large-scale net vertical field is necessary, since a large-scale poloidal field is self-consistently generated. The field connects the black hole horizon (as observed by a physical observer) to large distances. Apart from the existence of an overall poloidal sign, the results are insensitive to the details of the initial magnetic field geometry for any physically motivated geometry (McKinney & Gammie, 2004).
### 3.3. Floor vs. Pair Creation Model
Numerical models often “model” the injection physics in the Poynting-dominated jet by employing a so-called floor model. This model forces a minimum on the rest-mass density ($`\rho _{fl}`$) and internal energy density ($`u_{fl}`$), which are usually set to several orders of magnitude lower than the disk density. Actually, however, floor models have always been treated as a purely numerical invention in order to avoid numerical artifacts associated with the inability to numerically solve the equations of motion when the density is low. Indeed, the idea was one should convergence test by gradually lowering $`\rho _{fl}`$ and $`u_{fl}`$. This is perhaps a reasonable for numerical study of stars, but accretion flows are so hot (baryons with $`T10^{10}K`$ in the thick disk state near the black hole) that pairs are created when neutrinos annihilate (or when photons annihilate for x-ray binary and AGN systems).
In previous numerical models of Poynting jets, the funnel region is always completely evacuated, and floor-models necessarily generate mass at least where the poloidal velocity $`u^p=0`$ at the stagnation surface. That is, matter inside the stagnation surface falls into the black hole, while matter outside it is ejected as part of the jet. Indeed, arbitrary floor-models violate the ideal-MHD condition far away from the black hole where no pairs should be produced. For example, if a numerical model uses $`\rho _{fl}\mathrm{Const}.`$, then this leads to a completely unphysical model of pair creation and the ideal-MHD approximation is violated for the entire length of the jet in the funnel.
While pair creation has so far been ignored by those doing numerical models of jets from black hole accretion systems, it has been shown that the properties of the Poynting-dominated jet are almost completely determined by the pair creation physics, which thus cannot be ignored (Phinney, 1983; Punsly, 1991; Levinson, 2005).
Our collapsar model accounts for pair creation and Fick diffusion of neutrons. In this paper, there is an injected rest-mass, enthalpy, and momentum. For the fiducial collapsar-like model, rest-mass, enthalpy, and momentum are injected with energy at infinity fractions of, respectively, $`f_\rho =0.05`$, $`f_h=0.45`$, and $`f_m=0.5`$ as described in McKinney (2005b). The total energy at infinity injected is determined by neutrino annihilation rates and Fick diffusion rates. The energy injection rate from neutrino annihilation is based upon viscous disk models (Popham et al., 1999; MacFadyen & Woosley, 1999), where the rate of injection is determined by choosing a viscosity coefficient $`\alpha =0.01`$ and mass accretion rate $`\dot{M}_0=0.1\mathrm{M}_{}s^1`$ for the collapsar model. As discussed in that paper, the rest-mass is dominated by electron-positron pairs only very close to the black hole.
The value of $`\alpha `$ and $`\dot{M}_0`$ determine the energy injected in proportion to the disk density per unit light crossing time ($`t_g`$). Thus any system with comparable normalized energy injection rate would follow similar results. For example, the M87 model is similar, where there the electron-positron pairs do not annihilate and represent the true density in the jet. In the M87 model there is no Fick diffusion, yet the dimensionless model parameters are approximately the same.
The momentum direction is chosen by setting $`u_{inj}^\varphi =u^\varphi `$ and $`u_{inj}^\theta =0`$, so that $`f_m`$ determines $`u_{}^{r}{}_{inj}{}^{}`$. It turns out that the choice of $`f_h`$, $`f_m`$, and the injected momentum direction very weakly determine the jet structure for $`r6r_g`$. This is because a large fraction of the pairs are lost to the black hole and magnetic forces, rather than the injected momentum, dominate the motion of the plasma near the black hole.
Beyond this injection region, the rest-mass is dominated by electron-protons created in collision events from neutrons that Fick diffuse across the field lines into the jet region. For the collapsar model this coincidentally corresponds to the injected rest-mass injected in this model. Since the electron-positron pair annihilation only gives an additional $`10\%`$ of internal energy, then pair annihilation need not be considered, and the rest-mass injected well-models the rest-mass injected as an electron-proton plasma from Fick-diffused neutrons that collisionally decay.
This physical model of the injected particles is augmented when the rest-mass or internal energy density reaches very low values. If the density drops below some “floor” value, then the density is returned to the floor value. The floor model chosen has $`\rho _{fl}=10^7r^{2.7}\rho _{0,disk}`$ and $`u_{fl}=10^9r^{2.7}\rho _{0,disk}`$. HARM keeps track of how often the density goes below the floors and how this modifies the conservation of mass, energy, and angular momentum. The floor model contribution is negligible compared to the physical injection model at all spatial locations and at all times. The coefficient of the floor was chosen to be close to the resulting density near the black hole horizon. It was not chosen arbitrarily small in order to avoid numerical difficulties in integrating the equations when, rarely, the floor is activated in regions of large magnetic energy density per unit rest-mass density.
### 3.4. Boundary Conditions
Two different models are chosen for the outer boundary condition. One model is called an “outflow” boundary condition, for which all primitive variables are projected into the ghost zones while forbidding inflow. The other model is to inject matter at some specified rate at the outer boundary, unless there is outflow. This would correspond to a Bondi-like infall for AGN and x-ray binaries, or would correspond to the collapsing envelope of a massive star for collapsar models. In particular, unless there is outflow, the outer boundary is set to inject mass at the free-fall rate, with a density of $`\rho _0=10^4r^{3/2}\rho _{0,disk}`$ and $`u=10^6r^{5/2}\rho _{0,disk}`$ and no angular momentum. Presupernova models suggest that the infalling matter is at about $`30\%`$ the free-fall rate we have chosen above (MacFadyen & Woosley, 1999), but this difference is unlikely to significantly affect the jet formation. Also, for their collapsar model, the density structure in the equatorial plane varies between $`\rho _0r^{3/2}`$ to $`\rho _0r^2`$ and the internal energy is $`ur^{5/2}`$ to $`ur^{2.7}`$ (MacFadyen & Woosley, 1999). This is similar to our simplified model, which happens to also model a Bondi accretion for AGN and x-ray binaries. Thus the results are indicative of GRBs, AGN, and x-ray binaries. The outer grid radius corresponds to about $`20`$ presupernova core radii (Woosley & Weaver, 1995) or $`10^{10}\mathrm{km}`$ or about $`1/10`$th the entire star’s radius.
### 3.5. Fiducial Model
The overall character of the accretion flow is unchanged compared to the descriptions given in McKinney & Gammie (2004). The disk enters a long, quasi-steady phase in which the accretion rates of rest-mass, angular momentum, and energy onto the black hole fluctuate around a well-defined mean. Meanwhile, as in McKinney & Gammie (2004), a Poynting-dominated jet and Poynting-baryon jet (coronal outflow) have formed. The Poynting-dominated jet forms once the ram pressure of the funnel material is lower than the toroidal magnetic pressure. Afterwards the baryons are spread apart in the launch of a magnetic tube filled with pairs. This occurs within $`t500t_g`$.
The Poynting-dominated jet forms as the differential rotation of the disk and the frame-dragging of the black hole induce a significant toroidal field that launches material away from the black hole by the same force described in McKinney (2005b).
A coronal outflow is also generated between the disk and Poynting-dominated jet. In this model the coronal outflow has $`\mathrm{\Gamma }_{\mathrm{}}1.5`$. The coronal-funnel boundary contains shocks with a sonic Mach number of $`M_s100`$. The inner-radial interface between the disk and corona is a site of vigorous reconnection due to the magnetic buoyancy and convective instabilities present there. These two parts of the corona are about $`100`$ times hotter than the bulk of the disk. Thus these coronal components are a likely sites for Comptonization and nonthermal particle acceleration.
Figure 3 and figure 4 show the final time ($`t=1.4\times 10^4t_g`$) log of rest-mass density and magnetic field projected on the Cartesian z vs. x plane. For the purposes of properly visualizing the accretion flow and jet, we follow MacFadyen & Woosley (1999) and show both the negative and positive $`x`$-region by duplicating the axisymmetric result across the vertical axis. Color represents $`\mathrm{log}(\rho _0/\rho _{0,disk})`$ with dark red highest and dark blue lowest. The final state has a density maximum of $`\rho _02\rho _{0,disk}`$ and a minimum of $`\rho _010^{13}\rho _{0,disk}`$ at large radii. Grid zones are not smoothed to show grid structure. Outer radial zones are large, but outer $`\theta `$ zones are below the resolution of the figure.
Clearly the jet has pummelled its way through the surrounding medium, which corresponds to the stellar envelope in the collapsar model. By the end of the simulation, the field has been self-consistently launched in to the funnel region and has a regular geometry there. In the disk and at the surface of the disk the field is curved on the scale of the disk scale height. Within $`r10^2r_g`$ the funnel field is ordered and stable due to the poloidal field dominance. However, beyond $`r1010^2r_g`$ the poloidal field is relatively weak compared to the toroidal field and the field lines bend and oscillate erratically due to pinch instabilities. The radial scale of the oscillations is $`10^2r_g`$ (but up to $`10^3r_g`$ and as small as $`10r_g`$), where $`r10r_g`$ is the radius where poloidal and toroidal field strengths are equal. By the end of the simulation, the jet has only fully evolved to a state independent of the initial conditions at $`r5\times 10^3r_g`$, beyond which the jet features are a result of the tail-end of the initial launch of the field. The head of the jet has passed beyond the outer boundary of $`r=10^4r_g`$. Notice that the magnetic field near the black hole is in an X-configuration. This is due to the BZ-effect having power $`P_{jet}\mathrm{sin}^2\theta `$, which vanishes at the polar axis. The X-configuration is also related to the fact that the disk+corona is collimating the Poynting-dominated jet. The field is mostly monopolar near the black hole, and such field geometries decollimate for rapidly rotating black holes in force-free electrodynamics (Krasnopolsky et al., 2005).
Figures 5 and 6 show the energy structure of the disk, corona, and jet. This figure is comparable to the left panel of figure 2 in McKinney & Gammie (2004). Figure 5 shows contours for $`t1500t_g`$ and an outer scale of $`r=10^2r_g`$, while figure 6 shows contours for $`t1.4\times 10^4t_g`$ and an outer scale of $`r=10^3r_g`$. The disk-coronal boundary is represented as a cyan contour where $`\beta p_g/p_b2(\gamma 1)u/b^2=3`$, the coronal-wind boundary as a magenta contour where $`\beta =1`$, and the jet-wind boundary as a red contour where $`b^2/(\rho _0c^2)=1`$. The black contour denotes the boundary beyond where material is unbound and outbound (wind or jet). Beyond $`r10r_g`$ the jet undergoes poloidal oscillations due to toroidal pinch instabilities, which subside by $`r10^210^3r_g`$ where the jet is thermally supported. At large scales, the cyan and magenta contours closer to the equatorial plane are not expected to cleanly distinguish any particular structure.
Figure 7 shows the disk, corona, and jet magnetic field structure during the turbulent phase of accretion at $`t1500t_g`$. Compared to figure 4, this shows the turbulence in the disk, but is otherwise similar. The jet, disk, and coronal structures remains mostly unchanged at late times despite the decay of disk turbulence. That is, the current sheets in the disk do not decay and continue to support the field around the black hole that leads to the Blandford-Znajek effect. The corona thickness and radial extent do not require the disk turbulence, which only adds to the time-dependence of the coronal outflow.
Figure 8 shows a color plot of $`\mathrm{\Gamma }_{\mathrm{}}`$, where red is highest and reaches up to $`\mathrm{\Gamma }_{\mathrm{}}10^310^4`$ and yellow has $`\mathrm{\Gamma }_{\mathrm{}}10^210^3`$. Panel (A) shows outer scale of $`r=10^4r_g`$, while panel (B) shows outer scale of $`r=10^3r_g`$ with same color scale. The inner-radial region is not shown since $`\mathrm{\Gamma }_{\mathrm{}}`$ is divergent near injection region where ideal-MHD breaks down. Different realizations (random seed of perturbations in disk) lead to up to about $`\mathrm{\Gamma }_{\mathrm{}}10^4`$ as shown for the lower pole in the color figures. This particular model was chosen for presentation for its diversity between the two polar axes. The upper polar axis is fairly well-structured, while the lower polar axis has undergone an atypically strong magnetic pinch instability. Various realizations show that the upper polar axis behavior is typical, so this is studied in detail below. The strong hollow-cone structure of the lower jet is due to the strongest field being located at the interface between the jet and envelope, and this is related to the fact that the BZ-flux is $`\mathrm{sin}^2\theta `$, which vanishes identically along the polar axis. It is only the disk+corona that has truncated the energy extracted, otherwise the peak power would be at the equator.
Figure 9 shows the velocity structure of the Poynting-dominated jet along a mid-level field line. The top panel shows the late-time time-averaged values of $`\mathrm{\Gamma }_{\mathrm{}}^{(EM)}`$ (solid line), $`\mathrm{\Gamma }_{\mathrm{}}^{(MA)}`$ (dotted line), and $`\mathrm{\Gamma }=u^{\widehat{t}}[\mathrm{BL}\mathrm{coords}]`$ (dashed line) as a function of radius along a mid-level field line. For $`2r10`$, the value of $`\mathrm{\Gamma }_{\mathrm{}}^{(EM)}`$ is highly oscillatory. This shows the location of the stagnation surface, where the poloidal velocity $`u^p=0`$, is temporally variable. The stagnation surface varies between $`2r10`$, within the range studied in prior sections. This is where ideal-MHD approximation is breaking and thus at any moment the value of $`\mathrm{\Gamma }_{\mathrm{}}^{(EM)}`$ nearly diverges. Notice that while at $`r10^3r_g`$ the Poynting flux dominates, the Poynting flux energy is slowly converted into enthalpy flux due to nonlinear time-dependent shock heating. Shocks are expected beyond the magneto-fast surface (see, e.g., Bogovalov & Tsinganos 2005) and here are due to pinch instabilities (Eichler, 1993; Begelman, 1998). For $`r10^3r_g`$, the enthalpy flux and Poynting flux are in equipartition.
Thus a “magnetic fireball” has developed and the terminal Lorentz factor is of order $`\mathrm{\Gamma }_{\mathrm{}}400`$ for this choice of flow line. The Lorentz factor by $`r10^4`$ is still only $`5\mathrm{\Gamma }10`$, with smaller patches with $`\mathrm{\Gamma }25`$. This implies there will be an extended acceleration region where the magnetic fireball loses energy during adiabatic expansion. These results should also help motivate the boundary conditions used in jet simulations (Aloy et al., 2000; Zhang, Woosley, & MacFadyen, 2003; Zhang W. et al., 2004), which sometimes start out with $`\mathrm{\Gamma }50`$ and $`u/\rho _0150`$ near the black hole. Such a high enthalpy only occurs by $`10^3r_g`$ in our simulations. Their simulations are somewhat indicative of the Lorentz factor growth beyond our simulated time and radial range. This suggests that $`\mathrm{\Gamma }100`$ should occur by $`r10^6r_g`$ and that $`\mathrm{\Gamma }10^3`$ should occur by $`r10^7r_g`$. This is sufficient to avoid the compactness problem. A simulation of the acceleration region is left for future work.
The next lower panel of figure 9 shows the electromagnetic contribution to the specific angular momentum at infinity ($`L=\mathrm{\Phi }B_\varphi `$) (solid line), the matter contribution to the specific angular momentum at infinity ($`hu_\varphi `$) (dotted line), and the specific angular momentum $`u_\varphi `$ (dashed line).
The bottom panel of figure 9 shows the orthonormal radial ($`u^{\widehat{r}}=\sqrt{g_{rr}}u^r`$) (solid line) and $`\varphi `$ ($`u^{\widehat{\varphi }}=\sqrt{g_{\varphi \varphi }}u^\varphi `$) (dotted line) 4-velocities in Boyer-Lindquist coordinates. The directed motion becomes relativistic, while the $`\varphi `$-component of the 4-velocity becomes sub-relativistic. However, the rotation is still large enough that it may stabilize the jet against $`m=1`$ kink instabilities, as shown in the nonrelativistic case (Nakamura & Meier, 2004).
Figure 10 shows the $`\theta `$ cross-section of the jet at $`r=5\times 10^3r_g`$ for the upper polar axis according to figures 3, 4, and 8. This is a location by which the jet has stabilized in time, where farther regions are still dependent on the initial conditions. The hot fast “core” of the jet includes $`\theta 5^{}`$ at this radius and is marked by the left dashed vertical line. Also, it is useful to notice that $`\theta 0.14`$ corresponds to the radius obtained for the “mid-level” field line shown for radial cross-sections in figures 9 and 11.
The region within $`\theta 27^{}`$ is an expanded cold slow portion of the jet. It is possible that the gamma-ray burst photons are due to Compton drag from soft photons emitted by this jet “sheath” dumping into the faster spine (Begelman & Sikora, 1987; Ghisellini et al., 2000; Lazzati et al., 2004). Based upon the data fits described below for $`r120r_g`$ and a radiation-dominated plasma, the outer sheath’s ($`\theta 0.2`$) seed photon temperature as a function of radius is
$$T_{\gamma ,seed}50\mathrm{k}\mathrm{e}\mathrm{V}\left(\frac{r}{5\times 10^3r_g}\right)^{1/3}$$
(9)
These seed photons can be upscattered by the jet and produce a GRB and high energy components. A self-consistent study of a Compton dragged jet that determines $`\mathrm{\Gamma }`$ and the emission energy is left for future work.
The right dashed vertical line marks the boundary between the last field line that connects to the black hole. Beyond this region is the surrounding infalling medium. The upper-left panel shows $`\mathrm{\Gamma }_{\mathrm{}}`$ (solid line), $`\mathrm{\Gamma }`$ (dotted line) and Gaussian fits to these two as dashed and long-dashed overlapping lines. The lower left panel shows the angular energy structure of the jet where $`ϵ(\theta )r^2T_t^rr^2\rho _0u^r\mathrm{\Gamma }_{\mathrm{}}`$ with an overlapping Gaussian fit as a dotted line. The right panels show the density and magnetic structure of the jet. The upper right panel shows the density (solid line) and internal energy density (dotted line). The lower right panel shows the orthonormal Boyer-Lindquist coordinate toroidal (solid line) and radial (dotted line) field components.
This jet structure is weakly due to an interaction with the surrounding medium, and primarily due to internal evolution of the jet. Clearly the slower jet envelope is non-negligible, giving credence to the universal structured jet (USJ) model, but see Lamb et al. (2004) and see Zhang B. et al. (2004); Lloyd-Ronning et al. (2004). Gaussian jets have been used to explain a universal connection between GRBs and x-ray flashes (Zhang B. et al., 2004). In most of our simulated models the jet structure is Gaussian with $`ϵ(\theta )=ϵ_0e^{\theta ^2/2\theta _0^2}`$, where $`ϵ_00.18`$ and $`\theta _08^{}`$, within the range that they suggest fits observations. The total luminosity per pole is $`L_j0.023\dot{M}_0c^2`$, where $`10\%`$ of that is in the “core” peak Lorentz factor region of the jet within a half-opening angle of $`5^{}`$. Also, $`\mathrm{\Gamma }_{\mathrm{}}`$ is approximately Gaussian with $`\mathrm{\Gamma }_{\mathrm{},0}3\times 10^3`$ and $`\theta _04.3^{}`$. Also, $`\mathrm{\Gamma }`$ is approximately Gaussian with $`\mathrm{\Gamma }_05`$ and $`\theta _011^{}`$. This can be folded into various models, such as the probability of observing polarised emission in Compton drag emission models (Ghisellini et al., 2000; Lazzati et al., 2004).
For the collapsar model the above gives $`L_j4\times 10^{51}\mathrm{erg}\mathrm{s}^1`$ for this $`j=0.9375`$ black hole model. This is approximately equal to the luminosity emitted by the black hole plus the energy in pairs produced by annihilating neutrinos. Not all of this energy can be observed in $`\gamma `$-rays to obtain $`E_j2\times 10^{51}\mathrm{erg}`$ for a $`30`$ second event (Zhang B. et al., 2004), unless the black hole has spin $`j0.6`$ for most of the burst, which would suggest little spin evolution that requires a much thicker disk (Gammie, Shapiro, & McKinney, 2004) and so less neutrino emission and annihilation efficiency. In McKinney (2005b), we found that $`\mathrm{\Gamma }_{\mathrm{}}`$ is weakly dependent on the black hole spin for $`a0.6`$, so the Lorentz factor is likely unaffected by such changes.
It is more probable that the emission in $`\gamma `$-rays is inefficient or only a fraction of the total energy is observed (i.e. this requires $`2\%`$ for each if sole effect) ; or the true black hole mass accretion rate is lower than predicted by current collapsar models (i.e. $`\dot{M}_00.002\mathrm{M}_{}s^1`$ rather than $`\dot{M}_00.1\mathrm{M}_{}s^1`$ if sole effect). For the standard collapsar model, if one only observed the core of the jet, which has only $`10\%`$ of the luminosity, then an efficiency of converting to $`\gamma `$-rays of $`20\%`$ is required to fit the model of Zhang B. et al. (2004).
Notice that the disk thickness (and so mass accretion rate) may strongly determine the collimation angle. In higher mass accretion rate systems, the disk is thicker than described here, and the final opening angle may be smaller. Also, if the mass accretion rate is much lower, the disk is also thicker (Kohri et al., 2005). The collapsar type model gives the thinnest disk near the black hole and small changes in the mass accretion rate lead to a small change in the disk thickness (Kohri et al., 2005). Also, for the relevant black hole spins, the spin only weakly determines the final collimation angle (McKinney & Gammie, 2004). The universal jet model requires quite large opening angles that would not be supported by invoking stellar (and so black hole) spin dependence. It would also not be supported by invoking mass accretion rate dependence, since a system with a higher or lower accretion rate actually has a thicker disk (Kohri et al., 2005) and so stronger collimation.
There is a weak dependence on the density of the stellar envelope compared to that seen in relativistic hydrodynamic simulations of jets (Aloy et al., 2000; Zhang, Woosley, & MacFadyen, 2003; Zhang W. et al., 2004), since here the confinement is mostly magnetic. During this energetic phase of the jet, the medium plays little role in shaping the core of the jet, as has been suggested for other reasons (Lazzati et al., 2005).
The peak Lorentz factors are obtained in a narrow region within a core with half-opening angle of $`\theta 5^{}`$. Notice as well that the very inner core of the jet is denser and slower, a feature predicted in the theoretical discussion of McKinney (2005b). This feature is simply a result that the Blandford-Znajek flux is $`\mathrm{sin}^2\theta `$. Most of the Poynting energy is initially at the interface between the jet and surrounding medium, until the jet internally evolves to collimate into a narrower inner core. Notice that in McKinney (2005b), we predicted $`\mathrm{\Gamma }_{\mathrm{}}1000`$ for $`j=0.9`$, where the simulated results show a peak $`\mathrm{\Gamma }_{\mathrm{}}3\times 10^3`$ and typical $`10^2\mathrm{\Gamma }_{\mathrm{}}10^3`$. This is in reasonable agreement with analytic estimates given the jet structure was not taken into account.
Figure 11 shows the collimation angle along a mid-level field line in the Poynting-dominated jet, along the same field line as in figure 9, and for both poles of the simulation. Generally the two poles are similar but not necessarily identical. Many simulations that were performed show at least one polar jet develop some pinch instabilities. In the model shown, one side develops very strong instabilities by the end of the simulation. The development of the pinch instability mostly affects the far-radial Lorentz factor ($`\mathrm{\Gamma }`$ and $`\mathrm{\Gamma }_{\mathrm{}}`$ — and so densities). The value of $`\mathrm{\Gamma }`$ is up to 5 times larger in pinched regions than non-pinched regions, while $`\mathrm{\Gamma }_{\mathrm{}}`$ can be up to a factor 10 times larger in pinched regions compared to non-pinched regions.
In figure 11, for $`7r_gr100r_g`$ there is a region of collimation slightly faster than logarithmic. For the field line closest to the coronal wind (not shown) starting at $`\theta _j1.0`$, collimation is logarithmic with
$$\theta _j1\left(\frac{r}{2.8r_g}\right)^{1/3}.$$
(10)
Closer to the polar axis the collimation is faster. For the field line starting at $`\theta _j0.3`$, approximately
$$\theta _jr^{2/5}$$
(11)
up to $`r10^2r_g`$. The inner-radial collimation is due to confinement by the disk+corona, while in the outer-radial range the coronal outflow collimates the Poynting flow. The corona is likely required for the Poynting-dominated jet collimation, since without a disk, a monopole-like field near the black hole actually decollimates at higher black hole spin (Krasnopolsky et al., 2005). Far from the coronal outflow, the internal jet is collimated by internal hoop stresses. Note that the classical hoop-stress paradigm that jets can self-collimate is not fully tested here, but these results suggest that collimation is in large part due to, or at least requires, the coronal outflow.
Notice from figure 9 that there is little acceleration beyond $`r10^2r_g`$ while the magnetic fireball is forming. It is well known that magnetic acceleration requires field lines to collimate (Begelman & Li, 1994). Once the flow is pseudo-conical the flow is unable to transfer magnetic energy into kinetic energy and proceeds to convert it into enthalpy flux. After $`r10^3r_g`$ the flow oscillates around a pseudo-conical asymptote with a mean half-opening angle of $`\theta _j5^{}`$. The opening angle has been found to be weakly dependent on the black hole spin (McKinney & Gammie, 2004) and, as found in this paper, the details of the injection model. This resulting pseudo-conical asymptote results once the magnetic energy no longer dominates and the coronal wind no longer helps collimate the flow. This opening angle is in line with expectations built around afterglow achromatic break measurements and estimates of the opening angle based upon energy of the burst. During the evolution of the burst, the jet may continue to collimate due to a more extended coronal outflow. This would be observed as an overall hard-to-soft evolution of $`\gamma `$-rays since one is gradually exposed to the slower edges of the jet since one likely does not observe exactly emission along the core center. The additional loading of neutrons from the coronal outflow may also lead to a hard-to-soft evolution of the jet as the overall Lorentz factor decreases over the duration of the burst.
Figure 11 shows the toroidal field and pitch angle, which shows that eventually the toroidal field completely dominates the poloidal field, and the toroidal field remains ordered. Thus, large polarizations up to $`60\%`$ are possible (Granot & Königl, 2003). The inner radial toroidal field is well fit by
$$\frac{B^{\widehat{\varphi }}}{\sqrt{\rho _{0,disk}c^2}}[\mathrm{Gauss}]=0.0023\left(\frac{r}{390r_g}\right)^{0.7}.$$
(12)
The outer radial field is well fit by
$$\frac{B^{\widehat{\varphi }}}{\sqrt{\rho _{0,disk}c^2}}[\mathrm{Gauss}]=0.0023\left(\frac{r}{390r_g}\right)^{1.5}.$$
(13)
The transition radius is $`r390r_g`$. Notice that this is along one mid-level field line, and the coefficients and power laws are slightly different for each field line. The pitch angle shows that the field becomes toroidally dominated and at large radius the magnetic loops have a pitch angle of $`1^{}`$ by $`10^3r_g`$. For regions not on the axis there is no possibility of reconnection unless the flow is kink unstable (Drenkhahn, 2002; Drenkhahn & Spruit, 2002). Given the regularity of the flow field and the lack of randomized field, reconnection is not necessary or likely. Here, the pinch instability drives a nonlinear coupling such as shocks that converts the Poynting flux to enthalpy flux. This mechanism should be investigated in future work.
The toroidal field dominance leads to $`m=0`$ pinch magnetic instabilities. The $`m=1`$ kink mode may also operate, but studying the 3D jet structure is left for future work. As such oscillations develop, this leads to arbitrarily sized patches moving at arbitrary relative velocities as in the internal shock model. The typical size of a patch is $`100r_g`$ to $`10^3r_g`$ in the lab frame along the length of the jet. For long-duration GRBs lasting about 30 seconds, the number of pulses should be no larger than the ratio of the scale of the instability to the length of the event ($`10^6r_g`$), or about $`1001000`$ pulses. Different realizations of the same simulation show that the likelihood of an instability growing is random, which could lead to the large variations in the number of observed pulses and explain the diversity of observed pulses (Nakar & Piran, 2002). That is, some jets may have very few patches and so pulses. This type of instability is likely required to produce the diversity of GRBs. It is uncertain whether toroidal field instability induced variability dominates pair creation loading variability due to disk structure variability, which can be addressed by future work that self-consistently treats the neutrino emission from the disk.
Figure 11 also shows the rest-mass density per disk density, internal energy density per rest-mass density, and the ratio of (twice) the comoving magnetic energy density per unit rest-mass density. Notice that the density is close to that estimated in McKinney (2005b), so those calculations are likely reasonable approximations to the simulation results. The inner radial rest-mass density is well fit by
$$\frac{\rho _0}{\rho _{0,disk}}=1.5\times 10^9\left(\frac{r}{120r_g}\right)^{0.9}.(\mathrm{inner})$$
(14)
The outer radial rest-mass density is well fit by
$$\frac{\rho _0}{\rho _{0,disk}}=1.5\times 10^9\left(\frac{r}{120r_g}\right)^{2.2}.(\mathrm{outer})$$
(15)
The transition radius is $`r120r_g`$. Likewise, the inner radial internal energy density is moderately fit by
$$\frac{u}{\rho _{0,disk}c^2}=4.5\times 10^9\left(\frac{r}{120r_g}\right)^{1.8}.(\mathrm{inner})$$
(16)
The outer radial internal energy density is moderately fit by
$$\frac{u}{\rho _{0,disk}c^2}=4.5\times 10^9\left(\frac{r}{120r_g}\right)^{1.3}.(\mathrm{outer})$$
(17)
The transition radius is $`r120r_g`$ and is due to the presence of the thick disk at small radii. Notice that this is along one mid-level field line, and the coefficients and power laws are slightly different for each field line. One can compare this to the “envelope” density model used with
$$\frac{\rho _0}{\rho _{0,disk}}=8\times 10^8\left(\frac{r}{120r_g}\right)^{1.5}.(\mathrm{envelope})$$
(18)
The envelope has little impact on the jet structure.
Figure 12 shows the characteristic structure of the jet. The figure shows a log-log plot of one hemisphere of the time-averaged flow at late time. In such a log-log plot, $`45^{}`$ lines correspond to lines of constant $`\theta `$. Lines of constant spherical polar $`r`$ are horizontal near the $`z`$-axis and vertical near the $`R`$-axis. The field lines are shown as red lines. The disk and coronal regions have been truncated with a power-law cutoff for $`r100r_g`$ and a conical cutoff for larger radii. The inner-most blue line is the event horizon. Clearly the field lines are nearly logarithmic until $`r100r_g`$. After this point the field lines stretch out and oscillate around a conical asymptote.
Blue lines in figure 12 show the ingoing and outgoing fast surfaces. Clearly the transition to a supercritical (superfast) flow has occurred. Within $`r10r_g`$ the plotting cutoff creates the appearance that the fast surface and other lines terminate along the cutoff.
The inner magenta line is the ingoing slow surface, while the outer magenta line is the outgoing slow surface. The inner cyan line is the ingoing Alfvén surface. Energy can be extracted from the black hole if and only if the Alfvén surface lies inside the ergosphere (Takahashi et al., 1990). The ergosphere is shown by the green line, which is indeed outside the ingoing Alfvén surface. The outer cyan line is the outgoing Alfvén surface. The inner black line (next to the ergosphere) is the surface where $`B^{\widehat{\varphi }}=B^{\widehat{r}}`$ in Boyer-Lindquist coordinates. At larger radii there are two overlapping black lines. One corresponds to, again, where the toroidal field is equal to the poloidal field. The other corresponds to the light “cylinder” surface.
Other small artifacts in the plot are a result of unsteady nature of the flow. However, the despite the unsteady nature of the flow the characteristic structure is quite simple and relatively smooth in appearance. This indicates that the flow is mostly stationary. The stagnation surface is fairly unsteady, but stable.
All the results discussed in this paper are robust to increases or decreases in numerical resolution. Convergence testing has been performed by choosing different models of the $`\theta `$ grid, including different $`Q_{jet}`$ parameters that controls whether the disk or jet is more resolved at small radii. Another resolution of $`256\times 256`$ instead of $`512\times 256`$ was also chosen. No qualitative differences were found. The interpolation scheme was also changed from parabolic to linear and no qualitative differences for the far-field jet region were found.
We find that the pair injection model weakly determines the bulk structure of the flow at large radii. Other injection models were simulated by varying the total energy injected $`\dot{e}_{inj}`$, fraction of rest-mass created $`f_\rho `$, fraction of internal energy created $`f_h`$, fraction given to momentum energy $`f_m`$, and the $`\varphi `$-velocity of the injected particles. Only $`\dot{e}_{inj}`$ and $`f_\rho `$ strongly determine the bulk jet flow.
Most of these results are insensitive to the two different models of the surrounding medium. One model is a surrounding infall of material and the other is an “evacuated” exterior region. The jet structure is negligibly broader or narrower in the evacuated case due to magnetic confinement and collimation by the coronal outflow.
### 3.6. Simulation as Applied to AGN and X-ray Binaries
Notice that, quantitatively, all of these results can be simply rescaled by density in the jet in order to apply to other GRB models, AGN, and even approximately x-ray binary systems. This is because by the outer regions simulation, the jet has only reached about $`\mathrm{\Gamma }510`$ and $`\mathrm{\Gamma }_{\mathrm{}}1/\rho _{0,jet}`$.
In particular, numerical models for M87 give the same results with $`2\mathrm{\Gamma }10`$ patches by $`r10^3r_g`$. The key difference is that while the collapsar jet is very optically thick, the M87 jet is not so most of the shock-heated energy is lost to synchrotron radiation. The same structured jet is formed, and is useful to explain TeV BL Lac objects and radio galaxies (Ghisellini et al., 2005) and may help explain observations of blazars (Blandford & Levinson, 1995; Ghisellini & Madau, 1996; Chiaberge et al., 2000). The full opening angle of the jet core of $`10^{}`$ also agrees with the observations of the far-field jet in M87 (Junor et al., 1999). As discussed previously, the shocks are synchrotron cooled and the heat generated does not contribute to larger scale acceleration. While self-consistent synchrotron cooling should be included, these results suggest the shock zone (or “blazar zone” for blazars) should be quite extended between $`r10^2r_g`$ and up to about $`r10^4r_g`$.
Similarly, simulations were performed using an injection model for GRS 1915+105 giving patches with $`1.5\mathrm{\Gamma }5`$ by $`r10^2r_g`$. Due to the efficient pair-loading there is little extra Poynting flux to convert to enthalpy flux or kinetic energy flux by the time shocks occur at $`r10^2r_g`$. As discussed in McKinney (2005b), the Poynting-lepton jet in the GRS 1915+105 model is possibly destroyed by Compton drag by disk photons and synchrotron self-Compton and limited to at most $`\mathrm{\Gamma }2`$, but the jet could be completely destroyed by Compton drag.
### 3.7. Summary of Fits
A summary of the fits along a fiducial field line is given. Near the black hole the half-opening angle of the full Poynting-dominated jet is $`\theta _j1.0`$, while by $`r120r_g`$, $`\theta _j0.1`$. This can be roughly fit by
$$\theta _j\left(\frac{r}{r_g}\right)^{0.4}(\mathrm{inner})$$
(19)
for $`r<120r_g`$ and $`\theta _j0.14`$ beyond. The core of the jet follows a slightly stronger collimation with
$$\theta _jr^{2/5}$$
(20)
up to $`r<120r_g`$ and $`\theta _j0.09`$ beyond. Also, roughly for M87 and the collapsar model, the core of the jet has
$$\mathrm{\Gamma }_{bulk}\left(\frac{r}{5r_g}\right)^{0.44}(\mathrm{inner})$$
(21)
for $`5<r10^3r_g`$ and constant beyond for the M87 model if including synchrotron radiation, while the collapsar model should continue accelerating and the power law will truncate when most of the internal and Poynting energy is lost to kinetic energy and the jet becomes optically thin at about $`r10^9r_g`$ or internal shocks take the energy away. If the acceleration is purely thermal without any magnetic effect, then $`\mathrm{\Gamma }r`$ (Mészáros & Rees, 1997). However, it is not clear how the equipartition magnetic field affects the acceleration. Roughly for GRS 1915+105 the core of the jet has
$$\mathrm{\Gamma }_{bulk}\left(\frac{r}{5r_g}\right)^{0.14}(\mathrm{inner})$$
(22)
for $`5<r10^3r_g`$ and constant beyond, with no account for Compton drag or pair annihilation. Also, for any jet system the base of the jet has $`\rho _0r^{0.9}(\mathrm{inner})`$ for $`r120r_g`$ and $`\rho _0r^{2.2}(\mathrm{outer})`$ beyond. For the collapsar and M87 models
$$\frac{\rho _0}{\rho _{0,disk}}1.5\times 10^9\left(\frac{r}{120r_g}\right)^{0.9}(\mathrm{inner})$$
(23)
and
$$\frac{\rho _0}{\rho _{0,disk}}1.5\times 10^9\left(\frac{r}{120r_g}\right)^{2.2}(\mathrm{outer}),$$
(24)
while for GRS 1915+105 the inner-radial coefficient is $`10^5`$ and outer is $`6\times 10^3`$. For the collapsar model, the inner radial internal energy density is moderately fit by
$$\frac{u}{\rho _{0,disk}c^2}=4.5\times 10^9\left(\frac{r}{120r_g}\right)^{1.8}(\mathrm{inner}).$$
(25)
The outer radial internal energy density is moderately fit by
$$\frac{u}{\rho _{0,disk}c^2}=4.5\times 10^9\left(\frac{r}{120r_g}\right)^{1.3}(\mathrm{outer}).$$
(26)
The transition radius is $`r120r_g`$. For M87 the internal energy is near the rest-mass density times $`c^2`$ until $`r120r_g`$ when the dependence is as for the collapsar case. For GRS1915+105 the internal energy is near the rest-mass density times $`c^2`$ until $`r120r_g`$ and then rises to about $`2.5`$ times the rest-mass density times $`c^2`$. The inner radial toroidal lab field is well fit by
$$\frac{B^{\widehat{\varphi }}}{\sqrt{\rho _{0,disk}c^2}}[\mathrm{Gauss}]=0.0023\left(\frac{r}{390r_g}\right)^{0.7}(\mathrm{inner})$$
(27)
for $`5<r<390r_g`$. The outer radial toroidal lab field is well fit by
$$\frac{B^{\widehat{\varphi }}}{\sqrt{\rho _{0,disk}c^2}}[\mathrm{Gauss}]=0.0023\left(\frac{r}{390r_g}\right)^{1.5}(\mathrm{outer})$$
(28)
for $`r>390r_g`$.
For the typical jet with no atypical pinch instabilities, the energy and velocity structure of the jet follow
$$ϵ(\theta )=ϵ_0e^{\theta ^2/2\theta _0^2},$$
(29)
where $`ϵ_00.18`$ and $`\theta _08^{}`$. The total luminosity per pole is $`L_j0.023\dot{M}_0c^2`$, where $`10\%`$ of that is in the “core” peak Lorentz factor region of the jet within a half-opening angle of $`5^{}`$. Also, $`\mathrm{\Gamma }_{\mathrm{}}`$ is approximately Gaussian
$$\mathrm{\Gamma }_{\mathrm{}}(\theta )=\mathrm{\Gamma }_{\mathrm{},0}e^{\theta ^2/2\theta _0^2},$$
(30)
where $`\mathrm{\Gamma }_{\mathrm{},0}3\times 10^3`$ and $`\theta _04.3^{}`$. Also, $`\mathrm{\Gamma }`$ is approximately Gaussian
$$\mathrm{\Gamma }(\theta )=\mathrm{\Gamma }_0e^{\theta ^2/2\theta _0^2},$$
(31)
where $`\mathrm{\Gamma }_05`$ and $`\theta _011^{}`$. The outer sheath’s ($`\theta 0.2`$) seed photon temperature as a function of radius is
$$T_{\gamma ,seed}50\mathrm{k}\mathrm{e}\mathrm{V}\left(\frac{r}{5\times 10^3r_g}\right)^{1/3}.$$
(32)
## 4. Discussion
A collapsar-type GRMHD simulation with a neutrino annihilation model and Fick diffusion model has been studied. A self-consistent Poynting-dominated jet is produced and the Lorentz factor at large distances is $`\mathrm{\Gamma }_{\mathrm{}}10^3`$ in the core of the jet, but an equally important structured component exists with $`\mathrm{\Gamma }_{\mathrm{}}10^2`$. The Lorentz factor of the jet is determined by the electron-proton loading by Fick diffusion of neutrons. Notice that estimates of the baryonic mass-loading from the optical flash of GRB990123 suggest a baryon loading that gives $`\mathrm{\Gamma }10^3`$ (Soderberg & Ramirez-Ruiz, 2003), which is in basic agreement with the findings here. The half opening angle of the core of the jet is $`\theta _j5^{}`$, while there exists a significant structured component with a half opening angle of $`\theta _j25^{}`$. It is likely that this jet component survives traversing the remaining portion of the stellar envelope ($`r10^5r_g`$, another decade in radius).
The jet at large distances is unstable to, at least, an $`m=0`$ pinch instability due to the dominance of the toroidal field to the poloidal field (see, e.g. Begelman 1998). Here it is found that the energy of the jet is patchy. Typical realizations of the same model have a random number of patches with $`100\mathrm{\Gamma }_{\mathrm{}}10^3`$. This likely results in a pulsed sequence of magnetic fireballs that move at large relative Lorentz factors, as required by the internal shock model (Mészáros, 2002; Ghirlanda et al., 2003; Piran, 2005). The acceleration region and internal shock region was not simulated since the dynamical range required is another $`6`$ orders of magnitude in radius.
Some models assume the Poynting energy to be converted into radiation far from the collapsing star by internal dissipation (see, e.g., Thompson 1994; Mészáros & Rees 1997; Spruit, Daigne, & Drenkhahn 2001; Drenkhahn 2002; Drenkhahn & Spruit 2002; Sikora et al. 2003; Lyutikov, Pariev, & Blandford 2003) such as magnetic reconnection. However, in our model, significant internal dissipation occurs already by $`10^3r_g`$.
A very efficient model that reduces the compactness problem invokes Compton drag (bulk Comptonization) to generate the emission, where the stellar envelope or presupernova region provides soft photons (Ghisellini et al., 2000; Lazzati et al., 2004). Our numerical model shows a jet structure with a slow cold “sheath” necessary for supplying the seed photons, so this process might explain the GRB emission process.
As in McKinney & Gammie (2004), a mildly relativistic Poynting-baryon jet is launched from the inner edge of the disk with a half-opening angle of roughly $`16^{}`$ to $`45^{}`$, however long-time study of this jet component depends on sustaining long-time disk turbulence, which cannot be simulated in axisymmetry. For GRBs, this hot baryon-loaded jet might play a significant role in the subsequent supernova (by producing, e.g., $`{}_{}{}^{56}\mathrm{Ni}`$) and be an interesting site for the r-process (Kohri et al., 2005). The mass ejection rate is found to be $`\dot{M}_{0,coronal}0.03\dot{M}_0`$, which for a $`30`$ second event gives $`0.1\mathrm{M}_{}`$ of baryonic mass. This may be sufficient to power supernovae. The Lorentz factor is about $`\mathrm{\Gamma }1.53`$, such as observed in a component of SN1998bw (Kulkarni et al., 1998).
For GRBs, the picture that emerges is that Poynting flux is emitted from the black hole at $`rr_g`$ and is loaded with electron-positron pairs within $`r20r_g`$. A similar mass-fraction of neutrons Fick-diffuses into the jet, and they dominate the rest-mass once the temperature decreases to $`T6\times 10^9`$K by $`r10r_g`$. Magnetic acceleration occurs over $`r10^2r_g`$ up to $`\mathrm{\Gamma }10`$. Notice that this is in contrast to Mészáros & Rees (1997) who assumed the initial bulk Lorentz factor is $`\mathrm{\Gamma }[r=r_H]\mathrm{\Gamma }_{\mathrm{}}^{(EM)}`$, which assumed instantaneous magnetic acceleration.
Once the toroidal field dominates the poloidal field, magnetic instabilities develops around $`r1010^2r_g`$ which turns the jet into an equipartition “magnetic fireball” by $`r10^3r_g`$. The spatial structure of the magnetic fireball energy flux is patchy on the instability scale of $`r10^2r_g`$. Simulations that demonstrate how these features would later interact require much more radial dynamic range. Between $`10^4r_gr10^8r_g`$ the flow accelerates due to adiabatic expansion with a radial dependence up to $`\mathrm{\Gamma }r`$ (Mészáros & Rees, 1997), where the Poynting flux provides a reservoir of magnetic energy that is continuously shock-converted into thermal energy.
After passing the stellar surface at $`r10^5r_g`$, the plasma becomes transparent to $`\gamma `$-rays at $`10^6r_gr10^7r_g`$, where possibly Compton drag of the sheath seed photons produces the GRB emission (Ghisellini et al., 2000; Lazzati et al., 2004). By $`r10^7`$, patches of varying $`\mathrm{\Gamma }1001000`$ have reached their terminal velocity. Between $`10^8r_gr10^{10}r_g`$ the fireball is optically thin to Compton scattering, radiation escapes, and pairs no longer annihilate but can continue to be magnetically accelerated (Mészáros & Rees, 1997). Within this same radial range, internal shocks proceed to convert the remaining kinetic energy to nonthermal $`\gamma `$-rays. Beyond $`10^{10}r_g`$ a reverse shock may occur and external shocks with the interstellar medium (ISM) occur.
Simulations of AGN were performed, which show similar $`\mathrm{\Gamma }`$ vs. radius as the collapsar case. That is, there is an early transfer of Poynting flux to kinetic energy flux leading up to about $`\mathrm{\Gamma }510`$ by about $`r10^3r_g`$ for an M87 model. This is consistent with the lack of observed Comptonization features in blazars (see, e.g., Sikora et al. 2005), although this is also consistent with the fact that the optical depth to Comptonization is low in such systems.
While prior work assuming $`\mathrm{\Gamma }_{bulk}10`$ and $`\theta _j15^{}`$ found that the BZ power is insufficient to account for Blazar emission (Maraschi & Tavecchio, 2003), as shown here the structure of the Poynting-dominated jet is nontrivial. The region with $`\mathrm{\Gamma }_{bulk}10`$ is narrower with $`\theta _j5^{}`$ and jet emission is likely dominated by shock accelerated electrons with $`\mathrm{\Gamma }_e100`$ with an extended high energy tail. This lowers the necessary energy budget of the jet to be consistent with BZ power driving the jet.
### 4.1. Theoretical Contemporaneous Comparisons
Theoretical studies of ideal MHD jets focused on cold jets and the conversion of Poynting energy directly to kinetic energy (see, e.g., Li et al. 1992; Begelman & Li 1994; Daigne & Drenkhahn 2002). Indeed, much of the work has focused on self-similar models, of which the only self-consistent solution found is suggested to be R-self-similar models (Vlahakis, 2004). Unfortunately, such models result in only cylindrical asymptotic solutions, which is apparently not what is observed in AGN jets nor present in the simulations discussed in this paper. The previous section showed that some aspects of the jet are nearly self-similar and that there is some classic ideal-MHD acceleration occurring in this region. However, once the toroidal field dominates the poloidal field, the Poynting flux is converted into internal energy until they reach equipartition. The conversion is due to a time-dependent nonlinear coupling, such as shocks, due to the magnetic pinch instability. Cold ideal-MHD jet models would have no way of addressing this. This region also involves relatively rapid variations in the flow, so stationary jet models would have difficulty modelling this region.
Notice that GRMHD numerical models show that the magnetic field corresponding to the Blandford-Znajek effect dominates all other magnetic field geometries (Hirose et al., 2004; McKinney, 2005a). For example, there are no dynamically stable or important field lines that tie the black hole to the outer accretion disk as in Uzdensky (2005). All black hole field lines tie to large distances or tie to the horizon-crossing accretion disk. Also, there are no field lines that connect the inner-radial accretion disk and large distances. The Blandford-Znajek associated magnetic field is completely dominant, as shown in, e.g., figure 3 and figure 4.
We have not directly included those effects of neutron diffusion that induce a jet structure (Levinson & Eichler, 2003). However, there is probably strong mixing within the jet and the large-scale jet is probably not affected by the details of the interface where neutron-diffusion occurs. Alternatively, the Gaussian structure we find may dominate the power law structure they find since the Gaussian structure is due to a strong internal electromagnetic evolution of the jet and matter plays little role in setting the jet structure.
### 4.2. Numerical Contemporaneous Comparisons
MacFadyen & Woosley (1999) evolve a presupernova core with a black hole inserted to replace part of the core and evolve the nonrelativistic viscous hydrodynamic equations of motion for various values of viscosity coefficient ($`\alpha `$), mass accretion rate, nuclear burning, stellar angular momentum, and some models have an injected energy at the poles at some fixed energy rate. For models in which they inject energy at some specified rate, they find the baryon-contamination problem is somewhat alleviated by forming a hot bubble. As shown in their figure 28, they find the jet has $`u/(\rho _0c^2)10`$ and they suggest that this corresponds to $`\mathrm{\Gamma }_{\mathrm{}}10`$, insufficient to avoid the compactness problem. Their $`\alpha =0.1`$ model shows a significant disk outflow, which over the duration of the GRB would yield $`M\mathrm{M}_{}`$ in mass that could power a supernova. In their $`\alpha =0.01`$ model there are insignificant outflows.
Our MHD results are most comparable to their $`\alpha =0.01`$ model. We find a disk outflow yielding $`M0.1\mathrm{M}_{}`$, which may still be sufficient to produce a supernova component. We have self-consistently evolved the jet formation and included an approximate model for the pair creation physics. In our case the baryon-contamination problem is self-consistently avoided by magnetic confinement of the jet against baryons, where only a small number of neutrons Fick-diffuse into the jet and self-consistently determine the Lorentz factor at large distances. We find that neutrino annihilation energy is probably dominated by energy from the black hole. Notice that we find $`\mathrm{\Gamma }_{\mathrm{}}1001000`$, which is much larger than they find. The difference between their and our model is the presence of a magnetic field and a rotating black hole, which drive the BZ-effect and a stronger evacuation of the polar jet region.
Proga (2005) have performed nonrelativistic MHD simulations of collapsars with a realistic equation of state. They suggest MHD accretion is able to launch, collimate, and sustain a strong Poynting outflow, although they measure a jet velocity of $`v0.2c`$. They find the jet is Poynting flux dominated such that $`\mathrm{\Gamma }_{\mathrm{}}10`$, insufficient to avoid the compactness problem. The rotation of the black hole is crucial to generate a sufficiently Poynting-dominated jet. Similar fully relativistic simulations by Mizuno et al. (2004) have performed with a jet velocity of only $`v0.3c`$, which is due to their choice of the “floor” model in the polar regions and the short time of integration.
Zhang, Woosley, & MacFadyen (2003); Zhang W. et al. (2004) use a relativistic hydrodynamic model to simulate jet propagation through the stellar envelope and subsequent breakout through the stellar surface. They assume a highly relativistic jet is formed early near the black hole and they inject the matter with a large enthalpy per baryon of about $`u/\rho _0c^2150`$. They also tune the injected energy to compare with observations (Frail et al., 2001). In their view, variability is due to hydrodynamic instabilities between the cocoon and core of the jet. They find the stellar envelope can collimate the flow.
We have self-consistently evolved the formation of the jet along with the disk without having to assume a jet structure. We suggest that the injected energy per baryon is only $`u/\rho _0c^220`$ unless super-efficient neutrino mechanisms are invoked. We also find that the self-consistent energy released is larger than observed, suggesting an fairly inefficient generation of $`\gamma `$-rays and a dominant lower $`\mathrm{\Gamma }`$ jet component that may explain x-ray flashes. We suggest magnetic instabilities due to pinch or kink modes dominate the variability rather than hydrodynamic instabilities, which are known to be quenched by magnetic confinement. We find that the jet structure is negligibly broader for models with no stellar envelope due to magnetic confinement and we find that the outer portion of the Poynting-dominated jet is collimated by the coronal outflow.
Many hydrodynamic jet propagation simulations have been studied (for a review see Scheck et al. 2002). Scheck et al. (2002) evolve hydrodynamic relativistic hadronic and leptonic jets using a realistic equation of state. They find that the resulting morphology of the jets are similar, despite the different composition. This suggests jet gross morphology is not a useful tool to differentiate jet composition. In our model, the most relativistic jet component is lepton-dominated in AGN and x-ray binaries and baryon-dominated in GRB systems. Notice that magnetic confinement quenches most hydrodynamic instabilities.
De Villiers et al. (2005b) evolve a fully relativistic, black hole mass-invariant model, showing a jet with hot blobs moving with $`\mathrm{\Gamma }50`$. A primary result in agreement with the results here is that a patchy or pulsed “magnetic fireball” is produced. This is gratifying since it suggests that the development of a “magnetic fireball” is not an artifact of the numerical implementation but is a likely result of shock heating. We also agree in finding that the core of the jet is hot and fast and is surrounded by a cold slow flow. One difference is that they say they seem to find the flow is cylindrically collimated by $`r300r_g`$, while we find nearly logarithmic collimation until $`r10^2r_g`$ and an oscillatory conical asymptote beyond. Another difference is that they suggest temporal variability is due to injection events near the black hole, while we suggest it is due to pinch (or perhaps kink) instabilities at $`r10^2r_g`$.
Another difference is that they find a larger value of $`\mathrm{\Gamma }`$ at smaller radius than we find. Indeed, one should wonder why their black hole mass-invariant model applies to only GRBs and not AGN or x-ray binaries. This is a result of their much lower “floor” density of $`\rho _010^{12}\rho _{0,disk}`$, which is not consistent with self-consistent pair-loading or neutron diffusion loading. Also, because they use a constant density floor, the jet region at large radius must be additionally loaded with an arbitrary amount of rest-mass. The typical “floor” model adds rest-mass into the comoving frame, but this artificially loads the jet with extra mass moving at high $`\mathrm{\Gamma }`$. We suggest that rather than the acceleration occurring within $`r700r_g`$ of their simulation, that acceleration occurs much farther away before the emitting region at $`r10^9r_g`$. In their $`\mathrm{\Gamma }50`$ knots the gas has $`10^3u/\rho _0c^210^6`$. This implies that their actual terminal Lorentz factor is on the order of $`10^5\mathrm{\Gamma }_{\mathrm{}}10^7`$, which would imply that the external shocks occur before internal shocks could occur.
Komissarov (2005) study the BZ process and MHD Penrose process. They found that the prior “jet” results of (Koide, Shibata, Kudoh, & Meier, 2002), who evolve only for $`t100t_g`$, are only a transient phenomena associated with the initial conditions. They also suggest that there are serious problems for the MHD-driven model of jets since they do not find jets in their models except for contrived field geometries. However, this is potentially due to three effects. First, their outer radius is $`r50r_g`$, while here we study out to $`r10^4r_g`$. The magnetic acceleration only leads to a logarithmic increase in the velocity, so a large radial range is required to observe relativistic motion. Second, magnetic acceleration requires collimating field lines (Begelman & Li, 1994), and a disk or disk wind is probably required to collimate the magnetic outflow (Okamoto, 1999, 2000). The “disk” that forms in their models is relatively thin. Third, to avoid numerical errors, they limit their solution to $`b^2/(\rho _0c^2)100`$. Here we find that a self-consistent source of jet matter from pair creation and neutron diffusion leads to $`b^2/(\rho _0c^2)10^5`$ near the black hole. These three effects are probably why they find no relativistic jets.
### 4.3. Limitations
As has been pointed out by Komissarov (2005), nonconservative GRMHD schemes often overestimate the amount of thermal energy produced. All GRMHD numerical models suffer from some numerical error. Shock-conversion of magnetic energy to thermal energy is modelled by HARM in the perfect magnetic fluid approximation with total energy conserved exactly. However, our numerical model may overestimate the amount of magnetic dissipation in shocks, and so the shocks may more slowly convert magnetic energy to thermal energy. If this dissipation was delayed until the $`\gamma `$-ray photosphere at $`r10^710^8r_g`$, then those shocks could be directly responsible for the $`\gamma `$-ray emission from GRBs. Clarification of this issue is left for future work. However, we expect that toroidal field instabilities drive efficient magnetic dissipation as shown in the numerical results presented here.
In order to evolve for a longer time than simulated in this paper, other physics must be included. For long-term evolution of a GRB model, one must include disk neutrino cooling, photodisintegration of nuclei, and a realistic equation of state. If one wishes to track nuclear species evolution, a nuclear burning reactions network is required. For the neutrino optically thick region of the disk, radiative transport should be included. The self-gravity of the star should be included to evolve the core-collapse. This includes a numerical relativity study of the collapse of a rotating magnetized massive star into a black hole (see review by Stergioulas 2003, §4.3). The jet should be followed through the entire star and beyond penetration of the stellar surface (Aloy et al., 2000; Zhang, Woosley, & MacFadyen, 2003; Zhang W. et al., 2004).
As applied to all black hole accretion systems, some other limitations of the numerical models presented include the assumption of axisymmetry, ideal MHD, and a nonradiative gas. The assumption of axisymmetry is likely not important for the inner jet region since our earlier results (McKinney & Gammie, 2004) find quantitative agreement with 3D results (De Villiers, Hawley, & Krolik, 2003). The primary observed limitation of axisymmetry appears to be the decay of turbulence (Cowling, 1934), which we attempt to avoid by requiring a resolution that gives quasi-steady turbulence for much of the simulation. Also, the jet at large distances has already formed by the time turbulence decays, and by that time the jet at large radius is not in causal contact with the disk.
When the toroidal field dominates the poloidal field, eventually $`m=1`$ kink instabilities and higher modes may appear (see, e.g., Nakamura et al. 2001; Nakamura & Meier 2004). Thus our models may underestimate the amount of oscillation in the flow and the conversion of Poynting flux to enthalpy flux.
The limitation of axisymmetry may also limit the efficiency of the Blandford-Znajek process. The magnetic arrested disk (MAD) model suggests that any accretion flow likely accumulates a large amount of magnetic flux near the black hole. This means the efficiency of extracting energy would be higher (Igumenshchev et al., 2003; Narayan et al., 2003).
We have neglected low-energy jet and disk radiative processes in the numerical simulations. Future work should include a model of synchrotron radiation for an AGN model in order to simulate the “jet” emission to verify the claims made in McKinney (2005b) that the broad “jet” emission observed by Junor et al. (1999) is actually the coronal outflow rather than the Poynting-dominated jet.
Also for AGN, a self-consistent simulation with synchrotron emission would likely show the continuous loss of Poynting flux until the synchrotron cooling timescale is longer than the jet propagation timescale, which still suggests the jet Poynting flux is finally in equipartition with the enthalpy flux. This would suggest that the shock-zone, and so emission region, is more extended ($`r10^210^4r_g`$) than the simulated shock-zone ($`r10^210^3r_g`$).
For AGN and x-ray binaries, the radiatively inefficient disk approximation, which assumes electrons couple weakly to ions, may not hold. If the electrons and ions eventually couple near the black hole, then the disk might collapse into an unstable magnetically dominated accretion disk (MDAF) (Meier, 2005). Like the MAD model, this might drastically alter the results here, although it is uncertain whether jets are actually produced under the conditions specified by the MDAF model.
The single-fluid, ideal MHD approximation breaks down under various conditions, such as during the quiescent output of x-ray black hole binaries, where a two-temperature plasma likely forms near the black hole as ions and electrons decouple (see, e.g., Narayan & Yi 1995). Resistivity plays a role in current sheets where reconnection events may generate flares as on the sun, such as possibly observed in Sgr A (Genzel et al., 2003). Finally, radiative effects may introduce dynamically important instabilities in the accretion disk (e.g. Gammie 1998; Blaes & Socrates 2003).
## 5. Conclusions
Primarily two types of relativistic jets form in black hole (and perhaps neutron star) systems. The Poynting-dominated jet region is composed of field lines that connect the rotating black hole to large distances. Since the ideal MHD approximation holds very well, the only matter that can cross the field lines are neutral particles, such as neutrinos, photons, and free neutrons.
In McKinney (2005b), we showed that the primary differences between GRBs, AGN, and black hole x-ray binaries is the pair-loading of the Poynting-dominated jet, a similar mass-loading by free neutrons in GRB-type systems, the optical depth of the jet, and the synchrotron cooling timescale of the jet.
In McKinney (2005b), we showed that for GRB-type systems the neutron diffusion flux is sufficiently large to be dynamically important, but small enough to allow $`\mathrm{\Gamma }1001000`$. Beyond $`r10r_g`$ many of the electron-positron pairs annihilate, so the Poynting-dominated jet is dominated in mass by electron-proton pairs from collision-induced neutron decay. Most of the energy is provided by the BZ effect instead of neutrino-annihilation.
In McKinney (2005b), we showed that for AGN and x-ray binaries, the density of electron-positron pairs established near the black hole primarily determines the Lorentz factor at large distances. Radiatively inefficient AGN, such as M87, achieve $`2\mathrm{\Gamma }_{\mathrm{}}10`$ and are synchrotron cooling limited. The lower the $`\gamma `$-ray radiative efficiency of the disk, the more energy per particle is available in the shock-zone. Radiatively efficient systems such as GRS1915+105 likely have no Poynting-lepton jet due to strong pair-loading and destruction by Comptonization by the plentiful soft photons for x-ray binaries with optically thick jets. However, all these systems have a mildly relativistic, baryon-loaded jet when in the hard-low state when the disk is geometrically thick, which can explain jets in most x-ray binary systems.
A GRMHD code, HARM, with pair creation physics was used to evolve many black hole accretion disk models. The basic theoretical predictions made in McKinney (2005b) that determine the Lorentz factor of the jet were numerically confirmed. However, Poynting flux is not necessarily directly converted into kinetic energy, but rather Poynting flux is first converted into enthalpy flux into a “magnetic fireball” due to shock heating. Thus, at large distances the acceleration is primarily thermal, but most of that thermal energy is provided by shock-conversion of magnetic energy. In GRB systems this magnetic fireball leads to thermal acceleration over an extended radial range. The jets in AGN and x-ray binaries release this energy as synchrotron and inverse Compton emission and so the jet undergoes negligible thermal acceleration beyond $`r10^210^3r_g`$.
Based upon prior theoretical (McKinney, 2005b) and this numerical work, basic conclusions for collapsars include:
1. Black hole energy, not neutrino energy, typically powers GRBs.
2. Poynting-dominated jets are mostly loaded by $`e^{}e^+`$ pairs close to the black hole, and by $`e^{}p`$ pairs for $`r10r_g`$.
3. BZ-power and neutron diffusion primarily determines Lorentz factor.
4. Variability is due to toroidal field instabilities.
5. Poynting flux is converted into enthalpy flux and leads to the formation of a “magnetic fireball.”
6. Patchy jet develops $`10^2\mathrm{\Gamma }_{\mathrm{}}10^3`$, as required by internal shock model.
7. Random number of patches ($`<1000`$ for 30 second burst) and so random number of pulses.
8. Energy structure of jet is Gaussian with $`\theta _08^{}`$.
9. Core of jet with $`\theta _j5^{}`$ can explain GRBs.
10. Extended slower jet component with $`\theta _j25^{}`$ can explain x-ray flashes.
11. Coronal outflows with $`\mathrm{\Gamma }1.5`$ may power supernovae (by producing, e.g., $`{}_{}{}^{56}\mathrm{Ni}`$) with $`M0.1\mathrm{M}_{}`$ processed by corona.
Based upon prior theoretical (McKinney, 2005b) and this numerical work, basic conclusions for AGN or x-ray binaries include:
1. Poynting-dominated jets $`e^{}e^+`$ pair-loaded unless advect complicated field.
2. $`\gamma `$-ray radiative efficiency, and so pair-loading, determines maximum possible Lorentz factor.
3. Poynting-lepton jet is collimated with $`\theta _j5^{}`$.
4. Extended slow jet component with $`\theta _j25^{}`$.
5. For fixed accretion rate, variability is due to toroidal field instabilities.
6. Poynting flux is shock-converted into enthalpy flux.
7. In some AGN, shock heat in transonic transition lost to synchrotron emission and limits achievable Lorentz factor to $`2\mathrm{\Gamma }10`$ (e.g. in M87).
8. Coronal outflows produce broad inner-radial jet features in AGN together with well-collimated jet component (e.g. in M87).
9. In some x-ray binaries, Compton drag loads Poynting-lepton jets and limits Poynting-lepton jet to $`\mathrm{\Gamma }2`$ or jet destroyed.
10. In some x-ray binaries, Poynting-lepton jet optically thick and emits self-absorbed synchrotron.
11. Coronal outflows have collimated edge with $`\mathrm{\Gamma }1.5`$.
12. Coronal outflows may explain all mildly relativistic and nonrelativistic jets in radiatively efficient systems (most x-ray binaries).
For AGN and X-ray binaries, the coronal outflow collimation angle is strongly determined by the disk thickness. The above conclusions regarding the collimation angle assumed $`H/R0.2`$ near the black hole and $`H/R0.6`$ far from the black hole, while $`H/R0.9`$ (ADAF-like) is perhaps more appropriate for some systems. The sensitivity of these results to $`H/R`$ is left for future work.
## Acknowledgments
I thank Avery Broderick for an uncountable number of inspiring conversations. I also thank Charles Gammie, Brian Punsly, Amir Levinson, and Ramesh Narayan, with whom each I have had inspiring conversations. I thank Scott Noble for providing his highly efficient and accurate primitive variable solver. I thank Xiaoyue Guan for providing her implementation of parabolic interpolation. This research was supported by NASA-ATP grant NAG-10780 and an ITC fellowship.
## Appendix A GRMHD Equations of Motion with Pair Creation
A single-component GRMHD approximation that accounts for baryon conservation is summarized. Leptons are assumed to be conserved in the equation of state (EOS) and by accounting for free-streaming neutrinos. We assume an ideal gas EOS of relativistic particles and neglect radiative transport since we assume all particles either stream freely or are trapped in the fluid. Alternatively, we assume the initial conditions well-model the steady-state radiative equilibrium. We model the streaming photon/neutrino-annihilation into pairs by injecting rest-mass and energy-momentum at an appropriate fraction of the baryon density.
The black hole has a Kerr metric written in Kerr-Schild coordinates, where the Kerr metric in Kerr-Schild coordinates and the Jacobian transformation to Boyer-Lindquist coordinates is given in McKinney & Gammie (2004). We use Kerr-Schild rather than Boyer-Lindquist because in Kerr-Schild coordinates the inner-radial boundary can be placed inside the horizon and so out of causal contact with the flow. It is difficult to avoid interactions between the numerical inner boundary and the jet when using Boyer-Lindquist coordinates. This interaction leads to excessive variability in the jet since the ingoing superfast transition is not on the grid and then the details of the boundary condition significantly impact the jet. Numerical models of viscous flows have historically had similar issues (see, e.g., McKinney & Gammie 2002).
A single-component MHD approximation is assumed such that baryon number is conserved up to a source term due to pair creation due to either radiative annihilation or neutron diffusion, such that
$$(\rho _0u^\mu )_{;\mu }=S_\rho ,$$
(A1)
where $`\rho _0m_bn_b`$, $`m_bm_n`$ the neutron mass, $`n_b=n_n+n_p`$, and $`m_p`$ is the proton mass. One may choose an arbitrary mass weight to define $`\rho _0`$ in the single component fluid approximation.
For a magnetized plasma the conservation of energy-momentum equations are
$$T_{}^{\mu \nu }{}_{;\nu }{}^{}=\left(T_{\mathrm{MA}}^{\mu \nu }+T_{\mathrm{EM}}^{\mu \nu }\right)_{;\nu }=S_T^\mu .$$
(A2)
where $`T^{\mu \nu }`$ is the stress-energy tensor, which can be split into a matter (MA) and electromagnetic (EM) part. In the fluid approximation
$$T_{\mathrm{MA}}^{\mu \nu }=(\rho _0+u_g)u^\mu u^\nu +p_gP^{\mu \nu },$$
(A3)
with a relativistic ideal gas pressure $`p_g=(\gamma 1)u_g`$, $`\gamma =4/3`$, and $`P^{\mu \nu }=g^{\mu \nu }+u^\mu u^\nu `$ is the projection tensor. Either $`\gamma =5/3`$ or $`\gamma =4/3`$ for either the baryon-dominated component or lepton-dominated component does not change the results described in section 3.
### A.1. Injection Physics
A detailed pair creation model would self-consistently determine the distribution of rest-mass and energy-momentum injected, which is left for future work. In this paper, the rough results of Popham et al. (1999); MacFadyen & Woosley (1999) are used to approximate the neutrino annihilation and pair creation. They determine the energy density creation rate in pairs as measured by an observer at infinity ($`\dot{e}_{inj}`$). A monoenergetic, monomomentum injection of rest-mass and energy-momentum is assumed. That is, for a Lorentz invariant particle distribution function of
$$f=\frac{dN}{dx^{\widehat{1}}dx^{\widehat{2}}dx^{\widehat{3}}du^{\widehat{1}}du^{\widehat{2}}du^{\widehat{3}}},$$
(A4)
the fluid equations are derived from the Boltzmann equation
$$\frac{df}{d\tau }=\frac{dn_{inj}}{d\tau }\delta [u^\mu u_{inj}^\mu ]$$
(A5)
for a particle creation density rate $`dn_{inj}/d\tau `$ in the comoving frame of the existing fluid. There is no collisional term in the ideal case. After mass-weighting and taking 4-velocity moments of the Boltzmann equation one has $`S_\rho =(\rho _{0,inj}u_{inj}^\mu )_{;\mu }`$, where $`\rho _{0,inj}`$ is the injected rest-mass density and $`u_{inj}^\mu `$ is the 4-velocity of the injected particles. Also, $`S_T^\mu =T_{inj}^{\mu \nu }{}_{;\nu }{}^{}`$. Here,
$$\frac{dn_{inj}}{d\tau }\frac{m_e}{m_b}\frac{dn_{e^{}e^+,inj}}{d\tau }=\frac{1}{m_b}\frac{d\rho _{0,e^{}e^+,inj}}{d\tau },$$
(A6)
so the fluid is still treated as a single component with a single mass weight $`m_b`$. As discussed in McKinney (2005b), this is a good approximation since the late-time Poynting-dominated jet region is dominated by lepton mass while the remaining region is dominated by baryon mass. The simultaneous advected flux of injected energy is neglected, so then $`S_{T}^{}{}_{t}{}^{}=\sqrt{g}\dot{e}_{inj}`$ is valid in Boyer-Lindquist, Kerr-Schild, or modified Kerr-Schild coordinates. This two-step approach (similar to operator splitting), where particles are injected and then advected with the normal fluid, is a good approximation for typical numerical integrations that use a multi-step timestep approach. This is also a good approximation because the injection region has a small 4-velocity beyond the stagnation surface. Also, the error in the injection approximations being made is larger than the error in neglecting the advection term. Thus
$$S_\rho =/t(\sqrt{g}\rho _{0,inj}u_{inj}^t),$$
(A7)
and
$$S_{T}^{}{}_{\mu }{}^{}=/t\left(\sqrt{g}\left(q_{0,inj}u_{inj}^tu_\mu ^{inj}+p_{inj}\delta _\mu ^t\right)\right).$$
(A8)
where $`q_{0,inj}\rho _{0,inj}+u_{inj}+p_{inj}`$. This represents the source of energy-momentum, such as photon and neutrino losses emitted and absorbed and the pairs created by annihilation of photons or neutrinos. This accounts for, in the guise of the MHD approximation, “non-local” transport of energy and momentum. The injected particles are a relativistic gas with $`\gamma =4/3`$ such that $`p_{inj}=(\gamma 1)u_{inj}`$.
Two approximate injection methods are used to treat the injected momentum. One assumes that the momentum injected is mostly aligned with the axis with $`u_{inj}^\theta =0`$ and either $`v_{inj}^\varphi =0`$ or $`v_{inj}^\varphi =\mathrm{\Omega }[R=r_{stag}]`$ for some typical origin of the neutrinos $`r_{stag}`$. For the other, the particles are injected in the comoving frame of the existing fluid (i.e. $`u_{inj}^\mu =u^\mu `$). It turns out that these different approaches lead to qualitatively similar results for the solution beyond the region where injection is important ($`r6r_L`$ in the jet). The total energy density injected ($`\dot{e}_{inj}`$ as defined by an observer at infinity) is partitioned between rest-mass, enthalpy, and momentum energy. The fraction of rest-mass injected is defined as $`f_\rho `$ and so
$$/t(\sqrt{g}\rho _{0,inj}u_{inj}^t)=f_\rho \sqrt{g}\dot{e}_{inj}.$$
(A9)
The fraction of internal energy related injected energy is defined as $`f_h`$, and so
$$/t(\sqrt{g}((u_{inj}+p_{inj})u_{inj}^tu_t^{inj}+p_{inj}))=f_h\sqrt{g}\dot{e}_{inj}.$$
(A10)
The fraction of rest-mass plus momentum energy injected is defined as $`f_\rho +f_m`$, and so
$$/t(\sqrt{g}\rho _{0,inj}u_{inj}^tu_t^{inj})=(f_\rho +f_m)\sqrt{g}\dot{e}_{inj}.$$
(A11)
Here $`f_\rho +f_h+f_m=1`$.
McKinney (2005b) defines $`\dot{e}_{inj}`$, which along with the injection fractions and $`u^\mu u_\mu =1`$ completely define the injection process by allowing one to solve for $`\rho _{inj}`$, $`u_{inj}`$, $`u_{inj}^t`$, and $`u_t^{inj}`$ for a fixed time interval $`dt`$. Radiative cooling effects are important for studying hundreds of dynamical times, but otherwise can be neglected if the initial model is consistent with the time-averaged radiative properties. Future work will refine the treatment of the pair creation process, considering the creation effects in the locally flat comoving frame, once a self-consistent model is available.
### A.2. Pair Plasma Annihilation
The electron-positron pair plasma that forms may annihilate itself into a fireball if the pair annihilation rate is faster than the typical rate of the jet ($`c^3/GM`$) near the black hole. Also, if the pair annihilation timescale is shorter than the dynamical time, then pair annihilation would give a collisional term in the Boltzmann equation. From the pair annihilation rate given in McKinney (2005b), one finds that $`t_{pa}GM/c^3`$ for AGN and marginally so for x-ray binaries. Thus, pairs mostly do not annihilate, and so formally the pair plasma that forms in the low-density funnel region is collisionless so that the Boltzmann equation should be solved directly. Plasma instabilities and relativistic collisionless shocks are implicitly assumed to keep the pairs in thermal equilibrium so the fluid approximation remains mostly valid, as is a good approximation for the solar wind (see, e.g. Feldman & Marsch 1997; Usmanov et al. 2000). This same approximation has to be invoked for the thick disk state in AGN and x-ray binaries, such as for the ADAF model (McKinney, 2004). For regions that pair produce slower than the jet dynamical time, each pair-filled fluid element has a temperature distribution that gives an equation of state with $`P=\rho _{0,e^{}e^+}k_bT_e/m_e`$ rather than $`P=(11/12)aT^4`$, where $`a`$ is the radiation constant. So most of the particles have a Lorentz factor of $`\mathrm{\Gamma }_eu/(\rho _{0,e^{}e^+}c^2)`$ and little of the internal energy injected is put into radiation. This also allow the use of a single-component approximation. A self-consistent Boltzmann transport solution is left for future work.
On the contrary for GRB systems, due to the relatively high density of pairs, the time scale for pair annihilation is $`t_{pa}GM/c^3`$ along the entire length of the jet. Thus a pair fireball forms and the appropriate equation of state is that of an electron-positron-radiation fireball. Thus, formally the pair fireball rest-mass density is not independent of the pair fireball internal energy density. However, because the pairs are well-coupled to the radiation until a much larger radius of $`r10^810^{10}r_L`$, the radiation provides an inertial drag on the remaining pair plasma. That is, the relativistic fluid energy-momentum equation is still accurate. So the effective rest-mass density is $`\rho _0+u`$ ($`u`$ the total internal energy of the fireball), and so the effective rest-mass is independent of the cooling of the fireball until the fireball is optically thin (see, e.g., Mészáros & Rees 1997).
For GRB systems, the mass conservation equation is reasonably accurate. Even though the electron-positron pairs annihilate, the rest-mass of pairs injected is approximately that of the pairs that are injected due to Fick-diffusion of neutrons (see appendix A of McKinney 2005b). The annihilation energy from electron-positron pairs contributes a negligible additional amount of internal energy, so can be neglected, especially compared to the Poynting energy flux that emerges from the black hole. Thus, the rest-mass can always be assumed to be due to baryons rather than the electron-positron pairs. This also suggests that the neutrino annihilation is a negligible effect if the BZ power is larger than the neutrino annihilation power.
In summary, the rest-mass evolution discussed in section 3 is accurate for GRB, AGN, and marginally so for x-ray binaries. This is despite the lack of Boltzmann transport for the collisional system, or a collisional term due to pair annihilation.
### A.3. Electromagnetic Terms
In terms of $`F^{\mu \nu }`$, the Faraday (or electromagnetic field) tensor,
$$T_{\mathrm{EM}}^{\mu \nu }=F^{\mu \gamma }F_{}^{\nu }{}_{\gamma }{}^{}\frac{1}{4}g^{\mu \nu }F^{\alpha \beta }F_{\alpha \beta },$$
(A12)
where a factor of $`\sqrt{4\pi }`$ is absorbed into the definition of $`F^{\mu \nu }`$. The induction equation is given by the space components of $`{}_{}{}^{^{}}F_{}^{\mu \nu }{}_{;\nu }{}^{}=0`$, where $`{}_{}{}^{^{}}F`$ is the dual of the Faraday, and the time component gives the no-monopoles constraint. The other Maxwell equations, $`J^\mu =F_{}^{\mu \nu }{}_{;\nu }{}^{}`$, define the current density $`J^\mu `$. The comoving electric field is defined as
$$e^\nu u_\mu F^{\mu \nu }=\frac{1}{2}ϵ^{\mu \nu k\lambda }u_\nu {}_{}{}^{^{}}F_{\lambda k}^{}=\eta j^\nu ,$$
(A13)
where $`\eta `$ corresponds to a scalar resistivity for a comoving current density $`j^\mu =J_\nu P^{\nu \mu }`$, where $`P^{\nu \mu }g^{\nu \mu }+u^\nu u^\mu `$ projects any 4-vector into the comoving frame (i.e. $`P^{\nu \mu }u_\mu =0`$). The classical ideal MHD approximation that $`\eta =e^\mu =0`$ is assumed. The comoving magnetic field is defined as
$$b^\nu u_\mu {}_{}{}^{^{}}F_{}^{\mu \nu }=\frac{1}{2}ϵ^{\mu \nu k\lambda }u_\nu {}_{}{}^{^{}}F_{\lambda k}^{},$$
(A14)
and so the stress-energy tensor can be written as
$$T_{\mathrm{EM}}^{\mu \nu }=\frac{b^2}{2}(u^\mu u^\nu +P^{\mu \nu })b^\mu b^\nu ,$$
(A15)
and
$${}_{}{}^{^{}}F_{}^{\mu \nu }=b^\mu u^\nu b^\nu u^\mu $$
(A16)
and so
$$F^{\mu \nu }=ϵ^{\mu \nu \sigma ϵ}u_\sigma b_ϵ$$
(A17)
since $`{}_{}{}^{^{}}F_{}^{\mu \nu }=\frac{1}{2}ϵ^{\mu \nu \kappa \lambda }F_{\kappa \lambda }`$. Here $`ϵ`$ is the Levi-Civita tensor. Following the notation of MTW, $`ϵ^{\mu \nu \lambda \delta }=\frac{1}{\sqrt{g}}[\mu \nu \lambda \delta ]`$, where $`[\mu \nu \lambda \delta ]`$ is the completely antisymmetric symbol and $`=0,1,`$ or $`1`$. Notice that $`e^\nu u_\nu =b^\nu u_\nu =0`$, so they each have only 3 independent components and are space-like 4-vectors.
With $`B^i{}_{}{}^{^{}}F_{}^{it}`$ and $`E^iF^{it}`$, the no-monopoles constraint becomes
$$(\sqrt{g}B^i)_{,i}=0,$$
(A18)
and the magnetic induction equation becomes
$$(\sqrt{g}B^i)_{,t}=(\sqrt{g}(b^iu^jb^ju^i))_{,j}.$$
(A19)
The ideal MHD approximation assumes that $`e^\mu =0`$, and so the invariant $`e^\mu b_\mu =0`$. Since the Lorentz acceleration on a particle is $`f_l^\mu =qe^\mu `$, then this implies that the Lorentz force vanishes on a particle in the ideal MHD approximation. Equation A14 implies $`b^t=B^iu_i`$ and $`b^i=(B^i+u^ib^t)/u^t`$, so the magnetic induction equation becomes
$`(\sqrt{g}B^i)_{,t}`$ $`=`$ $`(\sqrt{g}(B^iv^jB^jv^i)_{,j}`$ (A20)
$`=`$ $`(\sqrt{g}(ϵ^{ijk}\epsilon _k))_{,j},`$
where $`v^i=u^i/u^t`$, $`\epsilon _i=ϵ_{ijk}v^jB^k=𝐯\times 𝐁`$ is the EMF, and $`ϵ^{ijk}`$ is the spatial permutation tensor. A more complete account of the relativistic MHD equations can be found in Anile (1989).
## Appendix B Characteristic (and other) Surfaces
The ideal MHD dispersion relation is given, e.g., in Gammie et al. (2003a) (there is a sign typo there), and summarized here. In the comoving frame, the dispersion relation is
$$\begin{array}{c}D(k^\mu )=0\hfill \\ \omega (\omega ^2(𝐤𝐯_\mathrm{A})^2)\times \hfill \\ \left(\omega ^4\omega ^2\left(K^2c_{ms}^2+c_s^2(𝐤𝐯_\mathrm{A})^2/c^2\right)+K^2c_s^2(𝐤𝐯_\mathrm{A})^2\right),\hfill \end{array}$$
(B1)
where $`c_{ms}^2=(𝐯_\mathrm{A}^2+c_s^2(1𝐯_\mathrm{A}^2/c^2))`$ is the magnetosonic speed, $`c_s^2=((\rho +u)/p)_s^1`$ is the relativistic sound speed, $`𝐯_\mathrm{A}=𝐁/\sqrt{}`$ is the relativistic Alfvén velocity, $`=b^2+w`$, and $`w\rho +u+p`$. Here $`c`$ is the (temporarily reintroduced) speed of light. The invariant scalars defining the comoving dispersion relation are $`\omega =k_\mu u^\mu `$, $`K^2=K_\mu K^\mu =k_\mu k^\mu +\omega ^2`$, where $`K_\mu =P_{\mu \nu }k^\nu =k_\mu +\omega u_\mu `$ is the projected wave vector normal to the fluid 4-velocity, $`𝐯_\mathrm{A}^2=b_\mu b^\mu /`$, and $`(𝐤𝐯_\mathrm{A})=k_\mu b^\mu /\sqrt{}`$. The terms in the dispersion relation correspond to, respectively from left to right, the zero frequency entropy mode, the left and right going Alfvén modes, and the left and right going fast and slow modes. The eighth mode is eliminated by the no-monopoles constraint.
The dispersion relation gives the ingoing and outgoing slow, Alfvén, and fast surfaces. Energy can be extracted from the black hole if and only if the Alfvén point lies inside the ergosphere (Takahashi et al., 1990). Optimal acceleration of the flow by conversion of Poynting flux to kinetic energy flux occurs beyond the outer fast surface (Begelman & Li, 1994). Other surfaces include: the horizon at $`r_H1+\sqrt{1j^2}`$ ; the ergosphere at $`r1+\sqrt{1(j\mathrm{cos}\theta )^2}`$ ; the coordinate basis light surface in where $`\sqrt{g_{\varphi \varphi }}=c/\mathrm{\Omega }_F`$, where asymptotically $`\sqrt{g_{\varphi \varphi }}=r\mathrm{sin}(\theta )`$ is the Minkowski cylindrical radius ; the surface in Boyer-Lindquist coordinates where the toroidal field equals the poloidal field where $`B^{\widehat{\varphi }}=B^{\widehat{r}}`$. Finally, there is a stagnation surface where the poloidal velocity $`u^p=0`$. In a field confined jet where no matter can cross field lines into the jet, and if the jet has inflow near the black hole and outflow far from the black hole, then this necessarily marks at least one location where rest-mass must be created either by charge starving the magnetosphere till the Goldreich-Julian charge density is reached, or pair production rates sustains the rest-mass density.
|
warning/0506/cond-mat0506443.html
|
ar5iv
|
text
|
# Femtosecond formation of collective modes due to meanfield fluctuations
## Abstract
Starting from a quantum kinetic equation including the mean field and a conserving relaxation-time approximation we derive an analytic formula which describes the time dependence of the dielectric function in a plasma created by a short intense laser pulse. This formula reproduces universal features of the formation of collective modes seen in recent experimental data of femtosecond spectroscopy. The presented formula offers a tremendous simplification for the description of the formation of quasiparticle features in interacting systems. Numerical demanding treatments can now be focused on effects beyond these gross features found here to be describable analytically.
The last ten years have been characterized by an enormous activity about ultrafast excitations in semiconductors, clusters, or plasmas by ultrashort laser pulses. The femtosecond spectroscopy has opened the exciting possibility to observe directly the formation of collective modes and quasiparticles in interacting many-body systems. This formation is reflected in the time dependence of the dielectric function Huber et al. (2001, 2002, 2005) or terahertz emission Lloyd-Hughes et al. (2004). For an overview over theoretical and experimental work see Axt and Kuhn (2004); Morawetz (2004). Such ultrafast excitations in semiconductors have been satisfactorily described by calculating nonequilibrium Green’s functions Bányai et al. (1998); Gartner et al. (1999). This approach allows one to describe the formation of collective modes Vu and Haug (2000); Huber et al. (2005) and even exciton population inversions Kira and Koch (2004).
The experimental data in semiconductors like GaAs Huber et al. (2002) or InP Huber et al. (2005) reveal similar features. These features are explained by several numerically demanding calculations. Since some common features are robust, i.e., independent of the actual used material parameters, it should be possible to describe them by a simple theory. The origin of such robust features likely rests in the short- or transient-time behavior itself. At short times higher-order correlations have no time yet to develop, therefore the dynamics is controlled exclusively by mean-field forces.
Based on the mean-field character of the short-time evolution we intend to derive a time-dependent response function from the mean-field linear response. We will start from a conserving relaxation-time approximation which dates back to an idea of Mermin Mermin (1970); Das (1975). Our aim is a simple analytic formula suitable for fits of experimental data.
The electron motion is controlled by the kinetic energy $`\widehat{}`$, the external perturbation $`\widehat{V}^{\mathrm{ext}}`$ and the induced mean-field potential $`\widehat{V}^{\mathrm{ind}}`$. The corresponding kinetic equation for the one-particle reduced density matrix $`\widehat{\rho }`$ reads
$`\dot{\widehat{\rho }}+i[\widehat{}+\widehat{V}^{\mathrm{ind}}+\widehat{V}^{\mathrm{ext}},\widehat{\rho }]={\displaystyle \frac{\widehat{\rho }^{\mathrm{l}.\mathrm{e}.}\widehat{\rho }}{\tau }}`$ (1)
with the relaxation time $`\tau `$ and a local equilibrium density matrix $`\widehat{\rho }^{\mathrm{l}.\mathrm{e}.}`$ determined in the following. We will calculate the response in the momentum representation
$`f(p,q,t)=p+{\displaystyle \frac{q}{2}}\left|\widehat{\rho }\right|p{\displaystyle \frac{q}{2}},`$ (2)
where $`|p`$ is an eigenstate of momentum $`p`$. For interpretations it is more convenient to transform the reduced density matrix to the Wigner distribution in phase space
$`f(p,R,t)={\displaystyle \underset{q}{}}\mathrm{e}^{iqR}f(p,q,t),`$ (3)
where $`R`$ is the spatial coordinate.
The relaxation process in Eq. (1) tends to establish a local equilibrium. Following Mermin, the local equilibrium can be characterized by a local variation of the chemical potential. Up to linear order it can be written as
$`p+{\displaystyle \frac{q}{2}}\left|\widehat{\rho }^{\mathrm{l}.\mathrm{e}.}\right|p{\displaystyle \frac{q}{2}}=f_p\delta (q){\displaystyle \frac{f_{p+\frac{q}{2}}f_{p\frac{q}{2}}}{\epsilon _{p+\frac{q}{2}}\epsilon _{p\frac{q}{2}}}}\delta \mu (q,t)`$ (4)
with the Fermi-Dirac distribution $`f_p=f_{\mathrm{FD}}\left(\epsilon _p\right)`$ and the kinetic energy $`p+\frac{q}{2}\left|\widehat{}\right|p\frac{q}{2}=\epsilon _p\delta (q)`$. The delta function $`\delta (q)`$ expresses that the corresponding terms are diagonal in momentum or translational invariant. The local deviation $`\delta \mu `$ from the global chemical potential is selected so that the density of particles is conserved at each point and time instant. Momentum and energy conservations can be included, too, leading to slightly more complicated formulas Morawetz and Fuhrmann (2000a); Atwal and Ashcroft (2002).
Let us now specify $`\delta \mu `$ from the condition of density conservation which requires that the local equilibrium distribution yields the same density as the actual distribution, $`n=_pf=_pf^{\mathrm{l}.\mathrm{e}.}`$. From Eq. (4) the deviation from the density can be expressed as
$`\delta n(q,t)=\stackrel{~}{\mathrm{\Pi }}^{\mathrm{RPA}}(t,\omega =0)\delta \mu (q,t),`$ (5)
where $`\stackrel{~}{\mathrm{\Pi }}^{\mathrm{RPA}}(t,\omega =0)`$ is the polarization in random-phase approximation (RPA) at zero frequency with material parameters like electron density and the effective temperature fixed to the value at time $`t`$. Note that this RPA-polarization enters only the right hand side of (1). The actual polarization which we will derive from (1) is more complex.
The induced potential is given by the Poisson equation
$`V^{\mathrm{ind}}(q,t)`$ $`=`$ $`p+{\displaystyle \frac{q}{2}}|\widehat{V}^{\mathrm{ind}}|p{\displaystyle \frac{q}{2}}={\displaystyle \frac{e^2}{ϵ_0q^2}}\delta n(q,t).`$ (6)
Now we can solve the kinetic equation (1) for a small external perturbation $`V^{\mathrm{ext}}`$. We split the distribution into the homogeneous part and a small inhomogeneous perturbation, $`f(p,q,t)=f_p(t)+\delta f(p,q,t)`$. Neglecting the time derivative of $`f_p`$ compared to $`\delta f`$ we can solve Eq. (1) up to linear order in $`V^{\mathrm{ext}}`$
$`\delta f(p,q,t)=i{\displaystyle \underset{t_0}{\overset{t}{}}}𝑑t^{}\mathrm{exp}\left[\left(i\epsilon _{p+\frac{q}{2}}i\epsilon _{p\frac{q}{2}}+{\displaystyle \frac{1}{\tau }}\right)(t^{}t)\right]`$
$`\times \{[f_{p+\frac{q}{2}}(t^{})f_{p\frac{q}{2}}(t^{})][{\displaystyle \frac{e^2}{ϵ_0q^2}}\delta n(q,t^{})+V_q^{\mathrm{ext}}(t^{})]`$
$`+{\displaystyle \frac{1}{i\tau \stackrel{~}{\mathrm{\Pi }}^{\mathrm{RPA}}(t^{},0)}}{\displaystyle \frac{f_{p+\frac{q}{2}}(t^{})f_{p\frac{q}{2}}(t^{})}{\epsilon _{p+\frac{q}{2}}\epsilon _{p\frac{q}{2}}}}\delta n(q,t^{})\}`$ (7)
where the perturbation starts at $`\delta f(p,q,t_0)=0`$.
The density response, $`\chi (t,t^{})`$, defined by
$`\delta n(q,t)={\displaystyle \underset{t_0}{\overset{t}{}}}𝑑t^{}\chi (t,t^{})V_q^{\mathrm{ext}}(t^{})`$ (8)
follows from Eq. (7) by integrating over $`p`$ with the result
$`\chi (t,t^{})=\mathrm{\Pi }(t,t^{})+{\displaystyle \underset{t^{}}{\overset{t}{}}}𝑑\overline{t}\left[\mathrm{\Pi }(t,\overline{t}){\displaystyle \frac{e^2}{ϵ_0q^2}}+I(t,\overline{t})\right]\chi (\overline{t},t^{}).`$ (9)
Here Mermin’s correction is represented by the term
$`I(t,t^{})=i{\displaystyle \underset{p}{}}{\displaystyle \frac{f_{p+\frac{q}{2}}(t^{})f_{p\frac{q}{2}}(t^{})}{\epsilon _{p+\frac{q}{2}}\epsilon _{p\frac{q}{2}}}}{\displaystyle \frac{\mathrm{e}^{\left(i\epsilon _{p+\frac{q}{2}}i\epsilon _{p\frac{q}{2}}+\frac{1}{\tau }\right)(t^{}t)}}{\tau \stackrel{~}{\mathrm{\Pi }}^{\mathrm{RPA}}(t^{},0)}}.`$
(10)
The polarization,
$`\mathrm{\Pi }(t,t^{})=i{\displaystyle \underset{p}{}}[f_{p+\frac{q}{2}}(t^{})f_{p\frac{q}{2}}(t^{})]\mathrm{e}^{\left(i\epsilon _{p+\frac{q}{2}}i\epsilon _{p\frac{q}{2}}+\frac{1}{\tau }\right)(t^{}t)},`$ (11)
describes the response of the system with respect to the induced field in contrast to the response function (8) which describes the response of the system with respect to the external field.
To link the derived formulas with familiar results we note that for time-independent $`f_p`$, the Fourier transform of the polarization with respect to the difference time
$`\stackrel{~}{\mathrm{\Pi }}(T,\omega )={\displaystyle 𝑑\tau \mathrm{e}^{i\omega \tau }\mathrm{\Pi }(T+\frac{\tau }{2},T\frac{\tau }{2})}`$ (12)
becomes the standard RPA polarization with the frequency argument shifted by the inverse relaxation time
$`\stackrel{~}{\mathrm{\Pi }}(T,\omega )=\stackrel{~}{\mathrm{\Pi }}^{\mathrm{RPA}}\left(\omega +{\displaystyle \frac{i}{\tau }}\right).`$ (13)
Moreover, in the limit $`t+t^{}\mathrm{}`$ the solution of Eq. (9) approaches the familiar form $`\chi =\mathrm{\Pi }^\mathrm{M}/(1\frac{e^2}{ϵ_0q^2}\mathrm{\Pi }^\mathrm{M})`$ where the Mermin-Das polarization reads Mermin (1970)
$`\mathrm{\Pi }^\mathrm{M}={\displaystyle \frac{\stackrel{~}{\mathrm{\Pi }}^{\mathrm{RPA}}(\omega +\frac{i}{\tau })}{1i(1+\frac{1}{\omega \tau })\left(1\frac{\stackrel{~}{\mathrm{\Pi }}^{\mathrm{RPA}}(\omega +\frac{i}{\tau })}{\stackrel{~}{\mathrm{\Pi }}^{\mathrm{RPA}}(0)}\right)}}.`$ (14)
Eq. (9) describes how the Mermin susceptibility is formed after a fast release of free carriers.
In the further analysis we will follow closely the experimental way of analyzing the two-time response function Huber et al. (2005). The pump pulse is creating charge carriers in the conduction band at time $`t_0`$ and the probe pulse is sent a time $`t_D`$ later. The time delay after this probe pulse $`T=tt_Dt_0`$ is then Fourier transformed into frequency. Of course everything starts at $`t_0`$ when the pump pulse has created the carriers in the conduction band, i.e. the maximum delay is $`Ttt_0`$. The frequency-dependent inverse dielectric function associated with the actual time $`t`$ thus reads
$`z(t)={\displaystyle \frac{1}{ϵ(\omega ,t)}}1={\displaystyle \frac{e^2}{ϵ_0q^2}}{\displaystyle \underset{0}{\overset{tt_0}{}}}𝑑T\mathrm{e}^{i\omega T}\chi (t,tT).`$ (15)
This is exactly the one-sided Fourier transform introduced in Ref. ElSayed et al. (1994). It is worth noting that Wigner’s form of the Fourier transform (12) would just result in a factor of two in the time evolution of the dielectric function.
Now we turn our attention to short times when the response is built up. In general the integral equation (9) together with Eqs. (10) and (11) has to be solved. The useful limit of long wave lengths $`q0`$ offers an appreciable simplification in that the leading terms of (11) and (10) are $`\mathrm{\Pi }(t,t^{})\frac{q^2n(t^{})}{m}(t^{}t)\mathrm{e}^{\frac{t^{}t}{\tau }}`$ and $`I(t,t^{})\frac{1}{\tau }\mathrm{e}^{\frac{t^{}t}{\tau }}`$. com It is practical to transform the integral equation (9) for the response function directly into one for the inverse dielectric function, Eq. (15), by differentiating twice with respect to time
$`z^{\prime \prime }(t)+({\displaystyle \frac{1}{\tau }}2i\omega )z^{}(t)+\left[\omega _p^2(t)\omega ^2i{\displaystyle \frac{\omega }{\tau }}\right]z(t)=\omega _p^2(t)`$
$`z(t_0)=0;z^{}(t)|_{t=t_0}=0.`$ (16)
Here $`z^{}(t)=dz(t)/dt`$ and we have denoted the time-dependent plasma frequency by $`\omega _p^2(t)=\frac{e^2n(t)}{ϵ_0ϵ_{\mathrm{}}m}`$.
In our simplified treatment the density of electrons $`n(t)`$ is assumed to be a known function of the time. Due to Rabi oscillations it is a nontrivial function of the laser pulse. If the pulse is short compared to the formation of the collective mode, all details of the density build-up become unimportant and we can approximate the time dependence of the density by a step function ElSayed et al. (1994). Then the plasma frequency is $`\omega _p(t)=\omega _p\mathrm{\Theta }(tt_0)`$, where $`t_0`$ is the time when the pulse reaches its maximum. For this approximation the differential equation (16) can be solved analytically, yielding
$`{\displaystyle \frac{1}{ϵ(\omega ,t)}}=1{\displaystyle \frac{\omega _p^2}{\mathrm{\Gamma }}}{\displaystyle \underset{0}{\overset{tt_0}{}}}𝑑t^{}\mathrm{e}^{(i\omega \frac{1}{2\tau })t^{}}\mathrm{sin}(\mathrm{\Gamma }t^{}),`$ (17)
where $`\mathrm{\Gamma }=\sqrt{\omega _p^2\frac{1}{4\tau ^2}}`$. The integral (17) can be expressed in terms of elementary functions, but we find the integral form more transparent.
Formula (17) has been derived from Mermin’s density-conserving approximation. The long-time limit yields the Drude formula
$`\underset{t\mathrm{}}{lim}{\displaystyle \frac{1}{ϵ}}=1{\displaystyle \frac{\omega _p^2}{\omega _p^2\omega (\omega +\frac{i}{\tau })}}.`$ (18)
This obvious limit is not so easy to achieve within short-time expansions. For example the approximate result from quantum kinetic theory presented in ElSayed et al. (1994) gives the long-time limit of the form $`1\omega _p^2/[\omega _p^2(\omega +i/\tau )^2]`$.
In spite of the numerous approximations, Eq. (17) fits well the experimental data, as shown in Fig. 1 for the polar semiconductor GaAs.
The resonance visible at about 8 THz is due to optical phonons. This feature is easily accounted for, see Fig. 1, by adding the intrinsic contribution of the crystal lattice Huber et al. (2002)
$`ϵ_{\mathrm{GaAs}}(\omega ,t)=ϵ_{\mathrm{}}\left(ϵ(\omega ,t)+{\displaystyle \frac{\omega _{\mathrm{LO}}^2\omega _{\mathrm{TO}}^2}{\omega _{\mathrm{TO}}^2\omega ^2i\gamma \omega }}\right),`$ (19)
with the longitudinal and transversal optical frequencies $`\omega _{\mathrm{LO}}/2\pi =8.8`$ THz and $`\omega _{\mathrm{TO}}/2\pi =8.1`$ THz, the lattice damping $`\gamma =`$0.2 ps<sup>-1</sup> and the non-resonant background polarizability of the ion lattice $`ϵ_{\mathrm{}}=11.0`$, see Ref. Huber et al. (2002).
The formula (17) results in a too fast build-up of the collective mode. This is seen in the second row of Fig. 1 at the time $`t=25`$ fs. This is, however, just the time duration of the experimental pulse and consequently the time of populating the conduction band. Since we have approximated this by the instant jump, this discrepancy at short times around $`25`$ fs can be expected. A more realistic smooth populating can be modeled by an arctan-function and then the numerical solution of (16) improves the description as we see in Fig. 2. Only the early development is plotted since the later stages agree with the simple estimate (17).
It is obvious that only the gross feature of formation of collective modes can be described by the mean-field fluctuations. For correlations beyond the mean field we should observe deviations. This is the case in the recently observed response of InP Huber et al. (2005), where a coupling between LO phonon and photon modes is reported. In Fig. 3 we see that the formation of the collective mode is delayed in the experiment during the first 100 fs when compared to the mean-field formula (17). Such behavior is due to scattering processes neglected here. At times 100 fs-200 fs the formation of collective modes is described again sufficiently well. For InP we use parameters according to Huber et al. (2005) namely $`\omega _{\mathrm{LO}}/2\pi =10.3`$ THz, $`\omega _{\mathrm{TO}}/2\pi =9`$ THz, $`\gamma =0.2`$ ps<sup>-1</sup>, and $`ϵ_{\mathrm{}}=9.6`$.
The aim of the present paper was to separate the gross feature of the formation of collective modes at transient times which are due to simple mean-field fluctuations. This has resulted in a simple analytic formula for the time dependence of the dielectric function. Subtracting this gross feature from the data allows one to extract the effects which are due to higher-order correlations and which have to be simulated by quantum kinetic theory Bányai et al. (1998); Gartner et al. (1999); Vu and Haug (2000); Kira and Koch (2004) and response functions with approximations beyond the mean field Kwong and Bonitz (2000). These treatments are numerically demanding such that analytic expressions for the time dependence of some variables Morawetz et al. (1998) are useful for controlling the numerics.
To conclude, we have derived a simple analytic formula for the formation of a collective mode. Being able to describe universal features of the formation of quasiparticles, the simplicity of the presented result is extremely practical and offers a wide range of applications. We see two prominent examples: (1) It could spare a lot of computational power to simulate ultrashort-time behavior of new nano-devices and (2) it can help to understand and describe the formation of collective modes during nuclear collisions which are not experimentally accessible in the early phase of collision.
We thank A. Huber and R. Leitenstorfer for providing us with the experimental data and H. N. Kwong for enlightening and clarifying discussions.
|
warning/0506/math0506110.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
Representations of quantum superalgebras are known to provide solutions to the Yang–Baxter equation and represent the symmetries that underly supersymmetric exactly solvable (or integrable) models. Many such examples have arisen in the context of modelling systems of strongly correlated electrons . More recently, the properties of solvability and supersymmetry have been applied to other areas, such as the solution of the Kondo model , integrable superconformal field theory and disordered systems . Developing the representation theory of the quantum superalgebras is a useful step towards the complete understanding of such models. However, in many respects the representation theory of quantum superalgebras is not a straightforward generalisation of the quantum algebra case, principally because not all representations of quantum superalgebras are unitary . This leads, for example, to the existence of indecomposable representations not arising in the quantum algebra case, which generally make the analysis of supersymmetric models problematic (e.g. see ).
In this paper we construct the Casimir invariants (central elements) of quantised orthosymplectic superalgebras. Our method of construction follows from the general results of and the explicit form of the Lax operator obtained in . A fundamental problem is to determine the eigenvalues of the Casimir invariants when acting on an arbitrary finite-dimensional irreducible module. To date, the eigenvalues have only been calculated for the type I quantum superalgebras , while the results for $`U_q[osp(1|n)]`$ follow from an isomorphism derived in . In this paper we perform the calculations for the remaining non-exceptional quantum superalgebras, namely $`U_q[osp(m|n)]`$ for $`m>2`$. The procedure we use for calculating the eigenvalues of the Casimir invariants when acting on any irreducible module is based on the early work by Perelomov and Popov and Nwachuku and Rashid . In doing so we follow the method used in and for the classical general and orthosymplectic superalgebras respectively, which was adapted in to cover $`U_q[gl(m|n)]`$. Although the concepts are much the same as in those cases, the combination of the $`q`$-deformation and the more complex root system of $`U_q[osp(m|n)]`$ makes the calculations in this paper more technically challenging.
In the following section we introduce our notation for $`U_q[osp(m|n)]`$ and state the Lax operator. In Section 3 we develop the formulae for the Casimir invariants of $`U_q[osp(m|n)]`$. The bulk of the calculations are in Section 4, where the eigenvalues of the Casimir invariants are derived in detail.
## 2 The quantised orthosymplectic superalgebra $`U_q[osp(m|n)]`$
The quantum superalgebra $`U_q[osp(m|n)]`$ is a $`q`$-deformation of the classical orthosymplectic superalgebra. A brief explanation of $`U_q[osp(m|n)]`$ is given below, with a more thorough introduction to $`osp(m|n)`$ and the $`q`$-deformation to be found in .
First we need to define the notation. The grading of $`a`$ is denoted by $`[a]`$, where
$$[a]=\{\begin{array}{cc}0,a=i,\hfill & 1im,\hfill \\ 1,a=\mu ,\hfill & 1\mu n.\hfill \end{array}$$
Throughout this paper we use Greek letters $`\mu ,\nu `$ etc. to denote odd indices and Latin letters $`i,j`$ etc. for even indices. If the grading is unknown, the usual $`a,b,c`$ etc. are used. Which convention applies will be clear from the context. Throughout the paper we also use the symbols $`\overline{a}`$ and $`\xi _a`$, which are given by:
$$\overline{a}=\{\begin{array}{cc}m+1a,\hfill & [a]=0,\hfill \\ n+1a,\hfill & [a]=1,\hfill \end{array}\text{and}\xi _a=\{\begin{array}{cc}1,\hfill & [a]=0,\hfill \\ (1)^a,\hfill & [a]=1.\hfill \end{array}$$
As a weight system for $`U_q[osp(m|n)]`$ we take the set $`\{\epsilon _i,\mathrm{\hspace{0.33em}1}im\}\{\delta _\mu ,\mathrm{\hspace{0.33em}1}\mu n\}`$, where $`\epsilon _{\overline{i}}=\epsilon _i`$ and $`\delta _{\overline{\mu }}=\delta _\mu `$. Conveniently, when $`m=2l+1`$ this implies $`\epsilon _{l+1}=\epsilon _{l+1}=0`$. Acting on these weights, we have the invariant bilinear form defined by:
$$(\epsilon _i,\epsilon _j)=\delta _j^i,(\delta _\mu ,\delta _\nu )=\delta _\nu ^\mu ,(\epsilon _i,\delta _\mu )=0,1i,jl,1\mu ,\nu k.$$
When describing an object with unknown grading indexed by $`a`$ the weight will be described generically as $`\epsilon _a`$. This should not be assumed to be an even weight.
The even positive roots of $`U_q[osp(m|n)]`$ are composed entirely of the usual positive roots of $`o(m)`$ together with those of $`sp(n)`$, namely:
$`\epsilon _i\pm \epsilon _j,`$ $`1i<jl,`$
$`\epsilon _i,`$ $`1il`$ $`\text{when }m=2l+1,`$
$`\delta _\mu +\delta _\nu ,`$ $`1\mu ,\nu k,`$
$`\delta _\mu \delta _\nu ,`$ $`1\mu <\nu k.`$
The root system also contains a set of odd positive roots, which are:
$$\delta _\mu +\epsilon _i,1\mu k,\mathrm{\hspace{0.33em}1}im.$$
Throughout this paper we choose to use the following set of simple roots:
$`\alpha _i=\epsilon _i\epsilon _{i+1},1i<l,`$
$`\alpha _l=\{\begin{array}{cc}\epsilon _l+\epsilon _{l1},\hfill & m=2l,\hfill \\ \epsilon _l,\hfill & m=2l+1,\hfill \end{array}`$
$`\alpha _\mu =\delta _\mu \delta _{\mu +1},1\mu <k,`$
$`\alpha _s=\delta _k\epsilon _1.`$
Note this choice is only valid for $`m>2`$.
In $`U_q[osp(m|n)]`$ the graded commutator is realised by
$$[A,B]=AB(1)^{[A][B]}BA$$
and tensor product multiplication is given by
$$(AB)(CD)=(1)^{[B][C]}(ACBD).$$
Using these conventions, we have:
###### Definition 2.1
The quantum superalgebra $`U_q[osp(m|n)]`$ is generated by simple generators $`e_a,f_a,h_a`$ subject to the relations:
$`[h_a,e_b]=(\alpha _a,\alpha _b)e_b,`$
$`[h_a,f_b]=(\alpha _a,\alpha _b)f_b,`$
$`[h_a,h_b]=0,`$
$`[e_a,f_b]=\delta _b^a{\displaystyle \frac{(q^{h_a}q^{h_a})}{(qq^1)}},`$
$`[e_a,e_a]=[f_a,f_a]=0`$ $`\text{for }(\alpha _a,\alpha _a)=0,`$
We remark that $`U_q[osp(m|n)]`$ has the structure of a quasi-triangular Hopf superalgebra. In particular, there is a linear mapping known as the coproduct, $`\mathrm{\Delta }:U_q[osp(m|n)]U_q[osp(m|n)]^2`$, which is defined on the simple generators by:
$`\mathrm{\Delta }(e_a)=q^{\frac{1}{2}h_a}e_a+e_aq^{\frac{1}{2}h_a},`$
$`\mathrm{\Delta }(f_a)=q^{\frac{1}{2}h_a}f_a+f_aq^{\frac{1}{2}h_a},`$
$`\mathrm{\Delta }(q^{\pm \frac{1}{2}h_a})=q^{\pm \frac{1}{2}h_a}q^{\pm \frac{1}{2}h_a},`$
and extends to arbitrary elements according to the homomorphism property, namely:
$`\mathrm{\Delta }(AB)=\mathrm{\Delta }(A)\mathrm{\Delta }(B).`$
There are further defining relations such as the $`q`$-Serre relations, but they are not needed in this paper.
The quasi-triangular property guarantees the existence of a universal $`R`$-matrix, which provides a solution to the Yang–Baxter equation. Before elaborating, we need to introduce the graded twist map.
The graded twist map $`T:U_q[osp(m|n)]^2U_q[osp(m|n)]^2`$ is given by
$$T(ab)=(1)^{[a][b]}(ba).$$
For convenience $`T\mathrm{\Delta }`$, the twist map composed with the coproduct, is denoted $`\mathrm{\Delta }^T`$. Then a universal $`R`$-matrix, $``$, is an even, non-singular element of $`U_q[osp(m|n)]^2`$ satisfying the following properties:
$`\mathrm{\Delta }(a)=\mathrm{\Delta }^T(a),aU_q[osp(m|n)],`$
$`(\text{id}\mathrm{\Delta })=_{13}_{12},`$
$`(\mathrm{\Delta }\text{id})=_{13}_{23}.`$ (1)
Here $`_{ab}`$ represents a copy of $``$ acting on the $`a`$ and $`b`$ components respectively of $`U_1U_2U_3`$, where each $`U`$ is a copy of the quantum superalgebra $`U_q[osp(m|n)]`$. When $`a>b`$ the usual grading term from the twist map is included, so for example $`_{21}=[^T]_{12}`$, where $`^T=T()`$ is the opposite universal $`R`$-matrix.
The $`R`$-matrix is significant because it is a solution to the Yang–Baxter equation, which is prominent in the study of integrable systems :
$$_{12}_{13}_{23}=_{23}_{13}_{12}$$
A superalgebra may contain many different universal $`R`$-matrices, but there is always a unique one belonging to $`U_q[osp(m|n)]^{}U_q[osp(m|n)]^+`$, with its opposite $`R`$-matrix in $`U_q[osp(m|n)]^+U_q[osp(m|n)]^{}`$. Here $`U_q[osp(m|n)]^{}`$ is the Hopf subsuperalgebra generated by the lowering generators $`\{f_a\}`$ and Cartan elements $`\{h_a\}`$, while $`U_q[osp(m|n)]^+`$ is generated by the raising generators $`\{e_a\}`$ and the Cartan elements. These particular $`R`$-matrices arise out of the $`_2`$-graded version of Drinfeld’s double construction . In this paper we consider the universal $`R`$-matrix belonging to $`U_q[osp(m|n)]^{}U_q[osp(m|n)]^+`$.
We also need to define the vector representation for $`U_q[osp(m|n)]`$. Let $`\text{End}V`$ be the space of endomorphisms of $`V`$, an $`(m+n)`$-dimensional vector space. Then the irreducible vector representation $`\pi :U_q[osp(m|n)]\text{End}V`$ acts on the $`U_q[osp(m|n)]`$ generators as given in Table 1, where $`E_b^a`$ is the elementary matrix with a 1 in the $`(a,b)`$ position and zeroes elsewhere.
One quantity that repeatedly arises in calculations for both classical and quantum Lie superalgebras is $`\rho `$, the graded half-sum of positive roots. In the case of $`U_q[osp(m|n)]`$ it is given by:
$$\rho =\frac{1}{2}\underset{i=1}{\overset{l}{}}(m2i)\epsilon _i+\frac{1}{2}\underset{\mu =1}{\overset{k}{}}(nm+22\mu )\delta _\mu .$$
This satisfies the property $`(\rho ,\alpha )=\frac{1}{2}(\alpha ,\alpha )`$ for all simple roots $`\alpha `$.
### 2.1 The Lax operator for $`U_q[osp(m|n)]`$
Let $``$ be the universal $`R`$-matrix of $`U_q[osp(m|n)]`$ and $`\pi `$ the vector representation. The Lax operator associated with $``$ is given by
$$R=(\pi \text{id})(\text{End}V)U_q[osp(m|n)].$$
It has been shown in that the Lax operator is given by
$$R=\underset{a}{}E_a^aq^{h_{\epsilon _a}}+(qq^1)\underset{\epsilon _a<\epsilon _b}{}(1)^{[b]}E_b^aq^{h_{\epsilon _a}}\widehat{\sigma }_{ba},$$
where the simple operators $`\widehat{\sigma }_{ba}`$ are given by:
$`\widehat{\sigma }_{ii+1}=\widehat{\sigma }_{\overline{i+1}\overline{i}}=q^{\frac{1}{2}}e_iq^{\frac{1}{2}h_i},`$ $`1i<l,`$
$`\widehat{\sigma }_{l1\overline{l}}=\widehat{\sigma }_{l\overline{l1}}=q^{\frac{1}{2}}e_lq^{\frac{1}{2}h_l},`$ $`m=2l,`$
$`\widehat{\sigma }_{l\overline{l}}=0,`$ $`m=2l,`$
$`\widehat{\sigma }_{ll+1}=q^{\frac{1}{2}}\widehat{\sigma }_{l+1\overline{l}}=e_lq^{\frac{1}{2}h_l},`$ $`m=2l+1,`$
$`\widehat{\sigma }_{\mu \mu +1}=\widehat{\sigma }_{\overline{\mu +1}\overline{\mu }}=q^{\frac{1}{2}}e_\mu q^{\frac{1}{2}h_\mu },`$ $`1\mu <k,`$
$`\widehat{\sigma }_{\mu =ki=1}=(1)^kq\widehat{\sigma }_{i=\overline{1}\overline{\mu }=\overline{k}}=q^{\frac{1}{2}}e_sq^{\frac{1}{2}h_s};`$
and the remaining operators can be calculated using
(i) the $`q`$-commutation relations
$$q^{(\alpha _c,\epsilon _b)}\widehat{\sigma }_{ba}e_cq^{\frac{1}{2}h_c}(1)^{([a]+[b])[c]}q^{(\alpha _c,\epsilon _a)}e_cq^{\frac{1}{2}h_c}\widehat{\sigma }_{ba}=0,\epsilon _b>\epsilon _a,$$
where neither $`\epsilon _a\alpha _c`$ nor $`\epsilon _b+\alpha _c`$ equals any $`\epsilon _x`$; and
(ii) the induction relations
$$\widehat{\sigma }_{ba}=q^{(\epsilon _b,\epsilon _a)}\widehat{\sigma }_{bc}\widehat{\sigma }_{ca}q^{(\epsilon _c,\epsilon _c)}(1)^{([b]+[c])([a]+[c])}\widehat{\sigma }_{ca}\widehat{\sigma }_{bc},\epsilon _b>\epsilon _c>\epsilon _a,$$
where $`c\overline{b}\text{ or }\overline{a}`$.
To define the opposite Lax operator $`R^T=(\pi \text{id})`$ we require the graded conjugation action $``$, which is defined on the simple generators by (see ):
$$e_a^{}=f_a,f_a^{}=(1)^{[a]}e_a,h_a^{}=h_a.$$
It is consistent with the coproduct and extends naturally to all remaining elements of $`U_q[osp(m|n)]`$, satisfying the properties:
$`(\sigma _b^a)^{}=(1)^{[a]([a]+[b])}\sigma _a^b,`$
$`(ab)^{}=(1)^{[a][b]}b^{}a^{},`$
$`(ab)^{}=a^{}b^{},`$
$`\mathrm{\Delta }(a)^{}=\mathrm{\Delta }(a^{}).`$
Then the opposite $`R`$-matrix is given by
$$R^T=\underset{a}{}E_a^aq^{h_{\epsilon _a}}+(qq^1)\underset{\epsilon _b>\epsilon _a}{}(1)^{[a]}E_a^b\widehat{\sigma }_{ab}q^{h_{\epsilon _a}},$$
where
$$\widehat{\sigma }_{ab}=(1)^{[b]([a]+[b])}\widehat{\sigma }_{ba}^{},\epsilon _b>\epsilon _a.$$
## 3 Casimir Invariants of $`U_q[osp(m|n)]`$
We now use the Lax operator to construct a family of Casimir invariants and then to calculate their eigenvalues when acting on an irreducible highest weight module. Before constructing the Casimir invariants, however, we need to define a new object. Let $`h_\rho `$ be the unique element of the Cartan subalgebra H satisfying
$$\alpha _i(h_\rho )=(\rho ,\alpha _i),\alpha _iH^{}.$$
Then from we have the following theorem:
###### Theorem 3.1
Let $`𝒱`$ be the representation space of $`\tau `$, an arbitrary finite-dimensional representation of $`U_q[osp(m|n)]`$. If $`\mathrm{\Gamma }(\text{End }𝒱)U_q[osp(m|n)]`$ satisfies
$$(a)\mathrm{\Gamma }=\mathrm{\Gamma }(a),aU_q[osp(m|n)],$$
(2)
where $`(\pi \text{id})\mathrm{\Delta }`$, then
$$C=(str\mathrm{id})(\tau (q^{2h_\rho })I)\mathrm{\Gamma }$$
belongs to the centre of $`U_q[osp(m|n)]`$. Above $`str`$ denotes the supertrace.
Now choose $`\tau `$ to be the vector representation $`\pi `$. Recalling that the universal $`R`$-matrix satisfies
$$\mathrm{\Delta }(a)=\mathrm{\Delta }^T(a),aU_q[osp(m|n)],$$
it is clear that
$$(a)R^TR=R^TR(a),aU_q[osp(m|n)].$$
Hence if we set $`A(\text{End}V)U_q[osp(m|n)]`$ to be
$$A=\frac{(R^TRII)}{(qq^1)},$$
the operators $`A^l`$ will satisfy condition (2) for all non-negative integers $`l`$. Thus the operators $`C_l`$ defined as
$$C_l=(str\text{id})(\pi (q^{2h_p})I)A^l,l^+,$$
form a family of Casimir invariants. Here $`A`$ coincides with the matrix of Jarvis and Green in the classical limit $`q1`$, as do the invariants $`C_l`$.
Now write the Lax operator $`R`$ and its opposite $`R^T`$ in the form
$`R`$ $`=II+(qq^1){\displaystyle \underset{\epsilon _b\epsilon _a}{}}E_b^aX_a^b,`$
$`R^T`$ $`=II+(qq^1){\displaystyle \underset{\epsilon _b\epsilon _a}{}}E_b^aX_a^b.`$
In terms of the operators $`\widehat{\sigma }_{ba}`$, this implies
$$X_a^b=\{\begin{array}{cc}\frac{q^{h_{\epsilon _a}}I}{qq^1},\hfill & a=b,\hfill \\ (1)^{[b]}q^{h_{\epsilon _a}}\widehat{\sigma }_{ba},\hfill & \epsilon _a<\epsilon _b,\hfill \\ (1)^{[b]}\widehat{\sigma }_{ba}q^{h_{\epsilon _b}},\hfill & \epsilon _a>\epsilon _b.\hfill \end{array}$$
Writing $`A`$ as
$$A=\underset{a,b}{}E_b^aA_a^b,$$
we obtain
$$A_a^b=(1+\delta _b^a)X_a^b+(qq^1)\underset{\epsilon _c\epsilon _a,\epsilon _b}{}(1)^{([a]+[c])([b]+[c])}X_a^cX_c^b.$$
This produces a family of Casimir invariants
$$C_l=\underset{a}{}(1)^{[a]}q^{(2\rho ,\epsilon _a)}A_{}^{(l)}{}_{a}{}^{a},$$
where the operators $`A_{}^{(l)}{}_{a}{}^{b}`$ are recursively defined as
$$A_{}^{(l)}{}_{a}{}^{b}=\underset{c}{}(1)^{([a]+[c])([b]+[c])}A_{}^{(l1)}{}_{a}{}^{c}A_c^b.$$
(3)
Note that $`A`$ corresponds to the matrix $`A`$ given for the non-graded case in . Following a line of reasoning similar to that in , it can be shown that when acting on an irreducible module $`V(\mathrm{\Lambda })`$, $`A`$ satisfies the following polynomial identity:
$$\underset{a=1}{\overset{m+n}{}}(A\alpha _a(\mathrm{\Lambda })I)=0$$
where
$$\alpha _a(\mathrm{\Lambda })=\frac{q^{(\epsilon _a,\epsilon _a+2\mathrm{\Lambda }+2\rho )C(\mathrm{\Lambda }_0)}1}{qq^1}$$
and $`C(\mathrm{\Lambda }_0)=(\delta _1,\delta _1+2\rho )=mn1.`$ In the limit $`q1`$ this reduces to the identity given in .
## 4 Eigenvalues of the Casimir invariants
Now that we have found a family of Casimir invariants, we wish to calculate their eigenvalues on a general irreducible finite-dimensional module. Let $`V(\mathrm{\Lambda })`$ be an arbitrary irreducible finite-dimensional module with highest weight $`\mathrm{\Lambda }`$ and highest weight state $`|\mathrm{\Lambda }`$. Define $`t_a^{(l)}`$ to be the eigenvalue of $`A_{}^{(l)}{}_{a}{}^{a}`$ on this state, so
$$A_{}^{(l)}{}_{a}{}^{a}|\mathrm{\Lambda }=t_a^{(l)}|\mathrm{\Lambda }.$$
Once we have calculated $`t_a^{(l)}`$ we will use this result to find the eigenvalues of the Casimir invariants $`C_l`$.
To evaluate $`t_a^{(l)}`$, note that if $`\epsilon _b>\epsilon _a`$ then $`A_{}^{(l)}{}_{a}{}^{b}`$ is a raising operator, implying $`A_{}^{(l)}{}_{a}{}^{b}|\mathrm{\Lambda }=0`$. Thus from equation (3) we deduce
$`t_a^{(l)}|\mathrm{\Lambda }`$ $`=t_a^{(l1)}t_a^{(1)}|\mathrm{\Lambda }+{\displaystyle \underset{\epsilon _a<\epsilon _b}{}}(1)^{[a]+[b]}A_{}^{(l1)}{}_{a}{}^{b}A_b^a|\mathrm{\Lambda }`$
$`=t_a^{(l1)}t_a^{(1)}|\mathrm{\Lambda }+{\displaystyle \underset{\epsilon _a<\epsilon _b}{}}(1)^{[a]+[b]}A_{}^{(l1)}{}_{a}{}^{b}\left[X_b^a+(qq^1)X_b^aX_a^a\right]|\mathrm{\Lambda }`$
$`=t_a^{(l1)}t_a^{(1)}|\mathrm{\Lambda }+{\displaystyle \underset{\epsilon _a<\epsilon _b}{}}(1)^{[a]+[b]}q^{(\mathrm{\Lambda },\epsilon _a)}A_{}^{(l1)}{}_{a}{}^{b}X_b^a|\mathrm{\Lambda }.`$
Now we know that
$$A^l(X_b^a)=(X_b^a)A^l.$$
(4)
This can be used to calculate $`A_{}^{(l)}{}_{a}{}^{b}X_b^a|\mathrm{\Lambda }`$ for $`\epsilon _a<\epsilon _b`$. First we need an expression for $`\mathrm{\Delta }(X_b^a)`$. The $`R`$-matrix properties give
$`(\mathrm{\Delta }I)R=R_{13}R_{23}`$
$``$ $`(I\mathrm{\Delta })R^T=R_{12}^TR_{13}^T.`$
In terms of $`X_b^a`$, this implies
$`III+`$ $`(qq^1){\displaystyle \underset{\epsilon _a\epsilon _b}{}}E_a^b\mathrm{\Delta }(X_b^a)`$
$`=\left(III+(qq^1){\displaystyle \underset{\epsilon _a\epsilon _b}{}}E_a^bX_b^aI\right)`$
$`\times \left(III+(qq^1){\displaystyle \underset{\epsilon _a\epsilon _b}{}}E_a^bIX_b^a\right)`$
$`=III+(qq^1){\displaystyle \underset{\epsilon _a\epsilon _b}{}}E_a^b(X_b^aI+IX_b^a)`$
$`+(qq^1)^2{\displaystyle \underset{\epsilon _a\epsilon _c\epsilon _b}{}}(1)^{([a]+[c])([b]+[c])}E_a^bX_b^cX_c^a.`$
Hence for all $`\epsilon _a<\epsilon _b`$
$$\mathrm{\Delta }(X_b^a)=X_b^aI+IX_b^a+(qq^1)\underset{\epsilon _a\epsilon _c\epsilon _b}{}(1)^{([a]+[c])([b]+[c])}X_b^cX_c^a.$$
We also need an expression for $`\pi (X_b^a)`$ for $`\epsilon _a\epsilon _b`$. In we found the generators for $`R^T`$ in the vector representation are given by
$$\widehat{\sigma }_{ab}q^{h_{\epsilon _a}}=E_b^a(1)^{[a]([a]+[b])}\xi _a\xi _bq^{(\rho ,\epsilon _a\epsilon _b)}E_{\overline{a}}^{\overline{b}},\epsilon _a<\epsilon _b.$$
From this we deduce that
$$\pi (X_b^a)=(1)^{[a]}E_b^a(1)^{[a][b]}\xi _a\xi _bq^{(\rho ,\epsilon _a\epsilon _b)}E_{\overline{a}}^{\overline{b}},\epsilon _a<\epsilon _b.$$
Also, we know
$`\pi (X_a^a)`$ $`=(qq^1)^1\pi (q^{h_{\epsilon _a}}I)`$
$`=(qq^1)^1(q^{(\epsilon _a,\epsilon _a)(E_a^aE_{\overline{a}}^{\overline{a}})}I).`$
Applying these, we find that if $`\epsilon _a<\epsilon _b`$ then
$`(X_b^a)`$ $`=(\pi I)\mathrm{\Delta }(X_b^a)`$
$`=\pi (X_b^a)\left(I+(qq^1)X_a^a\right)+\left(I+(qq^1)\pi (X_b^b)\right)X_b^a`$
$`+(qq^1){\displaystyle \underset{\epsilon _a<\epsilon _c<\epsilon _b}{}}(1)^{([a]+[c])([b]+[c])}\pi (X_b^c)X_c^a`$
$`=\left((1)^{[a]}E_b^a(1)^{[a][b]}\xi _a\xi _bq^{(\rho ,\epsilon _a\epsilon _b)}E_{\overline{a}}^{\overline{b}}\right)q^{h_{\epsilon _a}}+q^{(\epsilon _b,\epsilon _b)(E_b^bE_{\overline{b}}^{\overline{b}})}X_b^a`$
$`+(qq^1){\displaystyle \underset{\epsilon _a<\epsilon _c<\epsilon _b}{}}(1)^{([a]+[c])([b]+[c])}`$
$`\times \left((1)^{[c]}E_b^c(1)^{[b][c]}\xi _b\xi _cq^{(\rho ,\epsilon _c\epsilon _b)}E_{\overline{c}}^{\overline{b}}\right)X_c^a.`$
Substituting this expression into equation (4) and equating the $`(a,b)`$ entries, we find
$`(1)^{[a]}`$ $`A_{}^{(l)}{}_{a}{}^{a}q^{h_{\epsilon _a}}\delta _{\overline{b}}^a(1)^{[a][b]}\xi _a\xi _bq^{(\rho ,\epsilon _a\epsilon _b)}A_{}^{(l)}{}_{a}{}^{a}q^{h_{\epsilon _a}}+q^{(\epsilon _b,\epsilon _b)}A_{}^{(l)}{}_{a}{}^{b}X_b^a`$
$`+(qq^1){\displaystyle \underset{\epsilon _a<\epsilon _c<\epsilon _b}{}}\left((1)^{[c]}A_{}^{(l)}{}_{a}{}^{c}X_c^a\delta _{\overline{c}}^b(1)^{[b][c]}\xi _b\xi _cq^{(\rho ,\epsilon _c\epsilon _b)}A_{}^{(l)}{}_{a}{}^{\overline{b}}X_c^a\right)`$
$`=`$ $`(1)^{[a]}q^{h_{\epsilon _a}}A_{}^{(l)}{}_{b}{}^{b}\delta _{\overline{b}}^a(1)^{[a][b]}\xi _a\xi _bq^{(\rho ,\epsilon _a\epsilon _b)}q^{h_{\epsilon _a}}A_{}^{(l)}{}_{b}{}^{b}+(1)^{[a]+[b]}q^{(\epsilon _a,\epsilon _b)}X_b^aA_{}^{(l)}{}_{a}{}^{b}`$
$`(qq)^1\delta _{\overline{b}}^a{\displaystyle \underset{\epsilon _a<\epsilon _c<\epsilon _b}{}}(1)^{[b][c]}\xi _b\xi _cq^{(\rho ,\epsilon _c\epsilon _b)}X_c^aA_{}^{(l)}{}_{\overline{c}}{}^{b}.`$
Simplifying gives
$`(1)^{[a]+[b]}q^{(\epsilon _a,\epsilon _b)}X_b^a`$ $`A_{}^{(l)}{}_{a}{}^{b}q^{(\epsilon _b,\epsilon _b)}A_{}^{(l)}{}_{a}{}^{b}X_b^a`$
$`=`$ $`\left((1)^{[a]}\delta _{\overline{b}}^aq^{(\rho ,\epsilon _a\epsilon _b)}\right)q^{h_{\epsilon _a}}(A_{}^{(l)}{}_{a}{}^{a}A_{}^{(l)}{}_{b}{}^{b})`$
$`+(qq^1){\displaystyle \underset{\epsilon _a<\epsilon _c<\epsilon _b}{}}\left((1)^{[c]}\delta _{\overline{c}}^bq^{(\rho ,\epsilon _c\epsilon _b)}\right)A_{}^{(l)}{}_{a}{}^{c}X_c^a`$
$`+(qq^1)\delta _{\overline{b}}^a{\displaystyle \underset{\epsilon _a<\epsilon _c<\epsilon _b}{}}(1)^{[b][c]}\xi _b\xi _cq^{(\rho ,\epsilon _c\epsilon _b)}X_c^aA_{}^{(l)}{}_{\overline{c}}{}^{\overline{a}}.`$
Remembering that $`\epsilon _a<\epsilon _b`$, we apply this to the highest weight state $`|\mathrm{\Lambda }`$ to obtain
$$\begin{array}{c}q^{(\epsilon _b,\epsilon _b)}A_{}^{(l)}{}_{a}{}^{b}X_b^a|\mathrm{\Lambda }=q^{(\mathrm{\Lambda },\epsilon _a)}\left((1)^{[a]}\delta _{\overline{b}}^aq^{2(\rho ,\epsilon _a)}\right)(t_a^{(l)}t_b^{(l)})|\mathrm{\Lambda }\hfill \\ \hfill +(qq^1)\underset{\epsilon _a<\epsilon _c<\epsilon _b}{}\left((1)^{[c]}\delta _{\overline{c}}^bq^{2(\rho ,\epsilon _c)}\right)A_{}^{(l)}{}_{a}{}^{c}X_c^a|\mathrm{\Lambda }.\end{array}$$
(5)
The next step is to calculate $`A_{}^{(l)}{}_{a}{}^{b}X_b^a|\mathrm{\Lambda }`$ for $`\epsilon _a<\epsilon _b`$. It is first convenient to order the indices according to $`b>c\epsilon _b<\epsilon _c`$. With this ordering we say an element $`a>0`$ if $`\epsilon _a<0`$, $`a=0`$ if $`\epsilon _a=0`$, and $`a<0`$ if $`\epsilon _a>0`$. Using this convention, it is apparent the solution to (5) will be of the form
$$A_{}^{(l)}{}_{a}{}^{b}X_b^a|\mathrm{\Lambda }=q^{(\mathrm{\Lambda },\epsilon _a)}(1)^{[a]}\underset{a>cb}{}\alpha _{bc}^a(t_a^{(l)}t_c^{(l)})|\mathrm{\Lambda },$$
(6)
where $`\alpha _{bc}^a`$ is a function of $`a,b`$ and $`c`$. Now from equation (5) we have
$`(qq^1){\displaystyle \underset{a>c>b}{}}`$ $`(1)^{[c]}A_{}^{(l)}{}_{a}{}^{c}X_c^a|\mathrm{\Lambda }`$
$`=q^{(\epsilon _b,\epsilon _b)}A_{}^{(l)}{}_{a}{}^{b}X_b^a|\mathrm{\Lambda }+(qq^1){\displaystyle \underset{a>c>b}{}}\delta _{\overline{c}}^bq^{2(\rho ,\epsilon _b)}A_{}^{(l)}{}_{a}{}^{c}X_c^a|\mathrm{\Lambda }`$
$`(1)^{[a]}q^{(\mathrm{\Lambda },\epsilon _a)}\left(1\delta _{\overline{b}}^a(1)^{[a]}q^{2(\rho ,\epsilon _a)}\right)(t_a^{(l)}t_b^{(l)})|\mathrm{\Lambda }`$
$`=q^{(\epsilon _{b+1},\epsilon _{b+1})}A_{}^{(l)}{}_{a}{}^{b+1}X_{b+1}^a|\mathrm{\Lambda }`$
$`+(qq^1){\displaystyle \underset{a>c>b+1}{}}\delta _{\overline{c}}^{b+1}q^{2(\rho ,\epsilon _{b+1})}A_{}^{(l)}{}_{a}{}^{c}X_c^a|\mathrm{\Lambda }`$
$`(1)^{[a]}q^{(\mathrm{\Lambda },\epsilon _a)}\left(1\delta _{\overline{a}}^{b+1}(1)^{[a]}q^{2(\rho ,\epsilon _a)}\right)(t_a^{(l)}t_{b+1}^{(l)})|\mathrm{\Lambda }`$
$`+(qq^1)(1)^{[b+1]}A_{}^{(l)}{}_{a}{}^{b+1}X_{b+1}^a|\mathrm{\Lambda }.`$
Substituting in the form of the solution given in equation (6) produces
$`q^{(\epsilon _b,\epsilon _b)}`$ $`{\displaystyle \underset{a>db}{}}\alpha _{bd}^a(t_a^{(l)}t_d^{(l)})|\mathrm{\Lambda }`$
$`=(`$ $`q^{(\epsilon _{b+1},\epsilon _{b+1})}(qq^1)(1)^{[b+1]}){\displaystyle \underset{a>db+1}{}}\alpha _{(b+1)d}^a(t_a^{(l)}t_d^{(l)})|\mathrm{\Lambda }`$
$``$ $`\left(1\delta _{\overline{b}}^a(1)^{[a]}q^{2(\rho ,\epsilon _a)}\right)(t_a^{(l)}t_b^{(l)})|\mathrm{\Lambda }+\left(1\delta _{\overline{b+1}}^a(1)^{[a]}q^{2(\rho ,\epsilon _a)}\right)(t_a^{(l)}t_{b+1}^{(l)})|\mathrm{\Lambda }`$
$`+`$ $`(qq^1){\displaystyle \underset{a>c>b}{}}\delta _{\overline{c}}^bq^{2(\rho ,\epsilon _b)}{\displaystyle \underset{a>dc}{}}\alpha _{\overline{b}d}^a(t_a^{(l)}t_d^{(l)})|\mathrm{\Lambda }`$
$``$ $`(qq^1){\displaystyle \underset{a>c>b+1}{}}\delta _{\overline{c}}^{b+1}q^{2(\rho ,\epsilon _{b+1})}{\displaystyle \underset{a>dc}{}}\alpha _{(\overline{b+1})d}^a(t_a^{(l)}t_d^{(l)})|\mathrm{\Lambda }.`$ (7)
Set
$$\alpha _{bd}^a=\overline{\alpha }_{bd}(1\delta _{\overline{d}}^a(1)^{[a]}q^{2(\rho ,\epsilon _a)}).$$
Then from equation (7) we obtain
$$\overline{\alpha }_{bb}=q^{(\epsilon _b,\epsilon _b)}$$
and
$`\overline{\alpha }_{b(b+1)}`$ $`=q^{(\epsilon _b,\epsilon _b)}[(q^{(\epsilon _{b+1},\epsilon _{b+1})}(qq^1)(1)^{[b+1]})\overline{\alpha }_{(b+1)(b+1)}+1`$
$`+(qq^1)\delta _{\overline{b+1}}^bq^{2(\rho ,\epsilon _b)}\overline{\alpha }_{\overline{b}(b+1)}]`$
$`=q^{(\epsilon _b,\epsilon _b)(\epsilon _{b+1},\epsilon _{b+1})}(qq^1)\left((1)^{[b+1]}\delta _{\overline{b+1}}^bq^{2(\rho ,\epsilon _b)}\right).`$
To simplify this expression note that $`q^{2(\rho ,\epsilon _{b+1}\epsilon _b)}=q^{(\epsilon _b,\epsilon _b)(\epsilon _{b+1},\epsilon _{b+1})}`$ in all cases except for $`[b]=0,b=l,m=2l`$, in which case $`q^{2(\rho ,\epsilon _{b+1}\epsilon _b)}=q^2q^{(\epsilon _b,\epsilon _b)(\epsilon _{b+1},\epsilon _{b+1})}`$. However $`[b]=0,b=l,m=2l`$ if and only if $`\delta _{\overline{b+1}}^b=1`$, and in that case we find $`\overline{\alpha }_{b(b+1)}=0`$. Hence for all values of $`b`$ we can write
$$\overline{\alpha }_{b(b+1)}=(qq^1)q^{2(\rho ,\epsilon _b)}\left((1)^{[b+1]}q^{2(\rho ,\epsilon _{b+1})}\delta _{\overline{b+1}}^b\right).$$
Now that we have found $`\overline{\alpha }_{bb}`$ and $`\overline{\alpha }_{b(b+1)}`$, they can be used to calculate the remaining $`\overline{\alpha }_{bd}`$. From equation (7) we observe that if $`d>b+1`$ then
$`\overline{\alpha }_{bd}`$ $`=q^{(\epsilon _b,\epsilon _b)}\left(q^{(\epsilon _{b+1},\epsilon _{b+1})}(qq^1)(1)^{[b+1]}\right)\overline{\alpha }_{(b+1)d}`$
$`+(qq^1)q^{(\epsilon _b,\epsilon _b)}{\displaystyle \underset{dc>b}{}}\delta _{\overline{c}}^bq^{2(\rho ,\epsilon _b)}\overline{\alpha }_{\overline{b}d}`$
$`(qq^1)q^{(\epsilon _b,\epsilon _b)}{\displaystyle \underset{dc>b+1}{}}\delta _{\overline{c}}^{b+1}q^{2(\rho ,\epsilon _{b+1})}\overline{\alpha }_{(\overline{b+1})d}.`$ (8)
Now define $`\theta _{xy}`$ by
$$\theta _{xy}=\{\begin{array}{cc}1\hfill & x<y,\hfill \\ 0\hfill & xy.\hfill \end{array}$$
Then equation (8) can be rewritten as
$$\begin{array}{c}\overline{\alpha }_{bd}=q^{(\epsilon _b,\epsilon _b)}\left(q^{(\epsilon _{b+1},\epsilon _{b+1})}(qq^1)(1)^{[b+1]}\right)\overline{\alpha }_{(b+1)d}\hfill \\ \hfill +(qq^1)q^{(\epsilon _b,\epsilon _b)}q^{2(\rho ,\epsilon _c)}\left(\theta _{bc}\theta _{c(d+1)}\delta _{\overline{c}}^b\theta _{(b+1)c}\theta _{c(d+1)}\delta _{\overline{c}}^{b+1}\right)\overline{\alpha }_{cd},d>a+1.\end{array}$$
(9)
Consider $`\overline{\alpha }_{bd}`$ for any $`b>l`$. Both $`\theta _{b\overline{b}}`$ and $`\theta _{(b+1)(\overline{b+1})}`$ will equal $`0`$, so
$`\overline{\alpha }_{bd}`$ $`=q^{(\epsilon _b,\epsilon _b)}\left(q^{(\epsilon _{b+1},\epsilon _{b+1})}(qq^1)(1)^{[b+1]}\right)\overline{\alpha }_{(b+1)d}^a`$
$`=q^{(\epsilon _b,\epsilon _b)}q^{(\epsilon _{b+1},\epsilon _{b+1})}\overline{\alpha }_{(b+1)d}`$
$`=q^{2(\rho ,\epsilon _{b+1}\epsilon _b)}\overline{\alpha }_{(b+1)d}.`$
Since
$$\overline{\alpha }_{(d1)d}=(1)^{[d]}(qq^1)q^{2(\rho ,\epsilon _d\epsilon _{d1})},$$
we obtain
$$\overline{\alpha }_{bd}=(1)^{[d]}(qq^1)q^{2(\rho ,\epsilon _d\epsilon _b)},d>b>l.$$
Substituting this together with our expression for $`\overline{\alpha }_{bb}`$ into equation (9), we find
$`\overline{\alpha }_{bd}=`$ $`q^{(\epsilon _b,\epsilon _b)}\left(q^{(\epsilon _{b+1},\epsilon _{b+1})}\delta _{\overline{b+1}}^{b+1}(qq^1)\right)\overline{\alpha }_{(b+1)d}`$
$`+`$ $`(qq^1)^2q^{(\epsilon _b,\epsilon _b)}(1)^{[d]}q^{2(\rho ,\epsilon _d)}\left(\theta _{b\overline{b}}\theta _{\overline{b}d}\theta _{(b+1)(\overline{b+1})}\theta _{(\overline{b+1})d}\right)`$
$``$ $`(qq^1)q^{(\epsilon _b,\epsilon _b)}q^{(\epsilon _d,\epsilon _d)}q^{2(\rho ,\epsilon _d)}(\delta _d^{\overline{b}}\delta _d^{\overline{b+1}}),d>b+1.`$ (10)
But for $`d>b+1`$
$`\theta _{b\overline{b}}\theta _{\overline{b}d}\theta _{(b+1)(\overline{b+1})}\theta _{(\overline{b+1})d}`$ $`=\delta _l^b\theta _{\overline{l}d}\delta _d^{\overline{b}}\theta _{bl}`$
$`=\delta _l^b(1\delta _{\overline{l}}^d)\delta _d^{\overline{b}}(1\delta _l^b)`$
$`=\delta _l^b\delta _d^{\overline{b}}.`$
Also, $`[(1)^{[d]}(qq^1)+q^{(\epsilon _d,\epsilon _d)}]\delta _d^{\overline{b}}=q^{(\epsilon _d,\epsilon _d)}\delta _d^{\overline{b}}`$, so equation (10) reduces to
$`\overline{\alpha }_{bd}`$ $`=\left(q^{2(\rho ,\epsilon _{b+1}\epsilon _b)}q^{2\delta _{\overline{b+1}}^b}\delta _{\overline{b+1}}^{b+1}q^1(qq^1)\right)\overline{\alpha }_{(b+1)d}+\delta _l^bq^1(qq^1)^2(1)^{[d]}q^{2(\rho ,\epsilon _d)}`$
$`\delta _d^{\overline{b}}(qq^1)q^{2(\rho ,\epsilon _d)}+\delta _d^{\overline{b+1}}(qq^1)q^{2(\rho ,\epsilon _{b+1}\epsilon _b)}q^{2\delta _{\overline{b+1}}^b}q^{2(\rho ,\epsilon _d)}`$
$`=\left(q^{2(\rho ,\epsilon _{b+1}\epsilon _b)}q^{2\delta _{\overline{b+1}}^b}\delta _{\overline{b+1}}^{b+1}q^1(qq^1)\right)\overline{\alpha }_{(b+1)d}+\delta _l^bq^1(qq^1)^2(1)^{[d]}q^{2(\rho ,\epsilon _d)}`$
$`+(qq^1)q^{2(\rho ,\epsilon _b)}(\delta _d^{\overline{b+1}}\delta _d^{\overline{b}}),d>b+1.`$
Recall that for $`b>l`$ we have
$$\overline{\alpha }_{bd}=(1)^{[d]}(qq^1)q^{2(\rho ,\epsilon _d\epsilon _b)},d>b.$$
Then when $`b=l`$ we find
$`\overline{\alpha }_{bd}`$ $`=\left(q^{2(\rho ,\epsilon _{b+1}\epsilon _b)}q^{2\delta _{\overline{b+1}}^b}\delta _{\overline{b+1}}^{b+1}q^1(qq^1)\right)(1)^{[d]}(qq^1)q^{2(\rho ,\epsilon _d\epsilon _{b+1})}`$
$`+q^1(qq^1)^2(1)^{[d]}q^{2(\rho ,\epsilon _d)}(qq^1)q^{2(\rho ,\epsilon _b)}\delta _d^{\overline{l}}`$
$`=(1)^{[d]}(qq^1)q^{2(\rho ,\epsilon _d\epsilon _b)}`$
$`[\delta _{\overline{b+1}}^{b+1}(1(qq^1)+(qq^1))+\delta _{\overline{b+1}}^b(q^2+q^1(qq^1)]`$
$`(qq^1)q^{2(\rho ,\epsilon _b)}\delta _d^{\overline{l}}`$
$`=(qq^1)q^{2(\rho ,\epsilon _b)}\left((1)^{[d]}q^{2(\rho ,\epsilon _d)}\delta _{\overline{d}}^b\right)`$
for all $`d>b+1`$. Comparing this with our earlier results for $`d=b+1`$ and $`b>l`$, we have
$$\overline{\alpha }_{bd}=(qq^1)q^{2(\rho ,\epsilon _b)}\left((1)^{[d]}q^{2(\rho ,\epsilon _d)}\delta _{\overline{d}}^b\right),bl,d>b.$$
But for $`b<l`$ we know
$`\overline{\alpha }_{bd}`$ $`=q^{2(\rho ,\epsilon _{b+1}\epsilon _b)}\overline{\alpha }_{(b+1)d}+(qq^1)q^{2(\rho ,\epsilon _b)}(\delta _d^{\overline{b+1}}\delta _d^{\overline{b}}),d>b+1.`$
Hence for all $`b`$ we obtain
$`\overline{\alpha }_{bd}`$ $`=(qq^1)q^{2(\rho ,\epsilon _b)}\left((1)^{[d]}q^{2(\rho ,\epsilon _d)}{\displaystyle \underset{c=b}{\overset{d1}{}}}\delta _d^{\overline{c}}+{\displaystyle \underset{c=b}{\overset{d2}{}}}\delta _d^{\overline{c+1}}\right)`$
$`=(qq^1)q^{2(\rho ,\epsilon _b)}\left((1)^{[d]}q^{2(\rho ,\epsilon _d)}\delta _d^{\overline{b}}\right),d>b.`$
Thus for all $`a>b`$
$$A_{}^{(l)}{}_{a}{}^{b}X_b^a|\mathrm{\Lambda }=q^{(\mathrm{\Lambda },\epsilon _a)}(1)^{[a]}\underset{a>cb}{}\alpha _{bc}^a(t_a^{(l)}t_c^{(l)})|\mathrm{\Lambda },$$
(11)
where $`\alpha _{bc}^a`$ is given by
$$\alpha _{bc}^a=\{\begin{array}{cc}q^{(\epsilon _b,\epsilon _b)}(1\delta _{\overline{b}}^a(1)^{[a]}q^{2(\rho ,\epsilon _a)}),\hfill & c=b,\hfill \\ (qq^1)q^{2(\rho ,\epsilon _b)}\left((1)^{[c]}q^{2(\rho ,\epsilon _c)}\delta _c^{\overline{b}}\right)(1\delta _{\overline{c}}^a(1)^{[a]}q^{2(\rho ,\epsilon _a)}),\hfill & c>b.\hfill \end{array}$$
### 4.1 Constructing the Perelomov-Popov matrix equation
The expression (11) can now be substituted into the equation
$$t_a^{(l)}|\mathrm{\Lambda }=t_a^{(l1)}t_a^{(1)}|\mathrm{\Lambda }+\underset{\epsilon _a<\epsilon _b}{}(1)^{[a]+[b]}q^{(\mathrm{\Lambda },\epsilon _a)}A_{}^{(l1)}{}_{a}{}^{b}X_b^a|\mathrm{\Lambda }$$
to find a matrix equation for the various $`t_a^{(l)}`$. The matrix factor is an analogue of the Perelomov-Popov matrix introduced in and , which was used to calculate the eigenvalues of the Casimir invariants of various classical Lie algebras.
First recall that
$$A_a^b=(1+\delta _b^a)X_a^b+(qq^1)\underset{ca,b}{}(1)^{([a]+[c])([b]+[c])}X_a^cX_c^b,$$
where
$$X_a^b=\{\begin{array}{cc}\frac{q^{h_{\epsilon _a}}I}{qq^1},\hfill & a=b,\hfill \\ (1)^{[b]}q^{h_{\epsilon _a}}\widehat{\sigma }_{ba},\hfill & \epsilon _a<\epsilon _b,\hfill \\ (1)^{[b]}\widehat{\sigma }_{ba}q^{h_{\epsilon _b}},\hfill & \epsilon _a>\epsilon _b.\hfill \end{array}$$
Then
$`A_a^a|\mathrm{\Lambda }`$ $`=2X_a^a|\mathrm{\Lambda }+(qq^1)X_a^aX_a^a|\mathrm{\Lambda }`$
$`=(qq^1)^1(2(q^{h_{\epsilon _a}}1)+(q^{h_{\epsilon _a}}1)^2)|\mathrm{\Lambda }.`$
$``$ $`t_a^{(1)}`$ $`={\displaystyle \frac{q^{2(\mathrm{\Lambda },\epsilon _a)}1}{qq^1}}.`$
Hence we obtain
$`t_a^{(l)}`$ $`={\displaystyle \frac{(q^{2(\mathrm{\Lambda },\epsilon _a)}1)}{(qq^1)}}t_a^{(l1)}`$
$`+{\displaystyle \underset{b<a}{}}(1)^{[a]+[b]}q^{(\mathrm{\Lambda },\epsilon _a)}\left(q^{(\mathrm{\Lambda },\epsilon _a)}(1)^{[a]}{\displaystyle \underset{bc<a}{}}\alpha _{bc}^a(t_a^{(l1)}t_c^{(l1)})\right)`$
$`={\displaystyle \frac{(q^{2(\mathrm{\Lambda },\epsilon _a)}1)}{(qq^1)}}t_a^{(l1)}`$
$`q^{2(\mathrm{\Lambda },\epsilon _a)}{\displaystyle \underset{b<a}{}}(1)^{[b]}q^{(\epsilon _b,\epsilon _b)}(1\delta _{\overline{b}}^a(1)^{[a]}q^{2(\rho ,\epsilon _a)})(t_a^{(l1)}t_b^{(l1)})`$
$`+(qq^1)q^{2(\mathrm{\Lambda },\epsilon _a)}{\displaystyle \underset{c<b<a}{}}(1)^{[c]}q^{2(\rho ,\epsilon _c)}(1\delta _{\overline{b}}^a(1)^{[a]}q^{2(\rho ,\epsilon _a)})`$
$`((1)^{[b]}q^{2(\rho ,\epsilon _b)}\delta _{\overline{c}}^b)(t_a^{(l1)}t_b^{(l1)}).`$
Now consider the function $`\gamma _b`$ defined by:
$$\gamma _b=(1)^{[b]}q^{(\epsilon _b,\epsilon _b)}(qq^1)\underset{c<b}{}(1)^{[c]}q^{2(\rho ,\epsilon _c)}((1)^{[b]}q^{2(\rho ,\epsilon _b)}\delta _{\overline{c}}^b)).$$
We evaluate this for all $`b`$, remembering that $`C(\mathrm{\Lambda }_0)=(\delta _1,\delta _1+2\rho )=mn1`$ and
$$\rho =\frac{1}{2}\underset{i=1}{\overset{l}{}}(m2i)\epsilon _i+\frac{1}{2}\underset{\mu =1}{\overset{k}{}}(nm+22\mu )\delta _\mu .$$
We find
$$\gamma _b=(1)^{[b]}q^{2(\rho ,\epsilon _b)}q^{C(\mathrm{\Lambda }_0)}$$
for all values of $`b`$. We also consider the function
$$\beta _a=1(qq^1)\underset{b<a}{}\gamma _b\left(1\delta _{\overline{b}}^a(1)^{[a]}q^{2(\rho ,\epsilon _a)}\right),$$
so that
$$t_a^{(l)}=\frac{(q^{2(\mathrm{\Lambda },\epsilon _a)}\beta _a1)}{(qq^1)}t_a^{(l1)}+q^{2(\mathrm{\Lambda },\epsilon _a)}\underset{b<a}{}\gamma _b\left(1\delta _{\overline{b}}^a(1)^{[a]}q^{2(\rho ,\epsilon _a)}\right)t_b^{(l1)}.$$
(12)
Again, by considering the various cases individually we find
$$\beta _a=q^{(\epsilon _a,2\rho +\epsilon _a)C(\mathrm{\Lambda }_0)}$$
for any $`a`$, regardless of whether $`m`$ is even or odd. Substituting this result together with that for $`\gamma _b`$ into equation (12) gives
$$\begin{array}{c}t_a^{(l)}=\frac{(q^{(\epsilon _a,2\mathrm{\Lambda }+2\rho +\epsilon _a)C(\mathrm{\Lambda }_0)}1)}{(qq^1)}t_a^{(l1)}\hfill \\ \hfill +q^{(2\mathrm{\Lambda },\epsilon _a)C(\mathrm{\Lambda }_0)}\underset{b<a}{}(1)^{[b]}q^{(2\rho ,\epsilon _b)}\left(1\delta _{\overline{b}}^a(1)^{[a]}q^{(2\rho ,\epsilon _a)}\right)t_b^{(l1)}.\end{array}$$
This can be written in the matrix form
$$\underset{¯}{t}^{(l)}=M\underset{¯}{t}^{(l1)},$$
where $`M`$ is a lower triangular matrix with entries
$$M_{ab}=\{\begin{array}{cc}0,\hfill & a<b\hfill \\ (qq^1)^1(q^{(\epsilon _a,2\mathrm{\Lambda }+2\rho +\epsilon _a)C(\mathrm{\Lambda }_0)}1),\hfill & a=b,\hfill \\ q^{(2\mathrm{\Lambda },\epsilon _a)C(\mathrm{\Lambda }_0)}\left((1)^{[b]}q^{(2\rho ,\epsilon _b)}\delta _{\overline{b}}^a\right),\hfill & a>b.\hfill \end{array}$$
Then we have
$$\underset{¯}{t}^{(l)}=M^l\underset{¯}{t}^{(0)},\text{with}t_a^{(0)}=1a,$$
where $`M`$ is an analogue of the Perelomov-Popov matrix.
### 4.2 Solving the matrix equation
This matrix equation for $`t_a^{(l)}`$ can now be used to calculate the eigenvalues of $`C_l`$. Loosely speaking, the problem reduces to diagonalising the matrix $`M`$. Recall
$$C_l=\underset{a}{}(1)^{[a]}q^{(2\rho ,\epsilon _a)}A_{}^{(l)}{}_{a}{}^{a}.$$
Denote the eigenvalue of $`C_l`$ on $`V(\mathrm{\Lambda })`$ as $`\chi _\mathrm{\Lambda }(C_l)`$. Then we have
$$\chi _\mathrm{\Lambda }(C_l)=\underset{a}{}(1)^{[a]}q^{(2\rho ,\epsilon _a)}t_a^{(l)}=\underset{a,b}{}(1)^{[a]}q^{(2\rho ,\epsilon _a)}(M^l)_{ab}.$$
To calculate this we wish to diagonalise $`M`$. We assume the eigenvalues of $`M`$,
$$\alpha _a^\mathrm{\Lambda }=\frac{(q^{(\epsilon _a,2\mathrm{\Lambda }+2\rho +\epsilon _a)C(\mathrm{\Lambda }_0)}1)}{(qq^1)},$$
are distinct. Then we need a matrix $`N`$ satisfying
$$(N^1MN)_{ab}=\delta _b^a\alpha _a^\mathrm{\Lambda },$$
which implies
$$\chi _\mathrm{\Lambda }(C_l)=\underset{a,b,c}{}(1)^{[a]}q^{(2\rho ,\epsilon _a)}(\alpha _b^\mathrm{\Lambda })^lN_{ab}(N^1)_{bc}.$$
(13)
Now
$$(MN)_{ab}=\alpha _b^\mathrm{\Lambda }N_{ab}.$$
Substituting in the values for $`M_{ab}`$ gives
$$\alpha _a^\mathrm{\Lambda }N_{ab}+q^{(2\mathrm{\Lambda },\epsilon _a)C(\mathrm{\Lambda }_0)}\underset{c<a}{}\left((1)^{[c]}q^{(2\rho ,\epsilon _c)}\delta _{\overline{c}}^a\right)N_{cb}=\alpha _b^\mathrm{\Lambda }N_{ab}.$$
(14)
Since the eigenvalues $`\alpha _a^\mathrm{\Lambda }`$ are distinct, this implies
$$N_{ab}=0,a<b.$$
Set
$$P_{ab}=\underset{ca}{}(1)^{[c]}q^{(2\rho ,\epsilon _c)}N_{cb}.$$
(15)
Then equation (14) becomes
$`(\alpha _b^\mathrm{\Lambda }\alpha _a^\mathrm{\Lambda })N_{ab}=q^{(2\mathrm{\Lambda },\epsilon _a)C(\mathrm{\Lambda }_0)}P_{(a1)b}\theta _{0a}q^{(2\mathrm{\Lambda },\epsilon _a)C(\mathrm{\Lambda }_0)}N_{\overline{a}b}`$
$``$ $`(\alpha _b^\mathrm{\Lambda }\alpha _a^\mathrm{\Lambda })(1)^{[a]}q^{(2\rho ,\epsilon _a)}(P_{ab}P_{(a1)b})`$
$`=q^{(2\mathrm{\Lambda },\epsilon _a)C(\mathrm{\Lambda }_0)}P_{a1b}\theta _{0a}q^{(2\mathrm{\Lambda },\epsilon _a)C(\mathrm{\Lambda }_0)}N_{\overline{a}b},`$
which simplifies to
$$P_{ab}=\frac{(\alpha _b^\mathrm{\Lambda }\alpha _a^\mathrm{\Lambda }+(1)^{[a]}q^{2(\mathrm{\Lambda }+\rho ,\epsilon _a)C(\mathrm{\Lambda }_0)})}{(\alpha _b^\mathrm{\Lambda }\alpha _a^\mathrm{\Lambda })}P_{(a1)b}\frac{\theta _{0a}(1)^{[a]}q^{2(\mathrm{\Lambda }+\rho ,\epsilon _a)C(\mathrm{\Lambda }_0)}}{(\alpha _b^\mathrm{\Lambda }\alpha _a^\mathrm{\Lambda })}N_{\overline{a}b}.$$
Set
$$\psi _a^b=\alpha _b^\mathrm{\Lambda }\alpha _a^\mathrm{\Lambda }+(1)^{[a]}q^{2(\mathrm{\Lambda }+\rho ,\epsilon _a)C(\mathrm{\Lambda }_0)},$$
so this becomes
$$P_{ab}=\frac{\psi _a^b}{(\alpha _b^\mathrm{\Lambda }\alpha _a^\mathrm{\Lambda })}P_{(a1)b}\frac{\theta _{0a}(1)^{[a]}q^{2(\mathrm{\Lambda }+\rho ,\epsilon _a)C(\mathrm{\Lambda }_0)}}{(\alpha _b^\mathrm{\Lambda }\alpha _a^\mathrm{\Lambda })}N_{\overline{a}b}.$$
(16)
Without loss of generality we can choose $`N_{aa}=1a`$, so $`P_{bb}=(1)^{[b]}q^{2(\rho ,\epsilon _b)}`$. Then in the cases $`0a>b`$ and $`a>b0`$ the last term in equation (16) vanishes, giving
$$P_{ab}=(1)^{[b]}q^{2(\rho ,\epsilon _b)}\underset{c=b+1}{\overset{a}{}}\frac{\psi _c^b}{(\alpha _b^\mathrm{\Lambda }\alpha _c^\mathrm{\Lambda })}.$$
Similarly, for $`a>\overline{b}>0`$ we obtain
$$P_{ab}=P_{\overline{b}b}\underset{c=\overline{b}+1}{\overset{a}{}}\frac{\psi _c^b}{(\alpha _b^\mathrm{\Lambda }\alpha _c^\mathrm{\Lambda })}.$$
(17)
It remains to find $`P_{ab}`$ for $`\overline{b}a>0`$. In this case, the last term in equation (16) contributes, giving
$$\begin{array}{c}P_{ab}=(1)^{[b]}q^{2(\rho ,\epsilon _b)}\underset{c=b+1}{\overset{a}{}}\frac{\psi _c^b}{(\alpha _b^\mathrm{\Lambda }\alpha _c^\mathrm{\Lambda })}\frac{(1)^{[a]}q^{2(\mathrm{\Lambda }+\rho ,\epsilon _a)C(\mathrm{\Lambda }_0)}}{(\alpha _b^\mathrm{\Lambda }\alpha _a^\mathrm{\Lambda })}N_{\overline{a}b}\hfill \\ \hfill \underset{d=\overline{l}}{\overset{a1}{}}\frac{(1)^{[d]}q^{2(\mathrm{\Lambda }+\rho ,\epsilon _d)C(\mathrm{\Lambda }_0)}}{(\alpha _b^\mathrm{\Lambda }\alpha _d^\mathrm{\Lambda })}N_{\overline{d}b}\underset{c=d+1}{\overset{a}{}}\frac{\psi _c^b}{(\alpha _b^\mathrm{\Lambda }\alpha _c^\mathrm{\Lambda })}.\end{array}$$
(18)
Recall that if $`b<a<0`$, then
$`N_{ab}`$ $`={\displaystyle \frac{q^{(2\mathrm{\Lambda },\epsilon _a)C(\mathrm{\Lambda }_0)}}{(\alpha _b^\mathrm{\Lambda }\alpha _a^\mathrm{\Lambda })}}P_{(a1)b}`$
$`={\displaystyle \frac{(1)^{[b]}q^{2(\mathrm{\Lambda },\epsilon _a)+2(\rho ,\epsilon _b)C(\mathrm{\Lambda }_0)}}{(\alpha _b^\mathrm{\Lambda }\alpha _a^\mathrm{\Lambda })}}{\displaystyle \underset{c=b+1}{\overset{a1}{}}}{\displaystyle \frac{\psi _c^b}{(\alpha _b^\mathrm{\Lambda }\alpha _c^\mathrm{\Lambda })}}.`$
Substituting this into equation (18), we find
$`P_{\overline{b}b}`$ $`=(1)^{[b]}q^{2(\rho ,\epsilon _b)}{\displaystyle \underset{c=b+1}{\overset{\overline{b}}{}}}{\displaystyle \frac{\psi _c^b}{(\alpha _b^\mathrm{\Lambda }\alpha _c^\mathrm{\Lambda })}}{\displaystyle \frac{(1)^{[b]}q^{2(\mathrm{\Lambda }+\rho ,\epsilon _b)C(\mathrm{\Lambda }_0)}}{(\alpha _b^\mathrm{\Lambda }\alpha _{\overline{b}}^\mathrm{\Lambda })}}`$
$`{\displaystyle \underset{d=\overline{l}}{\overset{\overline{b}1}{}}}{\displaystyle \frac{(1)^{[d]+[b]}q^{2(\rho ,\epsilon _d+\epsilon _b)2C(\mathrm{\Lambda }_0)}}{(\alpha _b^\mathrm{\Lambda }\alpha _d^\mathrm{\Lambda })(\alpha _b^\mathrm{\Lambda }\alpha _{\overline{d}}^\mathrm{\Lambda })}}{\displaystyle \underset{c=b+1}{\overset{\overline{d}1}{}}}{\displaystyle \frac{\psi _c^b}{(\alpha _b^\mathrm{\Lambda }\alpha _c^\mathrm{\Lambda })}}{\displaystyle \underset{c=d+1}{\overset{\overline{b}}{}}}{\displaystyle \frac{\psi _c^b}{(\alpha _b^\mathrm{\Lambda }\alpha _c^\mathrm{\Lambda })}},`$
which can also be simplified to
$$P_{\overline{b}b}\underset{c=b+1}{\overset{\overline{b}}{}}\frac{(\alpha _b^\mathrm{\Lambda }\alpha _c^\mathrm{\Lambda })}{\psi _c^b}=(1)^{[b]}q^{2(\rho ,\epsilon _b)}\left[1\underset{d=\overline{l}}{\overset{\overline{b}}{}}\frac{(1)^{[d]}q^{2(\rho ,\epsilon _d)2C(\mathrm{\Lambda }_0)}}{\psi _d^b\psi _{\overline{d}}^b}\underset{c=\overline{d}+1}{\overset{d1}{}}\frac{(\alpha _b^\mathrm{\Lambda }\alpha _c^\mathrm{\Lambda })}{\psi _c^b}\right].$$
(19)
From this point we will consider the case $`m=2l+1`$. This is marginally more complicated than the case with even $`m`$. Define $`\mathrm{\Phi }_d^b`$ to be
$`\mathrm{\Phi }_d^b`$ $`={\displaystyle \underset{c=\overline{l}}{\overset{d1}{}}}{\displaystyle \frac{(\alpha _b\alpha _c)(\alpha _b\alpha _{\overline{c}})}{\psi _c^b\psi _{\overline{c}}^b}}`$
$`={\displaystyle \frac{(\alpha _b\alpha _{d1})(\alpha _b\alpha _{\overline{d1}})}{\psi _{d1}^b\psi _{\overline{d1}}^b}}\mathrm{\Phi }_{d1}^b,\mathrm{\Phi }_{\overline{l}}^b=1.`$
Then $`P_{\overline{b}b}`$ can be written as
$$P_{\overline{b}b}=(1)^{[b]}q^{2(\rho ,\epsilon _b)}\underset{_{c0}^{c=b+1}}{\overset{\overline{b}}{}}\frac{\psi _c^b}{(\alpha _b^\mathrm{\Lambda }\alpha _c^\mathrm{\Lambda })}\left[\frac{\psi _0^b}{\alpha _b\alpha _0}\underset{d=\overline{l}}{\overset{\overline{b}}{}}\frac{(1)^{[d]}q^{2(\rho ,\epsilon _d)2C(\mathrm{\Lambda }_0)}}{\psi _d^b\psi _{\overline{d}}^b}\mathrm{\Phi }_d^b\right].$$
Note that for $`c0`$,
$`\psi _c^b`$ $`={\displaystyle \frac{q^{C(\mathrm{\Lambda }_0)}}{(qq^1)}}\left(q^{(\epsilon _b,2\rho +2\mathrm{\Lambda }+\epsilon _b)}q^{(\epsilon _c,2\rho +2\mathrm{\Lambda }+\epsilon _c)}+(qq^1)(1)^{[c]}q^{(\epsilon _c,2\rho +2\mathrm{\Lambda })}\right)`$
$`={\displaystyle \frac{q^{C(\mathrm{\Lambda }_0)}\stackrel{~}{\psi }_c^b}{(qq^1)}}`$
where
$$\stackrel{~}{\psi }_c^b=q^{(\epsilon _b,2\rho +2\mathrm{\Lambda }+\epsilon _b)}q^{(\epsilon _c,2\rho +2\mathrm{\Lambda }\epsilon _c)}.$$
So
$`{\displaystyle \underset{d=\overline{l}}{\overset{\overline{b}}{}}}{\displaystyle \frac{(1)^{[d]}q^{2(\rho ,\epsilon _d)2C(\mathrm{\Lambda }_0)}}{\psi _d^b\psi _{\overline{d}}^b}}\mathrm{\Phi }_d^b`$ $`=(qq^1){\displaystyle \underset{d=\overline{l}}{\overset{\overline{b}}{}}}{\displaystyle \frac{(1)^{[d]}(qq^1)q^{2(\rho ,\epsilon _d)}}{\stackrel{~}{\psi }_d^b\stackrel{~}{\psi }_{\overline{d}}^b}}\mathrm{\Phi }_d^b`$
$`=(qq^1){\displaystyle \underset{d=\overline{l}}{\overset{\overline{b}}{}}}{\displaystyle \frac{(q^{2(\epsilon _d,\epsilon _d)}1)q^{2(\rho ,\epsilon _d)(\epsilon _d,\epsilon _d)}}{\stackrel{~}{\psi }_d^b\stackrel{~}{\psi }_{\overline{d}}^b}}\mathrm{\Phi }_d^b`$ (20)
and
$`\mathrm{\Phi }_{d+1}^b`$ $`={\displaystyle \frac{(\alpha _b\alpha _d)(\alpha _b\alpha _{\overline{d}})}{\psi _d^b\psi _{\overline{d}}^b}}\mathrm{\Phi }_d^b`$
$`={\displaystyle \frac{(q^{(\epsilon _b,\epsilon _b+2\rho +2\mathrm{\Lambda })}q^{(\epsilon _d,\epsilon _d+2\rho +2\mathrm{\Lambda })})(q^{(\epsilon _b,\epsilon _b+2\rho +2\mathrm{\Lambda })}q^{(\epsilon _d,\epsilon _d2\rho 2\mathrm{\Lambda })})}{\stackrel{~}{\psi }_d^b\stackrel{~}{\psi }_{\overline{d}}^b}}\mathrm{\Phi }_d^b`$
for $`d\overline{l}`$. Now
$`(q^{(\epsilon _b,\epsilon _b+2\rho +2\mathrm{\Lambda })}q^{(\epsilon _d,\epsilon _d+2\rho +2\mathrm{\Lambda })})(q^{(\epsilon _b,\epsilon _b+2\rho +2\mathrm{\Lambda })}q^{(\epsilon _d,\epsilon _d2\rho 2\mathrm{\Lambda })})`$
$`=q^{2(\epsilon _d,\epsilon _d)}(q^{(\epsilon _b,\epsilon _b+2\rho +2\mathrm{\Lambda })}q^{(\epsilon _d,\epsilon _d+2\rho +2\mathrm{\Lambda })})(q^{(\epsilon _b,\epsilon _b+2\rho +2\mathrm{\Lambda })}q^{(\epsilon _d,\epsilon _d+2\rho +2\mathrm{\Lambda })})`$
$`+q^{2(\epsilon _b,\epsilon _b+2\rho +2\mathrm{\Lambda })}(1q^{2(\epsilon _d,\epsilon _d)})+q^{2(\epsilon _d,\epsilon _d)}1`$
$`=q^{2(\epsilon _d,\epsilon _d)}\stackrel{~}{\psi }_d^b\stackrel{~}{\psi }_{\overline{d}}^b(q^{2(\epsilon _b,\epsilon _b+2\rho +2\mathrm{\Lambda })}1)(q^{2(\epsilon _d,\epsilon _d)}1).`$
Then, for $`d\overline{l}`$,
$$\frac{\mathrm{\Phi }_{d+1}^b}{(q^{2(\epsilon _b,\epsilon _b+2\rho +2\mathrm{\Lambda })}1)}=\left[\frac{q^{2(\epsilon _d,\epsilon _d)}}{(q^{2(\epsilon _b,\epsilon _b+2\rho +2\mathrm{\Lambda })}1)}\frac{(q^{2(\epsilon _d,\epsilon _d)}1)}{\stackrel{~}{\psi }_d^b\stackrel{~}{\psi }_{\overline{d}}^b}\right]\mathrm{\Phi }_d^b.$$
(21)
Now for $`d=\overline{b}`$
$`{\displaystyle \frac{(q^{2(\epsilon _d,\epsilon _d)}1)q^{2(\rho ,\epsilon _d)(\epsilon _d,\epsilon _d)}}{\stackrel{~}{\psi }_d^b\stackrel{~}{\psi }_{\overline{d}}^b}}`$
$`={\displaystyle \frac{(q^{2(\epsilon _b,\epsilon _b)}1)q^{2(\rho ,\epsilon _{\overline{b}})(\epsilon _b,\epsilon _b)}}{(q^{(\epsilon _b,2\rho +2\mathrm{\Lambda }+\epsilon _b)}q^{(\epsilon _b,2\rho +2\mathrm{\Lambda }+\epsilon _b)})q^{(\epsilon _b,2\rho +2\mathrm{\Lambda })}(q^{(\epsilon _b,\epsilon _b)}q^{(\epsilon _b,\epsilon _b)})}}`$
$`={\displaystyle \frac{q^{2(\rho ,\epsilon _{\overline{b}})+(\epsilon _b,\epsilon _b)}}{(q^{2(\epsilon _b,\epsilon _b+2\rho +2\mathrm{\Lambda })}1)}},`$
which can be written as
$$\frac{(q^{2(\epsilon _d,\epsilon _d)}1)q^{2(\rho ,\epsilon _d)(\epsilon _d,\epsilon _d)}}{\stackrel{~}{\psi }_d^b\stackrel{~}{\psi }_{\overline{d}}^b}=\frac{q^{2(\rho ,\epsilon _{\overline{b}1})(\epsilon _{\overline{b}1},\epsilon _{\overline{b}1})}}{(q^{2(\epsilon _b,\epsilon _b+2\rho +2\mathrm{\Lambda })}1)}$$
when $`b<l`$. Hence equation (21) can be used to pairwise cancel the terms in the sum in equation (20). Adding the first two terms ($`d=\overline{b},\overline{b}1`$), we find:
$`q^{2(\rho ,\epsilon _{\overline{b}1})(\epsilon _{\overline{b}1},\epsilon _{\overline{b}1})}[{\displaystyle \frac{\mathrm{\Phi }_{\overline{b}}^b}{(q^{2(\epsilon _b,\epsilon _b+2\rho +2\mathrm{\Lambda })}1)}}`$ $`+{\displaystyle \frac{(q^{2(\epsilon _{\overline{b}1},\epsilon _{\overline{b}1})}1)}{\stackrel{~}{\psi }_{\overline{b}1}^b\stackrel{~}{\psi }_{b+1}^b}}\mathrm{\Phi }_{\overline{b}1}^b]`$
$`=q^{2(\rho ,\epsilon _{\overline{b}1})(\epsilon _{\overline{b}1},\epsilon _{\overline{b}1})}{\displaystyle \frac{q^{2(\epsilon _{\overline{b}1},\epsilon _{\overline{b}1})}}{(q^{2(\epsilon _b,\epsilon _b+2\rho +2\mathrm{\Lambda })}1)}}\mathrm{\Phi }_{\overline{b}1}^b`$
$`={\displaystyle \frac{q^{2(\rho ,\epsilon _{\overline{b}2})(\epsilon _{\overline{b}2},\epsilon _{\overline{b}2})}}{(q^{2(\epsilon _b,\epsilon _b+2\rho +2\mathrm{\Lambda })}1)}}\mathrm{\Phi }_{\overline{b}1}^b.`$
Continuing to apply equation (21) in this manner gives
$`{\displaystyle \underset{d=\overline{l}}{\overset{\overline{b}}{}}}{\displaystyle \frac{(q^{2(\epsilon _d,\epsilon _d)}1)q^{2(\rho ,\epsilon _d)(\epsilon _d,\epsilon _d)}}{\stackrel{~}{\psi }_d^b\stackrel{~}{\psi }_{\overline{d}}^b}}\mathrm{\Phi }_d^b`$ $`={\displaystyle \frac{q^{2(\rho ,\epsilon _{\overline{l}})+(\epsilon _l,\epsilon _l)}}{(q^{2(\epsilon _b,\epsilon _b+\rho +\mathrm{\Lambda })}1)}}\mathrm{\Phi }_{\overline{l}}^b`$
$`={\displaystyle \frac{q^{2l+1m}}{(q^{2(\epsilon _b,\epsilon _b+2\rho +2\mathrm{\Lambda })}1)}}.`$ (22)
Hence in the case $`m=2l+1`$
$$P_{\overline{b}b}=(1)^{[b]}q^{2(\rho ,\epsilon _b)}\left[\frac{\psi _0^b}{\alpha _b\alpha _0}\frac{(qq^1)}{(q^{2(\epsilon _b,\epsilon _b+2\rho +2\mathrm{\Lambda })}1)}\right]\underset{_{c0}^{c=b+1}}{\overset{\overline{b}}{}}\frac{\psi _c^b}{(\alpha _b^\mathrm{\Lambda }\alpha _c^\mathrm{\Lambda })}.$$
By substituting in the formulae for $`\psi _c^b`$ and $`\alpha _b`$ and simplifying we obtain
$$P_{\overline{b}b}=(1)^{[b]}q^{2(\rho ,\epsilon _b)}\left[1+(qq^1)\frac{q^{(\epsilon _b,\epsilon _b+2\rho +2\mathrm{\Lambda })}}{(q^{2(\epsilon _b,\epsilon _b+2\rho +2\mathrm{\Lambda })}1)}\right]\underset{c=b+1}{\overset{\overline{b}}{}}\frac{(q^{(\epsilon _b,2\rho +2\mathrm{\Lambda }+\epsilon _b)}q^{(\epsilon _c,2\rho +2\mathrm{\Lambda }\epsilon _c)})}{(q^{(\epsilon _b,2\rho +2\mathrm{\Lambda }+\epsilon _b)}q^{(\epsilon _c,2\rho +2\mathrm{\Lambda }+\epsilon _c)})},$$
and thus for $`a\overline{b}>0`$
$$P_{ab}=(1)^{[b]}q^{2(\rho ,\epsilon _b)}\left[1+(qq^1)\frac{q^{(\epsilon _b,\epsilon _b+2\rho +2\mathrm{\Lambda })}}{(q^{2(\epsilon _b,\epsilon _b+2\rho +2\mathrm{\Lambda })}1)}\right]\underset{c=b+1}{\overset{a}{}}\frac{(q^{(\epsilon _b,2\rho +2\mathrm{\Lambda }+\epsilon _b)}q^{(\epsilon _c,2\rho +2\mathrm{\Lambda }\epsilon _c)})}{(q^{(\epsilon _b,2\rho +2\mathrm{\Lambda }+\epsilon _b)}q^{(\epsilon _c,2\rho +2\mathrm{\Lambda }+\epsilon _c)})}.$$
Similarly, we find from equations (17), (19), (20) and (22) that when $`m`$ is even then
$$P_{ab}=(1)^{[b]}q^{2(\rho ,\epsilon _b)}\left[1\frac{q(qq^1)}{(q^{2(\epsilon _b,\epsilon _b+2\rho +2\mathrm{\Lambda })}1)}\right]\underset{c=b+1}{\overset{a}{}}\frac{(q^{(\epsilon _b,2\rho +2\mathrm{\Lambda }+\epsilon _b)}q^{(\epsilon _c,2\rho +2\mathrm{\Lambda }\epsilon _c)})}{(q^{(\epsilon _b,2\rho +2\mathrm{\Lambda }+\epsilon _b)}q^{(\epsilon _c,2\rho +2\mathrm{\Lambda }+\epsilon _c)})}$$
for $`a\overline{b}>0`$. Hence we have found expressions for $`P_{ab}`$ for all $`a,b`$ satisfying $`a\overline{b}>0`$. At the end of the paper these, together with the earlier results for $`P_{ab}`$, will be used to calculate $`\chi _\mathrm{\Lambda }(C_l)`$.
Now we return to the diagonalisation of the matrix $`N`$. We know
$$(N^1M)_{ab}=\alpha _a^\mathrm{\Lambda }(N^1)_{ab}.$$
Substituting in the values for $`M_{ab}`$ gives
$$\alpha _b^\mathrm{\Lambda }(N^1)_{ab}+(1)^{[b]}q^{(2\rho ,\epsilon _b)C(\mathrm{\Lambda }_0)}\underset{c>b}{}q^{(2\mathrm{\Lambda },\epsilon _c)}(1\delta _{\overline{b}}^c(1)^{[b]}q^{2(\rho ,\epsilon _b)})(N^1)_{ac}=\alpha _a^\mathrm{\Lambda }(N^1)_{ab}.$$
(23)
Set
$$\widehat{Q}_{ab}=\underset{cb}{}q^{2(\mathrm{\Lambda },\epsilon _c)}(N^1)_{ac}.$$
We then solve for $`\widehat{Q}_{ab}`$, with the calculations being very similar to those for $`P_{ab}`$. For $`0b<a`$ and $`b<a0`$ we find
$$\widehat{Q}_{ab}=q^{2(\mathrm{\Lambda },\epsilon _a)}\underset{c=b}{\overset{a1}{}}\frac{\psi _c^a}{(\alpha _a^\mathrm{\Lambda }\alpha _c^\mathrm{\Lambda })}.$$
For $`m=2l+1`$ we obtain
$$\widehat{Q}_{ab}=q^{2(\mathrm{\Lambda },\epsilon _a)}\left[1+(qq^1)\frac{q^{(\epsilon _a,\epsilon _a+2\rho +2\mathrm{\Lambda })}}{(q^{2(\epsilon _a,\epsilon _a+2\rho +2\mathrm{\Lambda })}1)}\right]\underset{c=b}{\overset{a1}{}}\frac{(q^{(\epsilon _a,2\rho +2\mathrm{\Lambda }+\epsilon _a)}q^{(\epsilon _c,2\rho +2\mathrm{\Lambda }\epsilon _c)})}{(q^{(\epsilon _a,2\rho +2\mathrm{\Lambda }+\epsilon _a)}q^{(\epsilon _c,2\rho +2\mathrm{\Lambda }+\epsilon _c)})}.$$
for $`b\overline{a}<0`$. Similarly, for even $`m`$ we find
$$\widehat{Q}_{ab}=q^{2(\mathrm{\Lambda },\epsilon _a)}\left[1\frac{q(qq^1)}{(q^{2(\epsilon _a,\epsilon _a+2\rho +2\mathrm{\Lambda })}1)}\right]\underset{c=b}{\overset{a1}{}}\frac{(q^{(\epsilon _a,2\rho +2\mathrm{\Lambda }+\epsilon _a)}q^{(\epsilon _c,2\rho +2\mathrm{\Lambda }\epsilon _c)})}{(q^{(\epsilon _a,2\rho +2\mathrm{\Lambda }+\epsilon _a)}q^{(\epsilon _c,2\rho +2\mathrm{\Lambda }+\epsilon _c)})}.$$
for $`b\overline{a}<0`$.
To use these results to calculate $`\chi _\mathrm{\Lambda }(C_l)`$ we introduce a new function $`Q_{ab}`$, defined by:
$$Q_{ab}=\underset{cb}{}(N^1)_{ac}.$$
Then from equations (13) and (15) we deduce
$$\chi _\mathrm{\Lambda }(C_l)=\underset{a}{}(\alpha _a^\mathrm{\Lambda })^lP_{(\mu =\overline{1})a}Q_{a(\nu =1)}.$$
(24)
However we know
$$t_a^{(l)}=\frac{q^{2(\mathrm{\Lambda },\epsilon _a)}1}{qq^1},$$
and
$`{\displaystyle \underset{b}{}}(N^1)_{ab}t_b^{(1)}`$ $`={\displaystyle \underset{b,c}{}}(N^1)_{ab}M_{bc}t_c^{(0)}`$
$`{\displaystyle \underset{b}{}}(N^1)_{ab}{\displaystyle \frac{(q^{2(\mathrm{\Lambda },\epsilon _b)}1)}{(qq^1)}}`$ $`={\displaystyle \underset{b}{}}(N^1M)_{ab}`$
$`={\displaystyle \underset{b}{}}\alpha _a^\mathrm{\Lambda }(N^1)_{ab}`$
$`={\displaystyle \underset{b}{}}(N^1)_{ab}{\displaystyle \frac{(q^{(\epsilon _a,2\mathrm{\Lambda }+2\rho +\epsilon _a)C(\mathrm{\Lambda }_0)}1)}{(qq^1)}}.`$
Thus
$$Q_{a(\nu =1)}=q^{C(\mathrm{\Lambda }_0)(\epsilon _a,2\rho +2\mathrm{\Lambda }+\epsilon _a)}\widehat{Q}_{a(\nu =1)}.$$
### 4.3 Explicit formulae for the eigenvalues
Substituting our formulae for $`P_{(\mu =\overline{1})a}`$ and $`Q_{a(\nu =1)}`$ into equation (24), noting that for $`a0`$ exactly one of $`a<0`$ or $`a>0`$ is true, we find the eigenvalues of the Casimir invariants $`C_l`$ are given by:
$$\begin{array}{c}\chi _\mathrm{\Lambda }(C_l)=\underset{a}{}(1)^{[a]}q^{C(\mathrm{\Lambda }_0)(\epsilon _a,\epsilon _a)}f(a)\left[\frac{(q^{(\epsilon _a,2\rho +2\mathrm{\Lambda }+\epsilon _a)C(\mathrm{\Lambda }_0)}1)}{(qq^1)}\right]^l\hfill \\ \hfill \times \underset{ba}{}\frac{(q^{(\epsilon _a,2\rho +2\mathrm{\Lambda }+\epsilon _a)}q^{(\epsilon _b,2\rho +2\mathrm{\Lambda }\epsilon _b)})}{(q^{(\epsilon _a,2\rho +2\mathrm{\Lambda }+\epsilon _a)}q^{(\epsilon _b,2\rho +2\mathrm{\Lambda }+\epsilon _b)})},\end{array}$$
where
$$f(a)=\{\begin{array}{cc}1(qq^1)\frac{q}{(q^{2(\epsilon _a,\epsilon _a+2\rho +2\mathrm{\Lambda })}1)},\hfill & m=2l,\hfill \\ 1+(qq^1)\frac{q^{(\epsilon _a,\epsilon _a+2\rho +2\mathrm{\Lambda })}}{(q^{2(\epsilon _a,\epsilon _a+2\rho +2\mathrm{\Lambda })}1)},\hfill & a0,m=2l+1,\hfill \\ 1,\hfill & a=0,m=2l+1.\hfill \end{array}$$
Throughout we assumed the eigenvalues were distinct. If they are not, the calculations are more complicated but the result is the same. Thus we have found:
###### Theorem 4.1
The quantum superalgebra $`U_q[osp(m|n)]`$, for $`m>2,`$ has an infinite family of Casimir invariants of the form
$$C_l=(strI)(\pi (q^{2h_p})I)A^l,l^+,$$
where
$$A=\frac{(R^TRII)}{(qq^1)}.$$
The eigenvalues of the invariants when acting on an arbitrary irreducible finite-dimensional module with highest weight $`\mathrm{\Lambda }`$ are given by:
$$\begin{array}{c}\chi _\mathrm{\Lambda }(C_l)=\underset{a}{}(1)^{[a]}q^{C(\mathrm{\Lambda }_0)(\epsilon _a,\epsilon _a)}f(a)\left[\frac{(q^{(\epsilon _a,\epsilon _a+2\rho +2\mathrm{\Lambda })C(\mathrm{\Lambda }_0)}1)}{(qq^1)}\right]^l\hfill \\ \hfill \times \underset{ba}{}\frac{(q^{(\epsilon _a,2\rho +2\mathrm{\Lambda }+\epsilon _a)}q^{(\epsilon _b,2\rho +2\mathrm{\Lambda }\epsilon _b)})}{(q^{(\epsilon _a,2\rho +2\mathrm{\Lambda }+\epsilon _a)}q^{(\epsilon _b,2\rho +2\mathrm{\Lambda }+\epsilon _b)})},\end{array}$$
where
$$f(a)=\{\begin{array}{cc}1(qq^1)\frac{q}{(q^{2(\epsilon _a,\epsilon _a+2\rho +2\mathrm{\Lambda })}1)},\hfill & m=2l,\hfill \\ 1+(qq^1)\frac{q^{(\epsilon _a,\epsilon _a+2\rho +2\mathrm{\Lambda })}}{(q^{2(\epsilon _a,\epsilon _a+2\rho +2\mathrm{\Lambda })}1)},\hfill & a0,m=2l+1,\hfill \\ 1,\hfill & a=0,m=2l+1.\hfill \end{array}$$
This completes the calculation of the eigenvalues of an infinite family of Casimir invariants of $`U_q[osp(m|n)]`$ when acting on an arbitrary irreducible highest weight module, provided $`m>2`$. This had already been done for $`U_q[osp(2|n)]`$, using a different method, in . Also every finite-dimensional representation of $`U_q[osp(1|n)]`$ is isomorphic to a finite-dimensional representation of $`U_q[o(n+1)]`$ , whose central elements are well-understood. Hence the eigenvalues of a family of Casimir invariants, when acting on an arbitrary irreducible finite-dimensional highest weight module have now been calculated for all quantised orthosymplectic superalgebras. Together with the results for $`U_q[gl(m|n)]`$ , this covers all non-exceptional quantised superalgebras.
Acknowledgements – We gratefully acknowledge financial assistance from the Australian Research Council.
|
warning/0506/hep-ph0506126.html
|
ar5iv
|
text
|
# Particle Correlations in Saturated QCD Matter
## I Introduction
High-$`p_T`$ near-side and back-to-back particle correlations have become recently the focus of intensive study at the Relativistic Heavy Ion Collider (RHIC). Such correlations give access to the microscopic dynamics of two major phenomena searched for at RHIC in order to elucidate the properties of dense QCD matter : the phenomenon of perturbative saturation expected to determine the initial condition of hadronic collisions at sufficiently high center of mass energy, and the phenomenon of jet quenching established to suppress the high-$`p_T`$ hadronic yields due to final state interactions. To disentangle initial and final state effects, it is clearly important to compare data on nucleus-nucleus collisions to data on proton (deuteron)-nucleus collisions in which final state effects are absent.
Two-particle correlations may provide further tests of the dynamics underlying final state jet quenching, since it has been argued that their angular dependence is sensitive to the density of the produced matter and its collective flow . Moreover, it has been pointed out that in an ideal liquid, the energy radiated off a hard final state parton would propagate in a ‘sonic boom’ . This would lead to an azimuthal back-to-back correlation which is peaked away from $`180^{}`$, and may thus serve as a characteristic signature of a medium with (almost) vanishing viscosity . On the other hand, it has been suggested that initial state saturation could affect dramatically the two-gluon spectrum by strongly reducing the angular correlation at high transverse momentum, leading practically to a disappearance of the back-to-back correlations of jets in nuclear collisions. Although the effect is suggested to be more dramatic for large nuclear projectiles, the trend should also be seen for proton (deuteron)-nucleus collisions.
The main purpose of the present work is to study quantitatively saturation effects in the simplest process which gives rise to two-parton correlation functions. In the target rest frame, this is the emission of one gluon from a projectile quark $`qAqgX`$ with the transverse momentum of quark and gluon resolved. These results are presented in Section II.
The other goal of this work is to set up a formalism for calculating differential multi-parton cross sections in the eikonal approximation . In particular we shall show how to take properly account of the final state radiation by unitarily transforming gluonic (and in general partonic) observables with the operator which dresses the partons by the cloud of Weizsäcker-Williams gluons. We show that perturbative expansion of this cloud operator precisely generates the perturbative radiation in the final state.
The paper is organized as follows: In Section II, we discuss the physics of two-parton correlation functions, starting from basic formulae and numerical results. The subsequent sections discuss the general framework for calculating parton correlations in the eikonal formalism and some applications. Section III presents the general discussion. This formalism is then applied to parton correlations for $`qAq(𝐤)g(𝐩)X`$ in Section IV, and to $`qAqg(𝐩_1)g(𝐩_2)X`$ in Section V. The main results are summarized in the Conclusions.
## II Angular correlations in $`qAq(𝐤)g(𝐩)X`$
We consider the simplest case of a two-particle final state. The projectile consists of a single quark. It scatters on a hadronic target of sizeable saturation momentum and produces a gluon in the final state: $`qAq(𝐤)g(𝐩)X`$. For a related discussion with strong emphasis on breaking of $`k_T`$-factorization, see Refs. .
### II.1 Basic formulae and definitions
The starting point of our study is the following expression for the doubly inclusive spectrum $`qAq(𝐤)g(𝐩)X`$ (for derivation see Sections III and IV)
$`{\displaystyle \frac{dN}{dyd𝐤d𝐩}}`$ $`=`$ $`{\displaystyle \frac{\alpha _sC_F}{\pi ^2}}{\displaystyle \frac{1}{(2\pi )^4}}{\displaystyle 𝑑𝐳𝑑\overline{𝐳}𝑑𝐱e^{i𝐤𝐱i𝐩(𝐳\overline{𝐳})}\frac{(𝐱𝐳)\overline{𝐳}}{(𝐱𝐳)^2\overline{𝐳}^2}}`$ (1)
$`\times [Q(𝐳,𝐱,\mathrm{𝟎},\overline{𝐳})S(\overline{𝐳},𝐳)+S(𝐱,\mathrm{𝟎})`$
$`S(𝐱,\overline{𝐳})S(\overline{𝐳},\mathrm{𝟎})S(𝐱,𝐳)S(𝐳,\mathrm{𝟎})].`$
The entire information about the target nucleus is contained in two target averages of products of Wilson lines,
$`S(\overline{𝐮},𝐮)`$ $`=`$ $`{\displaystyle \frac{1}{N}}\mathrm{Tr}\left[W_{}^{F}{}_{}{}^{}(\overline{𝐮})W^F(𝐮)\right]_T,`$ (2)
$`Q(\overline{𝐮},𝐮,𝐳,\overline{𝐳})`$ $`=`$ $`{\displaystyle \frac{1}{N}}\mathrm{Tr}\left[W_{}^{F}{}_{}{}^{}(\overline{𝐮})W^F(𝐮)W_{}^{F}{}_{}{}^{}(𝐳)W^F(\overline{𝐳})\right]_T.`$ (3)
To model these target averages, we use the expressions
$`S(\overline{𝐮},𝐮)`$ $`=`$ $`\mathrm{exp}\left[v(\overline{𝐮}𝐮)\right],`$ (4)
$`Q(\overline{𝐲},𝐱,\overline{𝐱},𝐲)`$ $`=`$ $`{\displaystyle \frac{v(𝐱\overline{𝐱})+v(𝐲\overline{𝐲})v(𝐱𝐲)v(\overline{𝐱}\overline{𝐲})}{v(𝐱\overline{𝐱})+v(𝐲\overline{𝐲})v(𝐱\overline{𝐲})v(𝐲\overline{𝐱})}}`$ (5)
$`\times \mathrm{exp}\left[v(𝐱\overline{𝐱})v(𝐲\overline{𝐲})\right]`$
$`{\displaystyle \frac{v(𝐱\overline{𝐲})+v(𝐲\overline{𝐱})v(𝐱𝐲)v(\overline{𝐱}\overline{𝐲})}{v(𝐱\overline{𝐱})+v(𝐲\overline{𝐲})v(𝐱\overline{𝐲})v(𝐲\overline{𝐱})}}`$
$`\times \mathrm{exp}\left[v(𝐱\overline{𝐲})v(𝐲\overline{𝐱})\right],`$
$`v(𝐱)`$ $`=`$ $`𝐱^2{\displaystyle \frac{\stackrel{~}{Q}_s^2(𝐱)}{8}}𝐱^2{\displaystyle \frac{Q_{s,0}^2}{8}}\mathrm{log}\left[{\displaystyle \frac{1}{𝐱^2\mathrm{\Lambda }^2}}+a\right].`$ (6)
The function $`v(𝐱)`$ has the meaning of the target gluon field correlation function and is directly proportional to the gluon density in the target, hence the logarithmic dependence on the transverse separation.
For numerical evaluation, we shall take $`\mathrm{\Lambda }\mathrm{\Lambda }_{\mathrm{QCD}}=0.2\mathrm{GeV}`$. The small regulator $`a=1/(x_c^2\mathrm{\Lambda }^2)`$, $`x_c=3\mathrm{GeV}^1`$ is chosen such that the logarithm in (6) does not turn negative and that the sensitivity on the infrared cut-off is negligible for sufficiently large momentum . The saturation scale $`Q_s`$ is defined implicitly as
$$Q_s^2\stackrel{~}{Q}_s^2(𝐱^2=1/Q_s^2).$$
(7)
With this definition, $`Q_s^2=2\mathrm{GeV}^2`$ corresponds to $`Q_{s,0}^20.5\mathrm{GeV}^2`$ in Eq. (6).
In the model (4)-(6) the properties of the single inclusive gluon radiation spectrum $`qAqg(𝐩)X`$ and its small-$`x`$ evolution have been studied extensively in recent years and are well understood . This spectrum is obtained from the two-particle correlation function (1), integrating over the quark momentum $`𝐤`$,
$`{\displaystyle \frac{dN}{dyd𝐩}}`$ $`=`$ $`{\displaystyle \frac{\alpha _sC_F}{\pi ^2}}{\displaystyle \frac{1}{(2\pi )^2}}{\displaystyle 𝑑𝐳𝑑\overline{𝐳}e^{i𝐩(𝐳\overline{𝐳})}\frac{𝐳\overline{𝐳}}{𝐳^2\overline{𝐳}^2}}`$ (8)
$`\times \left[S^2(\overline{𝐳},𝐳)+1S^2(\overline{𝐳})S^2(𝐳)\right].`$
In the approximation (4)-(6), one finds
$`{\displaystyle \frac{dN}{dyd𝐩}}`$ $`=`$ $`{\displaystyle \frac{1}{\pi }}{\displaystyle 𝑑𝐳𝑑\overline{𝐳}\frac{1}{(2\pi )^2}\frac{\alpha _sC_F}{\pi }\frac{𝐳\overline{𝐳}}{𝐳^2\overline{𝐳}^2}e^{i𝐩(𝐳\overline{𝐳})}}`$ (9)
$`\times \left[1+e^{2v(𝐳\overline{𝐳})}e^{2v(𝐳)}e^{2v(\overline{𝐳})}\right],`$
which coincides with the quasiclassical expression of .
In the limit of small transverse momenta, the relevant values of the transverse coordinates $`𝐳`$, $`\overline{𝐳}`$ are large. In this case, the logarithm in (6) becomes unimportant and the Gaussian approximation $`Q_s^2(𝐱)=\overline{Q}_s^2`$ is justified. Formally, one finds in this approximation
$$\frac{dN}{dyd𝐩}|_{\mathrm{Gaussian}}=\frac{\alpha _sC_F}{\pi ^3\overline{Q}_s^2}𝑑𝐪e^{𝐪^2/\overline{Q}_s^2}\frac{𝐪^2}{𝐩^2\left(𝐩𝐪\right)^2}.$$
(10)
Here, $`\frac{𝐪^2}{𝐩^2\left(𝐩𝐪\right)^2}`$ is the typical momentum dependence of gluons radiated off a high-energy quark, which received a momentum transfer $`𝐪`$. For gluons of small transverse momentum $`𝐩`$, the momentum transfer $`𝐪`$ from the target is accumulated according to transverse Brownian motion $`𝐪^2\overline{Q}_s^2A^{1/3}`$. For a hard gluon $`|𝐩|Q_s`$, however, Eq. (9) shows the typical power law $`\frac{1}{𝐩^4}\mathrm{ln}\frac{𝐩^2}{\mathrm{\Lambda }^2}`$ characteristic for a single hard scattering off the power-law tail of a hard scattering center, see (15) below. In this way, the ansatz (6) for the saturation momentum characterizes a medium with a physically sensible interpolation between single hard and multiple soft scattering as a function of transverse momentum transfer. We note that this is the QCD analogue of QED Molière scattering theory , which achieves the analogous interpolation between soft multiple and single hard target momentum transfers for electromagnetic scattering processes.
### II.2 Perturbative baseline
As a baseline for comparison of our numerical results, we collect here the perturbative expressions for the single and double inclusive cross sections. These do not include the effects of multiple rescattering which are due to the existence of a large saturation scale in the target. The perturbative baseline is obtained by identifying the single hard interaction term in the target averages (4), (5). This corresponds to an expansion to first order in $`v(𝐱)`$:
$`S(\overline{𝐮},𝐮)`$ $`=`$ $`1v(\overline{𝐮}𝐮)+O(v^2),`$ (11)
$`Q(\overline{𝐮},𝐮,\overline{𝐰},𝐰)`$ $`=`$ $`1[v(\overline{𝐮}𝐮)+v(\overline{𝐰}𝐰)+v(\overline{𝐮}𝐰)+v(\overline{𝐰}𝐮)`$ (12)
$`v(𝐮𝐰)v(\overline{𝐮}\overline{𝐰})]+O(v^2).`$
Inserting this into Eq.(1), we find
$$\frac{dN}{dyd𝐤d𝐩}=\frac{\alpha _sC_F}{\pi ^2}\frac{Q_{s,0}^2}{\pi }\frac{1}{𝐩^2(𝐤+𝐩)^2𝐤^2}.$$
(13)
To obtain (13), one has to keep the logarithmic dependence in the scale entering $`v(𝐱)`$ in (6). Neglecting this dependence of $`Q_s`$ on $`𝐱`$ amounts to neglecting the possibly large (albeit more rare) momentum transfer due to the interaction with the high momentum tail of the target fields. In this case, $`Q_s^2(𝐱)=\overline{Q}_s^2`$, one finds instead
$$\frac{dN_{\mathrm{Gauss}}}{dyd𝐤d𝐩}=\frac{\alpha _sC_F}{\pi ^2}\frac{\overline{Q}_s^2}{𝐩^4}\delta ^{(2)}\left(𝐩+𝐤\right),$$
(14)
where the quark and gluon momenta are exactly balanced. This shows that as long as the interaction with the target can not impart a significant kick to a propagating parton, large momenta in the final state can only appear due to the large relative momentum between the quark and the gluon in the initial wave function of the projectile.
Integrating Eq.(13) over the quark mometum $`𝐤`$, we obtain the perturbative expression for the single gluon inclusive cross section
$`{\displaystyle \frac{dN}{dyd𝐩}}|_{|𝐩|Q_s}={\displaystyle \frac{2\alpha _sC_FQ_{s,0}^2}{\pi ^2}}{\displaystyle \frac{1}{𝐩^4}}(\mathrm{ln}[{\displaystyle \frac{𝐩^2}{4\mathrm{\Lambda }^2}}]+2\gamma _E1).`$ (15)
To determine the distribution of the total recoil momentum, we integrate (1) over the relative momentum of the quark-gluon pair in the final state. Defining
$$𝐊=𝐤+𝐩,𝐪=𝐤𝐩,$$
(16)
we find to leading logarithmic accuracy
$$\frac{dN}{dyd𝐊}=2\frac{\alpha _sC_F}{\pi ^2}\frac{Q_{s,0}^2}{(𝐊^2)^2}\mathrm{ln}\left[\frac{𝐊^2}{\mathrm{\Lambda }^2}\right].$$
(17)
This expression exhibits the typical perturbative power-law dependence for the total recoil momentum $`K`$ above the infrared regulator scale $`\mathrm{\Lambda }`$. This is consistent with the expectation, that for large recoil $`𝐊`$, the target behaves like a single hard perturbative scattering center.
We also give here the perturbative expression for the two gluon inclusive cross section $`qAqg(𝐩_1)g(𝐩_2)X`$, with gluons produced at rapidities $`\eta `$ and $`\xi `$. The full expression, including the effects of the saturation is given in Eq.(66) below. Its perturbative limit is
$$\frac{dN}{d\eta d𝐩_1d\xi d𝐩_2}=\frac{\alpha _s^2}{\pi ^6}Q_{s,0}^2\frac{1}{𝐩_1^2𝐩_2^2(𝐩_1+𝐩_2)^2}\mathrm{ln}\frac{(𝐩_1+𝐩_2)^2}{\mathrm{\Lambda }^2}.$$
(18)
### II.3 Distribution of the recoil momentum
The total transverse momentum $`𝐊`$ of the quark-gluon pair in the final state traces the distribution of recoil momentum transferred from the target. We calculate it by integrating (1) over the relative transverse momentum $`𝐪`$. After performing some angular integrations, one finds
$`{\displaystyle \frac{dN}{dyKdK}}`$ $`=`$ $`{\displaystyle \frac{\alpha _sC_F}{\pi ^2}}{\displaystyle _0^{\mathrm{}}}𝑑z{\displaystyle _0^{1/\mathrm{\Lambda }_{\mathrm{cut}}}}𝑑\overline{z}{\displaystyle _0^{2\pi }}𝑑\varphi {\displaystyle \frac{z}{\overline{z}}}J_0(Kz)`$ (19)
$`\times [Q(𝐳+\overline{𝐳},𝐳,\mathrm{𝟎},\overline{𝐳})S(𝐳)+S(𝐳)`$
$`S(𝐳\overline{𝐳})S(\overline{𝐳})S(\overline{𝐳})S(𝐳+\overline{𝐳})].`$
Here, we have introduced the infrared cut-off $`\mathrm{\Lambda }_{\mathrm{cut}}`$ to regulate the logarithmically infrared divergent $`\overline{z}`$-integral. For the discussion in this subsection, we choose $`\mathrm{\Lambda }_{\mathrm{cut}}=\mathrm{\Lambda }`$, the same infrared cut-off, which regulates the perturbative expression (17). However, as discussed below, $`\mathrm{\Lambda }_{\mathrm{cut}}`$ has a physical interpretation as regulator of the transverse size of the incoming projectile. We will discuss the implications of restricing the size of the projectile in more detail later.
We have evaluated expression (19) numerically. In Fig. 1 we plot the ratio of the distribution (19) to the perturbative expression (17). For momenta $`KQ_s`$ well above the saturation scale, the recoil distribution (19) approaches the perturbative power law $`\frac{1}{K^4}\mathrm{ln}\frac{K^2}{\mathrm{\Lambda }^2}`$ and is indistinguishable from the perturbative expression. For low momenta $`K<Q_s`$, the saturated distribution is suppressed relative to the perturbative one. This is the manifestation of the fact that a parton propagating through the saturated target is very unlikely to get a kick of momentum less than of order of $`Q_s`$, whereas in the interaction with the perturbative target on the contrary low momentum transfer processes are very important. Finally, Fig. 1 exhibits a fairly wide maximum around $`K=(2÷3)Q_s`$, which is due to the transfer of momentum of order $`Q_s`$ to one of the propagating partons ($`q`$ or $`g`$). The general features of Fig. 1 are very similar to those of the nuclear modification factor calculated within the same model in .
### II.4 Angular Correlations - large trigger momentum
We next study the angular dependence of the two-parton correlation function (1). Shifting $`𝐳𝐳+𝐱`$ and introducing radial coordinates, we find
$`{\displaystyle \frac{dN}{dykdkpdpd\mathrm{\Delta }\varphi _{kp}}}`$ $`=`$ $`{\displaystyle \frac{\alpha _sC_F}{\pi ^2}}{\displaystyle \frac{1}{(2\pi )^2}}{\displaystyle x𝑑x^{\frac{1}{\mathrm{\Lambda }_{\mathrm{cut}}}}𝑑z^{\frac{1}{\mathrm{\Lambda }_{\mathrm{cut}}}}𝑑\overline{z}}`$ (20)
$`\times {\displaystyle }d\varphi _zd\varphi _{\overline{z}}\mathrm{cos}(\varphi _z\varphi _{\overline{z}})J_0\left(\sqrt{A^2+B^2}\right)`$
$`\times [Q(𝐳+𝐱,𝐱,\mathrm{𝟎},\overline{𝐳})S(\overline{𝐳},𝐳+𝐱)+S(𝐱,\mathrm{𝟎})`$
$`S(𝐱,\overline{𝐳})S(\overline{𝐳},\mathrm{𝟎})S(\mathrm{𝟎},𝐳)S(𝐳+𝐱,\mathrm{𝟎})].`$
Here, $`𝐳\overline{𝐳}=z\overline{z}\mathrm{cos}(\varphi _z\varphi _{\overline{z}})`$, $`𝐱𝐳=xz\mathrm{cos}(\varphi _z)`$, $`\overline{𝐳}𝐱=\overline{z}x\mathrm{cos}(\varphi _{\overline{z}})`$ and the notational shorthands
$`A=kx\mathrm{cos}(\mathrm{\Delta }\varphi _{kp})+px+pz\mathrm{cos}(\varphi _z)p\overline{z}\mathrm{cos}(\varphi _{\overline{z}}),`$ (21)
$`B=kx\mathrm{sin}(\mathrm{\Delta }\varphi _{kp})pz\mathrm{sin}(\varphi _z)+p\overline{z}\mathrm{sin}(\varphi _{\overline{z}}).`$ (22)
The $`z`$ and $`\overline{z}`$ integrals in equation (20) are logarithmically IR divergent. Regulating these integrals by cutting off both integrations at distances greater than $`1/\mathrm{\Lambda }_{\mathrm{cut}}`$, we can numerically determine the correlation of the two partons as a function of their relative azimuthal angle $`\mathrm{\Delta }\varphi _{kp}`$ and their transverse momentum $`k`$ and $`p`$. Technically, it is sufficient to regulate either one of the integrations $`z`$ or $`\overline{z}`$, to arrive at a finite result. Physically, the cut-off $`1/\mathrm{\Lambda }_{\mathrm{cut}}`$ limits the transverse size of the incoming wave function to $`z<1/\mathrm{\Lambda }_{\mathrm{cut}}`$ in the amplitude and to $`\overline{z}<1/\mathrm{\Lambda }_{\mathrm{cut}}`$ in the complex conjugate amplitude. Thus, we choose to implement the cut-off in a symmetric way.
Fig. 2(a) shows the correlation function as a function of the azimuthal angle for a fixed and relatively large value of the quark momentum $`k=4Q_s`$ and values of the gluon momentum $`p`$ between $`Q_s`$ and $`4Q_s`$. We observe that for larger values of $`p`$ the distribution exhibits a strong back-to-back correlation. As $`p`$ decreases, the distribution becomes wider, and finally at $`p=Q_s`$ it is practically flat.
The main features of Fig. 2(a) can be understood in the following simple picture. The incoming wave function of the projectile in the present approximation contains only one quark and up to one gluon (see Sections III-V). The total transverse momentum of the incoming quark-gluon system is zero, and thus they are exactly correlated back-to-back. While travelling through the target, each parton gets a transverse kick which in most cases is close to $`Q_s`$. However, albeit with small probability, the impinging partons can scatter off the perturbative high momentum tail of the target field (Molière scattering). The final state configurations with large transverse momentum of both, quark and gluon, come mainly from two sources: either from the wave function components of the projectile where $`q`$ and $`g`$ have both large momentum and experience soft momentum transfer, or from a low momentum component of the incoming wave function where the quark experiences hard scattering off the Molière tail and subsequently radiates a gluon in the final state. \[In the present calculation only the scattered quark can radiate. The radiation off the gluon is higher order correction in $`\alpha _s`$, since the weight of the gluon component in the initial state itself is of order $`\alpha _s`$.\] These events are therefore mostly back-to-back correlated, mirroring correlations in the initial projectile wave function and the back-to-back nature of the final state radiation.
On the other hand, final states with a large transverse momentum of the quark but a gluon with $`p=Q_s`$ can only arise due to Molière scattering of the quark, since there are no asymmetric configurations in the initial projectile wave function. The direction of the hard kick the quark experiences is random and uncorrelated with the direction of momentum of the gluon, and thus the angular distribution in this regime is almost completely flat, see Fig. 2(a). For the intermediate values of $`p`$, the distribution interpolates between these two extremes.
The angular integrated intensity (20) is shown in Fig. 2(b) as a function of $`p`$ for a trigger quark momentum $`k=4Q_s`$. It exhibits a characteristic maximum at $`p=0`$ and another less sharp maximum at $`p=k`$. The maximum at $`p=0`$ originates from the momentum distribution of the incoming projectile wave function, which is peaked at small values of momentum as $`1/p^2`$. The behavior near this peak is unaffected by the presence of $`Q_s`$, but it is affected by the infrared cut-off $`\mathrm{\Lambda }_{\mathrm{cut}}`$, which eliminates the low relative momentum configurations from the initial wave function. The maximum at $`p=k`$ can be traced to the pertubative expression (13) which (after integration over $`\mathrm{\Delta }\varphi _{kp}`$) diverges linearly at $`p=k`$. This divergence originates from the back-to-back correlations between the quark and the gluon in the projectile wave function and the back-to-back final state radiation. For the saturated target however, the final state momentum of each parton is smeared around its initial state value within an interval of the width $`\pm Q_s`$. Thus the linearly divergent peak of (13) turns into a finite peak of width $`Q_s`$ in Fig. 2(b).
Figs. 2(c,d) present analogous plots, where the momentum of the outgoing gluon is fixed at $`p=4Q_s`$ and the quark momentum is varied. Although some fine details are different, the overall picture is qualitatively very similar to Figs. 2(a,b).
### II.5 Angular Correlations - trigger momenta close to $`Q_s`$
We consider now the regime where the trigger momenta are of the order of $`Q_s`$. We concentrate on the situation when the associated momenta are equal, $`p=k`$. In Fig. 3, we show angular correlations when the trigger momenta are varied between $`0.625Q_s`$ and $`1.5Q_s`$. These plots show a structure distinct from the expected back-to-back correlations. The angular correlation is not peaked at $`\varphi =\pi `$, but instead at $`\varphi =\pi \delta `$, where $`\delta `$ is clearly greater for smaller values of $`p`$. The dip at $`\varphi =\pi `$ disappears for trigger momenta lower than about $`0.6Q_s`$ and higher than about $`1.4Q_s`$.
The dip at $`\pi `$ arises due to a coherent scattering effect, namely due to soft multiple scattering of the initial quark-gluon components of small transverse size. If the size of the pair is smaller than the transverse correlation length of the target fields, then we expect that the pair scatters as one single object as it sees the same target fields. It then picks up a typical soft momentum which is shared equally between the quark and the gluon. \[In the model defined by (6), it seems reasonable to use $`Q_{s,0}`$ as the typical soft scale, since the logarithmic correction which shifts $`Q_{s,0}`$ to $`Q_s`$ is due to harder target gluons. Also, the correlation length of the target may be estimated to be of order $`1/Q_{s,0}`$.\] In the initial state the quark and the gluon have momenta equal in magnitude and opposite in direction. In our trigger momenta kinematics, where the magnitudes of the trigger momenta are also equal, the momentum transfer from the target must be in the direction perpendicular to the initial momenta. We can thus estimate the maximal correlation angle when $`p>Q_{s,0}`$ by the following argument. We start with the initial momenta of the quark and the gluon $`𝐤_{in}`$, $`𝐤_{in}`$. In the final state $`𝐤=𝐤_{in}+\delta 𝐤`$ and $`𝐩=𝐤_{in}+\delta 𝐤`$ with $`|\delta 𝐤|=Q_{s,0}/2`$ and $`𝐤_{in}\delta 𝐤=0`$. For $`kQ_{s,0}`$ we find the angle between the momenta in the final state
$$\varphi =\pi \frac{Q_{s,0}}{\sqrt{2}k}.$$
(23)
The magnitude of the effect seen in Fig. 3 is consistent with this rough estimate.
So far, we have discussed how coherent scattering of incoming parton pairs can lead to a shift of the backward peak away of zero. To understand why the dip in the two-parton correlator appears only for a narrow range of trigger momenta, we consider two other scattering mechanisms: i) the perturbative hard Molière scattering off the large momentum tails of the target, which leads to a back-to-back distribution, and ii) the incoherent soft scattering for those components of the wave function, whose transverse size is greater than the correlation length $`1/Q_{s,0}`$ of the target fields. Indeed, when the trigger momenta are much higher than $`Q_s`$, the phase space in the initial state from which one can get to these final momenta by the coherent scattering mechanism is very small. Thus the perturbative final state radiation dominates, which is exactly back-to-back. The phase space for coherent scattering of incoming parton pairs also disappears if $`k_{trigger}`$ is significantly smaller than $`Q_s`$, since the coherent scattering mechanism requires enough particles in the initial state with momenta $`k_{initial}<k_{trigger}\beta Q_s`$ with $`\beta `$ \- a number of order one. Thus, it seems natural that the coherent scattering mechanism is dominant only for trigger momenta around $`Q_s`$.
To further test the mechanism suggested to underly the dip in the two-parton correlation functions, one can exploit the dependence of the two-parton correlation function (20) on the infrared cut-off $`\mathrm{\Lambda }_{\mathrm{cut}}`$. In the derivation of this expression in Section IV, this cut-off dependence can be seen to restrict the cross section of the final state radiation. Thus, if one artificially raises the infrared cut-off so that it is close to the trigger momentum, one expects that most of the contribution of the back-to-back correlated, final state radiation is eliminated. One expects a dip at $`\varphi =\pi `$ for a much larger trigger momentum, with maxima at a position given by the simple estimate (23). We have performed various such consistency checks to further substantiate the explanation given above and find that the results conform with this simple picture.
The physical interpretation to the cut-off dependence of (20) suggests that one can use it to extract some additional information from our calculation. As mentioned above, and as can be seen from the derivation in Section IV, the cut-off on the $`z`$ and $`\overline{z}`$ integrals restricts the size of the projectile wave function to $`1/\mathrm{\Lambda }_{\mathrm{cut}}`$. Thus, varying the cut-off, we can probe the physics of small size projectiles. For example, since the scale of $`0.3`$ Fermi is the natural size of the constituent quark, one may ask what is the effect of restricting the gluon in the wave function to be emitted no further than $`0.3`$ Fermi from the valence quark. \[Alternatively, one may think of this excercise as a “poor man’s” way to mimick the doubly inclusive spectrum of a scattering of a dipole of this size.\] In any case, raising the cut-off to this value should give a reasonable indication of the upper limit on the effect that one might expect. The numerical results with the cut-off $`\mathrm{\Lambda }_{\mathrm{cut}}=Q_{s,0}=3.57\mathrm{\Lambda }`$ are shown in Fig. 4. In accordance with our expectations, the effect is more pronounced. The dip at $`\varphi =\pi `$ now persists up to higher values of the trigger momentum, about $`2.5Q_s`$. It is also much deeper than for the lower IR cut-off.
This concludes our discussion of the numerical results. We now turn to the derivation of the basic formulae used in this section and a more general discussion of differential cross sections in the eikonal formalism.
## III Differential cross sections in the eikonal approximation
Below we follow the basic formalism of . Consider an energetic hadronic projectile impinging on a large nuclear target. The projectile is characterized by a wave function, in which the relevant degrees of freedom are the transverse positions and color states of the partons,
$$|\mathrm{\Psi }_{in}=\underset{\{\alpha _i,𝐱_i\}}{}\psi (\{\alpha _i,𝐱_i\})|\{\alpha _i,𝐱_i\}.$$
(24)
The color index $`\alpha _i`$ can belong to the fundamental, antifundamental or adjoint representation of the color $`SU(N_c)`$ group, corresponding to quark, antiquark or gluon in the wavefunction. In what follows we will consider wave functions with a small number of partons.
At high energy, the propagation time through the target is short, and thus partons propagate independently of each other. For the same reason the transverse positions of the partons do not change during the propagation. The only effect of the propagation is that the wave function of each parton acquires an eikonal phase due to the interaction with the gluon field of the target. Thus the projectile emerges form the interaction region with the wave function
$$|\mathrm{\Psi }_{out}=𝒮|\mathrm{\Psi }_{in}=\underset{\{\alpha _i,𝐱_i\}}{}\psi (\{\alpha _i,𝐱_i\})\underset{i}{}W(𝐱_i)_{\alpha _i\beta _i}|\{\beta _i,𝐱_i\}.$$
(25)
Here $`𝒮`$ is the $`S`$-matrix, and the $`W`$’s are Wilson lines
$$W(𝐱_i)=𝒫\mathrm{exp}\{i𝑑z^{}T^aA_a^+(𝐱_i,z^{})\}$$
(26)
with $`A^+`$ \- the gauge field in the target and $`T^a`$ \- the generator of $`SU(N_c)`$ in a representation corresponding to a given parton. The relative phases between the components of the wave function change, and the state that emerges after the target is no longer an eigenstate of the strong interaction Hamiltonian (as the incoming state is assumed to be) but rather a superposition of such eigenstates. It is convenient to rewrite the outgoing wave function in terms of the second quantized operator corresponding to the gauge rotation Eq.(26)
$$|\mathrm{\Psi }_{out}=\widehat{W}|\mathrm{\Psi }_{in}$$
(27)
with
$$\widehat{W}=\mathrm{exp}\left[i\lambda ^a(𝐱)\rho ^a(𝐱)\right].$$
(28)
Here $`\rho ^a(𝐱)`$ is the color charge density operator, integrated over the rapidities of the projectile, and the parameters $`\lambda ^a`$ are the functions of the target field $`A^+`$ defined so that
$$\mathrm{exp}\left[iT^a\lambda ^a(𝐱_i)\right]=W(𝐱_i).$$
(29)
Various differential cross sections are given by the expectation values of gluonic observables $`O(a,a^{})`$ in the outgoing wave function at $`t\mathrm{}`$. The evolution between the time of scattering and the time of measurement ($`t\mathrm{}`$) is unitary, and thus does not affect inclusive observables like total or inelastic cross sections. But it certainly does affect the measured particle spectrum via the final state radiation and has to be taken into account. Therefore expectation values of gluonic observables $`O(a,a^{})`$ are not simply given by $`\mathrm{\Psi }_{out}|O(a,a^{})|\mathrm{\Psi }_{out}`$ with $`|\mathrm{\Psi }_{out}`$ of (27) since $`|\mathrm{\Psi }_{out}`$ has to be evolved to infinite time before the average can be taken . Another way to view the final state radiation is to realize that the state immediately after the scattering has a nonvanishing overlap with the incoming state $`|\mathrm{\Psi }_{in}`$ and similar “hadronic states”. \[By hadronic states in this context we mean the states constructed just like the incoming state, so that they are perturbative eigenstates of the QCD Hamiltonian, which in the lowest perturbative order are orthogonal to single gluon states. We do not imply of course that within the present approximation one has to take into account the nonperturbative hadronization process.\] This overlap has to be subtracted to get a result which counts only the produced free gluons, and not those which are contained in the incoming wave function or the wave function of outgoing “hadronic” states but not freed during the scattering process.
Consider for example the simplest possible projectile, namely the wave function of a quark at transverse coordinate zero and color index $`\alpha `$, which is “dressed” to first order in perturbation theory. The incoming quark state contains a component coming from the splitting $`\alpha \beta b`$, where the gluon labeled by adjoint index $`b`$ sits at a transverse position displaced by $`𝐳`$ from its parent quark. To first order, the quark wave function reads
$`|\alpha _D`$ $`=`$ $`|\alpha +{\displaystyle 𝑑𝐳𝑑\xi f_i(𝐳)T_{\alpha \beta }^b|\beta ;b(𝐳,i,\xi )}`$ (30)
$`=`$ $`|\alpha +{\displaystyle 𝑑𝐳\stackrel{}{f}(𝐳)T_{\alpha \beta }^b|\beta ;b(𝐳)}.`$
Here, the second line specifies a shorthand used below, and
$$f_i(𝐳)=\frac{g}{2\pi ^{3/2}}\frac{z_i}{𝐳^2}$$
(31)
denotes the perturbative Weizsäcker-Williams (WW) gluon field and the index $`i`$ labels the direction in the transverse plane. For the single inclusive gluon cross section, one has to calculate
$$\delta \mathrm{\Psi }|a_{i}^{d}{}_{}{}^{}(𝐩)a_i^d(𝐩)|\delta \mathrm{\Psi };|\delta \mathrm{\Psi }=|\mathrm{\Psi }_{out}\underset{\alpha }{}|\alpha _D\alpha _D|\mathrm{\Psi }_{out}.$$
(32)
The subtraction (32) properly accounts for the evolution of the scattered system after it emerges from the target , since the only ”hadronic state” in the final state to the first order in $`\alpha _s`$ is the dressed quark. Eq.(32) ensures that the observable does not include the gluons in the wave function of this dressed quark.
The above example illustrates the general statement that in calculations of $`O(a,a^{})`$, one should not count gluons that belong to the WW cloud of any of the fast charged partons that constitute the projectile. Rather, a ”quark” or a ”gluon” in the final state should not be thought of as a free Fock space quark or gluon, but the quark (or gluon) plus its WW cloud. The change of basis from free to dressed partons is described by the unitary ”gluon cloud” operator $`C`$
$$|\alpha _D=C|\alpha $$
(33)
with
$$C=P\mathrm{exp}\left(i𝑑𝐱𝑑𝐳𝑑\xi f_i(𝐳𝐱)[a_i^d(𝐳,\xi )+a_i^d(𝐳,\xi )]\rho _\xi ^d(𝐱)\right),$$
(34)
where $`P`$ stands for rapidity ordering, and $`\rho _\xi ^d(𝐱)`$ denotes the total charge density operator integrated from the rapidity of the projectile to $`\xi `$. The expression (34) is only valid to first order in $`\alpha _s`$, but we will not need higher orders in the calculations presented in this paper. The observable
$$Ca_{i}^{d}{}_{}{}^{}a_i^dC^{}$$
(35)
counts directly those low x gluons whose wave functions are orthogonal to those of dressed quarks (and dressed ”valence” gluons). Since the dressed gluons are eigenstates of the propagation outside the target, they can be counted directly in $`|\mathrm{\Psi }_{out}`$ without the need to account for the additional unitary evolution to infinite time. This is also true for any other gluonic observable. Thus the correct expression for calculating an arbitrary gluonic observable $`O`$ in the state $`t\mathrm{}`$ is
$$\mathrm{\Psi }_{out}|CO(a,a^{})C^{}|\mathrm{\Psi }_{out},$$
(36)
with the gluon cloud operator $`C`$ given by Eq.(34). One can explicitly verify that this procedure reproduces Eq.(32) in the simple example given above.
## IV Single gluon and quark-gluon inclusive cross sections
We now apply the formalism of Section III to the calculation of the one gluon inclusive emission cross section for an arbitrary projectile which contains a small number of partons, $`n1/\alpha _s`$. The restriction to a small number of partons is necessary to treat the projectile wave function perturbatively.
We take the wave function of the projectile in the form of a valence Fock space dressed by the WW field
$$|\mathrm{\Psi }_{in}=C|R,$$
(37)
where $`|R`$ is an arbitrary state in the free Fock space with a small number of large rapidity partons. The outgoing wave function after the propagation through the target is
$$|\mathrm{\Psi }_{out}=\widehat{W}C|R.$$
(38)
The gluon yield per unit rapidity at transverse momentum $`𝐩`$ is given by \[see Eq. (36)\]
$$\frac{dN}{dyd𝐩}=\mathrm{\Psi }_{out}|Ca_{i}^{d}{}_{}{}^{}(𝐩,y)a_i^d(𝐩,y)C^{}|\mathrm{\Psi }_{out}.$$
(39)
To evaluate this, we act with the gluon cloud operator $`C`$ on the gluon annihilation and creation operators $`a`$ and $`a^{}`$
$$Ca_i^dC^{}=a_i^db_i^d(\rho ),$$
(40)
where $`b_i^d(\rho )`$ is the classical WW field associated with the charge density $`\rho `$
$$b_i^d(𝐳,\rho )=𝑑𝐱f_i(𝐳𝐱)\rho ^d(𝐱).$$
(41)
To leading order in $`\alpha _s`$, the gluon cloud operator $`C`$ commutes with the charge density $`\rho ^d`$. We thus find
$`(a_ib_i(\rho ))|\mathrm{\Psi }_{out}=\widehat{W}\widehat{W}^{}(a_ib_i(\rho ))\widehat{W}C|R,`$ (42)
$`\widehat{W}(wa_ib_i(w\rho ))C|R=\widehat{W}C(wa_ib_i(w\rho ))|R.`$ (43)
We have used here the fact that the state $`|R`$ does not contain gluons with rapidities corresponding to the operator $`a`$, and hence it is annihilated by $`a`$. The shorthand notation in (43) stands for
$$wa_i^dW_{db}^A(𝐳)a_i^b(𝐳),etc.$$
(44)
with $`W^A`$ \- the eikonal Wilson factor in the adjoint representation. We thus have
$`{\displaystyle \frac{dN}{dyd𝐩}}`$ $`=`$ $`{\displaystyle \frac{1}{(2\pi )^2}}{\displaystyle 𝑑𝐳𝑑\overline{𝐳}e^{i𝐩(𝐳\overline{𝐳})}R|\left[w^{}(𝐳)b_i^d(𝐳,\rho )b_i^d(𝐳,w\rho )\right]}`$ (45)
$`\times \left[w(\overline{𝐳})b_i^d(\overline{𝐳},\rho )b_i^d(\overline{𝐳},w\rho )\right]|R.`$
Using
$`w(𝐳)b_i^d(𝐳,\rho )b_i^d(𝐳,w\rho )`$ $`=`$ $`W_{db}^A(𝐳){\displaystyle 𝑑𝐱f_i(𝐳𝐱)\rho ^b(𝐱)}`$ (46)
$`{\displaystyle 𝑑𝐱f_i(𝐳𝐱)W_{db}^A(𝐱)\rho ^b(𝐱)},`$
we find
$`{\displaystyle \frac{dN}{dyd𝐩}}={\displaystyle \frac{\alpha _sC_F}{\pi ^2}}{\displaystyle \frac{1}{(2\pi )^2}}{\displaystyle 𝑑𝐳𝑑\overline{𝐳}e^{i𝐩(𝐳\overline{𝐳})}𝑑𝐱𝑑\overline{𝐱}\rho ^a(𝐱)\rho ^b(\overline{𝐱})_P\frac{(𝐳𝐱)(\overline{𝐳}\overline{𝐱})}{(𝐳𝐱)^2(\overline{𝐳}\overline{𝐱})^2}}`$
$`\left[W_{}^{A}{}_{}{}^{}(𝐳)W^A(\overline{𝐳})+W_{}^{A}{}_{}{}^{}(𝐱)W^A(\overline{𝐱})W_{}^{A}{}_{}{}^{}(𝐳)W^A(\overline{𝐱})W_{}^{A}{}_{}{}^{}(𝐱)W^A(\overline{𝐳})\right]^{ab}_T,`$ (47)
where the averages of the charge density and of the products of eikonal factors are taken with respect to the projectile and the target wave functions, respectively. For a translationally invariant and gauge singlet but otherwise arbitrary target this expression reduces to the $`k_T`$-factorized form derived in .
### IV.1 Quark-gluon correlation function: $`qAq(𝐤)g(𝐩)X`$
We now turn to the simplest two-parton correlation function calculable in the eikonal formalism: a single quark projectile of final transverse momentum $`𝐤`$ which shares its recoil between a gluon of transverse momentum $`𝐩`$ and the target. The correlation function is
$$\frac{dN}{dyd𝐤d𝐩}=\mathrm{\Psi }_{out}|Ca_{i}^{b}{}_{}{}^{}(𝐩,y)a_i^b(𝐩,y)d_\delta ^{}(𝐤)d_\delta (𝐤)C^{}|\mathrm{\Psi }_{out}_P.$$
(48)
Here, $`d_\delta ^{}`$ is the quark creation operator at projectile rapidity, and $`|\mathrm{\Psi }_{out}(\alpha )=\widehat{W}C|\alpha `$. We average over the color charge $`\alpha `$ of the incoming quark,
$$\mathrm{\Psi }_{out}|O|\mathrm{\Psi }_{out}_P=\frac{1}{N}\mathrm{\Sigma }_\alpha \mathrm{\Psi }_{out}(\alpha )|O|\mathrm{\Psi }_{out}(\alpha ).$$
(49)
The quark color charge density operator is
$$\rho ^a(𝐱)=d_\alpha ^{}(𝐱)T_{\alpha \beta }^ad_\beta (𝐱).$$
(50)
This charge creates the Weizsäcker-Williams field according to Eq.(41). The calculation of the previous subsection is easily repeated. The action of the eikonal S-matrix operator $`\widehat{W}`$ amounts to rotating the quark and the gluon creation operators by appropriate eikonal factors. This amounts to attaching a fundamental Wilson line $`W_{\beta \delta }^F(𝐱)`$ at the transverse position of the quark and an adjoint Wilson line $`W_{bd}^A(𝐳)`$ at the transverse position of the gluon. We find
$`a_i^d(𝐳,y)d_\delta (𝐱)C^{}|\mathrm{\Psi }_{out}(\alpha )`$
$`=f_i(𝐳𝐱)\left[T_{\alpha \beta }^bW_{\beta \delta }^F(𝐱)W_{bd}^A(𝐳)T_{\beta \delta }^dW_{\alpha \beta }^F(𝐱)\right]|\alpha .`$ (51)
In terms of this expression we can write the quark-gluon correlator (48) explicitly. In the large-$`N`$ limit, the color algebra simplifies considerably and one obtains for the two-particle correlation integrated over impact parameter $`𝐛`$
$`{\displaystyle 𝑑𝐛\frac{dN}{dyd𝐤d𝐩}}`$ $`=`$ $`{\displaystyle \frac{1}{(2\pi )^4}}{\displaystyle _{𝐱\overline{𝐱}𝐳\overline{𝐳}}}e^{i𝐤(𝐱\overline{𝐱})i𝐩(𝐳\overline{𝐳})}\stackrel{}{f}(𝐳𝐱)\stackrel{}{f}(\overline{𝐳}\overline{𝐱})`$ (52)
$`\times [Q(𝐳,𝐱,\overline{𝐱},\overline{𝐳})S(\overline{𝐳},𝐳)+S(𝐱,\overline{𝐱})`$
$`S(𝐱,\overline{𝐳})S(\overline{𝐳},\overline{𝐱})S(𝐱,𝐳)S(𝐳,\overline{𝐱})].`$
Here, we use the shorthand $`_𝐱=𝑑𝐱`$. This quark-gluon correlation function is expressed in terms of two target averages, $`S(\overline{𝐳},𝐳)`$ and $`Q(𝐳,𝐱,\overline{𝐱},\overline{𝐳})`$ defined in eqs. (2)-(6). Integrated over $`𝐤`$, it reproduces the expression for the single gluon inclusive emission cross section given in (47). This can be checked explicitly by inserting in (47) the color charge density correlator $`\rho ^a(𝐱)\rho ^b(\overline{𝐱})_P=\frac{1}{2N}\delta ^{ab}\delta (𝐱)\delta (\overline{𝐱})`$ and writing the averages over adjoint Wilson lines in terms of fundamental ones.
## V The two gluon inclusive cross section: $`nAg(𝐩_1)g(𝐩_2)X`$
In this section we calculate the eikonal cross section for production of two gluons. We first give the derivation for an arbitrary perturbative projectile. Then we specialize to the case of a single quark projectile.
The cross section for production of two gluons with rapidities and transverse momenta $`(\eta ,𝐩_1)`$ and $`(\xi ,𝐩_2)`$ is
$`{\displaystyle \frac{dN}{d\eta d𝐩_1d\xi d𝐩_2}}`$ $`=`$ $`{\displaystyle \frac{1}{(2\pi )^4}}{\displaystyle _{𝐳\overline{𝐳}𝐮\overline{𝐮}}}e^{i𝐩_1(𝐳\overline{𝐳})i𝐩_2(𝐮\overline{𝐮})}R|C^{}\widehat{W}^{}C[a_{i}^{a}{}_{}{}^{}(𝐳,\xi )a_i^a(\overline{𝐳},\xi )`$ (53)
$`a_{j}^{b}{}_{}{}^{}(𝐮,\eta )a_j^b(\overline{𝐮},\eta )]C^{}\widehat{W}C|R.`$
Our convention is such that the rapidity $`\eta `$ is closer to the rapidities of the valence partons in the projectile, $`\eta \xi `$. We consider $`\eta `$ and $`\xi `$ to be sufficiently close, so that no evolution effects between $`\eta `$ and $`\xi `$ have to be taken into account.
For this case, the action of the cloud operator $`C`$ Eq.(34) on the gluon field operator can be restricted to the two rapidities $`\eta `$ and $`\xi `$. We can therefore write it schematically as
$$C=C_\xi C_\eta $$
(54)
with
$`C_\eta `$ $`=`$ $`\mathrm{exp}\left[i{\displaystyle _𝐳}b_i^d(𝐳)[a_i^d(𝐳,\eta )+a_i^d(𝐳,\eta )]\right],`$ (55)
$`C_\xi `$ $`=`$ $`\mathrm{exp}\left[i{\displaystyle _𝐳}[b_i^d(𝐳)+\delta b_i^d(𝐳)][a_i^d(𝐳,\xi )+a_i^d(𝐳,\xi )]\right].`$ (56)
Here, $`\rho ^d(𝐱)`$ is the charge density operator at the valence rapidity. It determines the classical WW field $`b_i^d(𝐳)`$ as specified in Eq. (41). The color charges produced at rapidity $`\eta `$ are measured by $`\rho _\eta ^d(𝐱)`$ and constitute an additional contribution $`\delta b_i^d(𝐳)`$ to the WW field which affects the gluon production at lower rapidities $`\xi `$,
$$\delta b_i^d(𝐳)=_𝐱f_i(𝐳𝐱)\rho _\eta ^d(𝐱),$$
(57)
where
$$\rho _\eta ^d(𝐱)=a_i^b(𝐱,\eta )T_{bc}^da_i^c(𝐱,\eta ),T_{bc}^a=if^{abc}.$$
(58)
To construct a state with up to two gluons, we have to expand the cloud operator (54) up to second order in $`\rho `$,
$`C`$ $`=`$ $`1+i{\displaystyle _𝐳}\{[a_i^d(𝐳,\eta )+a_{i}^{d}{}_{}{}^{}(𝐳,\eta )]b_i^d(𝐳)`$ (59)
$`+[a_i^d(𝐳,\xi )+a_{i}^{d}{}_{}{}^{}(𝐳,\xi )][b_i^d(𝐳)+\delta b_i^d(𝐳)]\}`$
$`{\displaystyle _{𝐳,𝐮}}[a_i^d(𝐳,\xi )+a_{i}^{d}{}_{}{}^{}(𝐳,\xi )][b_i^d(𝐳)+\delta b_i^d(𝐳)]`$
$`\times [a_i^d(𝐮,\eta )+a_{i}^{d}{}_{}{}^{}(𝐮,\eta )]b_i^d(𝐮).`$
It is now a straightforward albeit tedious matter to calculate $`C^{}\widehat{W}C`$ and to act with it on $`|R`$. \[Alternatively, one can construct $`|\mathrm{\Psi }_{\mathrm{in}}`$ and $`|\mathrm{\Psi }_{\mathrm{out}}`$ from the cloud operator and evaluate the gluon number operator in this $`|\mathrm{\Psi }_{\mathrm{out}}`$ state. This is done in Appendix A.\] Remembering that $`\widehat{W}^{}a_i^a(𝐳)\widehat{W}=w^{ab}(𝐳)a_i(𝐳)`$ and $`\widehat{W}^{}\rho ^a(𝐱)W=w^{ab}(𝐱)\rho ^b(𝐱)`$, we get after some algebra
$`\widehat{W}^{}a_i^a(\overline{𝐳},\xi )a_j^b(\overline{𝐮},\eta )C^{}\widehat{W}C|R`$
$`={\displaystyle _{\overline{𝐱}_1,\overline{𝐱}_2}}f_i(\overline{𝐳}\overline{𝐱}_1)f_j(\overline{𝐮}\overline{𝐱}_2)\left\{\left(w(\overline{𝐱}_1)w(\overline{𝐳})\right)\rho (\overline{𝐱}_1)\right\}^a\left\{w(\overline{𝐮})\rho (\overline{𝐱}_2)\right\}^b|R`$
$`{\displaystyle _{\overline{𝐱}_1,\overline{𝐱}_2}}f_i(\overline{𝐳}\overline{𝐱}_1)f_j(\overline{𝐮}\overline{𝐱}_2)\left\{w(\overline{𝐱}_2)\rho (\overline{𝐱}_2)\right\}^b\left\{\left(w(\overline{𝐱}_1)w(\overline{𝐳})\right)\rho (\overline{𝐱}_1)\right\}^a|R`$
$`+{\displaystyle _{\overline{𝐱}_1}}f_i(\overline{𝐳}\overline{𝐮})f_j(\overline{𝐮}\overline{𝐱}_1)\left\{\left(w(\overline{𝐳})w(\overline{𝐳})\right)w^{}(\overline{𝐮})T^bw(\overline{𝐮})\rho (\overline{𝐱}_1)\right\}^a|R.`$ (60)
With this state, the expectation value of the observable (53) can be calculated directly. It contains terms which are quadratic, cubic and quartic in the density of the projectile,
$$\frac{dN}{d\eta d𝐩_1d\xi d𝐩_2}=\frac{1}{(2\pi )^4}_{𝐳\overline{𝐳}𝐮\overline{𝐮}}e^{i𝐩_1(𝐳\overline{𝐳})i𝐩_2(𝐮\overline{𝐮})}\left[\mathrm{\Sigma }_2+\mathrm{\Sigma }_3+\mathrm{\Sigma }_4\right],$$
(61)
where
$`\mathrm{\Sigma }_2`$ $`=`$ $`{\displaystyle _{𝐱\overline{𝐱}}}\stackrel{}{f}(𝐮𝐳)\stackrel{}{f}(\overline{𝐳}\overline{𝐮})\stackrel{}{f}(\overline{𝐮}\overline{𝐱})\stackrel{}{f}(𝐱𝐮)\rho ^a(𝐱)\rho ^b(\overline{𝐱})_P`$ (62)
$`\times \{w^{}(𝐮)T^cw(𝐮)(w^{}(𝐳)w^{}(𝐮))`$
$`(w(\overline{𝐳})w(\overline{𝐮}))w^{}(\overline{𝐮})T^cw(\overline{𝐮})\}^{ab},`$
$`\mathrm{\Sigma }_3`$ $`=`$ $`{\displaystyle _{𝐱_1\overline{𝐱}_1\overline{𝐱}_2}}\stackrel{}{f}(𝐮𝐳)\stackrel{}{f}(\overline{𝐳}\overline{𝐱}_1)\stackrel{}{f}(𝐱_1𝐮)\stackrel{}{f}(\overline{𝐮}\overline{𝐱}_2)`$ (63)
$`\times [\rho ^c(𝐱_1)\rho ^e(\overline{𝐱}_2)\rho ^d(\overline{𝐱}_1)_P`$
$`\times \left\{[w^{}(\overline{𝐱}_1)w^{}(\overline{𝐳})][w(𝐳)w(𝐮)]T^cw^{}(𝐮)w(\overline{𝐱}_2)\right\}^{de}_T`$
$`\rho ^c(𝐱_1)\rho ^d(\overline{𝐱}_1)\rho ^e(\overline{𝐱}_2)_P`$
$`\times \left\{[w^{}(\overline{𝐱}_1)w^{}(\overline{𝐳})][w(𝐳)w(𝐮)]T^cw^{}(𝐮)w(\overline{𝐮})\right\}^{de}_T]`$
$`+`$ $`{\displaystyle _{𝐱_1𝐱_2\overline{𝐱}_1}}\stackrel{}{f}(\overline{𝐳}\overline{𝐮})\stackrel{}{f}(𝐱_1𝐳)\stackrel{}{f}(\overline{𝐮}\overline{𝐱}_1)\stackrel{}{f}(𝐱_2𝐮)`$
$`\times [\rho ^d(𝐱_1)\rho ^c(𝐱_2)\rho ^e(\overline{𝐱}_1)_P`$
$`\times \left\{w^{}(𝐱_2)w(\overline{𝐮})T^e[w^{}(\overline{𝐳})w^{}(\overline{𝐮})][w(𝐱_1)w(𝐳)]\right\}^{cd}_T`$
$`\rho ^c(𝐱_2)\rho ^d(𝐱_1)\rho ^e(\overline{𝐱}_1)_P`$
$`\times \left\{w^{}(𝐮)w(\overline{𝐮})T^e[w^{}(\overline{𝐳})w^{}(\overline{𝐮})][w(𝐱_1)w(𝐳)]\right\}^{cd}_T],`$
and
$`\mathrm{\Sigma }_4`$ $`=`$ $`{\displaystyle _{𝐱_1𝐱_2\overline{𝐱}_1\overline{𝐱}_2}}\stackrel{}{f}(\overline{𝐳}\overline{𝐱}_1)\stackrel{}{f}(𝐱_1𝐳)\stackrel{}{f}(\overline{𝐮}\overline{𝐱}_2)\stackrel{}{f}(𝐱_2𝐮)`$ (64)
$`\times [\rho ^g(𝐱_2)\rho ^k(𝐱_1)\rho ^l(\overline{𝐱}_1)\rho ^h(\overline{𝐱}_2)_P`$
$`\times \left\{[w^{}(𝐱_1)w^{}(𝐳)][w(\overline{𝐱}_1)w(\overline{𝐳})]\right\}^{kl}\left\{w^{}(𝐮)w(\overline{𝐮})\right\}^{gh}_T`$
$`\rho ^k(𝐱_1)\rho ^g(𝐱_2)\rho ^l(\overline{𝐱}_1)\rho ^h(\overline{𝐱}_2)_P`$
$`\times \left\{[w^{}(𝐱_1)w^{}(𝐳)][w(\overline{𝐱}_1)w(\overline{𝐳})]\right\}^{kl}\left\{w^{}(𝐱_2)w(\overline{𝐮})\right\}^{gh}_T`$
$`\rho ^g(𝐱_2)\rho ^k(𝐱_1)\rho ^h(\overline{𝐱}_2)\rho ^l(\overline{𝐱}_1)_P`$
$`\times \left\{[w^{}(𝐱_1)w^{}(𝐳)][w(\overline{𝐱}_1)w(\overline{𝐳})]\right\}^{kl}\left\{w^{}(𝐮)w(\overline{𝐱}_2)\right\}^{gh}_T`$
$`+\rho ^k(𝐱_1)\rho ^g(𝐱_2)\rho ^h(\overline{𝐱}_2)\rho ^l(\overline{𝐱}_1)_P`$
$`\times \left\{[w^{}(𝐱_1)w^{}(𝐳)][w(\overline{𝐱}_1)w(\overline{𝐳})]\right\}^{kl}\left\{w^{}(𝐱_2)w(\overline{𝐱}_2)\right\}^{gh}_T].`$
This is an explicit function of the color charge density correlators in the projectile, and the correlators of the eikonal factors in the target. The term $`\mathrm{\Sigma }_2`$ is the probability corresponding to the process when the valence component of $`|R`$ emits the gluon with rapidity $`\eta `$, which subsequently splits into two gluons with rapidities $`\eta `$ and $`\xi `$. The term $`\mathrm{\Sigma }_4`$ corresponds to the probability of emission of both gluons $`\xi `$ and $`\eta `$ from the valence component, and the term $`\mathrm{\Sigma }_3`$ is the interference of these two amplitudes. These expressions are general and valid for any perturbative projectile. Also note that we have not used the formal $`1/N_c`$ expansion to arrive at these expressions. In these aspects, our expression is more general than the one given in for a dipole as projectile.
### V.1 Two-gluon correlations of a single quark projectile: $`qAg(𝐩_1)g(𝐩_2)X`$
We now specify to the simplest projectile - a single quark. We take the incoming quark to be at the origin of the transverse plane. The color charge density operator is (50) and we average again over the color index of the incoming quark as in (49). The color charge density correlators are
$`\rho ^a(𝐱)\rho ^b(𝐲)_P`$ $`=`$ $`{\displaystyle \frac{1}{2N}}\delta ^{ab}\delta (𝐱)\delta (𝐲),`$
$`\rho ^a(𝐱)\rho ^b(𝐲)\rho ^c(𝐳)_P`$ $`=`$ $`{\displaystyle \frac{1}{4N}}\left(d^{abc}+if^{abc}\right)\delta (𝐱)\delta (𝐲)\delta (𝐳),`$
$`\rho ^a(𝐱)\rho ^b(𝐲)\rho ^c(𝐳)\rho ^d(𝐮)_P`$ $`=`$ $`\left[{\displaystyle \frac{1}{2N^2}}\delta ^{ab}\delta ^{cd}+{\displaystyle \frac{1}{8N}}\left(d^{abe}+if^{abe}\right)\left(d^{ecd}+if^{ecd}\right)\right]`$ (65)
$`\times \delta (𝐱)\delta (𝐲)\delta (𝐳)\delta (𝐮).`$
The two-gluon correlation function (53) is given explicitly in terms of these projectile-averaged correlators. In the large-$`N_c`$ limit, the expression simplifies. Using the $`SU(N_c)`$-identities compiled in the appendix of , the two-gluon correlation function can be expressed in terms of the target averages (2) and (3) of two and four fundamental Wilson lines, respectively. The final result takes the form
$`{\displaystyle \frac{dN}{d\eta d𝐩_1d\xi d𝐩_2}}={\displaystyle \frac{1}{(2\pi )^4}}{\displaystyle _{𝐳\overline{𝐳}𝐮\overline{𝐮}}}e^{i𝐩_1(𝐳\overline{𝐳})i𝐩_2(𝐮\overline{𝐮})}\stackrel{}{f}(𝐮)\stackrel{}{f}(\overline{𝐮})`$
$`\times [2\stackrel{}{f}(𝐳𝐮)\stackrel{}{f}(\overline{𝐳}\overline{𝐮})\{Q(\overline{𝐮},𝐮,𝐳,\overline{𝐳})S(𝐳,\overline{𝐳})S(𝐮,\overline{𝐮})+S^2(𝐮,\overline{𝐮})`$
$`S(\overline{𝐮},\overline{𝐳})S(\overline{𝐳},𝐮)S(𝐮,\overline{𝐮})S(𝐮,𝐳)S(𝐳,\overline{𝐮})S(\overline{𝐮},𝐮)\}`$
$`\stackrel{}{f}(𝐳𝐮)\stackrel{}{f}(\overline{𝐳})\{Q(𝐮,\overline{𝐮},\overline{𝐳},𝐳)S(𝐮,\overline{𝐮})S(𝐳,\overline{𝐳})+S(𝐮)S(\overline{𝐮})S(𝐮,\overline{𝐮})`$
$`Q(𝐮,\overline{𝐮},\mathrm{𝟎},𝐳)S(𝐳)S(𝐮,\overline{𝐮})S(\overline{𝐮},𝐮)S(𝐮,\overline{𝐳})S(\overline{𝐳},\overline{𝐮})`$
$`+Q(𝐮,\mathrm{𝟎},\overline{𝐳},𝐳)S(𝐳,\overline{𝐳})S(𝐮)+S^2(𝐮)`$
$`S(𝐮)S(𝐳)S(𝐳,𝐮)S(𝐮)S(\overline{𝐳},𝐮)S(\overline{𝐳})\}`$
$`\stackrel{}{f}(𝐳)\stackrel{}{f}(\overline{𝐳}\overline{𝐮})\{Q(\overline{𝐮},𝐮,𝐳,\overline{𝐳})S(𝐮,\overline{𝐮})S(𝐳,\overline{𝐳})+S(𝐮)S(\overline{𝐮})S(𝐮,\overline{𝐮})`$
$`Q(\overline{𝐮},𝐮,\mathrm{𝟎},\overline{𝐳})S(\overline{𝐳})S(𝐮,\overline{𝐮})S(\overline{𝐮},𝐮)S(𝐳,\overline{𝐮})S(𝐳,𝐮)`$
$`+Q(\overline{𝐮},\mathrm{𝟎},𝐳,\overline{𝐳})S(𝐳,\overline{𝐳})S(\overline{𝐮})+S^2(\overline{𝐮})`$
$`S(\overline{𝐮})S(\overline{𝐳})S(\overline{𝐳},\overline{𝐮})S(\overline{𝐮})S(𝐳)S(𝐳,\overline{𝐮})\}`$
$`+\stackrel{}{f}(𝐳)\stackrel{}{f}(\overline{𝐳})\{Q(\overline{𝐮},𝐮,𝐳,\overline{𝐳})S(𝐳,\overline{𝐳})S(𝐮,\overline{𝐮})+S^2(𝐮,\overline{𝐮})`$
$`Q(\overline{𝐮},𝐮,\mathrm{𝟎},\overline{𝐳})S(𝐮,\overline{𝐮})S(\overline{𝐳})+S^2(𝐳,\overline{𝐳})`$
$`Q(\overline{𝐮},𝐮,𝐳,\mathrm{𝟎})S(𝐮,\overline{𝐮})S(𝐳)+1S^2(\overline{𝐳})S^2(𝐳)\}].`$ (66)
We note that the two gluon correlation function of the perturbative $`q\overline{q}`$ dipole projectile was derived in Ref. in the large $`N_c`$ limit. We have checked that in the limit of arbitrary large dipole size, the final result of Ref. coincides with (66).
## VI Conclusions
In this paper, we have shown that the framework of perturbative saturation provides for a mechanism which can shift the maximum strength which of azimuthal particle correlations away from $`180^{}`$. We understand the physics of this phenomenon in the following way. There are two basic mechanisms by which a quark and a gluon can be produced in the final state. The first one is the QCD analog of Molière scattering. The quark in the projectile wave function scatters from the perturbative part of the target gluon field and then radiates a gluon in the final state. This final state radiation is predominantly back-to-back and generates maximal correlations at angle $`\pi `$. The second mechanism is the multiple scattering of the $`qg`$ component of the initial state on the target. The final states produced by this mechanism depend strongly on the separation between the initial quark and gluon in the transverse plane. For components with large separation (small relative transverse momentum $`|𝐤𝐩|Q_s`$ in the initial state), the quark and the gluon scatter independently off the target fields. This process produces uncorrelated $`qg`$ pairs in the final state and leads to the broadening of the angular distribution. However, the components of the incoming state which have transverse size smaller than the correlation length of the target fields scatter coherently, i.e. they scatter effectively as a single particle. These quark-gluon components pick up a typical soft momentum (of order $`Q_{s,0}`$, as explained in Section III), which is equally shared between the two partons. As a result, the two partons emerge from the interaction region with momenta shifted in the same direction by an equal amount of order $`Q_{s,0}/2`$. In the initial state, the total transverse momentum vanishes, and thus the momenta of the quark and the gluon are balanced. The equal momentum transfer therefore produces angular correlation in the final state which is peaked away from $`180^{}`$. The maximum correlation produced by this mechanism for large trigger momenta $`k`$ can be simply estimated to lie about $`Q_{s,0}/\sqrt{2}k`$ away from $`180^{}`$. Whether this shift in the maximal correlation angle is actually observable in the spectrum depends very much on the relative importance of the Molière component and the coherent soft scattering component. Our numerical results suggest that the coherent scattering component is dominant for the trigger momenta around $`Q_s`$.
We also observe that the coherent scattering component, which shifts the maximal correlation strength away from $`180^{}`$, is enhanced and dominates over a wider range of trigger momenta, if the projectile system has smaller transverse size. To establish this statement, we use the fact that our calculation allows us to regulate the transverse size of the incoming projectile via the cut-off $`\mathrm{\Lambda }_{\mathrm{cut}}`$. Although it is difficult to draw firm phenomenological conclusions, we believe that there is potential for an observable effect.
Also, it would be interesting to study how this picture discussed above is affected by low-$`x`$ evolution. One may expect that since $`Q_s`$ increases with rapidity, the maximal correlation angle at forward rapidities should move further away from $`180^{}`$. However, low-$`x`$ evolution does not only affect the value of the saturation scale, but also changes significantly the efficiency of the target as a function of momentum transfer. In particular, the momentum dependence of the target gluon distribution decreases slower at large momenta thus effectively increasing the Molière hard scattering component. Hence, it is not clear to us how the azimuthal shift of the maximal two-parton correlation strength will evolve with $`x`$.
## Appendix A Alternative calculation of two-gluon correlation function (66)
In this appendix, we derive the two-gluon correlation function (66) following the formulation given in . We start from the state $`|\psi _{\mathrm{in}}^\alpha `$ in (24) of a single quark with color $`\alpha `$ and its gluon cloud, expanded in perturbation theory up to $`O(g^2)`$,
This is the incoming state with two gluons of color $`b`$ and $`c`$. The crossed line indicates a probability conserving virtual correction, such that the state $`|\psi _{\mathrm{in}}^\alpha `$ is normalized to unity up to $`O(g^2)`$. In this approximation, the states do not depend on rapidity, and the rapidity labels are suppressed in the following.
The interaction of $`|\psi _{\mathrm{in}}^\alpha `$ with the target results in phase shifts by eikonal Wilson lines, which are in the fundamental ($`W^F`$) and adjoint ($`W^A`$) representation, respectively. This leads to the outgoing state
$`|\mathrm{\Psi }_{\mathrm{out}}^\alpha `$ $`=`$ $`\left(1{\displaystyle \frac{C_F}{2}}{\displaystyle 𝑑𝐱\stackrel{}{f}(𝐱)\stackrel{}{f}(𝐱)}\right)W_{\alpha \beta }^F(\mathrm{𝟎})|\beta `$ (67)
$`+i{\displaystyle 𝑑𝐱\stackrel{}{f}(𝐱)\left(T^aW^F(\mathrm{𝟎})\right)_{\alpha \beta }W_{ab}^A(𝐱)|\beta ;b(𝐱)}`$
$`{\displaystyle \frac{1}{2}}{\displaystyle }d𝐱d𝐲[\stackrel{}{f}(𝐱)\stackrel{}{f}(𝐲)\left(T^aT^dW^F(\mathrm{𝟎})\right)_{\alpha \beta }W_{ab}^A(𝐱)W_{dc}^A(𝐲)`$
$`+\stackrel{}{f}(𝐱)\stackrel{}{f}(𝐲𝐱)\left(T^eW^F(\mathrm{𝟎})\right)_{\alpha \beta }\left(T_{ad}^eW_{ab}^A(𝐱)W_{dc}^A(𝐲)\right)]|\beta ;b(𝐱),c(𝐲).`$
The quark propagates at the transverse position $`𝐱_q=0`$, and the gluons propagate at $`𝐱`$ and $`𝐲`$, respectively.
According to (32), we have to subtract the overlap with the dressed incoming state. This leads to the state $`|\delta \mathrm{\Psi }_\alpha `$ which to lowest order in $`\stackrel{}{f}\stackrel{}{f}`$ consists of the following contributions:
Here, we use the diagrammatic shorthand of Ref. in which thick lines denote partons propagating through the target and thus picking up eikonal Wilson lines in the corresponding representation. The first two graphs correspond to the case that the quark emits both gluons before the target and all three partons propagate through the target (thick lines). The third diagram can be viewed as emission after interaction since the outgoing lines do not carry eikonal factors (thin lines). This third diagram, as well as the remaining three arise from the subtraction of the overlap in (32).
A convenient expression for $`|\delta \mathrm{\Psi }_\alpha `$ is obtained by applying identities like
$$\left(T^bW^F(\mathrm{𝟎})T^d\right)_{\alpha \beta }=\left(T^bT^aW^F(\mathrm{𝟎})\right)_{\alpha \beta }W_{ad}^A(\mathrm{𝟎}).$$
(68)
This allows us to shift the Wilson operator $`W^F(\mathrm{𝟎})`$ such that we can write
$`|\delta \mathrm{\Psi }_\alpha `$ $`=`$ $`{\displaystyle }d𝐱d𝐲\stackrel{}{f}(𝐱)\stackrel{}{f}(𝐲)\left(T^aT^dW^F(\mathrm{𝟎})\right)_{\alpha \beta }[W_{ab}^A(𝐱)(W_{dc}^A(\mathrm{𝟎})W_{dc}^A(𝐲))`$ (69)
$`W_{db}^A(\mathrm{𝟎})(W_{ac}^A(\mathrm{𝟎})W_{ac}^A(𝐲))]|\beta ;b(𝐱),c(𝐲)`$
$`+{\displaystyle }d𝐱d𝐲\stackrel{}{f}(𝐱)\stackrel{}{f}(𝐲𝐱)\left(T^eW^F(\mathrm{𝟎})\right)_{\alpha \beta }[T_{ad}^eW_{ab}^A(𝐱)W_{dc}^A(𝐲)`$
$`W_{ea}^A(𝐱)T_{bc}^a]|\beta ;b(𝐱),c(𝐲).`$
This expression may be written in the shorthand form, introducing an operator $`𝒪(𝐱,𝐲)`$ as follows
$$|\delta \mathrm{\Psi }_\alpha =𝑑𝐱𝑑𝐲𝒪_{\alpha \beta }^{bc}(𝐱,𝐲)|\beta ;b(𝐱),c(𝐲).$$
(70)
This allows us to write the two-gluon correlation in a compact way by calculating the expectation value of the two-particle number operator in this state, averaged over the incoming quark color index $`\alpha `$,
$`{\displaystyle \frac{1}{N}}{\displaystyle \underset{\alpha }{}}\delta \mathrm{\Psi }_\alpha |a^{}(𝐩_1,\xi )a(𝐩_1,\xi )a^{}(𝐩_2,\eta )a(𝐩_2,\eta )|\delta \mathrm{\Psi }_\alpha `$
$`={\displaystyle \frac{1}{(2\pi )^4}}{\displaystyle _{𝐳\overline{𝐳}𝐮\overline{𝐮}}}e^{i𝐩_1(𝐳\overline{𝐳})i𝐩_2(𝐮\overline{𝐮})}\left[𝒪_{\alpha \beta }^{bc}(\overline{𝐳},\overline{𝐮})\right]^{}𝒪_{\alpha \beta }^{bc}(𝐳,𝐮)_T.`$ (71)
Working out the target averages as described in , we obtain after some color algebra the result given in (66).
|
warning/0506/q-bio0506034.html
|
ar5iv
|
text
|
# The Shapley Value of Phylogenetic Trees
## 1. Introduction
The Shapley value is arguably the most important solution concept for $`n`$-player cooperative games. Given a set of players $`N`$ of size $`n=|N|`$ in a cooperative game $`v`$, the Shapley value $`\phi (N,v)`$ is the unique imputation vector that satisfies four “fairness” criteria (the Shapley axioms) that we shall discuss later. In this paper we consider the game $`v_𝒯`$ induced by an unrooted $`n`$-leaf tree $`𝒯`$ in which each edge is assigned a positive number called an edge weight. In this context, the players are represented by the leaves of the tree and the value of any coalition $`S`$ is the total weight of the subtree spanned by the members of $`S`$.
In a more applied context, we consider games induced by a phylogenetic tree in which players are species and the tree represents a proposed evolutionary relationship among the species. We suggest that a biological interpretation for the Shapley value is a notion of the average marginal diversity that a species brings to any group, and we study how the Shapley value depends on the edge weights and topology of the tree.
One possible application of the Shapley value of a phylogenetic tree is the economic theory of biodiversity preservation . In such contexts, quantifying the biological diversity of a species or a group of species is of great interest; many measures have been proposed (see, e.g., ) . The Noah’s ark problem asks how to prioritize species in a population if only some limited number can be saved; we suggest that Shapley value provides a natural ranking criterion as it is provides a measure of the contribution each species brings to the diversity of a group.
The literature applying game-theoretic solution concepts to an analysis of trees appears to be limited. One closely related example is Kar , who studies cost-sharing in a network structure and characterizes the Shapley value of the minimum cost spanning tree game of an arbitrary graph. Also, as well as study values for games that arise from a tree structure. However, these three works differ from ours because there each node of a graph is considered as a player in the game, whereas we specifically study tree games and allow only leaves as players. Day and McMorris propose suitable axioms for a consensus rule that will aggregate several phylogenetic trees into one consensus tree; this differs from the thrust of our work, which is to consider one tree and explore the interpretation and properties of the Shapley value of the associated tree game.
In the next section we provide a biological interpretation for the Shapley value of phylogenetic trees. Then we discuss the mathematics of calculating the Shapley value on tree games, starting with some examples on small trees. We determine the linear transformation that shows how the Shapley value depends on the edge weights of the tree, and compute a null space basis that shows how to vary edge weights without changing the Shapley value. We also explain how these depend on the tree topology. We conclude this paper by developing an analogue of Shapley’s theorem that characterizes the Shapley value on games by four axioms. We show that on the smaller class of tree games, the Shapley value is characterized by those four axioms plus an additional axiom.
## 2. Phylogenetic Trees and the Shapley Value
### 2.1. Phylogenetic trees
Evolutionary relationships between species are frequently represented by a phylogenetic tree. Evidence for such relationships can come from a variety of sources, such as genomic data or morphological comparisons, and much work has been done to develop methods for constructing a phylogenetic tree from such data (for surveys, see Felsenstein and Semple-Steel ).
Phylogenetic trees are usually binary trees in which each internal node represents a bifurcation in some characteristic and the leaves are the species for which we have data. Each edge has a weight that represents some unit of distance between the nodes at its endpoints (for instance, it could be the time between speciation events). Figure 1 gives a small example of what a (rooted) phylogenetic tree could look like. However, in this paper we shall not be concerned with the location of the root of a tree, so all our trees will be unrooted.
Formally, we shall think of a phylogenetic tree $`𝒯`$ as an unrooted tree with leaf set $`N:=\{1,\mathrm{},n\}`$ (representing the species in the population), edge set $`E`$, and and an edge weight $`\alpha _k`$ for each edge $`k`$ in $`E`$.
### 2.2. The Shapley value
In cooperative game theory, a cooperative game is a pair $`(N,v)`$ consisting of a set of players $`N=\{1,2,\mathrm{},n\}`$ and a characteristic function $`v`$ that takes every subset of $`N`$ (called a coalition) to a real number (called the worth of the coalition). The subset consisting of all players is called the grand coalition. Formally, if $`2^N`$ is the set of all subsets of $`N`$, then $`v:2^N`$. For instance, $`N`$ could be a set of companies and $`v`$ could describe the profit that each coalition of companies could make if the members of that coalition worked together.
One of the basic questions in cooperative game theory is: if players work together to achieve some total worth (in our example, profit), how should players then distribute their worth (profit) among themselves?
As all (Pareto efficient) solution concepts from cooperative game theory do, the value introduced by Shapley suggests a “fair” distribution of the total worth of the entire set of players $`N`$ among the members of $`N`$. Given a cooperative game $`(N,v)`$, the Shapley value is a vector $`\phi =(\phi _i)`$ defined by the formula
(1)
$$\phi _i(N,v)=\frac{1}{n!}\underset{\begin{array}{c}SN\\ iS\end{array}}{}(s1)!(ns)!(v(S)v(Si))$$
where $`s=|S|`$ is the size of the coalition $`S`$ and $`n=|N|`$ is the total number of players.
The formula above has a sensible interpretation that suggests a rationale for the Shapley value to obtain a “fair” distribution. For a player $`iN`$ and a coalition $`SN`$ that contains $`i`$, the quantity $`v(S)v(Si)`$ describes $`i`$’s marginal contribution to the worth of $`S`$. Then, if we choose an ordering of the players (uniformly at random, in $`n!`$ ways) and if $`i`$ appears as the $`s`$-th person in that order, then $`i`$’s marginal contribution will be $`v(S)v(Si)`$ for each ordering in which the members of $`Si`$ appear before $`i`$ and the members of $`NS`$ appear after $`i`$. This may happen in $`(s1)!(ns)!`$ ways. Hence the combinatorial form of (1) reflects the Shapley value’s interpretation as the expected marginal contribution that $`i`$ makes.
### 2.3. The Phylogenetic Tree Game
Given a phylogenetic tree $`𝒯`$, we can define an associated cooperative game $`(N,v_𝒯)`$ that we call a phylogenetic tree game. Let $`N`$ be the set of leaves of the tree (species). For any subset $`SN`$ of species, consider the unique spanning subtree containing the members in $`S`$, and let $`v_𝒯(S)`$ be the sum of the edge weights of that spanning tree. Thus for each set $`S`$ we may think of $`v_𝒯(S)`$ as a measure of the phylogenetic diversity within $`S`$. This measure and its computational aspects have been studied much in recent years (see e.g., ).
Then the pair $`(N,v_𝒯)`$ naturally forms a cooperative game. Although species can hardly be compared with rationally acting agents (as usually assumed in theory of cooperative games), we may still ask for a meaningful re-interpretation of game-theoretic solution concepts such as the Shapley value in the context of phylogenetic trees.
Given a phylogenetic tree game $`(N,v_𝒯)`$, equation (1) suggests that the Shapley value of a given species may be thought of as its average marginal diversity, i.e., the average diversity the species can be expected to add to a group that it joins. So if $`\phi _i>\phi _j`$, then species $`i`$ can be thought to contribute a greater diversity to a group than species $`j`$ might.
###### Example 1.
From direct calculations using (1), the five-leaf tree $`𝒯`$ in Figure 2 has Shapley value
$$\phi =(\phi _A,\phi _B,\phi _C,\phi _D,\phi _E)=(5.28,6.78,4.2,4.95,2.78)$$
as we will show in Section 3.2.
### 2.4. The Shapley Value Axioms
Besides the interpretation of the Shapley value as an average expected marginal contribution, there is an axiomatization of the Shapley value (see ) that uniquely characterizes it by a set of (desirable) properties. We review the axioms presented by Shapley and discuss their plausibility in the present setting as properties of phylogenetic trees. Let therefore $`𝒱:=\{v:2^N|v(\mathrm{})=0\}`$ be the set of all cooperative games with $`n`$ players.
1. (Pareto Efficiency Axiom) The Shapley value is Pareto efficient, i.e., $`_{iN}\phi _i(N,v)=v(N)`$ for all $`v𝒱`$.
This axiom just states that the total diversity present within a phylogenetic tree will be distributed and ascribed to the species within it. This is a reasonable axiom, given that the purpose of a solution concept for a cooperative game is to distribute the worth of the grand coalition among its members. In this context, the natural interpretation is that the Shapley value answers the question of how much a specific species is responsible for the total diversity, or, put another way, what is its share of $`v_𝒯(N)`$.
2. (Symmetry Axiom) For any permutation of players $`\pi :NN`$ the Shapley value satisfies $`\phi (\pi v)=\pi \phi (v)`$, where $`\pi v`$ is the permuted game given by $`\pi v(S):=v(\pi ^1(S))`$ for all $`SN`$ and $`\pi \phi (v)`$ is the permuted solution vector, i.e., $`(\pi \phi (v))_i:=\phi _{\pi ^1(i)}(v)`$.
The symmetry axiom states that a player’s allocation should not be based on her name. Another consequence of the symmetry axiom is if exchanging two players causes no difference in the worth that each adds to any coalition, then they should have the same Shapley value. Biologically speaking, if two species play the same role within a tree then they should be ascribed the same responsibility for diversity, which seems to be a plausible requirement.
3. (Dummy Axiom) A dummy player is one that does not add worth to the value of any coalition. This axiom says that dummy players should have a Shapley value of zero.
This axiom is vacuously satisfied in the case of a phylogenetic tree game because there are no dummy species. To see this, note that every species $`i`$ adds worth to the coalition that consists of a single species $`ji`$, because the weight of the subtree containing $`i`$ and $`j`$ is the sum of the edge weights between $`i`$ and $`j`$ and is therefore non-zero, but the weight of the subtree consisting of the singleton $`j`$ is zero. (Even though there are no dummy species, this is still a reasonable axiom here, since any species that does not diversify any coalition should get value zero.)<sup>1</sup><sup>1</sup>1In Section 6 we will replace the dummy axiom by a different one to characterize the Shapley value on the class of games that actually come from trees.
4. (Additivity Axiom) Given two games $`(N,v)`$ and $`(N,w)`$ in $`𝒱`$ with the same set of players $`N`$, define the sum game $`(N,v+w)`$ with characteristic function $`(v+w)(S)=v(S)+w(S)`$ for every coalition $`S`$. This axiom stipulates that the Shapley value of the sum game should be the sum of the Shapley values of the individual games: $`\phi (N,v+w)=\phi (N,v)+\phi (N,w)`$.
As an example, suppose we are given nucleotide sequences for a set of species $`N`$, and each sequence has length 200. For each pair of species $`i,j`$ consider the (rather crude) measure of distance $`d(i,j)`$ to be the number of positions in which the sequences differ. The pairwise distance data can be used to construct a tree (using any standard method) and consequently, a tree game. Thus the first 100 positions of the sequences can be used to construct a tree game $`(N,v_1)`$, and the second 100 positions a tree game $`(N,v_2)`$. Then the Shapley value of the sum game $`(N,v_1+v_2)`$ is the sum of the Shapley values for each game. This seems plausible in this context, since if the pairwise distances $`d(i,j)`$ from both sets of 100 positions actually arise from a tree metrics on the same topological tree, then the sum game will arise from the tree reconstructed from all 200 positions.
## 3. Examples and Motivation: The Shapley Value for Small Trees
As can be seen from (1), the Shapley value of a tree game is a linear function of the edge weights of the tree. We call that linear transformation the Shapley transformation. Before deriving a general formula for this transformation in the subsequent section, we study the Shapley transformation for games induced by unrooted three-, four-, five- and six-leaf trees.
We will refer to the weights of edges incident to leaves as leaf weights and other edge weights as internal edge weights. Note that for an unrooted $`n`$-leaf tree, there are $`n2`$ internal nodes and $`n3`$ internal edges in $`E`$. In what follows, the superscript <sup>T</sup> denotes the transpose.
###### Definition 2.
Let $`𝒯`$ be an $`n`$-leaf tree with leaves $`N=\{1,\mathrm{},n\}`$, associated leaf weights $`\alpha _1,\mathrm{},\alpha _n`$ and internal edges $`I_1,\mathrm{},I_{n3}`$ with associated internal edge weights $`\alpha _{I_1},\mathrm{},\alpha _{I_{n3}}`$. Let $`\stackrel{}{E}`$ be a vector consisting of the edge weights in this order: $`(\alpha _1,\mathrm{},\alpha _n,\alpha _{I_1},\mathrm{},\alpha _{I_{n3}})^T`$. Define $`𝐌=𝐌(N,v_𝒯)`$ to be the $`n\times (2n3)`$ matrix that represents the Shapley transformation, so that the Shapley value of the game $`v_𝒯`$ is
$$\phi (N,v_𝒯)=(\phi _1,\phi _2,\mathrm{},\phi _n)^T=𝐌\stackrel{}{E}$$
where $`\phi _i`$ is the Shapley value associated with leaf $`i`$. Note that $`𝐌`$ depends on the topology of the $`n`$-leaf tree.
Later in Theorem 4 we determine a formula for $`𝐌[i,k]`$, which is the coefficient of edge weight $`k`$ in the calculation of the Shapley value of $`i`$. But first, we give a few examples.
### 3.1. Three-Leaf Trees
Topologically, there is only one unrooted three-leaf tree $`𝒯`$. Let the leaves represent players A, B, and C with corresponding leaf weights $`\alpha `$, $`\beta `$, and $`\gamma `$ as seen in Figure 3.
The characteristic function $`v_𝒯`$ for this game is
$$v_𝒯(A)=v_𝒯(B)=v_𝒯(C)=0,$$
$$v_𝒯(AB)=\alpha +\beta ,v_𝒯(AC)=\alpha +\gamma ,v_𝒯(BC)=\beta +\gamma ,$$
$$v_𝒯(ABC)=\alpha +\beta +\gamma .$$
Using Definition 2, we can calculate the Shapley value by $`\phi =(\phi _A,\phi _B,\phi _C)=𝐌\stackrel{}{\mathrm{}}`$ where $`\stackrel{}{\mathrm{}}`$ is the vector of leaf weights $`(\alpha ,\beta ,\gamma )^T`$ and
$$𝐌=\frac{1}{6}\left[\begin{array}{ccc}4& 1& 1\\ 1& 4& 1\\ 1& 1& 4\end{array}\right].$$
It is apparent that we can solve for $`\alpha `$, $`\beta `$, and $`\gamma `$ in terms of $`\phi `$ by inverting $`𝐌`$:
$$\stackrel{}{\mathrm{}}=\frac{1}{3}\left[\begin{array}{ccc}5& 1& 1\\ 1& 5& 1\\ 1& 1& 5\end{array}\right]\left(\begin{array}{c}\phi _A\\ \phi _B\\ \phi _C\end{array}\right).$$
This means the Shapley value of a 3-leaf tree uniquely determines the tree representing the game.
### 3.2. Four- and Five-Leaf Trees
Similarly, we can calculate the Shapley value for each player in the four- and five-leaf cases. There is a unique tree topology for each case, as shown in Figure 4.
The Shapley value for the general four-leaf tree game is
$$\frac{1}{24}\left[\begin{array}{ccccc}18& 2& 2& 2& 6\\ 2& 18& 2& 2& 6\\ 2& 2& 18& 2& 6\\ 2& 2& 2& 18& 6\end{array}\right]\left(\begin{array}{c}\alpha \\ \beta \\ \gamma \\ \delta \\ \mu \end{array}\right).$$
Similarly for the five-leaf tree game, the Shapley value is
$$\frac{1}{120}\left[\begin{array}{ccccccc}96& 6& 6& 6& 6& 36& 16\\ 6& 96& 6& 6& 6& 36& 16\\ 6& 6& 96& 6& 6& 16& 36\\ 6& 6& 6& 96& 6& 16& 36\\ 6& 6& 6& 6& 96& 16& 16\end{array}\right]\left(\begin{array}{c}\alpha \\ \beta \\ \gamma \\ \delta \\ ϵ\\ \mu \\ \rho \end{array}\right).$$
This formula produces the calculation in Example 1.
It is apparent from the fact that there are more variables (edge weights) than equations that there is not a unique set of (possibly negative) edge weights for a given Shapley value. That is, there is not a unique tree corresponding to a given Shapley value. The null space of $`𝐌`$ will therefore help us determine which weighted trees have the same Shapley value. A basis for the null space of $`𝐌`$ for the four-leaf tree is
$$\left\{\left(\begin{array}{c}1/4\\ 1/4\\ 1/4\\ 1/4\\ 1\end{array}\right)\right\}$$
This means that given a tree $`𝒯`$, we can produce other trees with the same Shapley value by reducing the leaf weights by $`1/4`$ for each unit increase in the internal edge weight.
Similarly, a null space basis for the five-leaf tree is
$$\{\left(\begin{array}{c}1/3\\ 1/3\\ 1/9\\ 1/9\\ 1/9\\ 1\\ 0\end{array}\right),\left(\begin{array}{c}1/9\\ 1/9\\ 1/3\\ 1/3\\ 1/9\\ 0\\ 1\end{array}\right)\}.$$
For example, the first basis element (multiplied by -3) shows us that the tree in Figure 5 has the same Shapley value as the tree in Figure 2.
### 3.3. Six-Leaf Trees
For our last direct calculation, let us consider the games represented by six-leaf trees. In this case there are two topologies for unrooted trees with six leaves (see figure 6).
The Shapley value for the first and second six-leaf trees are, respectively,
$$\phi (N,v_𝒯)=\frac{1}{720}\left[\begin{array}{ccccccccc}600& 24& 24& 24& 24& 24& 240& 60& 120\\ 24& 600& 24& 24& 24& 24& 240& 60& 120\\ 24& 24& 600& 24& 24& 24& 60& 240& 120\\ 24& 24& 24& 600& 24& 24& 60& 240& 120\\ 24& 24& 24& 24& 600& 24& 60& 60& 120\\ 24& 24& 24& 24& 24& 600& 60& 60& 120\end{array}\right]\left(\begin{array}{c}\alpha \\ \beta \\ \gamma \\ \delta \\ ϵ\\ \pi \\ \mu \\ \rho \\ \eta \end{array}\right),$$
$$\phi (N,v_𝒯^{})=\frac{1}{720}\left[\begin{array}{ccccccccc}600& 24& 24& 24& 24& 24& 240& 60& 60\\ 24& 600& 24& 24& 24& 24& 240& 60& 60\\ 24& 24& 600& 24& 24& 24& 60& 240& 60\\ 24& 24& 24& 600& 24& 24& 60& 240& 60\\ 24& 24& 24& 24& 600& 24& 60& 60& 240\\ 24& 24& 24& 24& 24& 600& 60& 60& 240\end{array}\right]\left(\begin{array}{c}\alpha \\ \beta \\ \gamma \\ \delta \\ ϵ\\ \pi \\ \mu \\ \rho \\ \eta \end{array}\right).$$
As with the four and five leaf cases, both topologies of the six leaf tree allow for many trees to possess the same Shapley value. The basis for the null space of the first six-leaf tree is
$$\{\left(\begin{array}{c}3/8\\ 3/8\\ 1/16\\ 1/16\\ 1/16\\ 1/16\\ 1\\ 0\\ 0\end{array}\right),\left(\begin{array}{c}1/16\\ 1/16\\ 3/8\\ 3/8\\ 1/16\\ 1/16\\ 0\\ 1\\ 0\end{array}\right),\left(\begin{array}{c}1/6\\ 1/6\\ 1/6\\ 1/6\\ 1/6\\ 1/6\\ 0\\ 0\\ 1\end{array}\right)\}$$
and for the second six-leaf tree is
$$\{\left(\begin{array}{c}3/8\\ 3/8\\ 1/16\\ 1/16\\ 1/16\\ 1/16\\ 1\\ 0\\ 0\end{array}\right),\left(\begin{array}{c}1/16\\ 1/16\\ 3/8\\ 3/8\\ 1/16\\ 1/16\\ 0\\ 1\\ 0\end{array}\right),\left(\begin{array}{c}1/16\\ 1/16\\ 1/16\\ 1/16\\ 3/8\\ 3/8\\ 0\\ 0\\ 1\end{array}\right)\}.$$
### 3.4. Notes on Relationship between Trees and Shapley Values
From these examples, we make a few observations.
1. Any Shapley value $`n`$-vector can be realized by adjusting the edge weights of an $`n`$-leaf tree. This may involve positive as well as nonpositive edge weights. However, the positive hull of the column vectors of the matrix $`𝐌`$ can be realized as the Shapley value of trees with nonnegative edge weights.
2. When $`n4`$, there is not a unique $`n`$-leaf tree corresponding to a given Shapley value because the null space is nontrivial.
3. The null spaces for the two six-leaf trees are different (since there is exactly one basis for each null space whose projection to the last 3 coordinates are the standard unit vectors in $`^3`$, and these bases are different for the given null spaces). Moreover, one may check that there is no permutation of the coordinates of one null space that will identify it with the other null space (we say such spaces are not permutation equivalent). As we shall see in Section 5, if the null spaces are not permutation equivalent, then the two trees must not be isomorphic (Theorem 8).
4. Under close inspection, one notices a relationship between the numbers of leaves on each side of an internal edge (the split counts) and quantities such as the entries of the Shapley transformation matrix and the null space basis vectors. We exhibit their explicit dependence in the following sections.
## 4. The Shapley Transformation
We first show the contribution of each edge weight to the Shapley value; these are the entries of the matrix $`𝐌`$ representing the Shapley transformation. The following theorem gives us a quick way of finding the $`(i,k)`$th entry of $`𝐌`$. Before we state and prove the theorem, we need a definition that will be instrumental throughout the rest of this paper.
###### Definition 3.
Let $`𝒯`$ be an $`n`$-leaf tree with leaves $`N`$ and edges $`E`$. For $`iN`$ and $`kE`$, the removal of edge $`k`$ splits $`𝒯`$ into two subtrees. Let $`𝒞(i,k)`$ denote the set of leaves in the subtree that contains $`i`$ (the “containing” subtree) and let $`(i,k)`$ denote the set of leaves in the other subtree that is “far” from $`i`$. We then denote the number of leaves of $`𝒞(i,k)`$ and $`(i,k)`$ as $`c(i,k)`$ and $`f(i,k)`$, respectively.
If it is obvious what leaf $`i`$ and edge $`k`$ we are referring to, we will simply write $`c,f`$ instead of $`c(i,k),f(i,k)`$. Note that $`n=c+f`$. We call $`c,f`$ the split counts associated with leaf $`i`$ and edge $`k`$. As we shall see, the split counts will arise frequently in our results on the Shapley transformation.
###### Theorem 4.
Let $`𝒯`$ be an $`n`$-leaf tree. The $`(i,k)`$th entry of the Shapley transformation matrix $`𝐌`$ is given by
$$𝐌[i,k]=\frac{f(i,k)}{nc(i,k)}.$$
###### Proof.
Fix leaf $`i`$. To count the number of times a given edge weight contributes to $`i`$’s Shapley value, we need to know how many times it is in the marginal contribution of $`i`$ for coalitions of size $`s`$. Edge weight $`\alpha _k`$ will be part of $`i`$’s marginal contribution if the other $`s1`$ members of the coalition are from the far side of the edge from $`i`$. So
$$𝐌[i,k]=\frac{1}{n!}\underset{s=2}{\overset{n}{}}(s1)!(ns)!\left(\genfrac{}{}{0pt}{}{f(i,k)}{s1}\right)=\frac{1}{n!}\underset{s=2}{\overset{n}{}}\frac{(ns)!f(i,k)!}{(f(i,k)s+1)!}.$$
Using the fact $`f=nc`$, the above expression can be rewritten:
$$\frac{1}{n!}\underset{s=2}{\overset{n}{}}(nc)!(c1)!\left(\genfrac{}{}{0pt}{}{ns}{c1}\right)=\frac{(nc)!(c1)!}{n!}\underset{j=1}{\overset{n1}{}}\left(\genfrac{}{}{0pt}{}{j1}{c1}\right).$$
We use the identity
$$\underset{j=1}{\overset{n}{}}\left(\genfrac{}{}{0pt}{}{j1}{c1}\right)=\left(\genfrac{}{}{0pt}{}{n}{c}\right)=\left(\genfrac{}{}{0pt}{}{n1}{c1}\right)\frac{f}{c}+\left(\genfrac{}{}{0pt}{}{n1}{c1}\right)$$
to obtain
$$𝐌[i,k]=\frac{(nc)!(c1)!}{n!}\left(\genfrac{}{}{0pt}{}{n1}{c1}\right)\frac{f}{c}=\frac{f}{nc}.$$
This result is particularly nice because it shows how the Shapley value’s dependence on any edge weight simply hinges on the number of leaves on either side of that edge. Consider the following example.
###### Example 5.
Using Theorem 4 we will calculate the coefficient of $`\mu `$ in player A’s Shapley value for a five-leaf tree. Let the edge with edge weight $`\mu `$ be $`I_1`$. There are three leaves in $`(A,I_1)`$ and two leaves in $`𝒞(A,I_1)`$. Thus,
$$𝐌[1,6]=\frac{3}{52}$$
which is the same as the $`(A,\mu )`$ entry $`36/120`$ in the Shapley transformation of the five-leaf tree given in section 3.2.
## 5. The Null Space of the Shapley Transformation
Now we will also use Theorem 4 to understand the dependence of the null space of the Shapley transformation on the split counts, as suggested in Section 3.4.
The following theorem exhibits a null space basis of $`𝐌`$ in terms of the split counts.
###### Theorem 6.
Let $`𝒯`$ be an $`n`$-leaf tree with leaves $`N=\{1,\mathrm{},n\}`$ and internal edges $`I_1,\mathrm{},I_{n3}`$. The dimension of the null space of $`𝐌=𝐌(N,v_𝒯)`$ is $`n3`$. A basis for the null space is the collection of vectors $`\{w_{I_k}\}`$ in $`^{2n3}`$, one for each internal edge $`I_k`$:
(2)
$$(w_{I_k})_i=\{\begin{array}{cc}\frac{f(i,k)1}{(n2)c(i,k)}\hfill & \text{ if }1in\hfill \\ 1\hfill & \text{ if }i=n+k\hfill \\ 0\hfill & \text{ otherwise}\hfill \end{array}$$
for all $`k\{1,\mathrm{},n3\}`$ and entries $`i\{1,\mathrm{},2n3\}`$, where the first $`n`$ entries correspond to leaves and the last $`n3`$ entries correspond to internal edges.
Before proving the theorem, we give an example.
###### Example 7.
Consider the five-leaf tree in Figure 4. Let $`I_1,I_2`$ be the internal edges with weight $`\mu ,\rho `$, respectively. We use Theorem 6 to determine the null space vector $`w_{I_1}`$. The $`5+1=6`$-th entry of $`w_{I_1}`$ is 1 and all entries after that are 0. To find the first five entries of the vector, consider the two subtrees obtained by removing $`I_1`$ from the tree, namely, the subtrees AB and CDE. By (2), the first two entries of the $`w_{I_1}`$ corresponding to A and B will be
$$\frac{31}{(52)2}=\frac{1}{3}$$
and the next three entries corresponding to C, D, and E are
$$\frac{21}{(52)3}=\frac{1}{9}.$$
This agrees with the first null space basis vector we exhibited in Section 3.2. (The other basis vector there is $`w_{I_2}`$.)
Now we prove Theorem 6.
###### Proof.
Let $`𝒯`$ be an $`n`$-leaf tree. Consider the $`i`$th leaf. If we let $`𝐌`$ be the matrix of Shapley value coefficients for $`𝒯`$ then we want to show
(3)
$$\underset{j=1}{\overset{2n3}{}}𝐌[i,j](w_{I_k})_j=0.$$
Fix $`k\{1,\mathrm{},n3\}`$. Note that by Theorem 4, for all leaves $`ji`$, $`𝐌[i,j]=\frac{1}{n(n1)}`$ and $`𝐌[i,i]=\frac{n1}{n}`$. The only other non-zero entry of $`𝐌`$ we need to consider is the one associated with the $`(n+k)`$-th edge of the tree, i.e., the internal edge $`I_k`$. The split counts for $`I_k`$ are $`c,f`$ and so
$$𝐌[i,n+k]=\frac{f}{nc}.$$
Thus, showing (3) is the same as showing that
$$f\frac{1}{n(n1)}\frac{c1}{(n2)f}(c1)\frac{1}{n(n1)}\frac{f1}{(n2)c}\frac{n1}{n}\frac{f1}{(n2)c}+\frac{f}{nc}=0,$$
which can be checked by algebraic manipulation and the fact that $`n=f+c`$.
Thus $`w_{I_k}`$ is in the null space of the Shapley transformation $`𝐌`$. It is apparent that the null space has dimension $`n3`$ and the $`w_{I_k}`$ are linearly independent. Therefore the $`w_{I_k}`$ form a basis of the null space of $`𝐌`$. ∎
We note that this basis $`\{w_{I_k}\}`$, $`k=1,\mathrm{},n3`$, is uniquely determined by fixing the last $`n3`$ coordinates to be the standard basis vectors in $`^{n3}`$. For this reason we refer to this basis as the standard null space basis for $`𝐌`$.
An immediate corollary of Theorem 6 is that the standard basis of $`Null(𝐌)`$ reveals which pairs of leaves form cherries. A pair of leaves $`(i,j)`$ is called a cherry if they have a common parent. This is the case if and only if the tree spanned by $`i`$ and $`j`$ does not include an internal edge. Therefore, removing the internal edge $`k`$ that contains the common parent will split the tree into a 2-leaf subtree and an $`(n2)`$-leaf subtree, and in such a situation (as long as $`n4`$) we expect to find exactly two entries in $`w_{I_k}`$ whose values $`(n3)/2(n2)`$ correspond to the two cherry leaves. The examples in Section 3.2 and Section 3.3 nicely illustrate this fact.
Call two trees isomorphic if there is a bijection between edges that takes one tree to the other and preserves the topological structure of the tree. Call two matrices permutation-equivalent if one can be obtained from the other by a permutation of the rows and a permutation of the columns. Call two subspaces of $`^n`$ permutation-equivalent if one set can be obtained from the other by some permutation of the coordinates.
Since the split counts of a tree only depend on the topology of the tree, Theorem 4 shows that isomorphic trees will produce the same Shapley transformation matrix $`𝐌`$ up to a permutation of the rows (given by permuting the order of leaves that define the rows) and a permutation of the columns (given by a permuting the order of the edges that define the columns). The null space of $`𝐌`$ is not affected by permuting the rows of $`𝐌`$, but permuting the columns of $`𝐌`$ has the effect of permuting the coordinates of the null space of $`𝐌`$. Therefore we summarize:
###### Theorem 8.
Isomorphic trees induce permutation-equivalent Shapley transformation matrices with permutation-equivalent null-spaces. Hence, if for two trees $`𝒯_1,𝒯_2`$, their Shapley transformation matrices $`𝐌_1,𝐌_2`$ or their null spaces are not permutation-equivalent, then $`𝒯_1,𝒯_2`$ must not be isomorphic.
## 6. Characterization of the Shapley Value of Tree Games
The Shapley axioms presented in Section 2.4 uniquely characterize the Shapley value on the class of all $`n`$-person games. However, the class of $`n`$-person games that are derived from a tree is smaller. Thus while the Shapley axioms still hold for this smaller class, they may no longer uniquely determine the Shapley value as a function on this class. In this section we will therefore strengthen the axioms so that they once again uniquely characterize the Shapley value on the class of $`n`$-person games derived from a tree.
By $`𝒱^{N,E}`$ we denote the class of games arising from some tree with set of leaves $`N`$ and edge set $`E`$. For games in $`𝒱^{N,E}`$ we will allow positive as well as non-positive edge weights. Thus, $`𝒱^{N,E}`$ is a linear space and we ask for its dimension.
For a fixed pair $`(N,E)`$ define games $`v_k`$ ($`kE)`$ in the following way: $`v_k`$ corresponds to the tree in which edge $`k`$ is weighted 1 and all other edges are weighted zero. We call such a game a basis game. It is readily checked that the game $`v`$ associated with the tree that exhibits edge weights $`\alpha _1,\mathrm{},\alpha _n,\alpha _{I_1},\mathrm{},\alpha _{I_{n3}}`$ is the linear combination $`v=_{kE}\alpha _kv_k`$. Moreover, the family $`(v_k)_{kE}`$ is linearly independent. Therefore these games form a basis of $`𝒱^{N,E}`$ and $`dim𝒱^{N,E}=2n3`$.
Note that in this context, the Shapley transformation $`𝐌`$, as a $`n\times (2n3)`$ matrix, can be viewed as a linear transformation from $`𝒱^{N,E}`$ to $`^n`$.
To characterize this transformation, we ask what properties (axioms) we might expect a ”diversity measure” $`\psi :𝒱^{N,E}^n`$ to satisfy. Thus, given a tree game $`v`$, $`\psi (v)`$ is a vector in $`^n`$ which specifies for each of $`n`$ leaves (players) a number that measures, in some sense, their contribution to the diversity of a group. For instance, for the basis game $`v_k`$, let us consider what a “reasonable” distribution $`\psi (v_k)^n`$ might be. We may interpret zero edge weights on either side of the edge $`k`$ in the basis game $`v_k`$ as having two groups of species, each one being homogeneous. So a natural property would be that the degree of diversity that we assign to one group only depends on the size of this group (and hence the size of the other group) relative to the whole population. It seems plausible that a given group on one side of an edge diversifies the population more if there are more species on the other side of that edge. Thus we may assume that $`\psi _i(v_k)`$ is described by a function that is increasing in the fraction $`f(i,k)/n`$. We formulate these considerations as an additional axiom.
Axiom (group proportionality on basis games): For fixed $`N`$ and $`E`$, a mapping $`\psi :𝒱^{N,E}`$ is said to satisfy group proportionality on basis games, if there is some constant $`d`$ such that $`\psi `$ satisfies $`_{j𝒞(i,k)}\psi _j(v_k)=d\frac{f(i,k)}{n}`$ for all $`iN,kE`$.
Thus, with $`\psi `$ satisfying this axiom, a group’s assigned diversity linearly changes with the other group’s fraction of the whole population. Using the new axiom, we get a characterization result for the Shapley value of games in $`𝒱^{N,E}`$, which may be regarded as a counterpart to Shapley’s original theorem characterizing the Shapley value on all games.
###### Theorem 9.
For each pair $`(N,E)`$ (consisting of leaf set $`N`$ and edge set $`E`$) there is one and only one mapping $`\psi :𝒱^{N,E}^n`$ that satisfies Pareto efficiency, symmetry, additivity and group proportionality. This mapping coincides with the Shapley value $`\phi `$ restricted to $`𝒱^{N,E}`$, i.e., $`\psi (v_𝒯)=\phi (N,v_𝒯)`$ based on the phylogenetic diversity function $`v_𝒯`$.
###### Proof.
It is immediately verified that the Shapley value satisfies all the axioms (for group proportionality, use Theorem 4).
Now, let $`(N,E)`$ be fixed and $`\psi `$ satisfy the axioms. First, we take a basis game $`v_k`$ and determine $`\psi `$. By symmetry, we may conclude $`\psi _i(v_k)=\psi _j(v_k)`$ as long as $`i,j`$ are on the same side of edge $`k`$. Hence
(4)
$$\underset{j𝒞(i,k)}{}\psi _j(v_k)=c(i,k)\psi _i(v_k).$$
Pareto efficiency and group proportionality imply
$$v_k(N)=1=\underset{jN}{}\psi _j(v_k)=\underset{j𝒞(i,k)}{}\psi _j(v_k)+\underset{j(i,k)}{}\psi _j(v_k)=d(\frac{f(i,k)}{n}+\frac{c(i,k)}{n})=d.$$
Hence $`d=1`$ and by group proportionality and (4), we obtain $`\psi _i(v_k)=\frac{f(i,k)}{nc(i,k)}`$ for any $`iN`$ and $`kE`$. Analogously, we get $`\psi _i(\lambda v_k)=\lambda \psi _i(v_k)`$ for $`\lambda `$. Using additivity and Theorem 4, $`\psi `$ coincides with the Shapley value on $`𝒱^{N,E}`$. ∎
We close this section with two remarks. First, note that any game arising from a tree with nonnegative edge weights is representable as a linear combination of basis games using nonnegative coefficients. Hence, we may derive a version of Theorem 9 for classes of games that actually arise from phylogenetic trees.
Second, Theorem 9 provides further justification for the use of the Shapley value to analyze phylogenetic trees. If one wants to distribute the total diversity of a population on its species and the distribution rule should satisfy the above (reasonable) axioms, then the Shapley value is the only possible choice. As symmetry, Pareto efficiency and additivity are rather “obligatory” requirements for a plausible rule, it is the proportionality axiom that provides further insight in the rationale behind the Shapley value. Of course, modification of the group proportionality axiom eventually leads to a different distribution rule based on a different rationale.
## 7. The Core of Tree Games
In prior sections, we have explored the Shapley value as a solution concept for tree games. However, another solution concept for $`n`$-player cooperative games that is frequently studied is the core of a game, which is the set of all imputations $`\stackrel{}{x}^n`$ such that for all coalitions $`SN`$, $`_{iS}x_iv(S)`$ and $`_{iN}x_i=v(N)`$. In this section we examine the core of phylogenetic tree games.
We start with three- and four-leaf tree games for intuition.
###### Example 10.
The characteristic function of the three-leaf tree game is given in Section 3.1, and yields the following system of inequalities for the core:
$`x_A+x_B+x_C`$ $`=\alpha +\beta +\gamma `$
$`x_A+x_B`$ $`\alpha +\beta `$
$`x_A+x_C`$ $`\alpha +\gamma `$
$`x_B+x_C`$ $`\beta +\gamma `$
Hence the core consists of the single element $`\stackrel{}{\mathrm{}}`$, the vector of leaf weights $`\left(\begin{array}{c}\alpha \\ \beta \\ \gamma \end{array}\right)`$.
Thus the three-leaf tree has only one element in its core, namely the vector of leaf weights. Now we consider the four-leaf tree game, which, unlike the three-leaf tree, has an internal edge.
###### Example 11.
The characteristic function of the four-leaf tree game in Figure 4 yields the following system of inequalities for the core:
$`x_A+x_B+x_C+x_D`$ $`=\alpha +\beta +\mu +\gamma +\delta `$
(5) $`x_A+x_C`$ $`\alpha +\mu +\gamma `$
(6) $`x_B+x_D`$ $`\beta +\mu +\delta `$
$`\mathrm{}`$
From (5) and (6) we see that
$$\alpha +\mu +\gamma x_A+x_C\alpha +\gamma .$$
So either $`\mu =0`$ in which case we have a degenerate tree (internal edge weight zero) and the core is $`\stackrel{}{\mathrm{}}`$, or the core has to be empty since the inequality cannot be satisfied.
These two examples illustrate the following theorem:
###### Theorem 12.
Let $`𝒯`$ be an $`n`$-leaf game tree $`𝒯`$ where $`n3`$. If the tree is degenerate (all internal edge weights are zero), then the core consists of the leaf weight vector $`\stackrel{}{\mathrm{}}`$. Otherwise the core is empty.
###### Proof.
Let $`𝒯`$ be an $`n`$-leaf tree with edge weights $`\alpha _i`$ for $`i\{1,\mathrm{},2n3\}`$. Every tree has at least two cherries, where a cherry is a set of two leaves with a common parent. Label the two leaves on one cherry 1 and 2 and label the two leaves on the other cherry 3 and 4 each with corresponding leaf weights $`\alpha _1`$, $`\alpha _2`$, $`\alpha _3`$ and $`\alpha _4`$. We know from the properties of the core that for the set of leaves $`N`$,
(7) $`{\displaystyle \underset{jN}{}}x_j`$ $`={\displaystyle \underset{i\{1,\mathrm{},2n3\}}{}}\alpha _i`$
(8) $`x_1+x_3`$ $`{\displaystyle \underset{kP}{}}\alpha _k`$
(9) $`{\displaystyle \underset{jN\{1,3\}}{}}x_j`$ $`={\displaystyle \underset{kQ}{}}\alpha _k`$
where $`P`$ is the set of edges in the subtree spanned by 1 and 3 and $`Q`$ is the subtree spanned by all other leaves. Note that $`Q`$ must contain all edges in the tree except for the leaf edges associated with 1 and 3. Thus from (7) and (9) we get
(10)
$$x_1+x_3\alpha _1+\alpha _3.$$
Then from (8) and (10) we must have
$$\underset{kP}{}\alpha _kx_1+x_3\alpha _1+\alpha _3.$$
However this cannot be satisfied (the core is empty) unless all of the internal edge weights in $`P`$ are zero. But that every internal edge is in a subtree spanned by pairs of cherries, hence all internal edges weights are zero. In the latter case, the tree is degenerate and the core is the single element $`\stackrel{}{\mathrm{}}`$. ∎
Notice that for $`n=3`$, $`𝒯`$ is always degenerate, and thus the core will never be empty.
Because the core of tree games is empty in most interesting cases, the Shapley value is a far more valuable solution concept to consider.
## 8. Conclusion
In this paper we have presented a biological interpretation of the Shapley value on games derived from phylogenetic trees. We have determined the linear transformation $`𝐌`$ that produces the Shapley value from the edge weights of the tree. We also determined its null space. It is worth noting again the dependence of these results on the split counts of the tree. Finally, we characterized the Shapley value on the space of tree games by four axioms, in much the same way as Shapley did for the space of all games.
We close the paper with some speculation. One of our primary motivations for studying properties of the Shapley value of phylogenetic tree games was for the possibility of using game-theoretic concepts to reconstruct trees from data. Our results on the properties of the Shapley transformation suggest several directions for further research. For instance:
* If there were a way to estimate the Shapley value from data (such as by quantifying the notion of diversity of populations), this would be enough to determine edge weights of a degenerate tree. Do the leaf weights of this tree have any significance?
* Is there a way to determine or estimate split counts from data, and can this assist in determining the correct tree topology?
* Does the converse of Theorem 8 hold, i.e., if two trees have permutation-equivalent Shapley transformation matrices or permutation-equivalent null spaces, are they isomorphic?
* For a given $`n`$-leaf tree topology, the Shapley transformation takes a vector of leaf weights to a vector of Shapley values. However, one may speak of the space of all weighted $`n`$-leaf trees (of various tree topologies), as in , and we can therefore view the Shapley transformation as a map (the Shapley map) from the space of trees to a vector of Shapley values. However, the space of trees is naturally embedded in $`^{\left(\genfrac{}{}{0pt}{}{n}{2}\right)}`$, the space of pairwise distances. Is there a “natural” extension of the Shapley map to this space? How does the kernel of the Shapley map extend the null spaces of Theorem 6? Can this map be used to reconstruct trees?
* If we use the Shapley value to rank the species in the Noah’s ark problem for preservation, to what extent can we guarantee that the diversity of the top $`k`$ species (i.e., the weight of the subtree spanning them) approximates the total diversity of all $`n`$ species? Determine a bound that depends on $`k`$ and $`n`$.
### Acknowledgements
The authors thank Susan Holmes and Bernd Sturmfels for helpful feedback regarding these ideas, and Kyle Kinneberg, Aaron Mazel-Gee, and an anonymous referee for valuable comments on an earlier draft.
|
warning/0506/hep-th0506042.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
The theory of the supermembrane is a key element in the intricate network of related systems which are expected to define the consistent non perturbative aspects of quantum superstrings. It was originally constructed as the $`D=11`$ extended object which propagates consistently on a supergravity background but its role in $`M`$-theory turn out to have more implications. It was shown in that when the supermembrane is immersed in a $`D=11`$ Minkowski space its spectrum is continuous from $`[0,\mathrm{}]`$ but it is still unknown if there is a massless sector corresponding to the $`D=11`$ supergravity. It was also argued in Ref. that the supermembrane on a compact target space should also have a continuous spectrum. In contrast the spectrum of the $`D=11`$ supermembrane wrapping a two cycle when the configuration space satisfies the topological condition which implies a non trivial central charge in the supersymmetric algebra, was shown to be a discrete set with finite multiplicity .
In this paper we focus in the low energy spectrum of the supermembrane wrapped on a two cycle by considering the covariant quantization of the associated superparticle. As we show below, the resulting action describes a massive superparticle with an additional spinorial term arising from the non trivial wrapping of the supermembrane on the two-cycle. In the massless case, as is well known, the covariant quantization of the Casalbuoni-Brink-Schwarz superparticle in $`D=10`$ present similar obstacles related to the mixing of first and second class constraints associated with the $`\kappa `$-symmetry, as the Green-Schwarz superstring. The covariant gauge fixing of the symmetry was performed in by introducing an infinite tower of auxiliary fields . More recently, a new formulation was advanced in terms of pure spinors which requires a finite number of fields. The dynamics of a superparticle with a central charge was first considered in and later in . However the central charges arose in these cases from considerations different to those presented here. The covariant quantization of this superparticle differs from the usual massive superparticle since the central charge implies the existence of a $`\kappa `$-symmetry additionally to the second class constraints already present for the massive superparticle. Several formulation to quantize the superparticle with central charges were proposed . For example in a $`BRST`$ charge was constructed also from a infinitely reducible set of first class constraints. Fortunately as we discuss in section 4 the structure of the system allows a direct approach in which the physical degrees of freedom are identified in a covariant way.
## 2 The Supermembrane in $`M^9\times \mathrm{\Sigma }`$
The supermembrane action immersed in $`D=11`$ target space was obtained originally in Ref. and is given by,
$`I={\displaystyle \frac{1}{8\pi ^2}}{\displaystyle d^3\xi \left[\sqrt{g}g^{ij}\pi _i^\mu \pi _{j\mu }\sqrt{g}+iϵ^{ijk}\pi _i^\mu \pi _j^\nu \overline{\psi }\mathrm{\Gamma }_{\mu \nu }_k\psi \right]}`$
$`+\left[ϵ^{ijk}\pi _i^\mu \overline{\psi }\mathrm{\Gamma }_{\mu \nu }_j\psi \overline{\psi }\mathrm{\Gamma }_\nu _k\psi {\displaystyle \frac{i}{3}}ϵ^{ijk}\overline{\psi }\mathrm{\Gamma }_{\mu \nu }_i\psi \overline{\psi }\mathrm{\Gamma }_\nu _j\psi \psi \overline{\psi }\mathrm{\Gamma }_\nu _k\psi \right],`$ (1)
where $`\psi `$ is a Majorana spinor and
$$\pi _i^\mu =_iX^\mu i\overline{\psi }\mathrm{\Gamma }^\mu _i\psi $$
(2)
with $`X^\mu `$ the space time coordinate of the membrane. The supermembrane tension has been fixed to 1.
We are interested in the case when the $`D=11`$ target space has a compactified sector admitting a minimal immersion from the base manifold $`\mathrm{\Sigma }\times R`$ into it. $`\mathrm{\Sigma }`$ is a Riemann surface of genus $`g>0`$ and $`R`$ corresponds to the range of the time variable $`\tau `$.
To be specific let us consider the case in which $`\mathrm{\Sigma }`$ is a torus and the compactified target space is $`M^9\times S^1\times S^1`$ , although more general compactified target spaces for which there exists a minimal immersion from $`\mathrm{\Sigma }`$ into the target space may also be treated along the lines we follow here .
The minimal immersion is constructed from the harmonic one forms over $`\sigma `$ denoted by $`d\widehat{x}^r`$, $`r=1,2`$. We consider a pair of harmonic one-forms over $`\mathrm{\Sigma }`$ satisfying
$$_{C_s}d\widehat{x}^r=2\pi m_s^r,r,s=1,2$$
(3)
where $`m_s^r`$ are integers and $`C_j`$ is a basis of the homology on $`\mathrm{\Sigma }`$, together with
$$_\mathrm{\Sigma }𝑑\widehat{x}^rd\widehat{x}^s=nA(\mathrm{\Sigma })0,$$
(4)
where $`n=detm_s^r`$ and $`A(\mathrm{\Sigma })`$ is the area of $`\mathrm{\Sigma }`$. The first condition implies that each $`\widehat{x}^r`$ may define a map over a circle $`S^1`$. It also implies the equality in (4). The condition $`A(\mathrm{\Sigma })0`$ is the non trivial part of this relation.
The map from $`\mathrm{\Sigma }`$ onto $`S^1\times S^1`$ is given, with $`P_0`$ a fixed point in $`\mathrm{\Sigma }`$, by
$$P\mathrm{\Sigma }_{P_0}^P𝑑\widehat{x}^r\mathrm{mod}(2\pi n^r),r=1,2.$$
(5)
Condition (4) implies that the algebra of the supermembrane has a non-trivial central charge $`Z^{rs}=ϵ^{rs}nA(\mathrm{\Sigma })`$ with $`n`$ the number of times the supermembrane wraps $`S^1\times S^1`$. It may be shown that the above map is a local minimum of the hamiltonian of the supermembrane and defines a minimal immersion from $`\mathrm{\Sigma }`$ to $`S^1\times S^1`$ . It is a solution of the supermembrane field equations which preserves one half of the original supersymmetry. The most general configuration space and hamiltonian for the supermembrane with the above base manifold and target space with a non-trivial central charge was obtained in and shown to have a discrete spectrum. We are interested here in the quantization of the corresponding superparticle with non-trivial central charges.
## 3 Compactification of the Supermembrane
We consider now the covariant quantization of the center of mass of the supermembrane which corresponds to a superparticle. To do so we restrict the configuration space by the following conditions,
$`X^m`$ $`=`$ $`X^m(\tau ),m=0,\mathrm{},8`$ (6)
$`\psi `$ $`=`$ $`\psi (\tau ),`$ (7)
$`X^{r+8}`$ $`=`$ $`\widehat{x}^r(\sigma ),r=1,2`$ (8)
where $`\sigma ^a,a=1,2`$ denote local coordinates over $`\mathrm{\Sigma }`$. On this class of configurations the supermembrane action reduces to
$$Sk𝑑\tau \left(e^1\omega ^m\omega _mei\psi ^\beta (\frac{1}{2}C\mathrm{\Gamma }_{rs}ϵ^{rs})_{\beta \gamma }\dot{\psi }^\gamma \right),$$
$$\omega ^m=\dot{X}^mi\overline{\psi }\mathrm{\Gamma }^m\dot{\psi }.$$
(9)
where $`C`$ is the charge conjugation matrix in $`D=11`$ and $`\mathrm{\Gamma }_{rs}`$ is the antisymmetric product $`\frac{1}{2}(\mathrm{\Gamma }_r\mathrm{\Gamma }_s\mathrm{\Gamma }_s\mathrm{\Gamma }_r)`$.(The conventions for the $`\mathrm{\Gamma }`$ matrices are given in the appendix. $`C\mathrm{\Gamma }_{rs}`$ is symmetric on the spinorial indices). The constant $`k`$ is given by $`k=\frac{nA(\mathrm{\Sigma })}{8\pi ^2}`$.
In the following section we perform the covariant quantization of this system following a direct approach by introducing a new action for the superparticle with central charges with first class constraints only, that allows to obtain the space of physical states in a covariant way.
We use now a particular representation of the Dirac matrices. We consider $`\gamma ^m,m=0,\mathrm{},8`$ the set of Dirac matrices in $`D=9`$ satisfying $`\{\gamma ^\mu ,\gamma ^\nu \}=2\eta ^{\mu \nu }`$. We take as in ,
$$\mathrm{\Gamma }^m=\left[\begin{array}{cc}0& i\gamma ^m\\ i\gamma ^m& 0\end{array}\right],\mathrm{\Gamma }^9=\left[\begin{array}{cc}0& i\\ i& 0\end{array}\right].$$
(10)
Then we have $`\{\gamma ^\mu ,\gamma ^\nu \}=2\eta ^{\mu \nu },\mu ,\nu =0,\mathrm{},9,11`$, with
$$\mathrm{\Gamma }^{11}=\mathrm{\Gamma }^0\mathrm{\Gamma }^1\mathrm{}\mathrm{\Gamma }^9=\left[\begin{array}{cc}𝕀& 0\\ 0& 𝕀\end{array}\right].$$
(11)
The charge conjugation matrix in $`D=11`$ in terms of the corresponding one in $`D=9`$ is (see the appendix)
$$C=\left[\begin{array}{cc}0& i\stackrel{~}{C}\\ i\stackrel{~}{C}& 0\end{array}\right],$$
(12)
where $`C^T=C`$ and $`\stackrel{~}{C}^T=\stackrel{~}{C}`$. Now we decompose the $`D=11`$ spinors in terms of the $`D=9`$ ones
$$\psi =\left(\begin{array}{c}\theta _1\\ \theta _2\end{array}\right).$$
(13)
The Majorana condition $`\overline{\psi }=\psi ^TC`$ in $`D=11`$ implies
$$\overline{\theta }_A=\theta _A^T\stackrel{~}{C},$$
(14)
where $`\overline{\theta }_A=\theta _A^{}\gamma ^0`$. The action for the superparticle with central charges reduces then to
$$S=k𝑑\tau \left(e^1\omega ^m\omega _mei\overline{\theta }_A\dot{\theta }_A\right)$$
(15)
where now
$$\omega ^m=\dot{X}^mi\overline{\omega }_A\gamma ^m\dot{\omega }_A.$$
(16)
## 4 The Hamiltonian for the Superparticle with central charges
Let us introduce here the mass parameter $`m=2k`$ and the projectors,
$$P\pm =\frac{1}{2m}(m\pm \gamma ^mp_m),$$
(17)
which satisfy
$`P_+P_+`$ $`=`$ $`{\displaystyle \frac{1}{4m^2}}(p^2+m^2)+P_+,P_{}P_{}={\displaystyle \frac{1}{4m^2}}(p^2+m^2)+P_{},`$
$`P_+P_{}`$ $`=`$ $`P_{}P_+={\displaystyle \frac{1}{4m^2}}(p^2+m^2),P_++P_{}=𝕀.`$ (18)
The conjugate momenta to $`X^m`$ are
$$p_m=2ke^1\omega _m,$$
(19)
and the conjugate momenta to $`\theta _A`$, with $`m=2k`$ are
$$\overline{\pi }_A=2im\overline{\theta }_AP_{}$$
(20)
or equivalently
$$\pi _A=2imP_{}\theta _A.$$
(21)
Introducing the Lagrange multiplier $`\lambda \frac{e}{4k}`$, the hamiltonian may then be expressed as
$$=\lambda (p^2+m^2)$$
(22)
subject to
$$\pi _A+2imP_{}\theta _A=0,$$
(23)
which are a mixture of first and second class constraints.
The set of constraints (23) together with the mass shell condition $`p^2+m^2=0`$ are equivalent to
$`p^2+m^2=0,`$ (24)
$`P_+\pi _A=0,`$ (25)
$`P_{}\pi _A+2imP_{}\theta _A=0.`$ (26)
Now (24) and (25) are first class constraints while (26) is a mixture of first and second class constraints.
We may then consider the set,
$`p^2+m^2=0`$ (27)
$`\phi _1=P_+(\pi _1+i\pi _2)=0`$ (28)
$`\phi _2=P_+(\pi _1i\pi _2)=0`$ (29)
$`\mathrm{\Omega }_1=P_{}(\pi _1+i\pi _2)+2imP_{}(\theta _1+i\theta _2)=0`$ (30)
$`\mathrm{\Omega }_2=P_{}(\pi _1i\pi _2)+2imP_{}(\theta _1i\theta _2)=0`$ (31)
with the non trivial bracket,
$`\{\mathrm{\Omega }_1,\mathrm{\Omega }_2\}=4imP_{}.`$ (32)
We also note that the symplectic terms in the canonical action $`\overline{\pi }_A\dot{\theta }_A`$ may be decomposed as
$`\overline{\pi }_A\dot{\theta }_A={\displaystyle \frac{1}{2}}\overline{(\pi _1+i\pi _2)}(\dot{\theta }_1i\dot{\theta }_2)+{\displaystyle \frac{1}{2}}\overline{(\pi _1i\pi _2)}(\dot{\theta }_1+i\dot{\theta }_2),`$ (33)
and we identify the conjugate pairs $`(\pi _1+i\pi _2)`$,$`(\dot{\theta }_1i\dot{\theta }_2)`$ and $`(\pi _1i\pi _2)`$,$`(\dot{\theta }_1+i\dot{\theta }_2)`$. We notice here that the pair $`\mathrm{\Omega }_1`$ and $`\mathrm{\Omega }_2`$ may be regarded as a first class constraint ($`\mathrm{\Omega }_1`$) and an associate gauge fixing condition ($`\mathrm{\Omega }_2`$). The contribution of this pair to the functional measure when taken as a pair of second class constraints or in the way proposed is exactly the same , . With this observation we can finally define our system as a gauge system restricted by the set of first class constraints (27), (28) and (30). If one then would like to impose (31) as a partial gauge fixing condition the original set of constraints is recovered, but there is freedom to impose a different set of admissible gauge fixing conditions since the functional integral is invariant to this choice (In this case there are no gauge anomalies).
To continue we consider the partial gauge fixing conditions,
$`P_+(\theta _1i\theta _2)=0,`$ (34)
$`P_{}(\theta _1i\theta _2)=0,`$ (35)
corresponding to the symmetry generated by $`\phi _1`$ and $`\mathrm{\Omega }_1`$ respectively. This gauge fixing condition contributes to the functional integral with a constant factor independent of the fields. The canonical variables $`(\pi _1+i\pi _2)`$ and $`(\theta _1i\theta _2)`$ may then be integrated out from the functional integral.
We are thus left with the canonical action,
$``$ $`=`$ $`p\dot{X}+\overline{\pi }_A\dot{\theta }_A`$ (36)
$`=`$ $`p\dot{X}+{\displaystyle \frac{1}{2}}\overline{(\pi _1i\pi _2)}(\dot{\theta }_1+i\dot{\theta }_2),`$ (37)
constrained by (27) and (29). Let us introduce the variables $`\stackrel{~}{\theta }=\theta _1+i\theta _2`$ and $`\stackrel{~}{\pi }=\frac{1}{2}(\pi _1i\pi _2)`$. Notice that $`\stackrel{~}{\theta }`$ is a complex spinor which does not satisfy the pseudo Majorana condition $`\overline{\theta }=\theta ^T\stackrel{~}{C}`$
We may finally consider the partial gauge fixing condition
$$P_+(\theta _1+i\theta _2)=0,$$
(38)
and perform a canonical reduction to
$`=p\dot{X}+\overline{(P_{}\stackrel{~}{\pi })}(P_{}\stackrel{~}{\theta }),`$ (39)
subject to the mass shell condition (24). The space of physical states is obtained by considering superfields depending in $`P_{}\stackrel{~}{\theta }`$ and not on its complex conjugate. Since we have that $`P_{}+P_+=1`$ without using $`p^2+m^2=0`$ the degrees of freedom in $`P_{}\stackrel{~}{\theta }`$ are exactly half of the ones in $`\stackrel{~}{\theta }`$. They are therefore $`2^8`$ bosonic and fermionic on-shell degrees of freedom. It corresponds as we discuss below to a $`D=9`$ $`KKB`$ multiplet.
## 5 Conclusion
We have covariantly quantized the $`D=9`$ superparticle associated to the $`D=11`$ supermembrane wrapped on a torus with a non trivial central charge.
The Hilbert space of states we obtained is described in terms of an on-shell superfield $`\mathrm{\Psi }(x^\mu ,\stackrel{~}{\theta }_{})`$. The spinorial variable $`\stackrel{~}{\theta }_{}`$ has 8 independent variables. The superfield has $`2^8`$ degrees of freedom that fit neatly in a $`D=9`$ massive supermultiplet with central charge. The general form of this central charge in $`D=9`$ arising from the supermembrane algebra in $`D=11`$ is
$`Z^{ij}=Z\delta ^{ij}(P_9\sigma ^3P_{10}\sigma ^1),`$ (40)
with a BPS mass given by
$`M=\sqrt{P_9^2+P_{10}^2}+|Z|.`$ (41)
In this paper we have taken $`P_9=P_{10}=0`$. The multiplet that we have obtained from the quantization of the $`D=9`$ superparticle corresponds to a ultrashort KKB supermultiplet .
Due to the nature of our central charge we are studying the winding modes of the supermembrane on a torus and neglecting the Kaluza-Klein modes. As is well known this states should correspond to the Kaluza-Klein modes of the IIB superstring wrapped on $`S^1`$. Our results confirm this correspondence.
## 6 Acknowledgments
This work was supported by Did-Usb grants Gid-30 and Gid-11 and by Fonacit grant G-2001000712.
## Appendix A Appendix:Dirac matrices and Central charges
We collect here some useful results about the supersymmetry algebra in $`D=9`$. We take the ”mostly plus” signature with the Dirac matrices satisfying $`\{\mathrm{\Gamma }^\mu ,\mathrm{\Gamma }^\nu \}=2\eta ^{\mu ,\nu }`$. We construct these matrices recursively. Let $`\gamma ^\mu `$ be a set of Dirac matrices in $`D1`$ dimensions ($`D`$ even), then in $`D`$ dimensions we have,
$$\mathrm{\Gamma }^\mu =\left[\begin{array}{cc}0& S^\mu \\ \overline{S}^\mu & 0\end{array}\right],\begin{array}{cc}S^0=\overline{S^0}=𝕀& S^i=\gamma ^i\gamma ^0\\ \overline{S^i}=S^i=𝕀& S^{D1}=\gamma ^0\end{array}.$$
(42)
In an even dimensional space there exist matrices $`B,\stackrel{~}{B},C`$ and $`\stackrel{~}{C}`$ such that
$$\begin{array}{ccc}B\mathrm{\Gamma }^\mu B^1=(\mathrm{\Gamma }^\mu )^{}& \stackrel{~}{B}\mathrm{\Gamma }^\mu \stackrel{~}{B}^1=(\mathrm{\Gamma }^\mu )^{}& B=\stackrel{~}{B}W\\ C\mathrm{\Gamma }^\mu C^1=(\mathrm{\Gamma }^\mu )^T& \stackrel{~}{C}\mathrm{\Gamma }^\mu \stackrel{~}{C}^1=(\mathrm{\Gamma }^\mu )^T& C=\stackrel{~}{B}\mathrm{\Gamma }^0\end{array}$$
(43)
with $`W=i^{[D/2+1]}\mathrm{\Gamma }^0\mathrm{}.\mathrm{\Gamma }^D`$. In odd dimensions there exist either $`(B,C)`$ or $`(\stackrel{~}{B},\stackrel{~}{C})`$. There exist also a matrix $`\stackrel{~}{B}_9`$ in $`D=9`$ and a matrix $`B_11`$ in $`D=11`$ defined by,
$$\stackrel{~}{B}_9=\gamma ^0\gamma ^1\gamma ^3\gamma ^5\gamma ^7,B_{11}=\mathrm{\Gamma }^2\mathrm{\Gamma }^4\mathrm{\Gamma }^6\mathrm{\Gamma }^8,$$
(44)
such that
$$B_{11}=\left[\begin{array}{cc}\stackrel{~}{B}_9& 0\\ 0& \stackrel{~}{B}_9\end{array}\right],C_{11}=\left[\begin{array}{cc}0& \stackrel{~}{C}_9\gamma ^0\\ \stackrel{~}{C}_9\gamma ^0& 0\end{array}\right].$$
(45)
To make contact with our notation of section 3, call that set of matrices $`\mathrm{\Gamma }_H^\mu `$ they are related to $`\mathrm{\Gamma }^\mu `$ by a unitary transformation
$$U\mathrm{\Gamma }_H^\mu U^{}=\mathrm{\Gamma }^\mu U=\left[\begin{array}{cc}i& 0\\ 0& \gamma ^0\end{array}\right].$$
(46)
From here follows directly equation (12).
The most general supersymmetry algebra in $`D=9`$ and $`N=2`$ with Lorentz invariant central charges is
$$\{Q_{ai},Q_{bj}\}=2\delta ijp_\mu (\gamma ^\mu \stackrel{~}{C}^1)_{ab}+Z_{ij}\stackrel{~}{C}_{ab},$$
(47)
with $`Z_{ij}`$ a real symmetric matrix. Note that we can write
$`\gamma ^0\stackrel{~}{C}=\left[\begin{array}{cc}0& J\\ J& 0\end{array}\right],\stackrel{~}{C}=\left[\begin{array}{cc}J& 0\\ 0& J\end{array}\right],J^T=J,J^2=I.`$ (48)
Since these two matrices commute they can be simultaneously diagonalized. Then the algebra in the rest frame takes the form
$`\{Q_{ai},Q_{bj}\}=2m\delta _{ij}\left[\begin{array}{cc}J& 0\\ 0& J\end{array}\right]+Z_{ij}\left[\begin{array}{cc}J& 0\\ 0& J\end{array}\right].`$ (49)
The algebra of the 32 supercharges splits into 4, $`8\times 8`$ blocks. In our case $`Z_{ij}=2m\delta _{ij}`$ and we find
$`\{Q_{ai},Q_{bj}\}=4m\delta _{ij}\left[\begin{array}{cc}J& 0\\ 0& 0\end{array}\right].`$ (50)
The entire representation may be obtained now as usual. We notice that half of the supersymmetries are not present and the other half build a representation of $`2^8`$ states.
|
warning/0506/math0506340.html
|
ar5iv
|
text
|
# Special Lagrangians of cohomogeneity one in the resolved conifold
## 1 Introduction
The notion of a special Lagrangian submanifold was introduced in 1984 by Harvey and Lawson in their seminal paper \[HL\] on calibrated geometries. Special Lagrangian geometry in Calabi-Yau manifolds has become an important subject due to a phenomenon in physics known as mirror symmetry. In 1996, Strominger, Yau and Zaslow \[SYZ\] conjectured that a compact Calabi-Yau $`3`$-fold and its mirror should be foliated by special Lagrangian $`3`$-tori with possibly singular fibres and the fibrations are dual to each other. This conjecture proposes a way to construct the mirror of a compact Calabi-Yau manifold, by an appropriate compactification of the dual of the special Lagrangian fibration. Since compact Calabi-Yau metrics cannot be written explicitly and also because singularities play a fundamental role, physicists have always been interested in finding concrete Calabi-Yau metrics on non-compact manifolds. One of the earliest examples of a pair of mirror Calabi-Yau metrics was found by Candelas and de la Ossa \[CO\] in 1990. The two manifolds arise from perturbing a singular cone on $`S^2\times S^3`$ and are respectively known as the deformed and the resolved conifold. In fact, there is a $`1`$-parameter family of Calabi-Yau metrics that passes through the singular metric and transforms the deformed conifold into the resolved conifold. The deformed conifold is a (trivial) $`^3`$-bundle over $`S^3`$ and the resolved conifold is a $`^2`$-bundle over $`S^2`$. In passing through the singularity the special Lagrangian $`S^3`$ in the deformed conifold is pinched to a point and reappears in the resolved conifold as a holomorphic $`P^1`$. This is known as the conifold transition (see \[CO\] and \[STY\]).
In our recent paper \[IM\] we obtained a description of cohomogeneity one special Lagrangians in the deformed conifold. In particular, we found that the deformed conifold is foliated by special Lagrangians which are generically $`T^2\times `$. In the present paper we investigate the analogous problem for the resolved conifold and exhibit a foliation of the resolved conifold by $`T^2`$-invariant special Lagrangians, where the generic leaf is topologically $`T^2\times `$. We also obtain an $`SO(3)`$-invariant family of special Lagrangians. It turns out that in both cases, the special Lagrangian families found in the resolved conifold have the same asymptotics as the special Lagrangians that we described in the deformed conifold in \[IM\], i.e. both families asymptotically approach the same special Lagrangians cones in the conifold.
## 2 Special Lagrangian Geometry
In what follows, we will review some definitions and set up notation.
###### Definition 2.1
A Calabi-Yau $`n`$-fold $`(M,J,\omega ,\mathrm{\Omega })`$ is a Kähler $`n`$-dimensional manifold $`(M,J,\omega )`$ with Ricci-flat Kähler metric $`g`$ and a nonzero holomorphic section $`\mathrm{\Omega }`$ which trivializes the canonical bundle $`K_M`$.
Since the metric $`g`$ is Ricci-flat, $`\mathrm{\Omega }`$ is a parallel tensor with respect to the Levi-Civita connection $`^g`$ \[J1\]. By rescaling $`\mathrm{\Omega }`$, we can take it to be the holomorphic $`(n,0)`$-form that satisfies:
$$\frac{\omega ^n}{n!}=(1)^{\frac{n(n1)}{2}}(\frac{i}{2})^n\mathrm{\Omega }\overline{\mathrm{\Omega }},$$
(2.1)
where $`\omega `$ is the Kähler form of $`g`$. The form $`\mathrm{\Omega }`$ is called the holomorphic volume form of the Calabi-Yau manifold $`M`$. Such a holomorphic volume form $`\mathrm{\Omega }`$ exists if and only if the canonical bundle $`K_M`$ is trivial, which implies that the first Chern class $`c_1(M)=0`$ in $`H^2(M,)`$. Also, the above definition is equivalent to the manifold $`(M,g)`$ having holonomy $`SU(n)`$.
###### Definition 2.2
Let $`M^n`$ be an $`n`$-dimensional Calabi-Yau manifold and $`LM`$ a real $`n`$-dimensional submanifold of $`M`$. Then $`L`$ is a special Lagrangian submanifold of $`M`$ if $`\omega _L0`$ (i.e. $`L`$ is Lagrangian) and $`\mathrm{Im}\mathrm{\Omega }_L0`$. Equivalently, $`\mathrm{Re}(\mathrm{\Omega })_L`$ is the volume form on $`L`$ with respect to the induced metric.
One important property of the special Lagrangian submanifolds is that they are absolutely area minimizing in their homology class, so they are in particular minimal submanifolds (see \[HL\]).
An example of a Calabi-Yau manifold is $`^n`$, with coordinates $`(z_1,\mathrm{},z_n)`$, endowed with the flat metric $`g`$, the Kähler form $`\omega _0`$ and the holomorphic volume form $`\mathrm{\Omega }_0=dz_1\mathrm{}dz_n`$. $`^n`$ is a trivial example of a special Lagrangian submanifold in $`^n`$. The following standard examples of cohomogeneity one special Lagrangians in $`^n`$ are due to Harvey-Lawson.
Example 1:
$$L_c=\{\lambda u|uS^{n1}^n,\lambda ,\mathrm{Im}(\lambda ^n)=c\},$$
(2.2)
are special Lagrangians submanifolds of $`^n`$, invariant under the action of $`SO(n)SU(n)`$, where $`c`$ is a constant.
Remark: The variety $`L_0`$ is an union of $`n`$ special Lagrangian $`n`$-planes and when $`c0`$, each component of $`L_c`$ is diffeomorphic to $`\times S^{n1}`$ and it is asymptotic to $`L_0`$.
Example 2: The special Lagrangian submanifolds in $`^n`$, invariant under the action of the maximal torus $`T^{n1}=\{\text{ diag}(e^{i\theta _1},\mathrm{},e^{i\theta _n})|\theta _1+\mathrm{}\theta _n=0\}`$ of $`SU(n)`$ are given by the equations:
$`|z_1|^2|z_j|^2=c_j,j=2,3,\mathrm{},n\text{ and}`$
$`\mathrm{Re}(z_1z_2\mathrm{}z_n)=c_1,\text{ if }n\text{ even}`$
$`\mathrm{Im}(z_1z_2\mathrm{}z_n)=c_1,\text{ if }n\text{ odd}`$
where $`c_1,c_2,\mathrm{}c_n`$ are any real constants.
Remark: These special Lagrangians are topologically $`T^{n1}\times `$ and they are asymptotic to cones on two flat tori in $`S^{2n1}`$ obtained by setting all constants equal to $`0`$.
## 3 The resolved conifold as a Calabi-Yau manifold
In \[IM\], we studied the cohomogeneity one special Lagrangian submanifolds in the deformed conifold $`T^{}S^3`$. In this paper we will answer the same question in the resolved conifold. Both manifolds are Calabi-Yau and asymptotic to the cone on $`S^2\times S^3`$ with a Ricci flat metric, called the conifold. Following the original paper of Candelas-de la Ossa \[CO\], we will first give a description of the resolved conifold and its Calabi-Yau structure.
Let $`Q_0`$ be the singular quadric in $`^4`$ defined by the equation:
$$\underset{i=0}{\overset{3}{}}z_i^2=0$$
(3.3)
This quadric, called the conifold, is a cone on $`S^2\times S^3`$. The deformed conifold is obtained by perturbing the conifold equation to $`_{i=0}^3z_i^2=ϵ^2`$, with $`ϵ`$ a positive constant, and it admits a Calabi-Yau structure.
The other way to deal with the singularity of the conifold is to resolve it. Making the linear change of variables:
$$\left(\begin{array}{c}X\\ Y\\ U\\ V\end{array}\right)=\frac{1}{\sqrt{2}}\left(\begin{array}{cccc}1& i& 0& 0\\ 1& i& 0& 0\\ 0& 0& i& 1\\ 0& 0& i& 1\end{array}\right)\left(\begin{array}{c}z_0\\ z_1\\ z_2\\ z_3\end{array}\right)$$
(3.4)
transforms the quadratic form $`_{i=0}^3z_i^2`$ to $`\mathrm{\hspace{0.17em}2}(XYUV)`$ and the conifold equation (3.3) becomes:
$$XYUV=0$$
(3.5)
We note that the matrix
$$P=\frac{1}{\sqrt{2}}\left(\begin{array}{cccc}1& i& 0& 0\\ 1& i& 0& 0\\ 0& 0& i& 1\\ 0& 0& i& 1\end{array}\right)$$
(3.6)
is not in $`SO(4,)`$ (because it transforms the standard quadratic form), but it is an element of $`U(4)`$, since $`P^1=P^{}=\overline{P}^t`$. This shows that the Euclidean length is still preserved:
$$r^2=\underset{i=0}{\overset{3}{}}|z_i|^2=|X|^2+|Y|^2+|U|^2+|V|^2$$
(3.7)
The $`SO(4,)`$ action on the $`z`$ variables is now conjugated to an action on the new variables via $`g\stackrel{~}{g}=PgP^1`$, i.e.,
$$\stackrel{~}{g}\left(\begin{array}{c}X\\ Y\\ U\\ V\end{array}\right)=PgP^{}\left(\begin{array}{c}X\\ Y\\ U\\ V\end{array}\right),gSO(4,)$$
(3.8)
Let
$$W=\left(\begin{array}{cc}X& U\\ V& Y\end{array}\right)=\frac{1}{\sqrt{2}}\left(\begin{array}{cc}z_0iz_1& z_3iz_2\\ z_3iz_2& z_0+iz_1\end{array}\right)$$
(3.9)
To resolve the conifold, one has to replace the equation $`(\text{3.5})`$ by the pair of equations:
$$\left(\begin{array}{cc}X& U\\ V& Y\end{array}\right)\left(\begin{array}{c}\lambda _1\\ \lambda _2\end{array}\right)=\left(\begin{array}{c}0\\ 0\end{array}\right)$$
(3.10)
where $`[\lambda _1:\lambda _2]`$ are homogeneous coordinates on $`P^1=S^2`$. In other words, the resolved conifold $`M`$ is the complex 3-fold defined by:
$$\{(X,Y,U,V,[\lambda _1:\lambda _2])^4\times P^1|XYUV=0,X\lambda _1+U\lambda _2=0,V\lambda _1+Y\lambda _2=0\}$$
(3.11)
Each point $`(X,Y,U,V)Q_0`$, except the origin, determines a unique point
$$[\lambda _1:\lambda _2]=[U:X]=[Y:V]P^1$$
and the singularity at the origin is replaced (blown-up) by a copy of $`P^1`$. Outside the origin, the resolved conifold is topologically the same as the conifold.
Remark: There is a dual way to resolve the singularity where we replace the equation (3.10) by its transpose
$$\left(\begin{array}{cc}\lambda _1& \lambda _2\end{array}\right)\left(\begin{array}{cc}X& U\\ V& Y\end{array}\right)=\left(\begin{array}{cc}0& 0\end{array}\right)$$
These are known as small resolutions of the conifold. To obtain the “full” resolution of the singularity, one replaces the origin with the set of all complex lines through the origin that lie on the cone and thus the singularity would be replaced by an $`S^2\times S^2`$ rather than a single $`S^2`$.
We will also use inhomogeneous coordinates $`\lambda _+=\frac{\lambda _2}{\lambda _1}`$ in the coordinate patch $`H_+`$ where $`\lambda _10`$ and $`\lambda _{}=\frac{\lambda _1}{\lambda _2}`$ in the coordinate patch $`H_{}`$, where $`\lambda _20`$. In $`H_+`$ we can take $`(U,Y,\lambda _+)`$ and in $`H_{}`$ we can take $`(X,V,\lambda _{})`$ as coordinates for the conifold. On the intersection $`H_+H_{}`$ we have:
$$(X,V,\lambda _{})=(\lambda _+U,\lambda _+Y,\lambda _+^1)$$
In \[CO\], Candelas and de la Ossa showed that the resolved conifold is in fact the $`𝒪(1)𝒪(1)`$ bundle over $`P^1S^2`$ with fibre $`^2`$. The radius in the fibre is given by
$$r^2=\text{tr}(W^{}W)=(1+|\lambda _+|^2)(|U|^2+|Y|^2)=(1+|\lambda _{}|^2)(|X|^2+|V|^2)$$
Asymptotically, as $`r\mathrm{}`$, the resolved conifold approaches the singular conifold $`Q_0`$.
The Ricci-flat metric on $`M`$ found in \[CO\] is:
$$g_{rc}=F_a^{}(r^2)\text{tr}(dW^{}dW)+F_a^{\prime \prime }(r^2)|\text{tr}(W^{}dW)|^2+4a^2g_{S^2}$$
(3.12)
where $`F_a`$ is a function of $`r^2`$ satisfying an appropriate differential equation and $`g_{S^2}`$ is the Fubini-Study metric on $`S^2`$ with area $`\pi `$. In the patch $`H_+`$, $`g_{S^2}`$ is given by:
$$g_{S^2}=\frac{|d\lambda _+|^2}{(1+|\lambda _+|^2)^2}$$
The resolution parameter $`a`$ measures the size of the bolt $`P^1`$ ($`a=0`$ for the conifold).
The differential equation satisfied by the function $`F_a`$ is given by imposing the Ricci-flat condition. It is convenient to write the equation in terms of $`\gamma =r^2F^{}`$, where the derivative is with respect to $`r^2`$. One gets:
$$\gamma ^{}\gamma (\gamma +4a^2)=\frac{2}{3}r^2$$
where $`\gamma ^{}=\frac{d\gamma }{dr^2}`$. Integrating yields:
$$\gamma ^3+6a^2\gamma ^2r^4=0$$
which gives the solution as in \[CO\]:
$$\gamma =2a^2+4a^4N^{\frac{1}{3}}+N^{\frac{1}{3}},N(r)=\frac{1}{2}(r^416a^6+\sqrt{r^832a^6r^4})$$
The function $`F_a^{}(r^2)`$ is therefore given by:
$$F_a^{}(r^2)=r^2(2a^2+4a^4N^{\frac{1}{3}}+N^{\frac{1}{3}},)\text{where}N(r)=\frac{1}{2}(r^416a^6+\sqrt{r^832a^6r^4})$$
(3.13)
For the conifold $`Q_0`$ with $`a=0`$, we have $`F_0^{}=r^{\frac{2}{3}}`$.
The Kähler form $`\omega _{rc}(v,w)=g_{rc}(Jv,w)`$ on the resolved conifold can be expressed as a sum of two terms $`\omega _{rc}=d\alpha _{rc}+4a^2\omega _{S^2}`$, where the one-form $`\alpha _{rc}`$ (compare \[IM\]) is given by:
$$\alpha _{rc}=F_a^{}(r^2)\mathrm{Im}\text{tr}(W^{}dW)=F_a^{}(r^2)\mathrm{Im}\left(\overline{X}dX+\overline{Y}dY+\overline{U}dU+\overline{V}dV\right)$$
(3.14)
and $`\omega _{S^2}`$ is the standard Kähler form of area $`\pi `$ on $`S^2`$. It can be expressed as $`\omega _{S^2}=d\alpha _\pm `$ in the two patches $`H_\pm `$ on $`S^2`$ where:
$$\alpha _\pm =\frac{1}{2}\text{Im}\frac{\lambda _\pm d\overline{\lambda _\pm }}{1+|\lambda _\pm |^2}\text{on }H_\pm $$
One can see that $`\alpha _{rc}`$ is invariant under the action of $`SO(4)=SO(4,)`$. To show that $`\alpha _\pm `$ is also $`SO(4)`$-invariant, we now compute the action of $`SO(4)`$ on $`P^1`$. The following three matrices acting on the $`z`$-coordinates generate the $`𝔰o(4)`$ as a Lie algebra:
$$B_1=\left(\begin{array}{cccc}0& 1& 0& 0\\ 1& 0& 0& 0\\ 0& 0& 0& 0\\ 0& 0& 0& 0\end{array}\right),B_2=\left(\begin{array}{cccc}0& 0& 0& 0\\ 0& 0& 0& 0\\ 0& 0& 0& 1\\ 0& 0& 1& 0\end{array}\right),A_3=\left(\begin{array}{cccc}0& 0& 0& 0\\ 0& 0& 1& 0\\ 0& 1& 0& 0\\ 0& 0& 0& 0\end{array}\right)$$
(3.15)
Conjugating by the change of coordinate matrix $`P`$ (3.4), we obtain the action on the $`X,Y,U,V`$-coordinates:
$$\stackrel{~}{B_1}=\left(\begin{array}{cccc}i& 0& 0& 0\\ 0& i& 0& 0\\ 0& 0& 0& 0\\ 0& 0& 0& 0\end{array}\right),\stackrel{~}{B_2}=\left(\begin{array}{cccc}0& 0& 0& 0\\ 0& 0& 0& 0\\ 0& 0& i& 0\\ 0& 0& 0& i\end{array}\right),\stackrel{~}{A_3}=\frac{1}{2}\left(\begin{array}{cccc}0& 0& 1& 1\\ 0& 0& 1& 1\\ 1& 1& 0& 0\\ 1& 1& 0& 0\end{array}\right)$$
(3.16)
In $`H_+`$, $`\lambda _+=\frac{X}{U}=\frac{V}{Y}`$ and so $`d\lambda _+=\frac{dX}{U}+\frac{XdU}{U^2}`$. The action of $`\stackrel{~}{B_1}`$ on $`\lambda _+`$ is given by evaluating $`d\lambda _+`$ on the vector field corresponding to $`\stackrel{~}{B_1}`$. We obtain: $`\stackrel{~}{B_1}.\lambda _+=i\lambda _+`$. Similarly, on $`H_{}`$ we compute that $`\stackrel{~}{B_1}.\lambda _{}=i\lambda _{}`$. The action of $`\stackrel{~}{B_2}`$ is the same as for $`\stackrel{~}{B_1}`$ and the action of $`\stackrel{~}{A_3}`$ is given by:
$$\stackrel{~}{A_3}.\lambda _+=\frac{1}{2}(1+\lambda _+^2),\stackrel{~}{A_3}.\lambda _{}=\frac{1}{2}(1+\lambda _{}^2)$$
A computation now shows that the form $`\alpha _\pm `$ is invariant under the action of these three generators and hence under the $`SO(4)`$-action. Therefore, the Kähler form and the metric are $`SO(4)`$-invariant.
The holomorphic volume $`3`$-form on $`M`$ has the form:
$$\mathrm{\Omega }_{rc}=dUdYd\lambda _+=dVdXd\lambda _{}$$
(3.17)
in local coordinates and it is also invariant under the action of $`SO(4)`$.
We now want to describe cohomogeneity one special Lagrangian in the resolved conifold $`M`$. As in our previous paper \[IM\], the only subgroups of $`SO(4)`$ that we need to consider are the maximal torus $`T^2`$ and the subgroup $`SO(3)`$. We will use the same moment map techniques as in the previous paper. We recall the following proposition which holds in general for any Kähler manifold $`M`$.
###### Proposition 3.1
Let $`GSO(4)`$ be a connected Lie subgroup with Lie algebra $`𝔤`$ and moment map $`\mu :M𝔤^{}`$ where $`(M,\omega _{rc})`$ is the resolved conifold and $`𝒪`$ an orbit of $`G`$ in $`M`$. Then the orbit is isotropic, i.e. $`\omega _{rc}|_𝒪0`$ if and only if $`𝒪\mu ^1(c)`$ for some $`cZ(𝔤^{})`$, where $`Z(𝔤^{})`$ is the centre of $`𝔤`$.
## 4 $`T^2`$-invariant special Lagrangians
Under the coordinate transform (3.8), the action of the maximal torus $`T^2`$ of $`SO(4)`$ on the $`z`$-coordinates, described by the matrices:
$$\left\{\left(\begin{array}{cccc}\mathrm{cos}\theta _1& \mathrm{sin}\theta _1& 0& 0\\ \mathrm{sin}\theta _1& \mathrm{cos}\theta _1& 0& 0\\ 0& 0& \mathrm{cos}\theta _2& \mathrm{sin}\theta _2\\ 0& 0& \mathrm{sin}\theta _2& \mathrm{cos}\theta _2\end{array}\right)\right\}$$
correspond to the following matrices with respect to the $`(X,Y,U,V)`$-coordinates:
$$\left\{\left(\begin{array}{cccc}e^{i\theta _1}& 0& 0& 0\\ 0& e^{i\theta _1}& 0& 0\\ 0& 0& e^{i\theta _2}& 0\\ 0& 0& 0& e^{i\theta _2}\end{array}\right)\right\}$$
The $`T^2`$-action on the patch $`H_+`$ with coordinates $`(U,Y,\lambda )`$ is given by:
$$g.(U,Y,\lambda _+)=(e^{i\theta _2}U,e^{i\theta _1}Y,e^{i(\theta _1+\theta _2)}\lambda _+)$$
and on the patch $`H_{}`$ with coordinates $`(X,V,\mu )`$ is given by:
$$g.(X,V,\lambda _{})=(e^{i\theta _1}X,e^{i\theta _2}V,e^{i(\theta _1+\theta _2)}\lambda _{})$$
The following result gives all the special Lagrangian $`3`$-folds of the resolved conifold invariant under the action of $`T^2`$.
###### Theorem 4.1
The special Lagrangian submanifolds in the resolved conifold $`M`$ with coordinates $`(X,Y,U,V,\lambda _1,\lambda _2)`$ described by equations (3.11) with the Calabi-Yau structure, which are invariant under the action of the maximal torus $`T^2`$ of $`SO(4)`$ are given by the equations:
$`{\displaystyle \frac{1}{2}}F_a^{}(r^2)(|X|^2|Y|^2)+4a^2{\displaystyle \frac{|\lambda _2|^2}{|\lambda _1|^2+|\lambda _2|^2}}=c_1`$
$`{\displaystyle \frac{1}{2}}F_a^{}(r^2)(|V|^2|U|^2)+4a^2{\displaystyle \frac{|\lambda _2|^2}{|\lambda _1|^2+|\lambda _2|^2}}=c_2`$ (4.18)
$`\mathrm{Im}(XY)=c_3`$
where $`F_a^{}`$ is given by (3.13) and $`c_1,c_2`$ and $`c_3`$ are real constants.
*Proof*: Using the two infinitesimal generators:
$$\stackrel{~}{B}_1=\left(\begin{array}{cccc}i& 0& 0& 0\\ 0& i& 0& 0\\ 0& 0& 0& 0\\ 0& 0& 0& 0\end{array}\right),\stackrel{~}{B}_2=\left(\begin{array}{cccc}0& 0& 0& 0\\ 0& 0& 0& 0\\ 0& 0& i& 0\\ 0& 0& 0& i\end{array}\right)$$
of the $`T^2`$-action on the resolved conifold, the moment map is given by:
$`\mu :M(𝔱^2)^{}^2,`$
$`\mu =({\displaystyle \frac{1}{2}}F_a^{}(r^2)(|X|^2|Y|^2)+4a^2\mu _{S^2},{\displaystyle \frac{1}{2}}F_a^{}(r^2)(|V|^2|U|^2)+4a^2\mu _{S^2})`$
where $`\mu _{S^2}(\lambda _1,\lambda _2)=\frac{|\lambda _2|^2}{|\lambda _1|^2+|\lambda _2|^2}`$ is the moment map of the standard $`S^1`$-action on $`P^1`$.
Since $`(𝔱^2)^{}=^2`$ is abelian, it follows from Proposition 3.1 that any $`T^2`$-invariant special Lagrangian $`3`$-fold $`L`$ in $`Q^3`$ lies in a level set $`\mu ^1(c)`$, where $`c=(c_1,c_2)^2`$. The first two equations enforce this condition and ensures that the submanifold is Lagrangian. In order to impose the special Lagrangian condition at a given point $`p=(U,Y,\lambda _+)`$, we compute $`\mathrm{\Omega }_{rc}`$ on the coordinate patch $`H_+`$ with coordinates on the three tangent vectors $`Y_1=\stackrel{~}{B}_1p=(0,iY,i\lambda _+),Y_2=\stackrel{~}{B}_2p=(iU,0,i\lambda _+)`$ and $`Y_3=\dot{p}=(\dot{U},\dot{Y},\dot{\lambda }_+)`$:
$`\mathrm{\Omega }_{rc}(Y_1,Y_2,Y_3)`$ $`=(dUdYd\lambda _+)(Y_1,Y_2,Y_3)`$
$`=\left|\begin{array}{ccc}0& iU& \dot{U}\\ iY& 0& \dot{Y}\\ i\lambda _+& i\lambda _+& \dot{\lambda }_+\end{array}\right|`$ (4.19)
$`=\dot{U}Y\lambda _++U\dot{Y}\lambda _++UY\dot{\lambda }_+=(UY\lambda _+)^{}`$
Integrating the condition $`\mathrm{Im}\mathrm{\Omega }_{rc}=0`$, we obtain $`\mathrm{Im}(UY\lambda _+)=c`$ for $`c`$. Using $`\lambda _+=\frac{X}{U}`$ we finally obtain the third equation of the theorem. Note that in the other coordinate patch $`H_{}`$ we will obtain $`\mathrm{Im}(VX\lambda _{})=`$ c which is the same equation since $`\lambda _{}=\frac{Y}{V}`$.
$`\mathrm{}`$
Remark 1: Equations (4.18) are obviously $`T^2`$-invariant and linearly independent at a generic point The above family of $`T^2`$-invariant special Lagrangian $`3`$-folds foliates the resolved conifold $`M=𝒪(1)𝒪(1)`$. The generic orbit is $`T^2\times `$ where $`T^2`$ is an orbit of the maximal torus in $`SO(4)`$. For the special values $`c_1=c_2,c_3=0`$, these special Lagrangians intersect the $`P^1`$ in circles.
Remark 2: Asymptotically, as $`r^2\mathrm{}`$ or equivalently as $`a0`$, these special Lagrangians approach the special Lagrangian cone on $`T^2`$ in the conifold described by the equations:
$`|X|^2|Y|^2=0`$
$`|V|^2|U|^2=0`$
$`\mathrm{Im}(XY)=0`$
which is the same asymptotic cone that was described in \[IM\]. In $`z`$ coordinates on the conifold, the equations become: $`\mathrm{Im}(z_0\overline{z_1})=\mathrm{Im}(z_2\overline{z_3})=\mathrm{Im}(z_0^2+z_1^2)=0`$
## 5 $`SO(3)`$-invariant special Lagrangians
Let $`G=SO(3)`$ be the subgroup of $`SO(4)`$ embedded as:
$$\left\{\left(\begin{array}{cc}1& 0\\ 0& A\end{array}\right),ASO(3)\right\}$$
The $`SO(3)`$-invariant special Lagrangian $`3`$-folds in the resolved conifold $`M`$ are characterized in the following theorem:
###### Theorem 5.1
The $`SO(3)`$-invariant special Lagrangian submanifolds in the resolved conifold $`M`$, with coordinates $`(X,Y,U,V,\lambda _1,\lambda _2)`$ satisfying equation (3.5) are given by:
$$L_c=\{\stackrel{~}{g}.pM|gSO(3),p=(0,Y,0,0),Y0,\text{R}e(Y^2)=c\}$$
where $`\stackrel{~}{g}`$ is the conjugated action on the $`(X,Y,U,V)`$\- coordinates.
*Proof*:
Let $`A_1=\left(\begin{array}{cccc}0& 0& 0& 0\\ 0& 0& 0& 0\\ 0& 0& 0& 1\\ 0& 0& 1& 0\end{array}\right),A_2=\left(\begin{array}{cccc}0& 0& 0& 0\\ 0& 0& 0& 1\\ 0& 0& 0& 0\\ 0& 1& 0& 0\end{array}\right),A_3=\left(\begin{array}{cccc}0& 0& 0& 0\\ 0& 0& 1& 0\\ 0& 1& 0& 0\\ 0& 0& 0& 0\end{array}\right)`$ be the infinitesimal generators of $`SO(3)`$. Using the change of coordinate matrix $`P`$ (3.4), the action on the $`(X,Y,U,V)`$-variables is given by:
$$\stackrel{~}{A}_1=\left(\begin{array}{cccc}0& 0& 0& 0\\ 0& 0& 0& 0\\ 0& 0& i& 0\\ 0& 0& 0& i\end{array}\right),\stackrel{~}{A}_2=\frac{1}{2}\left(\begin{array}{cccc}0& 0& i& i\\ 0& 0& i& i\\ i& i& 0& 0\\ i& i& 0& 0\end{array}\right),\stackrel{~}{A}_3=\frac{1}{2}\left(\begin{array}{cccc}0& 0& 1& 1\\ 0& 0& 1& 1\\ 1& 1& 0& 0\\ 1& 1& 0& 0\end{array}\right)$$
The moment map for these generators is $`\mu :M𝔰o(3)^{}`$ given by:
$$\mu =(\frac{1}{2}(|U|^2|V|^2),\frac{1}{2}\mathrm{Im}((UV)(\overline{Y}\overline{X})),\frac{1}{2}\mathrm{Im}((YX)(\overline{U}+\overline{V})))+\mu _{S^2}$$
where
$$\mu _{S^2}(\lambda _+)=(\frac{|\lambda _+|^2}{1+|\lambda _+|^2},\frac{1}{2(1+|\lambda _+|^2)}\mathrm{Re}(\lambda _+|\lambda _+|^2\overline{\lambda _+}),\frac{1}{2(1+|\lambda _+|^2)}\mathrm{Im}(\lambda _++|\lambda _+|^2\overline{\lambda _+}))$$
on the patch coordinate patch $`H_+`$. A similar expression can be computed on $`H_{}`$.
Let $`p=(X,Y,U,V)M`$. Since $`Z(𝔰o(3)^{})=0`$, we need to start at a point in $`\mu ^1(0)`$. We make the simplifying assumption that $`U=V=0`$. Using the fact that $`XY=UV`$, we can assume that $`X=0`$ or $`Y=0`$. If we choose $`Y=0`$, then the point $`(X,0,0,0)\mu ^1(0)`$ since $`\mu (X,0,0,0)`$ has the first component equal to 1. Hence we must choose $`p=(0,Y,0,0)`$, where $`Y0`$. Since $`\lambda _+=\frac{V}{Y}=0`$, $`p\mu ^1(0)`$. We now look for curves $`Y(s)`$ in the complex $`Y`$-plane, which after applying the $`SO(3)`$-action give rise to special Lagrangians of the form $`L=\stackrel{~}{g}.Y(s)`$, $`gSO(3)`$.
The tangent plane at $`p`$ to $`L`$ is spanned by the vectors: $`\{v_1=\stackrel{~}{A}_2p,v_2=\stackrel{~}{A}_3p,v_3=\dot{p}\}`$, where the dot denotes differentiation with respect to the parameter $`s`$. The generic orbit of the $`SO(3)`$-action is an $`S^2`$, the three infinitesimals generators are linearly dependent (Note that $`\stackrel{~}{A}_1p=0`$). The tangent space of $`L`$ at the point $`p`$ is therefore spanned by the vectors:
$$\left\{v_1=\frac{iY}{2}\left(\begin{array}{c}0\\ 0\\ 1\\ 1\end{array}\right),v_2=\frac{Y}{2}\left(\begin{array}{c}0\\ 0\\ 1\\ 1\end{array}\right),v_3=\left(\begin{array}{c}0\\ \dot{Y}\\ 0\\ 0\end{array}\right)\right\}$$
$`L`$ is invariant under the $`SO(3)`$-action and $`\omega _{rc}_L=0`$, since $`L`$ lies in the zero level set of the moment map. We now impose the special Lagrangian condition $`\text{ Im }\mathrm{\Omega }_{rc}_L=0`$. Working on the patch $`H_+`$ with coordinates $`(U,Y,\lambda _+)`$, the holomorphic volume form is given by equation (3.17):
$$\mathrm{\Omega }_{rc}=dUdYd\lambda _+$$
Since $`\lambda +=\frac{V}{Y}`$, we can instead use coordinates $`(U,Y,V)`$ on and the holomorphic volume form in these coordinates becomes:
$$\mathrm{\Omega }_{rc}=dUdYd\lambda _+=dUdY\frac{VdYYdV}{Y^2}=\frac{1}{Y}dUdYdV$$
Therefore,
$$\mathrm{\Omega }_{rc}(v_1,v_2,v_3)=\frac{1}{Y}dUdYdV(v_1,v_2,v_3)$$
Computing yields:
$$\frac{1}{Y}\left|\begin{array}{ccc}\frac{iY}{2}& \frac{Y}{2}& 0\\ 0& 0& \dot{Y}\\ \frac{iY}{2}& \frac{Y}{2}& 0\end{array}\right|=\frac{i}{2}Y\dot{Y}$$
Integrating, the condition $`\text{Im }\mathrm{\Omega }_{rc}=0`$ becomes:
$$\text{Re}(Y^2)=c$$
(5.20)
where $`c`$ is a real constant. Let $`Y=u+iv`$. Then equation (5.20), gives a family of hyperbolas in the $`(u,v)`$-plane:
$$u^2v^2=c$$
The special Lagrangian $`L_c`$ is topologically $`S^2\times `$ and has two components, each asymptotic to the special Lagrangian cones on $`S^2`$ given by setting $`c=0`$ (obtained from the lines $`u+v=0`$ and $`uv=0`$) in the conifold. This case is reminiscent of the flat case studied in \[HL\], but of a different dimension.
Remark 1: $`L_c`$ does not intersect the zero-section (the bolt) $`S^2`$ of the resolved conifold, since $`r^2=|Y|^2`$ is preserved under the $`SO(3)`$.
Remark 2: Note that asymptotically, the $`SO(3)`$-invariant special Lagrangian in the resolved conifold approach the same special Lagrangian cones in the conifold as the $`SO(3)`$-invariant special Lagrangian in the deformed conifold that we found in our previous paper \[IM\].
Department of Mathematics & Statistics, McMaster University
e-mail address: ionelm@@math.mcmaster.ca
Department of Mathematics & Statistics, McMaster University
e-mail address: minoo@@mcmaster.ca
|
warning/0506/math0506408.html
|
ar5iv
|
text
|
# The Thom isomorphism in gauge-equivariant 𝐾-theory
## 1. Introduction
In this paper we establish a Thom isomorphism theorem for gauge equivariant $`K`$-theory. Let $`p:𝒢B`$ be a bundle of compact groups. Recall that this means that each fiber $`𝒢_b:=p^1(b)`$ is a compact group and that, locally, $`𝒢`$ is of the form $`U\times G`$, where $`UB`$ open and $`G`$ a fixed compact group. Let $`X`$ and $`B`$ be locally compact spaces and $`\pi _X:XB`$ be a continuous map. In the present paper, as in , this map will be supposed to be a locally trivial bundle. A part of present results can be extended to the case of a general map, this, as well as the proof of a general index theorem, will be the subject of a forthcoming paper.
Assume that $`𝒢`$ acts on $`X`$. This action will be always be fiber-preserving. Then we can associate to the action of $`𝒢`$ on $`X`$ $`𝒢`$-equivariant $`K`$-theory groups $`K_𝒢^i(X)`$ as in . We shall review and slightly generalize this definition in Section 2.
For $`X`$ compact, the group $`K_𝒢^0(X)`$ is defined as the Grothendieck group of $`𝒢`$-equivariant vector bundles on $`X`$. If $`X`$ is not compact, we define the groups $`K_𝒢^0(X)`$ using fiberwise one-point compactifications. We shall call these groups simply gauge-equivariant $`K`$-theory groups of $`X`$ when we do not want to specify $`𝒢`$. The reason for introducing the gauge-equivariant $`K`$-theory groups is that they are the natural range for the index of a gauge-invariant families of elliptic operators. In turn, the motivation for studying gauge-invariant families and their index is due to their connection to spectral theory and boundary value problems on non-compact manifolds. Some possible connections with Ramond-Ramond fields in String Theory were mentioned in . See also .
In this paper, we continue our study of gauge-equivariant $`K`$-theory. We begin by providing two alternative definitions of the relative $`K_𝒢`$–groups, both based on complexes of vector bundles. (In this paper, all vector bundles are complex vector bundles, with the exception of the tangent bundles and where explicitly stated.) These alternative definitions, modeled on the classical case , provide a convenient framework for the study of products, especially in the relative or non-compact cases. The products are especially useful for the proof of the Thom isomorphism in gauge-equivariant theory, which is one of the main results of this paper. Let $`EX`$ be a $`𝒢`$-equivariant complex vector bundle. Then the Thom isomorphism is a natural isomorphism
(1)
$$\tau _E:K_𝒢^i(X)K_𝒢^i(E).$$
(There is also a variant of this result for $`\mathrm{spin}^\mathrm{c}`$-vector bundles, but since we will not need it for the index theorem , we will not discuss this in this paper.) The Thom isomorphism allows us to define Gysin (or push-forward) maps in $`K`$-theory. As it is well known from the classical work of Atiyah and Singer , the Thom isomorphism and the Gysin maps are some of the main ingredients used for the definition and study of the topological index. In fact, we shall proceed along the lines of that paper to define the topological index for gauge-invariant families of elliptic operators. Some other approaches to Thom isomorphism in general settings of Noncommutative geometry were the subject of and many other papers.
Gauge-equivariant $`K`$-theory behaves in many ways like the usual equivariant $`K`$-theory, but exhibits also some new phenomena. For example, the groups $`K_𝒢^0(B)`$ may turn out to be reduced to $`K^0(B)`$ when $`𝒢`$ has “a lot of twisting” \[15, Proposition 3.6\]. This is never the case in equivariant $`K`$-theory when the action of the group is trivial but the group itself is not trivial. In , we addressed this problem in two ways: first, we found conditions on the bundle of groups $`p:𝒢B`$ that guarantee that $`K_𝒢^0(X)`$ is not too small (this condition is called finite holonomy and is recalled below), and, second, we studied a substitute of $`K_𝒢^0(X)`$ which is never too small (this substitute is $`K(C^{}(𝒢))`$, the $`K`$-theory of the $`C^{}`$-algebra of the bundle of compact groups $`𝒢`$).
In this paper, we shall again need the finite holonomy condition, so let us review it now. To define the finite holonomy condition, we introduced the representation covering of $`𝒢`$, denoted $`\widehat{𝒢}B`$. As a space, $`\widehat{𝒢}`$ is the union of all the representation spaces $`\widehat{𝒢}_b`$ of the fibers $`𝒢_b`$ of the bundle of compact groups $`𝒢`$. One measure of the twisting of the bundle $`𝒢`$ is the holonomy associated to the covering $`\widehat{𝒢}B`$. We say that $`𝒢`$ has representation theoretic finite holonomy if $`\widehat{𝒢}`$ is a union of compact-open subsets. (An equivalent conditions can be obtained in terms of the fundamental groups when $`B`$ is path-connected, see Proposition 2.3 below.)
Let $`C^{}(𝒢)`$ be the enveloping $`C^{}`$-algebra of the bundle of compact groups $`𝒢`$. We proved in \[15, Theorem 5.2\] that
(2)
$$K_𝒢^j(B)K_j(C^{}(𝒢)),$$
provided that $`𝒢`$ has representation theoretic finite holonomy. This guarantees that $`K_𝒢^j(B)`$ is not too small. It also points out an alternative, algebraic definition of the groups $`K_𝒢^i(X)`$.
The structure of the paper is as follows. We start from the definition of gauge-equivariant $`K`$-theory and some basic results from , most of them related to the “finite holonomy condition,” a condition on bundles of compact groups that we recall in Section 2. In Section 3 we describe an equivalent definition of gauge-equivariant $`K`$-theory in terms of complexes of vector bundles. This will turn out to be especially useful when studying the topological index. In Section 4 we establish the Thom isomorphism in gauge-equivariant $`K`$-theory, and, in Section 5, we define and study the Gysin maps. The properties of the Gysin maps allow us to define in Section 6 the topological index and establish its main properties.
## 2. Preliminaries
We now recall the definition of gauge-equivariant $`K`$-theory and some basic results from . An important part of our discussion is occupied by the discussion of the finite holonomy condition for a bundle of compact groups $`p:𝒢B`$.
All vector bundles in this paper are assumed to be complex vector bundles, unless otherwise mentioned and excluding the tangent bundles to the various manifolds appearing below.
### 2.1. Bundles of compact groups and finite holonomy conditions
We begin by introducing bundles of compact and locally compact groups. Then we study finite holonomy conditions for bundles of compact groups.
###### Definition 2.1.
Let $`B`$ be a locally compact space and let $`G`$ be a locally compact group. We shall denote by $`\mathrm{Aut}(G)`$ the group of automorphisms of $`G`$. A bundle of locally compact groups $`𝒢`$ with typical fiber $`G`$ over $`B`$ is, by definition, a fiber bundle $`𝒢B`$ with typical fiber $`G`$ and structural group $`\mathrm{Aut}(G)`$.
We fix the above notation. Namely, from now on and throughout this paper, unless explicitly otherwise mentioned, $`B`$ will be a compact space and $`𝒢B`$ will be a bundle of compact groups with typical fiber $`G`$.
We need now to introduce the representation theoretic holonomy of a bundle of Lie group with compact fibers $`p:𝒢B`$. Let $`\mathrm{Aut}(G)`$ be the group of automorphisms of $`G`$. By definition, there exists then a principal $`\mathrm{Aut}(G)`$-bundle $`𝒫B`$ such that
$$𝒢𝒫\times _{\mathrm{Aut}(G)}G:=(𝒫\times G)/\mathrm{Aut}(G).$$
We shall fix $`𝒫`$ in what follows.
Let $`\widehat{𝒢}`$ be the (disjoint) union of the sets $`\widehat{𝒢}_b`$ of equivalence classes of irreducible representations of the groups $`𝒢_b`$. Using the natural action of $`\mathrm{Aut}(G)`$ on $`\widehat{G}`$, we can naturally identify $`\widehat{𝒢}`$ with $`𝒫\times _{\mathrm{Aut}(G)}\widehat{G}`$ as fiber bundles over $`B`$.
Let $`\mathrm{Aut}_0(G)`$ be the connected component of the identity in $`\mathrm{Aut}(G)`$. The group $`\mathrm{Aut}_0(G)`$ will act trivially on the set $`\widehat{G}`$, because the later is discrete. Let
$$H_R:=\mathrm{Aut}(G)/\mathrm{Aut}_0(G),𝒫_0:=𝒫/\mathrm{Aut}_0(G),\text{ and }\widehat{𝒢}𝒫_0\times _{H_R}\widehat{G}.$$
Above, $`\widehat{𝒢}`$ is defined because $`𝒫_0`$ is an $`H_R`$-principal bundle. The space $`\widehat{𝒢}`$ will be called the representation space of $`𝒢`$ and the covering $`\widehat{𝒢}B`$ will be called the representation covering associated to $`𝒢`$.
Assume now that $`B`$ is a path-connected, locally simply-connected space and fix a point $`b_0B`$. We shall denote, as usual, by $`\pi _1(B,b_0)`$ the fundamental group of $`B`$. Then the bundle $`𝒫_0`$ is classified by a morphism
(3)
$$\rho :\pi _1(B,b_0)H_R:=\mathrm{Aut}(G)/\mathrm{Aut}_0(G),$$
which will be called the holonomy of the representation covering of $`𝒢`$.
For our further reasoning, we shall sometimes need the following finite holonomy condition.
###### Definition 2.2.
We say that $`𝒢`$ has representation theoretic finite holonomy if every $`\sigma \widehat{𝒢}`$ is contained in a compact-open subset of $`\widehat{𝒢}`$.
In the cases we are interested in, the above condition can be reformulated as follows
###### Proposition 2.3.
Assume that $`B`$ is path-connected and locally simply-connected. Then $`𝒢`$ has representation theoretic finite holonomy if, and only if $`\pi _1(B,b_0)\sigma \widehat{G}`$ is a finite set for any irreducible representation $`\sigma `$ of $`G`$.
From now on we shall assume that $`𝒢`$ has representation theoretic finite holonomy.
### 2.2. Gauge-equivariant $`K`$-theory
Let us now define the gauge equivariant $`K`$-theory groups of a “$`𝒢`$-fiber bundle” $`\pi _Y:YB`$. All our definitions are well known if $`B`$ is reduced to a point (cf. ). First we need to fix the notation.
If $`f_i:Y_iB`$, $`i=1,2`$, are two maps, we shall denote by
(4)
$$Y_1\times _BY_2:=\{(y_1,y_2)Y_1\times Y_2,f_1(y_1)=f_2(y_2)\}$$
their fibered product. Let $`p:𝒢B`$ be a bundle of locally compact groups and let $`\pi _Y:YB`$ be a continuous map. We shall say that $`𝒢`$ acts on $`Y`$ if each group $`𝒢_b`$ acts continuously on $`Y_b:=\pi ^1(b)`$ and the induced map $`\mu `$
$$𝒢\times _BY:=\{(g,y)𝒢\times Y,p(g)=\pi _Y(y)\}(g,y)\mu (g,y):=gyY$$
is continuous. If $`𝒢`$ acts on $`Y`$, we shall say that $`Y`$ is a $`𝒢`$-space. If, in addition to that, $`YB`$ is also locally trivial, we shall say that $`Y`$ is a $`𝒢`$-fiber bundle, or, simply, a $`𝒢`$-bundle. This definition is a particular case of the definition of the action of a differentiable groupoid on a space.
Let $`\pi _Y:YB`$ be a $`𝒢`$-space, with $`𝒢`$ a bundle of compact groups over $`B`$. Recall that a vector bundle $`\stackrel{~}{\pi }_E:EY`$ is a $`𝒢`$-equivariant vector bundle (or simply a $`𝒢`$–equivariant vector bundle) if
$$\pi _E:=\pi _Y\stackrel{~}{\pi }_E:EB$$
is a $`𝒢`$-space, the projection
$$\stackrel{~}{\pi }_E:E_b:=\pi _E^1(b)Y_b:=\pi _Y^1(b)$$
is $`𝒢_b:=p^1(b)`$ equivariant, and the induced action $`E_yE_{gy}`$ of $`g𝒢`$, between the corresponding fibers of $`EY`$, is linear for any $`yY_b`$, $`g𝒢_b`$, and $`bB`$.
To define gauge-equivariant $`K`$–theory, we first recall some preliminary definitions from . Let $`\stackrel{~}{\pi }_E:EY`$ be a $`𝒢`$-equivariant vector bundle and let $`\stackrel{~}{\pi }_E^{}:E^{}Y^{}`$ be a $`𝒢^{}`$-equivariant vector bundle, for two bundles of compact groups $`𝒢B`$ and $`𝒢^{}B^{}`$. We shall say that $`(\gamma ,\phi ,\eta ,\psi ):(𝒢^{},E^{},Y^{},B^{})(𝒢,E,Y,B)`$ is a $`\gamma `$–equivariant morphism of vector bundles if the following five conditions are satisfied:
1. $`\gamma :𝒢^{}𝒢,`$ $`\phi :E^{}E,`$ $`\eta :Y^{}Y,`$ and $`\psi :BB^{},`$
2. all the resulting diagrams are commutative,
3. $`\phi (ge)=\gamma (g)\phi (e)`$ for all $`eE_b^{}`$ and all $`g𝒢_b^{},`$,
4. $`\gamma `$ is a group morphism in each fiber, and
5. $`f`$ is a vector bundle morphism.
We shall say that $`\varphi :EE^{}`$ is a $`\gamma `$–equivariant morphism of vector bundles if, by definition, it is part of a morphism $`(\gamma ,\phi ,\eta ,\psi ):(𝒢^{},E^{},Y^{},B^{})(𝒢,E,Y,B)`$. Note that $`\eta `$ and $`\psi `$ are determined by $`\gamma `$ and $`\varphi `$.
As usual, if $`\psi :B^{}B`$ is a continuous \[respectively, smooth\] map, we define the inverse image $`(\psi ^{}(𝒢),\psi ^{}(E),\psi ^{}(Y),B^{})`$ of a $`𝒢`$-equivariant vector bundle $`EY`$ by $`\psi ^{}(𝒢)=𝒢\times _BB^{}`$, $`\psi ^{}(E)=E\times _BB^{}`$, and $`\psi ^{}(Y)=Y\times _BB^{}`$. If $`B^{}B`$ and $`\psi `$ is the embedding, this construction gives the restriction of a $`𝒢`$-equivariant vector bundle $`EY`$ to a closed, invariant subset $`B^{}B`$ of the base of $`𝒢`$, yielding a $`𝒢_B^{}`$–equivariant vector bundle. Usually $`𝒢`$ will be fixed, however.
Let $`p:𝒢B`$ be a bundle of compact groups and $`\pi _Y:YB`$ be a $`𝒢`$–space. The set of isomorphism classes of $`𝒢`$–equivariant vector bundles $`\stackrel{~}{\pi }_E:EY`$ will be denoted by $`_𝒢(Y)`$. On this set we introduce a monoid operation, denoted “$`+`$,” using the direct sum of vector bundles. This defines a monoid structure on the set $`_𝒢(Y)`$ as in the case when $`B`$ consists of a point.
###### Definition 2.4.
Let $`𝒢B`$ be a bundle of compact groups acting on the $`𝒢`$-space $`YB`$. Assume $`Y`$ to be compact. The $`𝒢`$-equivariant $`K`$-theory group $`K_𝒢^0(Y)`$ is defined as the group completion of the monoid $`_𝒢(Y)`$.
When working with gauge-equivariant $`K`$–theory, we shall use the following terminology and notation. If $`EY`$ is a $`𝒢`$-equivariant vector bundle on $`Y`$, we shall denote by $`[E]`$ its class in $`K_𝒢^0(Y)`$. Thus $`K_𝒢^0(Y)`$ consists of differences $`[E][E^1]`$. The groups $`K_𝒢^0(Y)`$ will also be called gauge equivariant $`K`$-theory groups, when we do not need to specify $`𝒢`$. If $`B`$ is reduced to a point, then $`𝒢`$ is group, and the groups $`K_𝒢^0(Y)`$ reduce to the usual equivariant $`K`$-groups.
We have the following simple observations on gauge-equivariant $`K`$–theory. First, the familiar functoriality properties of the usual equivariant $`K`$-theory groups extend to the gauge equivariant $`K`$-theory groups. For example, assume that the bundle of compact groups $`𝒢B`$ acts on a fiber bundle $`YB`$ and that, similarly, $`𝒢^{}B^{}`$ acts on a fiber bundle $`Y^{}B^{}`$. Let $`\gamma :𝒢𝒢^{}`$ be a morphism of bundles of compact groups and $`f:YY^{}`$ be a $`\gamma `$-equivariant map. Then we obtain a natural group morphism
(5)
$$(\gamma ,f)^{}:K_𝒢^{}^0(Y^{})K_𝒢^0(Y).$$
If $`\gamma `$ is the identity morphism, we shall denote $`(\gamma ,f)^{}=f^{}`$.
A $`𝒢`$-equivariant vector bundle $`EY`$ on a $`𝒢`$-space $`YB`$, $`Y`$ compact, is called trivial if, by definition, there exists a $`𝒢`$-equivariant vector bundle $`E^{}B`$ such that $`E`$ is isomorphic to the pull-back of $`E^{}`$ to $`Y`$. Thus $`EY\times _BE^{}`$. If $`𝒢B`$ has representation theoretic finite holonomy and $`Y`$ is a compact $`𝒢`$-bundle, then every $`𝒢`$–equivariant vector bundle over $`Y`$ can be embedded into a trivial $`𝒢`$–equivariant vector bundle. This embedding will necessarily be as a direct summand.
If $`𝒢B`$ does not have finite holonomy, it is possible to provide examples if $`𝒢`$–equivariant vector bundles that do not embed into trivial $`𝒢`$–equivariant vector bundles . Also, a related example from shows that the groups $`K_𝒢^0(Y)`$ can be fairly small if the holonomy of $`𝒢`$ is “large.”
A further observation is that it follows from the definition that the tensor product of vector bundles defines a natural ring structure on $`K_𝒢^0(Y)`$. We shall denote the product of two elements $`a`$ and $`b`$ in this ring by $`ab`$ or, simply, $`ab`$, when there is no danger of confusion. In particular the groups $`K_𝒢^i(X)`$ for $`\pi _X:XB`$ are equipped with a natural structure of $`K_𝒢^0(B)`$–module obtained using the pull-back of vector bundles on $`B`$, namely, $`ab:=\pi _X^{}(a)bK_𝒢^0(X)`$ for $`aK_𝒢^0(B)`$ and $`bK_𝒢^0(X)`$.
The definition of the gauge-equivariant groups extends to non-compact $`𝒢`$-spaces $`Y`$ as in the case of equivariant $`K`$–theory. Let $`Y`$ be a $`𝒢`$-bundle. We shall denote then by $`Y^+:=YB`$ the compact space obtained from $`Y`$ by the one-point compactification of each fiber (recall that $`B`$ is compact). The need to consider the space $`Y^+`$ is the main reason for considering also non longitudinally smooth fibers bundles on $`B`$. Then
$$K_𝒢^0(Y):=\mathrm{ker}\left(K_𝒢^0(Y^+)K_𝒢^0(B)\right).$$
Also as in the classical case, we let
$$K_𝒢^n(Y,Y^{}):=K_𝒢^0((YY^{})\times ^n)$$
for a $`𝒢`$-subbundle $`Y^{}Y`$. Then we have the following periodicity result
###### Theorem 2.5.
We have natural isomorphisms
$$K_𝒢^n(Y,Y^{})K_𝒢^{n2}(Y,Y^{}).$$
Gauge-equivariant $`K`$-theory is functorial with respect to open embeddings. Indeed, let $`UX`$ be an open, $`𝒢`$-equivariant subbundle. Then the results of \[15, Section 3\] provide us with a natural map morphism
(6)
$$i_{}:K_𝒢^n(U)K_𝒢^n(X).$$
In fact, $`i_{}`$ is nothing but the composition $`K_𝒢^n(U)K_𝒢^n(X,XU)K_𝒢^n(X).`$
### 2.3. Additional results
We now prove some more results on gauge-equivariant $`K`$-theory.
Let $`𝒢B`$ and $`B`$ be two bundles of compact groups over $`B`$. Recall that an $``$-bundle $`\pi _X:XB`$ is called free if the action of each group $`_b`$ on the fiber $`X_b`$ is free (i.e., $`hx=x`$, $`xX_b`$, implies that $`h`$ is the identity of $`𝒢_b`$.) We shall need the following result, which is an extension of a result in \[10, page 69\]. For simplicity, we shall write $`𝒢\times `$ instead of $`𝒢\times _B`$.
###### Theorem 2.6.
Suppose $`\pi _X:XB`$ is a $`𝒢\times `$-bundle that is free as an $``$-bundle. Let $`\pi :XX/`$ be the (fiberwise) quotient map. For any $`𝒢`$–equivariant vector bundle $`\stackrel{~}{\pi }_E:EX/`$, we define the induced vector bundle
$$\pi ^{}(E):=\left\{(x,\epsilon )X\times E,\pi (x)=\stackrel{~}{\pi }_E(\epsilon )\right\}X,$$
with the action of $`𝒢\times `$ given by $`(g,h)(x,\epsilon ):=((g,h)x,g\epsilon ).`$ Then $`\pi ^{}`$ is gives rise to a natural isomorphism $`K_𝒢^0(X/)K_{𝒢\times }^0(X).`$
###### Proof.
Let $`\pi ^{}:K_𝒢^0(X/)K_{𝒢\times }^0(X)`$ be the induction map, as above. We will construct a map $`r:K_{𝒢\times }^0(X)K_𝒢^0(X/)`$ satisfying $`\pi ^{}r=\mathrm{Id}`$ and $`r\pi ^{}=\mathrm{Id}`$. Let $`\pi _F:FX`$ be a $`𝒢\times `$-vector bundle. Since the action of $``$ on $`X`$ is free, the induced map $`\overline{\pi }_F:F/X/`$ of quotient spaces is a (locally trivial) $`𝒢`$-bundle. Clearly, this construction is invariant under homotopy, and hence we can define $`r[F]:=[F/]`$.
Let us check now that $`r`$ is indeed an inverse of $`\pi ^{}`$. Denote by $`FffF/`$ the quotient map. Let $`FX`$ be a $`𝒢\times `$-vector bundle. To begin with, the total space of $`\pi ^{}r(F)`$ is
$$\left\{(x,f)X\times (F/),\pi ^{}(x)=\overline{\pi }_F(f)\right\},$$
by definition. Then the map $`Ff(\pi _F(f),f)\pi ^{}r(F)`$ is an isomorphism. Hence, $`\pi ^{}r=\mathrm{Id}`$.
Next, consider a $`𝒢`$–vector bundle $`\pi _E:EX/`$. The total space of $`r\pi ^{}(E)`$ is then
$$\{(y,\epsilon )(X/)\times E,y=\pi _E(\epsilon ))\},$$
because $``$ acts only on the first component of $`\pi ^{}(E)`$. Then
$$Er\pi ^{}(E),\epsilon (\pi (\epsilon ),\epsilon )$$
is an isomorphism. Hence, $`r\pi ^{}=\mathrm{Id}`$. ∎
See also \[15, Theorem 3.5\].
###### Corollary 2.7.
Let $`P`$ be a $`𝒢\times `$–bundle that is free as an $``$–bundle. Also, let $`W`$ be an $``$–bundle, then there is a natural isomorphism
$$K_{𝒢\times }(P\times _BW)K_𝒢(P\times _{}W).$$
###### Proof.
Take $`X:=P\times _BW`$ in the previous theorem. Then $`X`$ is a free $``$–bundle, because $`P`$ is, and $`X/=:P\times _{}W`$. ∎
In the following section, we shall also need the following quotient construction associated to a trivialization of a vector bundle over a subset. Namely, if $`YX`$ is a $`𝒢`$–invariant, closed subbundle, then we shall denote by $`X/_BY`$ the fiberwise quotient space over $`B`$, that is the quotient of $`X`$ with respect to the equivalence relation $``$, $`xy`$ if, and only if, $`x,yY_b`$, for some $`bB`$.
If $`E`$ is a $`𝒢`$–equivariant vector bundle over a $`𝒢`$–bundle $`X`$, together with a $`𝒢`$–equivariant trivialization over a $`𝒢`$–subbundle $`YX`$, then we can generalize the quotient (or collapsing) construction of \[2, §1.4\] to obtain a vector bundle over $`X/_BY`$, where by $`X/_BY`$ we denote the fiberwise quotient bundle over $`B`$, as above.
###### Lemma 2.8.
Suppose that $`X`$ is a $`𝒢`$–bundle and that $`YX`$ is a closed, $`𝒢`$–invariant subbundle. Let $`EX`$ be a $`𝒢`$–vector bundle and $`\alpha :E|_YY\times _B𝒱`$ be a $`𝒢`$–equivariant trivialization, where $`𝒱B`$ is a $`𝒢`$–equivariant vector bundle. Then we can naturally associate to $`(E,\alpha )`$ a naturally defined vector bundle $`E/\alpha X/_BY`$ that depends only on homotopy class of $`\alpha `$.
###### Proof.
Let $`p:Y\times _B𝒱𝒱`$ be the natural projection. Introduce the following equivalence relation on $`E`$:
$$ee^{}e,e^{}E|_Y\text{ and }p\alpha (e)=p\alpha (e^{}).$$
Let then $`E/\alpha `$ be equal to $`E/`$. This is locally trivial vector bundle over $`X/_BY`$. Indeed, it is necessary to verify this only in a neighborhood of $`Y/_BYB`$. Let $`U`$ be a $`𝒢`$ invariant open subset of $`X`$ such that $`\alpha `$ can be extended to an isomorphism $`\stackrel{~}{\alpha }:E|_UU\times _B𝒱`$. We obtain an isomorphism
$$\alpha ^{}:(E|_U)/\alpha (U/_BY)\times _B𝒱,\alpha ^{}(e)=\stackrel{~}{\alpha }(e).$$
Moreover, $`(U/_BY)\times _B𝒱`$ is a locally trivial $`𝒢`$–equivariant vector bundle.
Suppose that $`\alpha _0`$ and $`\alpha _1`$ are homotopic trivializations of $`E|_Y`$, that is, trivializations such that there exists a trivialization $`\beta :E\times I|_{Y\times I}Y\times I\times _B𝒱`$, $`\beta |_{E\times \{0\}}=\alpha _0`$ and $`\beta |_{E\times \{1\}}=\alpha _1`$. Let
$$f:X/_BY\times I(X\times I)/_{(B\times I)}(Y\times I).$$
Then the bundle $`f^{}((E\times I)/\beta )`$ over $`X/_BY\times I`$ satisfies $`f^{}((E\times I)/\beta )|_{(X/_BY)\times \{i\}}=E/\alpha _i`$, $`i=0,1`$. Hence, $`E/\alpha _0E/\alpha _1`$. ∎
## 3. $`K`$-theory and complexes
For the purpose of defining the Thom isomorphism, it is convenient to work with an equivalent definition of gauge-equivariant $`K`$-theory in terms of complexes of vector bundles. This will turn out to be especially useful when studying the topological index.
The statements and proofs of this section, except maybe Lemma 3.4, follow the classical ones , so our presentation will be brief.
### 3.1. The $`L_𝒢^n`$–groups
We begin by adapting some well known concepts and constructions to our settings.
Let $`XB`$ be a locally compact, paracompact $`𝒢`$–bundle. A finite complex of $`𝒢`$–equivariant vector bundles over $`X`$ is a complex
$$(E^{},d)=\left(\mathrm{}\stackrel{d_{i1}}{}E^i\stackrel{d_i}{}E^{i+1}\stackrel{d_{i+1}}{}\mathrm{}\right)$$
$`i,`$ of $`𝒢`$–equivariant vector bundles over $`X`$ with only finitely many $`E^i`$’s different from zero. Explicitly, $`E^i`$ are $`𝒢`$–equivariant vector bundles, $`d_i`$’s are $`𝒢`$–equivariant morphisms, $`d_{i+1}d_i=0`$ for every $`i`$, and $`E^i=0`$ for $`|i|`$ large enough. We shall also use the notation $`(E^{},d)=(E^0,\mathrm{},E^n,d_i:E^i|_YE^{i+1}|_Y),`$, if $`E^i=0`$ for $`i<0`$ and for $`i>n`$.
As usual, a morphism of complexes $`f:(E^{},d)(F^{},\delta )`$ is a sequence of morphisms $`f_i:E^iF^i`$ such that $`f_{i+1}d_i=\delta _{i+1}f_i,`$ for all $`i`$. These constructions yield the category of finite complexes of $`𝒢`$–equivariant vector bundles. Isomorphism in this category will be denoted by $`(E^{},d)(F^{},\delta ).`$
In what follows, we shall consider a pair $`(X,Y)`$ of $`𝒢`$–bundles with $`X`$ is a compact $`𝒢`$–bundle, unless explicitly otherwise mentioned.
###### Definition 3.1.
Let $`X`$ be a compact $`𝒢`$–bundle and $`Y`$ be a closed $`𝒢`$–invariant subbundle. Denote by $`C_𝒢^n(X,Y)`$ the set of (isomorphism classes of) sequences
$$(E^{},d)=(E^0,E^1,\mathrm{},E^n,d_k:E^k|_YE^{k+1}|_Y)$$
of $`𝒢`$–equivariant vector bundles over $`X`$ such that $`(E^k|_Y,d)`$ is exact if we let $`E^j=0`$ for $`j<0`$ or $`j>n`$.
We endow $`C_𝒢^n(X,Y)`$ with the semigroup structure given by the direct sums of complexes. An element in $`C_𝒢^n(X,Y)`$ is called elementary if it is isomorphic to a complex of the form
$$\mathrm{}0E\stackrel{\mathrm{Id}}{}E0\mathrm{},$$
Two complexes $`(E^{},d),(F^{},\delta )C_𝒢^n(X,Y)`$ are called equivalent if, and only if, there exist elementary complexes $`Q^1,\mathrm{},Q^k,P^1,\mathrm{},P^mC_𝒢^n(X,Y)`$ such that
$$EQ_1\mathrm{}Q_kFP_1\mathrm{}P_m.$$
We write $`EF`$ in this case. The semigroup of equivalence classes of sequences in $`C_𝒢^n(X,Y)`$ will be denoted by $`L_𝒢^n(X,Y)`$.
We obtain, from definition, natural injective semigroup homomorphisms
$$C_𝒢^n(X,Y)C_𝒢^{n+1}(X,Y)\text{and}C_𝒢(X,Y):=\underset{n}{}C_𝒢^n(X,Y).$$
The equivalence relation $``$ commutes with embeddings, so the above morphisms induce morphisms $`L_𝒢^n(X,Y)L_𝒢^{n+1}(X,Y).`$ Let $`L_𝒢^{\mathrm{}}(X,Y):=\underset{}{lim}L_𝒢^n(X,Y)`$.
###### Lemma 3.2.
Let $`EX`$ and $`FX`$ be $`𝒢`$–vector bundles. Let $`\alpha :E|_YF|_Y`$ and $`\beta :EF`$ be surjective morphisms of $`𝒢`$–equivariant vector bundles. Also, assume that $`\alpha `$ and $`\beta |_Y`$ are homotopic in the set of surjective $`𝒢`$–equivariant vector bundle morphisms. Then there exists a surjective morphism of $`𝒢`$–equivariant vector bundles $`\stackrel{~}{\alpha }:EF`$ such that $`\stackrel{~}{\alpha }|_Y=\alpha .`$ The same result remains true if we replace “surjective” with “injective” or “isomorphism” everywhere.
###### Proof.
Let $`Z:=(Y\times [0,1])(X\times \{0\})`$ and $`\pi :ZX`$ be the projection. Let $`\pi ^{}(E)Z`$ and $`\pi ^{}(F)Z`$ be the pull-backs of $`E`$ and $`F`$. The homotopy in the statement of the Lemma defines a surjective morphism $`a:\pi ^{}(E)\pi ^{}(F)`$ such that $`a|_{Y\times \{1\}}=\alpha `$ and $`a|_{X\times \{0\}}=\beta `$. By \[15, Lemma 3.12\], the morphism $`a`$ can be extended to a surjective morphism over $`(U\times [0,1])(X\times \{0\})`$, where $`U`$ is an open $`𝒢`$–neighborhood of $`Y`$. (In fact, in that Lemma we considered only the case of an isomorphism, but the case of a surjective morphism is proved in the same way.) Let $`\phi :X[0,1]`$ be a continuous function such that $`\phi (Y)=1`$ and $`\phi (XU)=0`$. By averaging, we can assume $`\phi `$ to be $`𝒢`$–equivariant. Then define $`\stackrel{~}{\alpha }(x)=a(x,\phi (x))`$, for all $`xX`$. ∎
###### Remark 3.3.
Suppose that $`X`$ is a compact $`𝒢`$–space and $`Y=\mathrm{}.`$ Then we have a natural isomorphism $`\chi _1:L_𝒢^1(X,\mathrm{})K_𝒢^0(X)`$ taking the class of $`E^1,E^0`$ to the element $`[E^0][E^1].`$
We shall need the following lemma.
###### Lemma 3.4.
Let $`p:𝒢B`$ be a bundle of compact groups and $`\pi _X:XB`$ be a compact $`𝒢`$–bundle. Assume that $`\pi _X`$ has a cross-section, which we shall use to identify $`B`$ with a subset of $`X`$. Then the sequence
$$0L_𝒢^1(X,B)L_𝒢^1(X)L_𝒢^1(B)$$
is exact.
###### Proof.
Suppose that $`E=(E^1,E^0,\phi )`$ defines an element of $`L_𝒢^1(X)`$ such that its image in $`L_𝒢^1(B)`$ is zero. Then the definition of $`E0`$ in $`L_𝒢^1(B)`$ shows that the restrictions of $`E^1`$ and $`E^0`$ to $`B`$ are isomorphic over $`B`$. Hence, the above sequence is exact at $`L_𝒢^1(X)`$.
Suppose now that $`(E^1,E^0,\phi )`$ represents a class in $`L_𝒢^1(X,B)`$ such that its image in $`L_𝒢^1(X)`$ is zero. This means (keeping in mind Remark 3.3) that there exists a $`𝒢`$–equivariant vector bundle $`\stackrel{~}{P}`$ and an isomorphism $`\stackrel{~}{\psi }:E^1\stackrel{~}{P}E^0\stackrel{~}{P}`$. Let us define $`P:=\stackrel{~}{P}\pi _X^{}(E^0|_B)\pi _X^{}(\stackrel{~}{P}|_B)`$, where $`\pi _X:XB`$ is the canonical projection, as in the statement of the Lemma. Also, define $`\psi =\stackrel{~}{\psi }\mathrm{Id}:E^1PE^0P`$, which is also an isomorphism.
We thus obtain that $`T:=\psi (\phi \mathrm{Id})^1`$ is an automorphism of $`(E^0P)|_B`$. which has the form $`\left(\begin{array}{cc}\beta & 0\\ 0& \mathrm{Id}\end{array}\right)`$ with the respect to the decomposition
$$(E^0P)|_B=(E^0\stackrel{~}{P})|_B(E^0\stackrel{~}{P})|_B.$$
The automorphism $`T:=\psi (\phi \mathrm{Id})^1`$ is homotopic to the automorphism $`T_1`$ defined by the matrix
$$\left(\begin{array}{cc}\mathrm{Id}& 0\\ 0& \beta \end{array}\right).$$
Since $`T_1`$ extends to an automorphism of $`E^0P`$ over $`X`$, namely $`\left(\begin{array}{cc}\mathrm{Id}& 0\\ 0& \pi _X^{}(\beta )\end{array}\right)`$, Lemma 3.2 gives that the automorphism $`\psi (\phi \mathrm{Id})^1`$ also can be extended to $`X`$, such that over $`B`$ we have the following commutative diagram:
Hence, $`(E^1,E^0,\phi )(P,P,\mathrm{Id})(E^0P,E^0P,\mathrm{Id})`$ and so is zero in $`L_𝒢^1(X,B)`$. ∎
### 3.2. Euler characteristics
We now generalize the above construction to other groups $`L_𝒢^n`$, thus proving the existence and uniqueness of Euler characteristics.
###### Definition 3.5.
Let $`X`$ be a compact $`𝒢`$–space and $`YX`$ be a $`𝒢`$–invariant subset. An Euler characteristic $`\chi _n`$ is a natural transformation of functors $`\chi _n:L_𝒢^n(X,Y)K_𝒢^0(X,Y)`$, such that for $`Y=\mathrm{}`$ it takes the form
$$\chi _n(E)=\underset{i=0}{\overset{n}{}}(1)^i[E^i],$$
for any sequence $`E=(E^{},d)L_𝒢^n(X,Y)`$.
###### Lemma 3.6.
There exists a unique natural transformation of functors $`(`$i.e., an Euler characteristic$`)`$
$$\chi _1:L_𝒢^1(X,Y)K_𝒢^0(X,Y),$$
which, for $`Y=\mathrm{}`$, has the form indicated in 3.3.
###### Proof.
To prove the uniqueness, suppose that $`\chi _1`$ and $`\chi _1^{}`$ are two Euler characteristics on $`L_𝒢^1`$. Then $`\chi _1^{}\chi _1^1`$ is a natural transformation of $`K_𝒢^0`$ that is equal to the identity on each $`K_𝒢^0(X)`$. Let us consider the long exact sequence a long exact sequence
(7)
$$\begin{array}{c}\mathrm{}K_𝒢^{n1}(Y,Y^{})K_𝒢^{n1}(Y)K_{𝒢_1}^{n1}(Y^{})\hfill \\ \hfill K_𝒢^n(Y,Y^{})K_𝒢^n(Y)K_𝒢^n(Y^{})\mathrm{}\end{array}$$
associated to a pair $`(Y,Y)`$ of $`𝒢`$–bundles (see , Equation (10) for a proof of the exactness of this sequence). The map $`K_𝒢^0(X,B)K_𝒢^0(X)`$ from this exact sequence is induced by $`(X,\mathrm{})(X,B)`$, and hence, in particular, it is natural. Assume that $`\pi _X:XB`$ has a cross-section. Then the exact sequence (7) for $`(Y,Y^{})=(X,B)`$ yields a natural exact sequence $`0K_𝒢^0(X,B)K_𝒢^0(X)`$. This in conjunction with Lemma 3.4 shows that $`\chi _1^{}\chi _1^1`$ is the identity on $`K_𝒢^0(X,B)`$. Recall now that we agreed to denote by $`X/_BY`$ the fiberwise quotient space over $`B`$, that is the quotient of $`X`$ with respect to the equivalence relation $``$, $`xy`$ if, and only if, $`x,yY_b`$, for some $`bB`$. Finally, since the map $`(X,Y)(X/_BY,B)`$ induces an isomorphism of $`K_𝒢^0`$-groups \[15, Theorem 3.19\], $`\chi _1^{}\chi _1^1`$ is the identity on $`K_𝒢^0(X,Y)`$ for all pairs $`(X,Y)`$.
To prove the existence of the Euler characteristic $`\chi _1`$, let $`(E^1,E^0,\alpha ):=(\alpha :E^1E^0)`$ represent an element of $`L_𝒢^1(X,Y)`$. Suppose that $`X_0`$ and $`X_1`$ are two copies of $`X`$ and $`Z:=X_0_YX_1B`$ is the $`𝒢`$–bundle obtained by identifying the two copies of $`YX_i`$, $`i=0,1`$. The identification of $`E^1|_Y`$ and $`E^0|_Y`$ with the help of $`\alpha `$ gives rise to an element $`[F^0][F^1]K_𝒢^0(Z)`$ defined as follows. By adding some bundle to both $`E^i`$’s, we can assume that $`E^1`$ is trivial (that is, it is isomorphic to the pull–back of a vector bundle on $`B`$). Then $`E^1`$ extends to a trivial $`𝒢`$–vector bundle $`\stackrel{~}{E}^1Z`$. We define $`F^0:=E^0_\alpha E^1`$ and $`F^1:=\stackrel{~}{E}^1`$ .
The exact sequence (7) and the natural $`𝒢`$-retractions $`\pi _i:ZX_i`$, give natural direct sum decompositions
(8)
$$K_𝒢^0(Z)=K_𝒢^0(Z,X_i)K_𝒢^0(X_i),i=0,1.$$
The natural map $`(X_0,Y)(Z,X_1)`$ induces an isomorphism
$$k:K_𝒢^0(Z,X_1)K_𝒢^0(X_0,Y).$$
Let us define then $`\chi _1(E^0,E^1,\alpha )`$ to be equal to the image under $`k`$ of the $`K_𝒢^0(Z,X_1)`$-component of $`[E^1,\alpha ,E^0]`$ (with the respect to (8)). It follows from its definition that this map is natural, respects direct sums, and is independent with respect to the addition of elementary elements. Our proof is completed by observing that $`\chi _1(E^1,E^0,\alpha )=[E^0][E^1]`$ when $`Y=\mathrm{}`$. ∎
We shall also need the following continuity property of the functor $`L_𝒢^1`$. Recall that we have agreed to denote by $`X/_BY`$ the fiberwise quotient bundle over $`B`$.
###### Lemma 3.7.
The natural homomorphism
$$\mathrm{\Pi }^{}:L_𝒢^1(X/_BY,Y/_BY)=L_𝒢^1(X/_BY,B)L_𝒢^1(X,Y)$$
is an isomorphism for all pairs $`(X,Y)`$ of compact $`𝒢`$-bundles.
###### Proof.
Lemmata 3.6 and 3.4 give the following commutative diagram
From this we obtain injectivity.
To prove surjectivity, suppose that $`E^1`$ and $`E^0`$ are $`𝒢`$–equivariant vector bundles over $`X`$ and $`\alpha :E^1|_YE^0|_Y`$ is an isomorphism of the restrictions. Let $`PX`$ be a $`𝒢`$-bundle such that there is an isomorphism $`\beta :E^1PF`$, where $`F`$ is a trivial bundle (i.e., isomorphic to a pull back from $`B`$). Then $`(E^1,E^0,\alpha )(F,E^0P,\gamma )`$, where $`\gamma =(\alpha \mathrm{Id})\beta ^1`$. The last object is the image of $`(F,(E^0P)/\gamma ,\gamma /\gamma )`$ (see Lemma 2.8). ∎
We obtain the following corollaries.
###### Corollary 3.8.
The Euler characteristic $`\chi _1:L_𝒢^1(X,Y)K_𝒢^0(X,Y)`$ is an isomorphism and hence it is defines an equivalence of functors.
###### Proof.
This follows from Lemmas 3.7 and 3.6. ∎
###### Lemma 3.9.
The class of $`(E^1,E^0,\alpha )`$ in $`L_𝒢^1(X,Y)`$ depends only on the homotopy class of the isomorphism $`\alpha `$.
###### Proof.
Let $`Z=X\times [0,1]`$, $`W=Y\times [0,1]`$. Denote by $`p:ZX`$ the natural projection and assume that $`\alpha _t`$ is a homotopy, where $`\alpha _0=\alpha `$. Then $`\alpha _t`$ gives rise an isomorphism $`\beta :p^{}(E^1)|_Wp^{}(E^0)|_W`$, and hence to an element $`(p^{}(E^1),p^{}(E^0),\beta )`$ of $`L_𝒢^1(Z,W)`$. If
$$i_t:(X,Y)(X\times \{t\},Y\times \{t\})(Z,W),$$
are the standard inclusions, then $`(E^1,E^0,\alpha _t)=i_t^{}(p^{}(E^1),p^{}(E^0),\beta )`$. Consider the commutative diagram
The vertical morphisms and and the morphisms of the bottom line of the above diagram are isomorphisms. Hence, the arrows of the top line are isomorphisms too. The composition $`i_0^{}(i_1^{})^1`$ is identity for the bottom line, hence it is the identity for the top line too. ∎
The following theorem reduces the study of the functors $`L_𝒢^n`$, $`n>1`$, to the study of $`L_𝒢^1`$.
###### Theorem 3.10.
The natural map $`j_n:L_𝒢^n(X,Y)L_𝒢^{n+1}(X,Y)`$ is an isomorphism.
###### Proof.
Let $`E=(E^0,E^1,\mathrm{},E^{n+1};d_k)`$, $`d_k:E^k|_YE^{k+1}|_Y`$ represent an element of the semigroup $`L_𝒢^{n+1}(X,Y)`$. To prove the surjectivity of $`j_n`$, let us first notice that $`E`$ is equivalent to the complex
$$\begin{array}{c}(E^0,\mathrm{},E^{n2},E^{n1}E^{n+1},E^nE^{n+1},E^{n+1};\hfill \\ \hfill d_0,\mathrm{},d_{n2}0,d_{n1}\mathrm{Id},d_n0).\end{array}$$
The maps $`d_n\mathrm{\hspace{0.17em}0}:(E^nE^{n+1})|_YE^{n+1}|_Y`$ and $`0\mathrm{Id}:(E^nE^{n+1})|_YE^{n+1}|_Y`$ are homotopic within the set of surjective, $`𝒢`$-equivariant vector bundle morphisms $`(E^nE^{n+1})|_YE^{n+1}|_Y`$. Hence, by Lemma 3.2, $`d_n0`$ can be extended to a surjective morphism $`b:E^nE^{n+1}E^{n+1}`$ of $`𝒢`$–equivariant vector bundles (over the whole of $`X`$). So, the bundle $`E^nE^{n+1}`$ is isomorphic to $`\mathrm{ker}(b)E^{n+1}`$. Hence, the $`E`$ is equivalent to
$$(E^0,\mathrm{},E^{n2},E^{n1}E^{n+1},\mathrm{ker}(b),\mathrm{\hspace{0.17em}0};d_0,\mathrm{},d_{n2}0,d_{n1},\mathrm{\hspace{0.17em}0}).$$
This proves the surjectivity of $`j_n`$.
To prove the injectivity of $`j_n`$, it is enough to define, for any $`n`$, a left inverse $`q_n:L_𝒢^n(X,Y)L_𝒢^1(X,Y)`$ to $`s_n:=j_{n1}\mathrm{}j_1`$. Suppose that $`(E^{};d)`$ represents an element of semigroup $`L_𝒢^n(X,Y)`$. Choose $`𝒢`$-invariant Hermitian metrics on $`E^i`$ and let $`d_i^{}:E^{i+1}|_YE^i|_Y`$ be the adjoint of $`d_i`$. Let
$$F^0:=\underset{i}{}E^{2i},F^1:=\underset{i}{}E^{2i+1},b:F^0|_YF^1|_Y,b=\underset{i}{}(d_{2i}+d_{2i+1}^{}).$$
A standard verification shows that $`b`$ is an isomorphism. Since all invariant metrics are homotopic to each other, Lemma 3.9 shows that $`(E,d)(F,b)`$ defines a morphism $`q_n:L_𝒢^n(X,Y)L_𝒢^1(X,Y)`$. This is the desired left inverse for $`s_n`$. ∎
Let us observe that the proof of the above theorem and Lemma 3.9 give the following corollary.
###### Corollary 3.11.
The class of $`E=(E^i,d_i)`$ in $`L_𝒢^n(X,Y)`$ does not change if we deform the differentials $`d_i`$ continuously.
We are now ready to prove the following basic result.
###### Theorem 3.12.
For each $`n`$ there exists a unique Euler characteristic
$$\chi _n:L_𝒢^n(X,Y)K_𝒢^0(X,Y).$$
In particular, $`L_𝒢^{\mathrm{}}(X,Y)K_𝒢^0(X,Y)`$ and $`L_𝒢^n(X,Y)`$ has a natural group structure for any closed, $`𝒢`$–invariant subbundle $`YX`$.
###### Proof.
The statement is obtained from the lemmas we have proved above as follows. First of all, Theorem 3.10 allows us to define
$$\chi _n:=\chi _1j_1^1\mathrm{}j_{n1}^1:L_𝒢^n(X,Y)K_𝒢^0(X,Y).$$
Lemma 3.7 shows that $`\chi _n`$ is an isomorphism. The uniqueness of $`\chi _n`$ is proved in the same way as the uniqueness of $`\chi _1`$ (Lemma 3.6). ∎
### 3.3. Globally defined complexes
The above theorem provides us with an alternative definition of the groups $`K_𝒢^0(X,Y)`$. We now derive yet another definition of these groups that is closer to what is needed in applications and is based on differentials defined on $`X`$, not just on $`Y`$.
Let $`(E,d)`$ be a complex of $`𝒢`$–equivariant vector bundles over a $`𝒢`$–space $`X`$. A point $`xX`$ will be called a point of acyclicity of $`(E,d)`$ if the restriction of $`(E,d)`$ to $`x,`$ i.e., the sequence of linear spaces
$$(E,d)_x=\left(\mathrm{}\stackrel{(d_i)_x}{}E_x^i\stackrel{(d_{i+1})_x}{}E_x^{i+1}\stackrel{(d_{i+2})_x}{}\mathrm{}\right),$$
is exact. The support $`\mathrm{supp}(E,d)`$ of the finite complex $`(E,d)`$ is the complement in $`X`$ of the set of its points of acyclicity. This definition and the following lemma hold also for $`X`$ non-compact.
###### Lemma 3.13.
The support $`\mathrm{supp}(E,d)`$ is a closed $`𝒢`$–invariant subspace of $`X.`$
###### Proof.
The fact that $`\mathrm{supp}(E,d)`$ is closed is classical (see for example). The invariance should be checked up over one fiber of $`X`$ at $`bB`$. But this is once again a well known fact of equivariant $`K`$-theory (see e.g. ). ∎
###### Lemma 3.14.
Let $`E^n,\mathrm{},E^0`$ be $`𝒢`$–equivariant vector bundles over $`X`$ and $`Y`$ be a closed, $`𝒢`$–invariant subbundle of $`X`$. Suppose there are given morphisms $`d_i:E^i|_YE^{i1}|_Y`$ such that $`(E^i|_Y,d_i)`$ is an exact complex. Then the morphisms $`d_i`$ can be extended to morphisms defined over $`X`$ such that we still have a complex of $`𝒢`$–equivariant vector bundles.
###### Proof.
We will show that we can extend each $`d_i`$ to a morphism $`r_i:E^iE^{i1}`$ such that $`r_{i1}r_i=0`$. Let us find a $`𝒢`$-invariant open neighborhood $`U`$ of $`Y`$ in $`X`$ such that for any $`i`$ there exist an extension $`s_i`$ of $`d_i`$ to $`U`$ with $`(E,s)`$ still an exact sequence. The desired $`r_i`$ will be then be defined as $`r_i=\rho s_i`$, where $`\rho :X[0,1]`$ is a continuous function, $`\rho =1`$ on $`Y`$ and $`\mathrm{supp}\rho U`$.
Let us construct $`U`$ by induction over $`i`$. Assume that for the closure $`\overline{U}_i`$ of some open $`𝒢`$-neighborhood of $`Y`$ in $`X`$ we can extend $`d_j`$ to $`s_j`$ ($`j=1,\mathrm{},i`$) such that on $`\overline{U}_i`$ the sequence
$$E^i\stackrel{s_i}{}E^{i1}\stackrel{s_{i1}}{}\mathrm{}E^00$$
is exact. Suppose, $`K_i:=\mathrm{ker}(s_i|\overline{U}_i)`$. Then $`d_{i+1}`$ determines a cross section of the bundle $`\mathrm{Hom}_𝒢(E^i,K_i)|_Y`$. This section can be extended to an open $`𝒢`$-neighborhood $`V`$ of $`Y`$ in $`\overline{U}_i`$. We hence obtain an extension $`s_{i+1}:E^{i+1}K_i`$ of $`d_{i+1}:E^{i+1}K_i`$ over $`V`$. Since $`d_{i+1}|_Y`$ is surjective (with range $`K_i`$), the morphism $`s_{i+1}`$ will be surjective $`\overline{U}_{i+1}`$ for some open $`U_{i+1}U_i`$. ∎
The above lemma suggests the following definition.
###### Definition 3.15.
Let $`X`$ be a compact $`𝒢`$–bundle and $`YX`$ be a $`𝒢`$–invariant subbundle. We define $`E_𝒢^n(X,Y)`$ to be the semigroup of homotopy classes of complexes of $`𝒢`$–equivariant vector bundles of length $`n`$ over $`X`$ such that their restrictions to $`Y`$ are acyclic (i.e., exact).
We shall say that two complexes are homotopic if they are isomorphic to the restrictions to $`X\times \{0\}`$ and $`X\times \{1\}`$ of a complex defined over $`X\times I`$ and acyclic over $`Y\times I`$.
###### Remark 3.16.
By Corollary 3.11, the restriction of morphisms induces a morphism $`\mathrm{\Phi }^n:E_𝒢^n(X,Y)L_𝒢^n(X,Y)`$.
###### Theorem 3.17.
Let $`X`$ be a compact $`𝒢`$–bundle and $`YX`$ be a $`𝒢`$–invariant subbundle. Then the natural transformation $`\mathrm{\Phi }_n`$, defined in the above remark, is an isomorphism.
###### Proof.
The surjectivity of $`\mathrm{\Phi }_n`$ follows from 3.14. The injectivity of $`\mathrm{\Phi }_n`$ can be proved in the same way as \[2, Lemma 2.6.13\], keeping in mind Lemma 3.14.
More precisely, we need to demonstrate that differentials of any complex over $`𝒢`$-subbundle $`(X\times \{0\})(X\times \{1\})(Y\times I)`$ of $`X\times I`$, which is acyclic over $`Y\times I`$, can be extended to a complex over the entire $`X\times I`$. The desired construction has the following three stages. First, let $`V`$ be a $`𝒢`$-invariant neighborhood of $`Y`$ such that the restriction of our complex is still acyclic on $`(V\times \{0\})(V\times \{1\})(Y\times I)`$ as well as on its closure $`(\overline{V}\times \{0\})(\overline{V}\times \{1\})(Y\times I)`$. By Lemma 3.14, one can extend the differentials $`d_i`$ to $`𝒢`$–equivariant morphisms $`r_i`$ over $`\overline{V}\times I`$ that still define a complex. Second, let $`\rho _1`$, $`\rho _2`$ be a $`𝒢`$-invariant partition of unity subordinated to covering $`V\times I`$, $`(XY)\times I`$ of $`X\times Y`$. Let us extend original differentials to $`(X\times [0,1/4])(X\times [3/4,1])(V\times I)`$ by taking $`d_i(x,t):=d_i(x,0)`$ for $`t\frac{1}{4}\rho _2(x)`$, $`xXY`$. Similarly near $`t=1`$. Also,
$$d_i(x,t):=r_i(x,\frac{\left(t\frac{1}{4}\rho _2(x)\right)}{\left(1\frac{1}{2}\rho _2(x)\right)}),xV.$$
Third, multiplying this differential by a function $`\tau :X\times II`$ equal to $`1`$ on the original subset of definition of the differential and to $`0`$ off $`(X\times [0,1/4])(X\times [3/4,1])(V\times I)`$ we obtain the desired extension. ∎
### 3.4. The non-compact case
In the case of a locally compact, paracompact $`𝒢`$–bundle $`X`$, we change the definitions of $`L_𝒢^n`$ and $`E_𝒢^n`$ as follows. In the definition of $`L_𝒢^n`$, the morphisms $`d_i`$ have to be defined and form an exact sequence off the interior of some compact $`𝒢`$–invariant subset $`C`$ of $`XY`$ (the complement of $`Y`$ in $`X`$). In the definition of $`E_𝒢^n`$, the complexes have to be exact outside some compact $`𝒢`$–invariant subset of $`XY`$. In other words, $`L_𝒢^n(X,Y)=L_𝒢^n(X^+,Y^+)`$.
Since the proof of Lemma 3.14 is still valid, we have the analogue of Theorem 3.17: there is a natural isomorphism
$$L_𝒢^n(X,Y)E_𝒢^n(X,Y).$$
The proof of the other statement also can be extended to the non-compact case. The only difference is that we have to replace $`Y`$ with $`XU`$, where $`U`$ is an open, $`𝒢`$–invariant subset with compact closure. Then, when we study two element sequences $`E=(E^i,d_i)`$, we have to take the unions of the corresponding open sets. Of course, these sets are not bundles, unlike $`Y`$, but for our argument using extensions this is not dangerous. This ultimately gives
(9)
$$K_𝒢^0(X,Y)L_𝒢^n(X,Y)E_𝒢^n(X,Y),n1.$$
As we shall see below, the liberty of using these equivalent definitions of $`K_𝒢^0(X,Y)`$ is quite convenient in applications, especially when studying products.
## 4. The Thom isomorphism
In this section, we establish the Thom isomorphism in gauge-equivariant $`K`$-theory. We begin with a discussion of products and of the Thom morphism.
### 4.1. Products
Let $`\pi _X:XB`$ be a $`𝒢`$–space, $`\stackrel{~}{\pi }_F:FX`$ be a complex $`𝒢`$–vector bundle over $`X`$, and $`s:XF`$ a $`𝒢`$–invariant section. We shall denote by $`\mathrm{\Lambda }^iF`$ the $`i`$-th exterior power of $`F`$, which is again a complex $`𝒢`$–equivariant vector bundle over $`X`$. As in the proof of the Thom isomorphism for ordinary vector bundles, we define the complex $`\mathrm{\Lambda }(F,s)`$ of $`𝒢`$–equivariant vector bundles over $`X`$ by
(10)
$$\mathrm{\Lambda }(F,s):=(0\mathrm{\Lambda }^0F\stackrel{\alpha ^0}{}\mathrm{\Lambda }^1F\stackrel{\alpha ^1}{}\mathrm{}\stackrel{\alpha ^{n1}}{}\mathrm{\Lambda }^nF0),$$
where $`\alpha ^k(v_x)=s(x)v_x`$ for $`v_x\mathrm{\Lambda }^kF^x`$ and $`n=dimF`$. It is immediate to check that $`\alpha ^{j+1}(x)\alpha ^j(x)=0`$, and hence that $`(\mathrm{\Lambda }(F,s),\alpha )`$ is indeed a complex.
The Künneth formula shows that the complex $`\mathrm{\Lambda }(F,s)`$ is acyclic for $`s(x)0`$, and hence $`\mathrm{supp}(\mathrm{\Lambda }(F,s)):=\{xX|s(x)=0\}`$. If this set is compact, then the results of Section 3 will associate to the complex $`\mathrm{\Lambda }(F,s)`$ of Equation (10) an element
(11)
$$[\mathrm{\Lambda }(F,s)]K_𝒢^0(X).$$
Let $`X`$ be a $`𝒢`$–bundle and $`\pi _F:FX`$ be a $`𝒢`$–equivariant vector bundle over $`X`$. The point of the above construction is that $`\pi _F^{}(F)`$, the lift of $`F`$ back to itself, has a canonical section whose support is $`X`$. Let us recall how this is defined. Let $`\pi _{FF}:\pi _F^{}(F)F`$ be the $`𝒢`$–vector bundle over $`F`$ with total space
$$\pi _F^{}(F):=\{(f_1,f_2)F\times F,\pi _F(f_1)=\pi _F(f_2)\}$$
and $`\pi _{FF}(f_1,f_2)=f_1`$. The vector bundle $`\pi _{FF}:\pi _F^{}(F)F`$ has the canonical section
$$s_F:F\pi _F^{}F,s_F(f)=(f,f).$$
The support of $`s_F`$ is equal to $`X`$. Hence, if $`X`$ is a compact space, using again the results of Section 3, especially 3.17, we obtain an element
(12)
$$\lambda _F:=[\mathrm{\Lambda }(\pi _F^{}(F),s_F)]K_𝒢^0(F).$$
Recall that the tensor product of vector bundles defines a natural product $`ab=abK_𝒢^0(X)`$ for any $`aK_𝒢^0(B)`$ and any $`bK_𝒢^0(X)`$, where $`\pi _X:XB`$ is a compact $`𝒢`$–space, as above.
Recall that all our vector bundles are assumed to be complex vector bundles, except for the ones coming from geometry (tangent bundles, their exterior powers) and where explicitly mentioned. Due to the importance that $`F`$ be complex in the following definition, we shall occasionally repeat this assumption.
###### Definition 4.1.
Let $`\pi _F:FX`$ be a (complex) $`𝒢`$–equivariant vector bundle. Assume the $`𝒢`$–bundle $`XB`$ is compact and let $`\lambda _FK_𝒢^0(F)`$ be the class defined in Equation (12), then the mapping
$$\phi ^F:K_𝒢^0(X)K_𝒢^0(F),\phi ^F(a)=\pi _F^{}(a)\lambda _F.$$
is called the Thom morphism.
As we shall see below, the definition of the Thom homomorphism extends to the case when $`X`$ is not compact, although the Thom element itself is not defined if $`X`$ is not compact.
The definition of the Thom isomorphism immediately gives the following proposition. We shall use the notation of Proposition 4.1.
###### Proposition 4.2.
The Thom morphism $`\phi ^F:K_𝒢^0(X)K_𝒢^0(F)`$ is a morphism of $`K_𝒢^0(B)`$–modules.
Let $`\iota :XF`$ be the zero section embedding of $`X`$ into $`F`$. Then $`\iota `$ induces homomorphisms
$$\iota ^{}:K_𝒢^0(F)K_𝒢^0(X)\text{and}\iota ^{}\phi ^F:K_𝒢^0(X)K_𝒢^0(X).$$
It follows from the definition that $`\iota ^{}\phi ^F(a)=a_{i=0}^n(1)^i\mathrm{\Lambda }^iF.`$
### 4.2. The non-compact case
We now consider now the case when $`X`$ is locally compact, but not necessarily compact. The complex $`\mathrm{\Lambda }(\pi _F^{}(F),s_F)`$ has a non-compact support, and hence it does not define an element of $`K_𝒢^0(F)`$. However, if $`a=[(E,\alpha )]K_𝒢^0(X)`$ is represented by the complex $`(E,\alpha )`$ of vector bundles with compact support (Section 3), then we can still consider the tensor product complex
$$(\pi _F^{}(),\pi _F^{}(\alpha ))\mathrm{\Lambda }(\pi _F^{}F,s_F).$$
From the Künneth formula for the homology of a tensor product we obtain that the support of a tensor product complex is the intersection of the supports of the two complexes. In particular, we obtain
(13)
$$\begin{array}{c}\mathrm{supp}\{(\pi _F^{}E,\pi _F^{}\alpha )\mathrm{\Lambda }(\pi _F^{}F,s_F)\}\mathrm{supp}(\pi _F^{}E,\pi _F^{}\alpha )\mathrm{supp}\mathrm{\Lambda }(\pi _F^{}F,s_F)\hfill \\ \hfill \mathrm{supp}(\pi _F^{}E,\pi _F^{}\alpha )X=\mathrm{supp}(E,\alpha ).\end{array}$$
Thus, the complex $`(\pi _F^{},\pi _E^{}\alpha )\mathrm{\Lambda }(\pi _F^{}F,s_F)`$ has compact support and hence defines an element in $`K_𝒢^0(F)`$.
###### Proposition 4.3.
The homomorphism of $`K_𝒢^0(B)`$-modules
(14)
$$\phi ^F:K_𝒢^0(X)K_𝒢^0(F),\phi ^F(a)=[(\pi _F^{},\pi _F^{}\alpha )\mathrm{\Lambda }(\pi _F^{}F,s_F)],$$
defined in Equation (13) extends the Thom morphism to the case of not necessarily compact $`X`$. The Thom morphism $`\phi ^F`$ satisfies
(15)
$$i^{}\phi ^F(a)=a\underset{i=0}{\overset{n}{}}(1)^i\mathrm{\Lambda }^iF$$
in the non-compact case as well.
Let $`FX`$ be a $`𝒢`$–equivariant vector bundle and $`F^1=F\times `$, regarded as a vector bundle over $`X\times `$. The periodicity isomorphisms in gauge-equivariant $`K`$-theory groups \[15, Theorem 3.18\]
$$K_𝒢^{i\pm 1}(X\times ,Y\times )K_𝒢^i(X,Y)$$
can be composed with $`\phi ^{F^1}`$, the Thom morphism for $`F^1`$, giving a morphism
(16)
$$\phi ^F:K_𝒢^i(X)K_𝒢^i(F),i=0,1.$$
This morphism is the Thom morphism for $`K^1`$.
Let $`p_X:XB`$ and $`p_Y:YB`$ be two compact $`𝒢`$-fiber bundles. Let $`\pi _E:EX`$ and $`\pi _F:FY`$ be two complex $`𝒢`$–equivariant vector bundles. Denote by $`p_1:X\times _BYX`$ and by $`p_2:X\times _BYY`$ the projections onto the two factors and define $`EF:=p_1^{}Ep_2^{}F`$. The $`𝒢`$–equivariant vector bundle $`EF`$ will be called the external tensor product of $`E`$ and $`F`$ over $`B`$. It is a vector bundle over $`X\times _BY`$. Then the formula
$$K_𝒢^0(X)K_𝒢^0(Y)[E][F][E][F]:=[EF]K_𝒢^0(X\times _BY)$$
defines a product $`K_𝒢^0(X)K_𝒢^0(Y)K_𝒢^0(X\times _BY)`$.
In particular, consider two complex $`𝒢`$–equivariant vector bundles $`\pi _E:EX`$ and $`\pi _F:FX`$. Then $`EF=E\times _XF`$ and we obtain a product
$$K_𝒢^i(E)K_𝒢^j(F)[E][F][E][F]:=[EF]K_𝒢^{i+j}(EF).$$
Using also periodicity, we obtain the product
(17)
$$:K_𝒢^i(E)K_𝒢^j(F)K_𝒢^{i+j}(EF).$$
This product is again seen to be given by the tensor product of the (lifted) complexes (when representing $`K`$-theory classes by complexes) as in the classical case.
The external product $``$ behaves well with respect to the “Thom construction,” in the following sense. Let $`F^1`$ and $`F^2`$ be two complex bundles over $`X,`$ and $`s_1`$, $`s_2`$ two corresponding sections of these bundles. Then
(18)
$$\mathrm{\Lambda }(F^1F^2,s_11+1s_2)=\mathrm{\Lambda }(F^1,s_1)\mathrm{\Lambda }(F^2,s_2).$$
In particular, if $`X`$ is compact, we obtain
(19)
$$\lambda _E\lambda _F=\lambda _{EF}.$$
We shall write $`s_1+s_2=s_11+1s_2`$, for simplicity.
The following theorem states that the Thom class is multiplicative with respect to direct sums of vector bundles (see also ).
###### Theorem 4.4.
Let $`E,FX`$ be two $`𝒢`$–equivariant vector bundles, and regard $`EFE`$ as the $`𝒢`$–equivariant vector bundle $`\pi _E^{}(F)`$ over $`E`$. Then $`\phi ^{\pi _E^{}(F)}\phi ^E=\phi ^{EF}.`$
The above theorem amounts to the commutativity of the diagram
(20)
###### Proof.
Let $`F^1:=\pi _E^{}(F)=EF`$, regarded as a vector bundle over $`E`$. Consider the projections
$$\pi _E:EX,\pi _F:FX,\pi _{EF}:EFX,t=\pi _{F^1}:EFE.$$
Let $`xK_𝒢^0(X).`$ Then $`\phi ^E(x)=\pi _E^{}(x)\mathrm{\Lambda }(\pi _E^{}(E),s_E).`$ Now we use that $`t^{}\pi _E^{}(x)=\pi _{EF}^{}(x)`$ and $`t^{}\mathrm{\Lambda }(\pi _E^{}(E),s_E)=\mathrm{\Lambda }(\pi _{EF}^{}(E),s_Et)`$. Since $`s_Et+s_{F^1}=s_{EF},`$ Equations (18) and (19) then give
$$\mathrm{\Lambda }(\pi _{EF}^{}(EF),s_{EF})=\mathrm{\Lambda }(\pi _{EF}^{}(E),s_Et)\mathrm{\Lambda }(\pi _{EF}^{}(F),s_{F^1}).$$
Putting together the above calculations we obtain
$$\begin{array}{c}\phi ^{F^1}\phi ^E(x)=t^{}(\phi ^E(x))\mathrm{\Lambda }(t^{}(F^1),s_{F^1})\hfill \\ \hfill =t^{}\pi _E^{}(x)t^{}(\mathrm{\Lambda }(\pi _E^{}(E),s_E))\mathrm{\Lambda }(t^{}(F^1),s_{F^1})\\ \hfill =\pi _{EF}^{}(x)\mathrm{\Lambda }(\pi _{EF}^{}E,s_Et)\mathrm{\Lambda }(\pi _{EF}^{}F,s_{F^1})\\ \hfill =\pi _{EF}^{}(x)\mathrm{\Lambda }(\pi _{EF}^{}(EF),s_{EF})=\phi _1^{EF}(x).\end{array}$$
The proof is now complete. ∎
We are now ready to formulate and prove the Thom isomorphism in the setting of gauge-equivariant vector bundles.
###### Theorem 4.5.
Let $`XB`$ be a $`𝒢`$-bundle and $`FX`$ a complex $`𝒢`$–equivariant vector bundle, then $`\phi ^F:K_𝒢^i(X)K_𝒢^i(F)`$ is an isomorphism.
###### Proof.
Assume first that $`F`$ is a trivial bundle, that is, that $`F=X\times _B𝒱`$, where $`𝒱B`$ is a complex, finite-dimensional $`𝒢`$–equivariant vector bundle. We continue to assume that $`B`$ is compact.
Let us denote by $`\underset{¯}{}:=B\times `$ the $`1`$-dimensional $`𝒢`$-bundle with the trivial action of $`𝒢`$ on $``$. Also, let us denote by $`P(𝒱\underset{¯}{})`$ the projective space associated to $`𝒱\underset{¯}{}`$. As a topological space, $`P(𝒱\underset{¯}{})`$ identifies with the fiberwise one-point compactification of $`𝒱`$. The embeddings $`𝒱P(𝒱\underset{¯}{})`$ and $`𝒱\times _BXP(𝒱\underset{¯}{})\times _BX`$ then gives rise to the following natural morphism (Equation (6))
$$j:K_𝒢^0(𝒱)K_𝒢^0(P(𝒱\underset{¯}{})),j:K_𝒢^0(𝒱\times _BX)K_𝒢^0(P(𝒱\underset{¯}{})\times _BX).$$
Let $`X`$ be compact and let $`xK_𝒢^0(P(𝒱\underset{¯}{})\times _BX)`$ be arbitrary. The fibers of the projectivization $`P(V\underset{¯}{})`$ are complex manifolds, so we can consider the analytical index of the correspondent family of Dolbeault operators over $`P(V1)`$ with coefficients in $`x`$ (cf. \[3, page 123\]). This index is an element of $`K_𝒢^0(X)`$ by the results of . Taking the composition with $`j`$ (cf. \[3, page 122-123\]) we get a family of mappings $`\alpha _X:K_𝒢^0(𝒱\times _BX)K_𝒢^0(X)`$, having the following properties:
1. $`\alpha _X`$ is functorial with the respect to $`𝒢`$–equivariant morphisms;
2. $`\alpha _X`$ is a morphism of $`K_𝒢^0(X)`$-modules;
3. $`\alpha _B(\lambda _V)=1K_𝒢^0(B)`$.
Let $`X^+:=XB`$ be the fiberwise one-point compactification of $`X`$. The commutative diagram
with exact lines allows us to define $`\alpha _X`$ for $`X`$ non-compact.
Let $`xK_𝒢^0(X)`$, then by (ii)
(21)
$$\alpha _X(\lambda _Vx)=\alpha _X(\lambda _𝒱)x=x,\alpha \phi =\mathrm{Id}.$$
Let $`q:=\pi _F:F=𝒱\times _BXX`$, $`p:X\times _B𝒱X`$, $`q_1:𝒱\times _BX\times _B𝒱𝒱`$ (onto the first entry), $`q_2:X\times _B𝒱B`$, and by $`\stackrel{~}{y}K_𝒢^0(X\times _B𝒱)`$ we denote the element, obtained from $`y`$ under the mapping $`X\times _B𝒱𝒱\times _BX`$, $`(x,v)(v,x)`$ (such that $`𝒱\times _BX\times _B𝒱𝒱\times _BX\times _B𝒱`$, $`(u,x,v)(v,x,u)`$ is homotopic to the identity).
Let $`yK_𝒢^0(𝒱\times _BX)`$, then once again by (i), (ii) and then by (iii)
(22)
$$\begin{array}{c}\phi (\alpha _X(y))=\pi _E^{}\alpha _X(y)q^{}\lambda _𝒱=\alpha _{X\times _B𝒱}(p_1^{}y)q^{}\lambda _𝒱=\alpha _{X\times _B𝒱}(p_1^{}yq^{}\lambda _V)\hfill \\ \hfill =\alpha _{X\times _B𝒱}(y\lambda _V)=\alpha _{X\times _B𝒱}(\lambda _V\stackrel{~}{y})=\alpha _{X\times _B𝒱}(q_1^{}\lambda _V\stackrel{~}{y})\\ \hfill =\alpha _{X\times _B𝒱}(q_1^{}\lambda _V)\stackrel{~}{y}=q_2^{}\alpha _B(\lambda _V)\stackrel{~}{y}=q_2^{}(1)\stackrel{~}{y}=\stackrel{~}{y}K_𝒢^0(X\times _B𝒱),\end{array}$$
We obtain that $`\phi \alpha _X`$ is an isomorphism. Since $`\alpha _X\phi =\mathrm{Id}`$, $`\alpha _X`$ is the two-sides inverse of $`\phi `$ and the automorphism $`\phi \alpha _X`$ is the identity.
The proof for a general (complex) $`𝒢`$–equivariant vector bundle $`FX`$ can be done as in \[3, p. 124\]. However, we found it more convenient to use the following argument. Embed first $`F`$ into a trivial bundle $`E=𝒱\times _BX`$. Let $`\phi ^1`$ and $`\phi ^2`$ be the Thom maps associated to the bundles $`EF`$ and $`EX`$. Then by Theorem 4.4 the diagram
is commutative, while $`\phi ^2`$ is an isomorphism by the first part of the proof. Therefore $`\phi _F`$ is injective. The same argument show that $`\phi ^1`$ is injective. But $`\phi ^1`$ must also be surjective, because $`\phi ^2`$ is an isomorphism. Thus, $`\phi ^1`$ is an isomorphism, and hence $`\phi ^2`$ is an isomorphism too. ∎
## 5. Gysin maps
We now discuss a few constructions related to the Thom isomorphism, which will be necessary for the definition of the topological index. The most important one is the Gysin map. For several of the constructions below, the setting of $`𝒢`$-spaces and even $`𝒢`$-bundles is too general, and we shall have to consider longitudinally smooth $`𝒢`$–fiber bundles $`\pi _X:XB`$. The main reason why we need longitudinally smooth bundles to define the Gysin map is the same as in the definition of the Gysin map for embeddings of smooth manifolds. We shall denote by $`T_{vert}X`$ the vertical tangent bundle to the fibers of $`XB`$. All tangent bundles below will be vertical tangent bundles.
Let $`X`$ and $`Y`$ be longitudinally smooth $`𝒢`$-fiber bundles, $`i:XY`$ be an equivariant fiberwise embedding, and $`p_T:T_{vert}XX`$ be the vertical tangent bundle to $`X.`$ Assume $`Y`$ is equipped with a $`𝒢`$-invariant Riemannian metric and let $`p_N:N_{vert}X`$ be the fiberwise normal bundle to the image of $`i`$
Let us choose a function $`\epsilon :X(0,\mathrm{})`$ such that the map of $`N_{vert}`$ to itself
$$n\epsilon \frac{n}{1+|n|}$$
is $`𝒢`$-equivariant and defines a $`𝒢`$-diffeomorphism $`\mathrm{\Phi }:N_{vert}W`$ onto a bundle of open tubular neighborhoods $`WX`$ in $`Y.`$
Let $`(NN)_{vert}:=N_{vert}N_{vert}`$. The embedding $`i:XY`$ can be written as a composition of two fiberwise embeddings $`i_1:XW`$ and $`i_2:WY`$. Passing to differentials we obtain
$$T_{vert}X\stackrel{di_1}{}T_{vert}W\stackrel{di_2}{}T_{vert}Y\text{and}d\mathrm{\Phi }:T_{vert}NT_{vert}W,$$
where we use the simplified notation $`T_{vert}N=T_{vert}N_{vert}`$.
###### Lemma 5.1.
(cf. \[10, page 112\]) The manifold $`T_{vert}N`$ can be identified with $`p_T^{}(NN)_{vert}`$ with the help of a $`𝒢`$-equivariant diffeomorphism $`\psi `$ such that makes the following diagram commutative
###### Proof.
The vertical tangent bundle $`T_{vert}NN_{vert}`$ and the vector bundle
$$p_N^{}(T_{vert}X)p_N^{}(N_{vert})N_{vert}$$
are isomorphic as $`𝒢`$–equivariant vector bundles over $`N_{vert}`$.
Indeed, a point of the total space $`T_{vert}N`$ is a pair of the form $`(n_1,t+n_2),`$ where both vectors are from the fiber over the point $`xX.`$ Similarly, we represent elements $`p_T^{}(NN)_{vert}`$ as pairs of the form $`(t,n_1+n_2).`$ Let us define $`\psi `$ by the equality $`\psi (n_1,t+n_2)=(t,n_1+n_2).`$
With the help of the relation $`i(n_1,n_2)=(n_2,n_1)`$, we can equip
$$p_T^{}(NN)_{vert}=p_T^{}(N_{vert})p_T^{}(N_{vert})$$
with a structure of a complex manifold. Then we can consider the Thom homomorphism
$$\phi :K_𝒢^0(T_{vert}X)K_𝒢^0(p_T^{}(NN)_{vert}).$$
Since $`T_{vert}W`$ is an open $`𝒢`$-stable subset of $`T_{vert}Y`$ and $`di_2:T_{vert}WT_{vert}Y`$ is a fiberwise embedding, by Equation 6, there is the homomorphism $`(di_2)_{}:K_𝒢^0(T_{vert}W)K_𝒢^0(T_{vert}Y).`$
###### Definition 5.2.
Let $`i:XY`$ be an equivariant embedding of $`𝒢`$-bundles. The Gysin homomorphism is the mapping
$$i_!:K_𝒢^0(T_{vert}X)K_𝒢^0(T_{vert}Y),i_!=(di_2)_{}(d\mathrm{\Phi }^1)^{}\psi ^{}\phi .$$
In other words, it is obtained by passage to $`K`$-groups in the upper part of the diagram
Another choice of metric and neighborhood $`W`$ induces the homotopic map and (by the item 3 of Theorem 5.3 below) the same homomorphism.
###### Theorem 5.3 (Properties of Gysin homomorphism).
Let $`i:XY`$ be a $`𝒢`$-embedding.
1. $`i_!`$ is a homomorphism of $`K_𝒢^0(B)`$-modules.
2. Let $`i:XY`$ and $`j:YZ`$ be two fiberwise $`𝒢`$-embeddings, then $`(ji)_!=j_!i_!.`$
3. Let fiberwise embeddings $`i_1:XY`$ and $`i_2:XY`$ be $`𝒢`$-homotopic in the class of embeddings. Then $`(i_1)_!=(i_2)_!.`$
4. Let $`i_!:XY`$ be a fiberwise $`𝒢`$-diffeomorphism, then $`i_!=(di^1)^{}.`$
5. A fiberwise embedding $`i:XY`$ can be represented as a compositions of embeddings $`X`$ in $`N_{vert}`$ (as the zero section $`s_0:xN`$) and $`N_{vert}Y`$ by $`i_2\mathrm{\Phi }:N_{vert}Y.`$ Then $`i_!=(i_2\mathrm{\Phi })_!(s_0)_!.`$
6. Consider the complex bundle $`p_T^{}(N_{vert})`$ over $`T_{vert}X`$. Let us form the complex $`\mathrm{\Lambda }(p_T^{}(N_{vert}),0):`$
$$0\mathrm{\Lambda }^0(p_T^{}(N_{vert}))\stackrel{0}{}\mathrm{}\stackrel{0}{}\mathrm{\Lambda }^k(p_T^{}(N_{vert}))0$$
with noncompact support. If $`aK_𝒢^0(T_{vert}X),`$ then the complex
$$a\mathrm{\Lambda }(p_T^{}(N_{vert}),0)$$
has compact support and defines an element of $`K_𝒢^0(T_{vert}X)`$. Then
$$(di)^{}i_!(a)=a\mathrm{\Lambda }(p_T^{}(N_{vert}),0),$$
where $`di`$ is the differential of the embedding $`i`$.
7. $`i_!(x(di)^{}y)=i_!(x)y,`$ where $`xK_𝒢^0(T_{vert}X)`$ and $`yK_𝒢^0(T_{vert}Y)`$.
###### Proof.
(i) This follows from the definition of $`i_!.`$
(ii) To simplify the argument, let us identify the tubular neighborhood with the normal bundle. Then $`(ji)_!`$ is the composition
$$K_𝒢^0(T_{vert}X)\stackrel{\phi }{}K_𝒢^0(T_{vert}NT_{vert}N_{vert}^{\prime \prime })K_𝒢^0(T_{vert}Z),$$
where $`N_{vert}^{}`$ is the fiberwise normal bundle of $`Y`$ in $`Z,`$ $`N_{vert}^{\prime \prime }=N_{vert}^{}|_X,`$ and for the sum of tangent bundles to the vertical normal bundles $`T_{vert}NT_{vert}N_{vert}^{\prime \prime }`$ is considered in the same way as on page 5, that is, as a complex bundle over $`T_{vert}X`$. On the other hand, $`j_!i_!`$ represents the composition
$$K_𝒢^0(T_{vert}X)\stackrel{\phi }{}K_𝒢^0(T_{vert}N)K_𝒢^0(T_{vert}Y)\stackrel{\phi }{}K_𝒢^0(T_{vert}N_{vert}^{})K_𝒢^0(T_{vert}Z).$$
By the properties of $`\phi `$, the following diagram is commutative
This completes the proof of (ii).
(iii) The morphism $`q_T`$ depends only on the homotopy class of the embeddings used to define it. The assertion thus follows from the homotopy invariance of $`K`$-theory.
(iv) In this case $`N=X,W=Y,\mathrm{\Phi }=i,i_2=\mathrm{Id}_Y,`$ and the formula is obvious.
(v) This follows from (ii).
(vi) By definition,
$$(di)^{}i_!=(di_1)^{}(di_2)^{}(di_2)_{}(d\phi ^1)^{}\psi ^{}\phi ^{}=(\psi d\mathrm{\Phi }^1di_1)^{}\phi ,$$
where $`i_1:XW,i_2:WY.`$ Let $`(n_1,t+n_2)T_{vert}N=p_N^{}(T_{vert}X)p_N^{}(N_{vert})`$, where $`n_1`$ is the shift under the exponential mapping and $`t+n_2`$ is a vertical tangent vector to $`W.`$ If $`d\mathrm{\Phi }(n_1,t+n_2)`$ is in $`T_{vert}X`$, then $`n_1=n_2=0.`$ Hence,
$$d\mathrm{\Phi }^1di_1(t)=(0,t+0),\psi d\mathrm{\Phi }^1di_1(t)=(t,0+0).$$
Therefore, $`\psi d\mathrm{\Phi }^1di_1:T_{vert}Xp_T^{}(N_{vert}N_{vert})`$ is the embedding of the zero section. Since $`\phi (a)=a\mathrm{\Lambda }(q_T^{}p_T^{}(N_{vert}),s_{p_T^{}(N_{vert})}),`$ it follows that $`(di)^{}i_!(a)=a\mathrm{\Lambda }(p_T^{}(N_{vert}),0).`$
(vii) The mapping $`di_1q_T\psi d\mathrm{\Phi }^1:T_{vert}WT_{vert}W`$ is homotopic to the identical mapping. Hence,
$$\begin{array}{c}i_!(x(di)^{}y)=(di_2)_{}(d\mathrm{\Phi }^1)^{}\psi ^{}\phi (x(di)^{}y)=\hfill \\ =(di_2)_{}(d\mathrm{\Phi }^1)^{}\psi ^{}[(q_T^{}(x)\lambda _{p_T^{}(N_{vert})})(q_T^{}(di)^{}y)]=\hfill \\ =(di_2)_{}[(d\mathrm{\Phi }^1)^{}\psi ^{}(q_T^{}(x)\lambda _{p_T^{}(N_{vert})})\underset{\mathrm{Id}}{\underset{}{(d\mathrm{\Phi }^1)^{}\psi ^{}q_T^{}(di_1)^{}}}(di_2)^{}y]=\hfill \\ =[(di_2)_{}(d\mathrm{\Phi }^1)^{}\psi ^{}(q_T^{}(x)\lambda _{p_T^{}(N_{vert})})][(di_2)_{}(di_2)^{}y]=i_!(x)y.\hfill \end{array}$$
The proof is now complete. ∎
We shall need also the following properties of the Gysin map. If $`X=B`$, the trivial longitudinally smooth $`𝒢`$-bundle, we shall identify $`T_{vert}X=B`$ and $`T_{vert}𝒱=𝒱`$ for a real bundle $`𝒱B`$.
###### Theorem 5.4.
Suppose that $`𝒱B`$ is a $`𝒢`$-equivariant real vector bundle and that $`X=B`$. Then the mapping
$$i_!:K_𝒢^0(B)=K_𝒢^0(TX_{vert})K_𝒢^0(T_{vert}𝒱)=K_𝒢^0(𝒱)$$
coincides with the Thom homomorphism $`\phi ^𝒱`$.
###### Proof.
The assertion follows from the definition of $`i_!`$. More precisely, let $`X=B𝒱,N=𝒱`$ be the zero section embedding. In the definition of the Thom isomorphism, $`W`$ can be chosen to be equal to the bundle $`D_1`$ of interiors of the balls of radius 1 in $`𝒱`$ with respect to an invariant metric. In this case, the diagram from the definition of the Gysin homomorphism 5.2 takes the following form
In our case $`\mathrm{\Psi }=\mathrm{Id}`$ and $`di_2d\mathrm{\Phi }`$ is homotopic to $`\mathrm{Id},`$ since this map has the form $`vz(vz)/(1+|vz|).`$ Hence, $`i_!=\phi .`$
###### Theorem 5.5.
Suppose that $`𝒱^{}`$ and $`𝒱^{\prime \prime }`$ are $`𝒢`$-equivariant $``$-vector bundles over $`B`$ and that $`i:X𝒱^{}`$ is an embedding. Let $`k:X𝒱^{}𝒱^{\prime \prime }`$, $`k(x)=i(x)+0.`$ Let $`\phi `$ be the Thom homomorphism of the complex bundle
$$T_{vert}(𝒱^{}𝒱^{\prime \prime })=𝒱^{}𝒱^{\prime \prime }\stackrel{}{}T_{vert}𝒱^{}=𝒱^{}.$$
Then the following diagram is commutative
###### Proof.
To prove the statement, let us consider the embedding $`i:X𝒱^{}`$ and denote, as before, by $`N_{vert}`$ the fiberwise normal bundle and by $`W`$ the tubular neighborhood appearing in the definition of the Thom isomorphism associated to $`i`$. Then $`N_{vert}𝒱^{\prime \prime }`$ is a fiberwise normal bundle for the embedding $`k`$ with the tubular neighborhood $`WD(𝒱^{\prime \prime }),`$ where $`D(𝒱^{\prime \prime })`$ is a ball bundle. If $`aK_𝒢^0(T_{vert}X),`$ then
$$k_!(a)=(di_21)_{}(d\mathrm{\Phi }^11)^{}(\psi 1)^{}\phi ^{NN𝒱^{\prime \prime }𝒱^{\prime \prime }}(a).$$
We have $`\phi ^{NN𝒱^{\prime \prime }𝒱^{\prime \prime }}=\phi ^{NN}\phi ^{𝒱^{\prime \prime }𝒱^{\prime \prime }},`$ by Theorem 4.4. Since $`a=a\underset{¯}{},`$ where $`\underset{¯}{}`$ is the trivial line bundle, we obtain
(23)
$$\begin{array}{c}k_!(a)=(di_2)_{}(d\mathrm{\Phi }^1)^{}\mathrm{\Psi }^{}\phi ^{N_{vert}N_{vert}}(a)\phi ^{𝒱^{\prime \prime }𝒱^{\prime \prime }}(\underset{¯}{})\hfill \\ \hfill =i_!(a)\lambda _{T(𝒱^{}𝒱^{\prime \prime })_{vert}}=\phi (i_!(a)).\end{array}$$
The proof is now complete. ∎
## 6. The topological index
We begin with a “fibered Mostow-Palais theorem” that will be useful in defining the index.
###### Theorem 6.1.
Let $`\pi _X:XB`$ be a compact $`𝒢`$-fiber bundle. Then there exists a real $`𝒢`$–equivariant vector bundle $`𝒱B`$ and a fiberwise smooth $`𝒢`$-embedding $`X𝒱`$. After averaging one can assume that the action of $`𝒢`$ on $`𝒱`$ is orthogonal.
###### Proof.
Fix $`bB`$ and let $`U`$ be an equivariant trivialization neighborhood of $`b`$ for both $`X`$ and $`𝒢`$. By the Mostow-Palais theorem, there exists a representation of $`𝒢_b`$ on a finite dimensional vector space $`V_b`$ and a smooth $`𝒢_b`$-equivariant embedding $`i_b:X_bV_b`$. This defines an embedding
(24)
$$\psi :\pi _X^1(U_b)U_b\times X_bU_b\times V_b$$
which is $`𝒢`$-equivariant in an obvious sense.
We can cover $`B`$ with finitely many open sets $`U_{b_j}`$, as above, corresponding to the points $`b_j`$, $`j=1,\mathrm{},N`$. Denote by $`V_j`$ the corresponding representations and by $`\psi _j`$ the corresponding embeddings, as in Equation (24). Let $`W:=V_j`$. Also, Let $`\varphi _j`$ be a partition of unity subordinated to the covering by $`U_j=U_{b_j}`$. We define then
$$\mathrm{\Psi }:=(\varphi _j\pi _X)\psi _j:XB\times W,$$
which is a $`𝒢`$-equivariant embedding of $`X`$ into the trivial $`𝒢`$–equivariant vector bundle $`𝒱:=B\times W`$, as desired. ∎
Let us now turn to the definition of the topological index. Let $`XB`$ be a compact, longitudinally smooth $`𝒢`$-bundle. From Theorem 6.1 it follows that there exists an $`𝒢`$–equivariant real vector bundle $`𝒱B`$ and a fiberwise smooth $`𝒢`$-equivariant embedding $`i:X𝒱`$. We can assume that $`𝒱`$ is endowed with an orthogonal metric and that $`𝒢`$ preserves this metric. Thus, the Gysin homomorphism
$$i_!:K_𝒢^0(T_{vert}X)K_𝒢^0(T_{vert}𝒱)=K_𝒢^0(𝒱)$$
is defined (see Section 4). Since $`T_{vert}𝒱=𝒱`$ is a complex vector bundle, we have the following Thom isomorphism (see Section 4):
$$\phi :K_𝒢^0(B)\stackrel{}{}K_𝒢^0(T_{vert}𝒱).$$
###### Definition 6.2.
The topological index is by definition the morphism:
$$\mathrm{t}\text{-}\mathrm{ind}_𝒢^X:K_𝒢^0(T_{vert}X)K_𝒢^0(B),\mathrm{t}\text{-}\mathrm{ind}_𝒢^X:=\phi ^1i_!.$$
The topological index satisfies the following properties.
###### Theorem 6.3.
Let $`XB`$ be a longitudinally smooth bundle and
$$\mathrm{t}\text{-}\mathrm{ind}_𝒢^X:K_𝒢^0(T_{vert}X)K_𝒢^0(B)$$
be its associated topological index. Then
1. $`\mathrm{t}\text{-}\mathrm{ind}_𝒢^X`$ does not depend on the choice of the $`𝒢`$–equivariant vector bundle $`𝒱`$ and on the embedding $`i:X𝒱`$.
2. $`\mathrm{t}\text{-}\mathrm{ind}_𝒢^X`$ is a $`K_𝒢^0(B)`$-homomorphism.
3. If $`X=B`$, then the map
$$\mathrm{t}\text{-}\mathrm{ind}_𝒢^X:K_𝒢^0(B)=K_𝒢^0(T_{vert}X)K_𝒢^0(B)$$
coincides with $`\mathrm{Id}_{K_𝒢^0(B)}`$.
4. Suppose $`X`$ and $`Y`$ are compact longitudinally smooth $`𝒢`$-bundles, $`i:XY`$ is a fiberwise $`𝒢`$-embedding. Then the diagram
commutes.
###### Proof.
To prove (i), let us consider two embeddings
$$i_1:X𝒱^{},i_2:X𝒱^{\prime \prime }$$
into $`𝒢`$–equivariant vector bundles. Denote by $`j=i_1+i_2`$ the induced embedding $`j:X𝒱^{}𝒱^{\prime \prime }`$. It is sufficient to show that $`i_1`$ and $`j`$ define the same topological index. Let us define a homotopy of $`𝒢`$-embeddings by the formula
$$j_s(x)=i_1(x)+si_2(x):X𝒱^{}𝒱^{\prime \prime },0s1.$$
Then, by Theorems 5.3(iii) and 5.5, the indices for $`j`$ and $`j_0`$ coincide. Let us show now that $`j_0=i_1+0`$ and $`i_1`$ define the same topological indexes. For this purpose consider the diagram
where $`\phi _i`$ are the corresponding Thom homomorphisms. The upper triangle is commutative by Theorem 5.4.2, and the lower is commutative by Theorem 4.4. Hence $`\phi _1^1(i_1)_!=\phi _3^1(j_0)_1`$ as desired.
(ii) follows from 4.2 and 5.3(i).
Property (iii) follows from the definition of the index and from 5.4.
To prove (iv), let us consider the diagram
We now use 5.3(ii). This gives the commutative diagram
or
This completes the proof. ∎
We now investigate the behavior of the topological index with respect to fiber products of bundles of compact groups.
###### Theorem 6.4.
Let $`\pi :PX`$ be a principal right $``$-bundle with a left action of $`𝒢`$ commuting with $``$. Suppose $`F`$ is a longitudinally smooth $`(𝒢\times )`$-bundle. Let us denote by $`Y`$ the space $`P\times _{}F`$. Let $`j:X^{}X`$ and $`k:F^{}F`$ be fiberwise $`𝒢`$\- and $`(𝒢\times )`$-embeddings, respectively. Let $`\pi ^{}:P^{}X^{}`$ be the principal $``$-bundle induced by $`j`$ on $`X^{}`$. Assume that $`Y^{}:=P^{}\times _{}F^{}`$. The embeddings $`j`$ and $`k`$ induce $`𝒢`$-embedding $`jk:Y^{}Y`$. Then the diagram
is commutative.
Let us remark that there in the statement of this theorem there is no compactness assumption on $`X,X^{},F,`$ and $`F^{}`$, since there is no compactness assumption in the definition of the Gysin homomorphism. This is unlike in the definition of the topological index where we start with a compact $`𝒢`$-bundle $`XB`$.
###### Proof.
Let us use the definition of $`\gamma `$:
(25)
$$\begin{array}{c}K_𝒢^0((\pi _1^{}T_{vert}X)\times (P\times _{}T_{vert}F))\\ \text{4}\\ K_𝒢^0((\pi _1^{})^{}T_{vert}X^{}\times (P^{}\times _{}T_{vert}F^{}))\end{array}$$
$$\begin{array}{c}K_𝒢^0(\pi _1^{}T_{vert}X(P\times _{}T_{vert}F))=K_𝒢^0(T_{vert}Y)\\ \text{4}(jk)_!\\ K_𝒢^0((\pi _1^{})^{}T_{vert}X^{}(P^{}\times _{}T_{vert}F^{}))=K_𝒢^0(TY_{vert}^{}),\end{array}$$
where projections $`\pi _1:Y=P\times _{}FX`$ and $`\pi _1^{}:Y^{}=P\times _{}F^{}X^{}`$ are defined as above. Here we use the isomorphism $`K_{𝒢\times }^0(P\times W)K_𝒢^0(P\times _{}W)`$ for a free $``$-bundle $`P`$ (see Theorem 2.6). Let us remind the diagram, which was used for the definition of the Gysin homomorphism of an embedding $`j:X^{}X`$:
From the similar diagrams for $`k_!`$ and $`(jk)_!`$ and the explicit form of these maps, it follows that the square 4 in (25) is commutative if, and only if, $`\alpha `$ has the following form:
$$\begin{array}{c}\alpha (\sigma \rho )=(\pi _1^{})\left\{(dj_2)_{}(d\mathrm{\Phi }_X^{}^1)^{}\psi _X^{}^{}\right\}\phi ^S(\sigma )\hfill \\ (\pi ^{}j_2\times _{}dk_2)_{}\left((\pi ^{}\mathrm{\Phi }_X^{}\times _{}d\mathrm{\Phi }_F^{})^1\right)^{}(1\times _{}\psi _F^{})^{}\phi ^R(\rho ),\hfill \end{array}$$
where $`S`$ and $`T`$ are bundles of the form
$$\begin{array}{cccc}& \pi _1^{}\left((p_T^X^{})^{}\{N_X^{}N_X^{}\}\right)& & \pi ^{}N_X^{}\times _{}(p_T^F^{})^{}(N_F^{}N_F^{})\\ S:& (\pi _1^{})^{}q_T^X^{}& R:& (\pi ^{})^{}(p_{N_X^{}})\times _{}q_T^F^{}\\ & (\pi _1^{})^{}(T_{vert}X^{}),& & \pi ^{}X^{}\times _{}T_{vert}F^{}=P^{}\times _{}T_{vert}F^{}.\end{array}$$
Hence the square 3 in (25) is commutative iff the homomorphism $`\beta `$ has the form
$$\begin{array}{c}\beta (\tau \rho )=j_!(\tau )\hfill \\ (\pi ^{}j_2\times _{}dk_2)_{}\left((\pi ^{}\mathrm{\Phi }_X^{}\times _{}d\mathrm{\Phi }_F^{})^1\right)^{}(1\times _{}\psi _F^{})^{}\phi ^R(\rho ),\hfill \end{array}$$
$$\tau K_𝒢^0(TX^{}),\rho K_𝒢^0(P^{}\times _{}TF^{}).$$
In turn, the square 2 in (25) is commutative iff the homomorphism $`\epsilon `$ has the form
$$\begin{array}{c}\epsilon (\tau \delta )=j_!(\tau )\hfill \\ (\pi ^{}j_2\times _{}dk_2)_{}\left((\pi ^{}\mathrm{\Phi }_X^{}\times _{}d\mathrm{\Phi }_F^{})^1\right)^{}(1\times _{}\psi _F^{})^{}\phi _{}^{\stackrel{~}{R}}(\delta ),\hfill \end{array}$$
$$\tau K_𝒢^0(TX^{}),\delta K_{G\times }^0(P^{}\times TF^{}),$$
where $`\stackrel{~}{R}`$ is the following bundle:
$$\begin{array}{cc}& \pi ^{}N_X^{}\times (p_T^F^{})^{}(N_F^{}N_F^{})\\ \stackrel{~}{R}:& (\pi ^{})^{}(p_N)\times q_T^F^{}\\ & P^{}\times TF^{}.\end{array}$$
Suppose $`\delta =[\underset{¯}{}]\widehat{}\omega `$, where $`[\underset{¯}{}]K_{𝒢\times }^0(P^{})`$, $`\underset{¯}{}`$ is the one-dimensional trivial bundle and $`\omega K_{𝒢\times }^0(TF^{})`$. Then
$`\epsilon (\tau \delta )`$ $`=`$ $`j_!(\tau )\left\{\pi ^{}(j_2)_{}(\mathrm{\Phi }_X^{}^1)^{}[\underset{¯}{}]\widehat{}k_!(\omega )\right\}=`$
$`=`$ $`j_!(\tau )\left\{[\underset{¯}{}]\widehat{}k_!(\omega )\right\}.`$
Since the map $`K_{𝒢\times }^0(TF)K_{𝒢\times }^0(P\times TF)`$ (as well as the lower line in (25)) has the form $`\omega [\underset{¯}{}]\widehat{}\omega ,`$ we have proved the commutativity of 1 in (25). ∎
From this theorem we obtain the following corollary.
###### Corollary 6.5.
Let $`M`$ be a compact smooth $`H`$-manifold, let $`=B\times H`$, and let $`P`$ be a principal longitudinally smooth $``$-bundle over $`X`$ carrying also an action of $`𝒢`$ commuting with the action of $``$. Also, let $`XB`$ be a compact longitudinally smooth $`𝒢`$-bundle. Let $`Y:=P\times _HMX`$ be associated longitudinally smooth $`𝒢`$-bundle. Taking $`F=B\times M`$, we define $`T_MY:=T_FY`$. Then $`T_MY`$ is a $`𝒢`$-invariant real subbundle of $`T_{vert}Y`$ and $`T_MY=P\times _HTM`$. Let $`j:X^{}X`$ be a fiberwise $`𝒢`$–equivariant embedding and let $`k:M^{}M`$ be an $`H`$-embedding. Denote by $`\pi ^{}:P^{}X^{}`$ the principal $``$-bundle induced by $`j`$ on $`X^{}`$ and assume that $`Y^{}:=P^{}\times _HM^{}`$. The embeddings $`j`$ and $`k`$ induce $`𝒢`$-embedding $`jk:Y^{}Y`$. Then the diagram
is commutative.
|
warning/0506/math0506542.html
|
ar5iv
|
text
|
# Untitled Document
Integrability of graph combinatorics
via random walks and heaps of dimers
P. Di Francesco and E. Guitter
Service de Physique Théorique, CEA/DSM/SPhT
Unité de recherche associée au CNRS
CEA/Saclay
91191 Gif sur Yvette Cedex, France
philippe@spht.saclay.cea.fr
guitter@spht.saclay.cea.fr
Abstract
We investigate the integrability of the discrete non-linear equation governing the dependence on geodesic distance of planar graphs with inner vertices of even valences. This equation follows from a bijection between graphs and blossom trees and is expressed in terms of generating functions for random walks. We construct explicitly an infinite set of conserved quantities for this equation, also involving suitable combinations of random walk generating functions. The proof of their conservation, i.e. their eventual independence on the geodesic distance, relies on the connection between random walks and heaps of dimers. The values of the conserved quantities are identified with generating functions for graphs with fixed numbers of external legs. Alternative equivalent choices for the set of conserved quantities are also discussed and some applications are presented.
06/05
1. Introduction
Graph combinatorics seems to provide a remarkable source for (discrete) integrable systems. So far, the underlying integrability seemed to be intimately related to the existence of a matrix model formulation for the counting of these graphs . In the context of matrix model solutions to 2D Quantum Gravity , the generating functions for possibly decorated graphs of arbitrary genus may indeed be interpreted as tau-functions of various integrable hierarchies .
More recently, an alternative purely combinatorial and bijective approach to the enumeration of planar graphs was developed, based on the transformation of graphs into decorated trees \[5-7\]. In this context, another remarkable, apparently unrelated integrable structure was observed, now involving the geodesic distance on the graphs .
More precisely, in Ref., special attention was paid to the case of planar graphs with vertices of even valence only, and to their generating function $`R_n`$ with two distinguished points at geodesic distance at most $`n`$, with $`n`$ a non-negative integer. The latter was shown to obey a self-consistent master equation turning into a recursion relation on $`n`$ for graphs with bounded valences. This non-linear recursion relation turned out to be exactly solvable in terms of discrete soliton-like tau-functions. The precise solution involved integration constants which may be rephrased into conserved quantities, thus displaying explicitly the integrability of the master equation. The knowledge of these conserved quantities provides a powerful tool for generating explicit expressions for various generating functions of interest in the graph combinatorial framework. In particular, exploiting the conserved quantities allows to bypass the quite tedious use of the exact solutions of Ref..
The aim of this paper is to construct ab initio a set of conserved quantities for the master equation in a purely combinatorial setting, and to interpret them in the language of graphs. As will be recalled below, the master equation has a compact expression in terms of discrete random walks. Our construction relies therefore on properties of partition functions for such random walks, as well as for “hard dimers” on a line, both with suitable weights involving the $`R_n`$’s. Both objects are known to be related via some boson/fermion type inversion relation. The origin of this relation is best understood by relating random walks to so-called “heaps of dimers”, a boson-like counterpart of the mutually excluding, thus fermion-like, hard dimers.
The paper is organized as follows. In Section 2, we first recall the existing bijection between on the one hand planar graphs with two external legs and inner vertices of even valence, and on the other hand planted trees with two kinds of leaves, the so-called blossom trees (Sect.2.1). This bijection is used to re-derive the master equation for $`R_n`$ in Sect.2.2. Other possible applications of the master equation are discussed in Sect.2.3, as well as its integrable nature in Sect.2.4. Section 3 is devoted to the study of generating functions for random walks, heaps of dimers, and one-dimensional hard dimer configurations and to their relationships. In Section 3.1, we first present a useful fundamental relation for the generating function of random walks, which is then identified with that of a particular class of heaps of dimers called pyramids. Details of this identification are gathered in Appendix A. This identification allows to derive a crucial inversion relation involving the partition function for hard dimers on a segment. In Section 4, we first give explicit formulas for the conserved quantities $`\mathrm{\Gamma }_{2i}(n)`$, $`i=1,2,\mathrm{}`$ in terms of partition functions for random walks (Sect.4.1). The actual conservation of these quantities, i.e. their independence of $`n`$ whenever the master equation is satisfied, is proved in Sect.4.2 by extensive use of the fundamental and inversion formulas of Section 3. Section 5 concerns the graph interpretation of the conserved quantities $`\mathrm{\Gamma }_{2i}(n)`$ in terms of $`2i`$-point functions, i.e. generating functions for graphs with $`2i`$ external legs (Sect.5.1). In the case of graphs with bounded valences, say up to $`2m`$, the infinite set of conserved quantities is reducible to the finite set of the $`m1`$ first ones. The corresponding (linear) relations between their constant values are derived in Sect.5.2. Finally, Sect.5.3 presents an alternative bijective proof for the conservation of $`\mathrm{\Gamma }_2(n)`$, interpreted as the generating function for graphs with the two legs in the same face. Possible extensions of this bijective proof to higher order conserved quantities are also discussed, in view of particularly suggestive identities equivalent to the conservation of $`\mathrm{\Gamma }_{2i}`$ for $`i=2,3,4`$. In Section 6, we first present two equivalent alternative sets of conserved quantities, one respecting the “time-reversal” ($`nn`$) symmetry of the master equation (Sect.6.1) and the other compacted, i.e. involving the least possible number of $`R_n`$’s (Sect.6.2). Proofs of their conservation rely on generalized inversion formulas, all displayed in Appendix B. To illustrate the interest of using conserved quantities, we give in Sect.6.3 an example of application to the statistics of neighbors of the external face in pure tetravalent and pure hexavalent graphs. We gather a few concluding remarks in Section 7.
2. Generating functions for planar graphs: master equation
2.1. From planar graphs to blossom trees
Fig. 1: A two leg-diagram with distinguished incoming and outcoming legs (a). The incoming leg is adjacent to the external face. All inner vertices have even valence. Here the geodesic distance between the legs is $`1`$. This diagram is transformed into a blossom tree (e) by cutting edges into bud-leaf pairs as explained in the text. Buds (resp. leaves) are represented by black (resp. white) arrows. Here the cutting procedure requires performing two turns around the graph: (b)$``$(c) and (c)$``$(d). The outcoming leg serves as root for the blossom tree while the incoming one is replaced by a leaf. In the blossom tree, each inner vertex of valence $`2k`$ is adjacent to exactly $`k1`$ buds.
Let us first recall the bijection between planar graphs and blossom trees. More precisely, by planar graphs, we mean here graphs with planar topology, with inner vertices of even valences only and with two univalent endpoints (see Fig.1-(a) for an example). We shall refer to the endpoints as legs and to the graphs as two-leg diagrams. The two legs are distinguished into an incoming and an outcoming leg and need not be adjacent to the same face. We call geodesic distance between the two legs the minimal number of edges to be crossed in a path joining both legs in the plane. In the planar representation, we choose to have the incoming leg adjacent to the external face.
On the other hand, by blossom trees, we mean planted plane trees satisfying
(B1) all the inner vertices of the tree have even valences;
(B2) the endpoints of the tree are of two types: buds and leaves;
(B3) each $`2k`$-valent inner vertex is adjacent to exactly $`k1`$ buds;
(see Fig.1-(e) for an example). In this characterization, it is assumed that the tree has at least one inner vertex but, by convention, the tree reduced to a single leaf attached to the root is also considered as a blossom tree. Note that from property (B3), we deduce that the total number of leaves in a blossom tree is equal to the total number of buds plus one.
The bijection may be obtained by cutting each two-leg diagram into a blossom tree as follows. We first visit successively each edge bordering the external face in counterclockwise direction around the diagram (see Fig.1-(b)), starting from the incoming leg. Any visited edge is cut into a bud-leaf pair if and only if this operation does not disconnect the diagram. After one turn, the net result has been to merge a number of internal faces with the external one. We repeat the process on this newly obtained diagram (see Fig.1-(c)) until all faces are merged, resulting into a tree (see Fig.1-(d)). We finally replace the incoming leg by a leaf and plant the tree at the outcoming one (see Fig.1-(e)). As shown in Ref., the resulting tree is a blossom tree and the above construction establishes a bijection between, on the one hand two-leg diagrams and, on the other hand blossom trees. Note that the inner vertices of the obtained tree are clearly in one-to-one correspondence with those of the corresponding diagram and that their valence is preserved. This allows to keep track of vertex valences in both pictures.
Fig. 2: The closing procedure from a blossom tree (a) to a two-leg diagram (d). Each pair of a bud immediately followed counterclockwise by a leaf is glued into an edge (b). This is iterated (c) until all leaves but one are glued. This remaining leaf is replaced by an incoming leg while the root becomes an outcoming one. The closing procedure here encircles the root once, hence the geodesic distance between the legs is $`1`$.
Fig. 3: The contour and surrounding walks of the blossom tree of Fig.1(e), or equivalently of Fig.2(a). The contour walk, represented in solid line, is here to be read from right to left and records the succession of buds and leaves clockwise around the tree from the root. Starting from height $`0`$, it makes an ascending step for each encountered leaf and a descending one for each bud, hence ends at height $`1`$. Its depth is here $`1`$, equal to the absolute value of the minimum visited height. The surrounding walk, represented in dotted line, is to be read from left to right and records to sequence of buds and blossom subtrees counterclockwise around the root vertex from the root. To check whether the tree at hand contributes to $`R_n`$, we let the surrounding walk start from height $`n`$ and make an ascending step for each encountered bud and a descending one for each blossom subtree. Each descending step may be viewed as the net contribution of the subtree within the contour walk. Here each blob corresponds to such a subtree. The depth of the contour walk will remain less or equal to $`n`$ if the depth of each (vertically shifted) contour walk in a blob starting at height $`i`$ remains less or equal to $`i`$.
The inverse of the cutting procedure consists in reading the sequence of buds and leaves in counterclockwise order around the tree (see Fig.2-(a)) and connecting each pair of a bud immediately followed by a leaf into an edge (see Fig.2-(b)) The process is repeated until only one leaf is left (see Fig.2-(c)), which is chosen as the incoming leg while the root becomes the outcoming one (see Fig.2-(d)). Note that this root is usually encircled by a number of edges separating it from the outer face. This number is nothing but the geodesic distance between the legs. A simple way to measure it is to forbid to encircle the root in the closing process. The geodesic distance is then given by the number of excess buds, i.e. those buds which remain unmatched. Alternatively, we may define the contour walk of a blossom tree by reading the sequence of buds and leaves clockwise around the tree starting from the root and performing a down (resp. up) step for each encountered bud (resp. leaf) (see Fig.3). Starting from height $`0`$ and making steps of $`\pm 1`$, the contour walk ends at height $`1`$ and the geodesic distance between the legs of the corresponding diagram is given by the depth of the walk, defined as the absolute value of the minimum height reached by the walk.
2.2. Master equation
We may now define the generating function $`R_n`$ for two-leg diagrams with weight $`g_k`$ per $`2k`$-valent vertex and with legs at a geodesic distance less or equal to $`n`$. From the above bijection, $`R_n`$ is also the generating function for blossom trees with weight $`g_k`$ per $`2k`$-valent inner vertex and whose contour walk has depth less or equal to $`n`$ or equivalently with at most $`n`$ excess buds in the closing process that forbids encircling the root.
It is now easy to write down a master equation for $`R_n`$ by decomposing each blossom tree according to the environment of the inner vertex attached to the root (root vertex). This environment may be decomposed into a sequence of buds and descendent subtrees. All these subtrees are themselves blossom trees either made of a single leaf or satisfying the above characterizations (B1-B3). Note that the contour walk of the original tree is the concatenation of down steps for the buds and of the (vertically shifted) contour walks of the blossom subtrees. Hence, the constraint of depth less than $`n`$ translates into constraints on the depth reached by the contour walk of each blossom subtree. This turns into a closed relation between the $`R_n`$’s.
More precisely, the environment of the root vertex is best described by yet another walk, the surrounding walk, now obtained by reading the sequence of buds and subtrees in the opposite, counterclockwise direction around the root vertex. Starting from height $`n`$ at the root, we make a step $`+1`$ for each encountered bud and $`1`$ for each encountered blossom subtree (see Fig.3). Each $`1`$ step may be seen as the net contribution of the blossom subtree at hand within the total (reversed) contour walk. To ensure that the depth of the entire contour walk remains less or equal to $`n`$, we must require that each subtree encountered at height $`i`$ in the surrounding walk has a contour walk with depth less or equal to $`i`$. This leads us to attach a weight $`R_i`$ to each $`1`$ step from height $`i`$ to height $`(i1)`$.
The description of the root vertex environment by its surrounding walk and the attached weights are efficiently encoded into a “transfer matrix” $`Q`$ acting on a formal orthonormal basis $`|i`$ ($`i\text{ZZ}`$) as
$$Q|i=|i+1+R_i|i1$$
The generating function $`Z_{a,b}(k)`$ for walks of $`k`$ steps going from, say height $`a`$ to height $`b`$, and with weight $`R_i`$ for each descent $`i(i1)`$ is easily expressed in terms of $`Q`$ as
$$Z_{a,b}(k)b|Q^k|a$$
Note that $`Z_{a,b}(k)`$ is non zero only for $`(ba)=k\mathrm{mod}2`$. From the above analysis, the quantity $`g_kZ_{n,n1}(2k1)`$ is the generating function for blossom trees with a $`2k`$-valent root vertex and with contour walk of depth less or equal to $`n`$. To incorporate arbitrary (even) valences, we introduce the quantities
$$V^{}(Q)=\underset{k1}{}g_kQ^{2k1}$$
and
$$V_{a,b}^{}b|V^{}(Q)|a=\underset{k1}{}g_kZ_{a,b}(2k1)$$
(which is non zero only for $`(ba)`$ odd). The quantity $`V_{a,b}^{}`$ may be viewed as a “grand-canonical” generating function for walks from $`a`$ to $`b`$ and with arbitrary odd length.
With these notations, we may finally write the master equation
$`R_n=1+V_{n,n1}^{}`$
where the first contribution ($`1`$) corresponds to having a single leaf connected to the root. This relation is valid for all $`n0`$ with the convention that $`R_i=0`$ for all $`i<0`$. This equation determines all the $`R_n`$’s as formal power series of the $`g_k`$’s upon imposing that $`R_n=1+𝒪(\{g_k\})`$ for all $`n0`$.
Fig. 4: A schematic representation of the truncated master equation (2.6). The generating function $`R_n`$ is decomposed according to the environment of the root vertex, with $`0`$, $`1`$, $`2`$ buds and $`1`$, $`2`$, $`3`$ blossom subtrees according to its valence $`2`$, $`4`$, $`6`$ respectively. To ensure that the depth of the contour walk does not exceed $`n`$, we have to impose bounds on the maximal number of excess buds in each subtree. We have represented (here for $`n=3`$) these excess buds in maximal number for each subtree. The index of $`R`$ for each subtree is nothing but this maximal number.
For illustration, let us consider the simple case of diagram with only bi-, tetra- and hexavalent inner vertices, i.e. with $`g_k=0`$ for all $`k4`$. The master equation then reads
$$\begin{array}{cc}\hfill R_n& =1+g_1R_n+g_2R_n(R_{n1}+R_n+R_{n+1})\hfill \\ & +g_3R_n(R_{n2}R_{n1}+R_{n1}^2+R_n^2+2R_n(R_{n1}+R_{n+1})\hfill \\ & +R_{n1}R_{n+1}+R_{n+1}^2+R_{n+1}R_{n+2})\hfill \end{array}$$
This relation is illustrated in Fig.4. In the following, we will also make use of the generating function $`R`$ for two-leg diagrams without restriction on the geodesic distance. This may be recovered as the limit $`n\mathrm{}`$ of $`R_n`$, hence $`R`$ satisfies the relation
$$R=1+\underset{k1}{}g_k\left(\genfrac{}{}{0pt}{}{2k1}{k}\right)R^k$$
obtained by counting walks of length $`2k1`$ with global height difference equal to $`1`$, thus having $`k`$ descending steps each weighted by $`R`$. In the truncated case above, Eq.(2.1) reduces to
$$R=1+g_1R+3g_2R^2+10g_3R^3$$
2.3. Other applications of the master equation
The master equation may be rewritten as
$$R_n=\frac{1}{1V_{n1,n}^{}}$$
by noting that $`V_{n,n1}^{}=R_nV_{n1,n}^{}`$. The latter is a direct consequence of the general reflection relation
$$Z_{a,a1}(2k1)=R_aZ_{a1,a}(2k1)$$
obtained by first transferring the weights $`R_i`$ to the ascending steps $`(i1)i`$ and then reflecting the walks, thus interpreting them as walks from $`(a1)a`$. Note that there are as many descending steps $`i(i1)`$ as ascending steps $`(i1)i`$ except for $`i=a`$ where there is one more descent (before reflection), hence the extra factor $`R_a`$ needed to reproduce the correct weight.
Fig. 5: An example of labeled mobile (left) made of (inflated) polygons attached by their vertices into a tree-like structure. The vertex labels around each polygon obey the rule displayed on the right that they either decrease by $`1`$ or increase weakly clockwise.
In its alternative form (2.1), the master equation appeared in a slightly different enumeration problem, that of so-called labeled mobiles , generalizing the so-called well labeled trees of Ref.. These mobiles are made of rigid polygons, with weight $`g_k`$ per $`k`$-gon, glued by their vertices into a rooted tree-like object (see Fig.5 for an example). The labels are subject to local rules around each $`k`$-gon: going clockwise around a $`k`$-gon, the labels must either decrease by $`1`$ or increase weakly. This rule again is efficiently described by the transfer matrix $`Q`$ of Eq.(2.1) by reading the labels around $`k`$-gons from the vertex at which they are suspended and translating the sequence of labels into heights forming a closed walk of $`k`$ steps around the $`k`$-gon. This walk is then transformed into a closed boundary walk of length $`2k`$ by transforming each non-negative step $`(+p)`$ into a descent followed by $`p+1`$ unit ascending steps. Denoting by $`R_n`$ the generating function for planted mobiles with root labeled $`n`$ and inspecting the possible environments of the root, we have the relation
$$R_n=\frac{1}{1\underset{k1}{}g_kM_n(k)}$$
where $`M_n(k)`$ is the generating function for $`k`$-gons rooted at a vertex labelled $`n`$ and with $`k1`$ sub-mobiles dangling at the other vertices. In $`M_n(k)`$, we have to assign a weight $`R_i`$ to each vertex of the $`k`$-gon labelled $`i`$, except for the root vertex. This amounts equivalently to assign a weight $`R_i`$ to each $`1`$ step from height $`i`$ in the boundary walk, except for the $`1`$ step from the root. This leads to
$$M_n(k)=Z_{n1,n}(2k1)=n|Q^{2k1}|n1>$$
Eqs.(2.1) and (2.1) boil down to Eq.(2.1). Again this equation must be supplemented with some initial conditions, for instance that $`R_i=0`$ for $`i<0`$ if we demand that the labels remain non-negative.
Returning to graph enumeration problems, it was shown that mobiles with non-negative labels are in one to one correspondence with planar graphs with faces of even valences only and with a distinguished origin vertex . The vertices of the mobile are in correspondence with vertices of the associated graph, and the labels represent the geodesic distance (on the associated graph) of these vertices from the origin vertex. In this framework, $`R_n`$ is now interpreted as the generating function for the graphs with a distinguished edge at geodesic distance less than $`n`$ from the origin vertex.
Finally, a third domain of application of the master equation comes from the study of spatially extended branching processes, i.e. probabilistic models for the evolution and spreading of a population. In this language, the index $`n`$ stands for the (discrete one-dimensional) position of individuals and the blossom trees or mobiles are interpreted as the genealogical structure of families. As shown in Ref., the generating function $`R_n`$ is then related to the probability for a population with germ at position $`n`$ never to spread up to position $`0`$.
2.4. Integrability of the master equation
A remarkable feature of the master equation is that it may be solved exactly. More precisely, we will consider truncated versions of Eq.(2.1) by considering only diagrams with valences up to, say $`2m`$. This amounts to set $`g_k=0`$ for $`k>m`$ in which case $`V^{}(Q)`$ is an odd polynomial of degree $`2m1`$. As explained in Ref., the general solution of Eq.(2.1), now with arbitrary initial conditions but with the requirement that it converges at $`n\mathrm{}`$ takes the surprisingly simple form
$$R_n=R\frac{u_{n+1}u_{n+4}}{u_{n+2}u_{n+3}}$$
where $`u_n`$ has a multi-soliton structure involving $`m1`$ integration constants $`\lambda _1,\mathrm{},\lambda _{m1}`$. Those constants may be fixed by the initial conditions. This structure is characteristic of (discrete) integrable systems and implies the existence of conserved quantities. More precisely, by inverting the system $`R_i=R_i(\{\lambda \})`$ for any set $`i=n,n+1,\mathrm{},n+m2`$, one may in principle construct $`m1`$ conserved quantities for Eq.(2.1) in the form $`\lambda _j=\mathrm{\Lambda }_j(R_n.R_{n+1},\mathrm{},R_{n+m2})=`$ const. independently of $`n`$. In practice, this construction is however difficult to implement. The remainder of the paper is devoted to the explicit construction of conserved quantities ab initio, i.e. without reference to the above solution (2.1). This construction will be carried out by use of simple combinatorial tools only.
3. Random walks and heaps of dimers
3.1. From random walks to heaps of dimers
In this Section, we shall investigate a number of properties of the partition function of random walks $`Z_{a,b}(k)`$ above. In particular, we will emphasize the connection between random walks and so-called heaps of dimers, leading us eventually to an inversion relation involving partition functions for hard dimers on a segment. All the present relations will be instrumental for proving the conservation statements of Section 4 below.
Fig. 6: A schematic representation of the fundamental identity (3.1) for walks. The walks are decomposed according to either their first step (first line) or their last one (second line).
A first fundamental identity is obtained by expressing the generating function for paths of length $`k+1`$ in two different ways: (i) as the concatenation of a path of length $`1`$ (first step) and a path of length $`k`$, or (ii) as the concatenation of a path of length $`k`$ and a path of length $`1`$ (last step). According to whether the first (resp. the last) step is up or down, we have respectively
$$Z_{a,b}(k+1)=Z_{a+1,b}(k)+R_aZ_{a1,b}(k)=Z_{a,b1}(k)+R_{b+1}Z_{a,b+1}(k)$$
as illustrated in Fig.6.
Fig. 7: An example of a heap of dimers (a) and its canonical representation (b). The thickened dimers correspond to the right projection of the heap, forming in (b) a hard dimer configuration on the vertical line $`x=0`$. By rotating the picture (b) and extending each dimer into a $`2\times 1`$ rectangle, we obtain a piling up of these rectangles (c) whose bottom row is formed by the former right projection.
A second important property is that we may interpret $`Z_{a,b}(k)`$ as a generating function for heaps of dimers, a particular case of the so-called heaps of pieces introduced in Ref.. These heaps of dimers also occur in relation to lattice animals and so-called Lorentzian gravity and may be defined as follows. Let us decompose the plane into parallel (say horizontal) stripes of width $`1`$ by drawing horizontal lines at integer vertical positions. The stripes are labeled by the position of their top boundary. We then place within the stripes a number of dimers, i.e. vertical segments of length $`1`$ (see Fig.7-(a)). All dimers may freely slide within their stripe provided they do not cross a dimer within the same stripe or within the stripe immediately above and below. A heap of dimers is defined modulo this sliding freedom and only records the relative positions of the dimers. In a canonical representation, we may place all the dimers at integer horizontal coordinates within their respective stripes with the requirement that these coordinates be negative or zero and maximal (see Fig.7-(b)). This is realized by pushing all dimers as much to the right as possible while staying at integer horizontal positions in the left half plane. This in turn allows upon rotating the picture by $`90^{}`$ clockwise to view a configuration as a piling up of rectangles of size $`2\times 1`$ (see Fig.7-(c)) deposited on the $`x=0`$ line. For each heap of dimers, we may define its right projection as the configuration made of those dimers which may slide freely all the way to infinity to the right. In the canonical representation, these are those dimers with horizontal position $`0`$. Clearly, the right projection of any heap of dimers is a configuration of hard dimers on a (vertical) line, i.e. a set of dimers occupying (vertical) segments $`[i1,i]`$ with the restriction that no two dimers may come in contact (the occupied segments must be disjoint).
Fig. 8: An example of a pyramid (a) and a half-pyramid (b), both with base $`[a,b]`$.
For $`(ba)`$ odd and positive, we may consider the maximally occupied hard dimer configuration of the segment $`[a,b]`$. It is made of $`(ba+1)/2`$ dimers occupying the elementary segments $`[a+2i,a+2i+1]`$, $`i=0,\mathrm{},(ba1)/2`$. Any heap of dimer having this maximally occupied segment as right projection will be called a pyramid of base $`[a,b]`$ (see Fig.8-(a) for an example). We then define the generating function $`H_{a,b}(k)`$ for pyramids of base $`[a,b]`$ made of a total number $`k`$ of dimers, with weight $`R_i`$ per dimer in the stripe labeled $`i`$, except for the dimers of the right projection. Note that $`H_{a,b}(k)`$ is non zero only if $`k(ba+1)/2`$ and that $`H_{a,b}((ba+1)/2)=1`$.
Fig. 9: A schematic representation (first line) of the equality (3.2) between generating functions $`Z_{a,b}(2k1)`$ for walks from $`a`$ to $`b`$ with $`2k1`$ steps and $`H_{a,b}(k)`$ for pyramids of base $`[a,b]`$ with a total of $`k`$ dimers. The similar representation (second line) for the relation (3.3) between the generating function $`Z_{a,b}^+(2k1)`$ of positive walks and that $`H_{a,b}^+((k)`$ of half-pyramids.
We these definitions, we have the remarkable relation
$$Z_{a,b}(2k1)=H_{a,b}(k)$$
valid for $`(ba)`$ odd and positive, and $`k1`$. This relation is illustrated in Fig.9-(a). It follows from a general bijection between walks and heaps of pieces given in Ref.. This bijection is recalled in detail in Appendix A below for the particular case at hand.
In the next section, we shall also make use of the generating function $`Z_{a,b}^+(k)`$ for walks of $`k`$ steps going from height $`a`$ to height $`b`$, with weight $`R_i`$ per descent $`i(i1)`$, and which stay at or above height $`a`$. In the following, we shall refer to these walks as positive walks. The generating function $`Z_{a,b}^+(k)`$ may be obtained from $`Z_{a,b}(k)`$ by simply taking $`R_i0`$ for all labels $`ia`$. In particular, we deduce that, for $`(ba)`$ odd and positive and $`k1`$
$$Z_{a,b}^+(2k1)=H_{a,b}^+(k)$$
where $`H_{a,b}^+(k)`$ denotes the generating function for half-pyramids of base $`[a,b]`$ with a total of $`k`$ dimers, i.e. pyramids with dimers only in stripes with labels $`i>a`$ (see Fig.8-(b) for an example). This relation (3.1) is illustrated in Fig.9-(b).
3.2. Inversion relation
Viewing hard dimers on a line as mutually excluding and fermionic objects (with at most one dimer in a unit segment), heaps of dimers appear as their interacting bosonic counterpart in which arbitrarily many dimers may be piled up in each stripe. It is therefore natural to expect some boson/fermion inversion relations to hold between their respective generating functions. An example of such inversion relation in a “grand-canonical” ensemble, i.e. for arbitrarily many dimers, may be found in Refs.. In this Section, we shall derive another inversion formula, now in the canonical ensemble, i.e. with fixed numbers of dimers.
Introducing the generating function $`\mathrm{\Pi }_{a,b}(k)`$ for configurations of $`k`$ hard dimers in the segment $`[a,b]`$ and with weight $`R_i`$ per dimer in the segment $`[i1,i]`$, we have the remarkable inversion relation
$$\underset{\mathrm{}=0}{\overset{k}{}}(1)^k\mathrm{}\mathrm{\Pi }_{a,a+2k1}(k\mathrm{})Z_{a,a+2j}^+(2\mathrm{})=\delta _{k,j}$$
for $`k,j0`$, with the convention that $`\mathrm{\Pi }_{a,a1}(0)=1`$. This relation may be written in a more compact matrix form by defining a lower-triangular, semi-infinite matrix $`𝐙(a)`$ with entries
$$𝐙(a)_{i,j}Z_{a,a+2j}^+(2i)$$
for $`0ji`$, and a lower-triangular, semi-infinite matrix $`𝐃(a)`$ with entries
$$𝐃(a)_{k,i}=(1)^{ki}\mathrm{\Pi }_{a,a+2k1}(ki)$$
for $`0ik`$ while all other entries vanish. The relation (3.1) now reads
$$𝐃(a)𝐙(a)=𝐈$$
with $`𝐈`$ the (semi-infinite) identity matrix. To prove the relation (3.1), let us consider pairs $`𝒫`$ made of
(i) a hard dimer configuration in the segment $`[a,a+2k1]`$
(ii) a half-pyramid with projection $`[a,a+2j1]`$
with a total number of $`k`$ dimers. In the half-pyramid, each dimer in the stripe $`i`$ (including the dimers of the right projection) receives a weight $`R_i`$. On the contrary, in the hard dimer configuration, each dimer in the segment $`[i1,i]`$ receives a weight $`R_i`$. Denoting by $`\mathrm{}`$ the number of dimers in the half-pyramid, the generating function $`P_a(j,k)`$ for the above pairs reads
Fig. 10: An example (left) of pair $`𝒫`$ of a hard-dimer configuration in the segment $`[a,a+2k1]`$ and a half-pyramid of base $`[a,a+2j1]`$ (with $`jk`$), with a total number $`k`$ of dimers (here $`k=7`$). These are concatenated (right) into a larger heap $``$. The equivalence class $`𝒞()`$ of those pairs leading to the same heap $``$ is constructed by considering in $``$ the dimers which belong to the left projection of the heap but do not belong to the right projection. In the present example, there is exactly one such (encircled) dimer. The equivalence class $`𝒞()`$ is generated by distributing in all possible ways these dimers either in the hard-dimer configuration or in the half-pyramid. If we assign opposite weights to dimers in the hard-dimer configuration and in the half-pyramid, this results in a vanishing net contribution of the class $`𝒞()`$ unless the left and right projections of $``$ are identical. This happens only when $``$ is the maximally occupied hard-dimer configuration made of $`k`$ dimers in the segment $`[a,a+2k1]`$.
$$P_a(j,k)=\mathrm{\Pi }_{a,a+2j1}(j)\underset{\mathrm{}=0}{\overset{k}{}}(1)^k\mathrm{}\mathrm{\Pi }_{a,a+2k1}(k\mathrm{})H_{a,a+2j1}^+(\mathrm{})$$
Note that $`P_a(j,k)`$ is non zero only if $`kj`$. For each such pair $`𝒫`$, we may absorb all dimers of the hard dimer configuration into a larger heap $`(𝒫)`$ of $`k`$ dimers by simply adding them to the left of the original half-pyramid (see Fig.10). We can then regroup all pairs leading to the same larger heap $``$ into an equivalence class $`𝒞()`$. Conversely, given $`=(𝒫)`$, we may reconstruct all pairs in the class $`𝒞()`$ by considering the set of dimers which belong to the left projection of $``$ (those dimers which may be pushed all the way to infinity to the left) but which do not belong to the right projection. If this set contains at least one dimer, this dimer may be incorporated either in the hard dimer configuration of the pair or in its half-pyramid, contributing with opposite weights. This causes the contribution to $`P_a(j,k)`$ of the class $`𝒞()`$ to vanish. The only case where no such vanishing takes place is when the left projection of $``$ is identical to its right projection, in which case $``$ itself is a hard dimer configuration with $`k`$ dimers on the segment $`[a,a+2k1]`$, therefore the unique maximally occupied hard dimer configuration, with contribution $`(1)^{kj}\mathrm{\Pi }_{a,a+2k1}(k)`$ as the $`j`$ lower dimers with label less than $`a+2j1`$ belong to the half-pyramid and the $`(kj)`$ upper ones belong to the hard dimer configuration. We immediately deduce that
$$P_a(j,k)=(1)^{kj}\mathrm{\Pi }_{a,a+2k1}(k)\delta _{kj}$$
with $`\delta _{kj}=1`$ for $`kj`$ and zero otherwise. Using $`\delta _{k,j}=\delta _{kj}\delta _{kj+1}`$, we deduce
$$\begin{array}{cc}\hfill \delta _{k,j}\mathrm{\Pi }_{a,a+2k1}(k)& =P_a(j,k)+P_a(j+1,k)\hfill \\ & =\underset{\mathrm{}=0}{\overset{k}{}}(1)^k\mathrm{}\mathrm{\Pi }_{a,a+2k1}(k\mathrm{})\hfill \\ & \times \left(H_{a,a+2j1}^+(\mathrm{})\mathrm{\Pi }_{a,a+2j1}(j)+H_{a,a+2j+1}^+(\mathrm{})\mathrm{\Pi }_{a,a+2j+1}(j+1)\right)\hfill \end{array}$$
Using $`H_{a,a+2j1}^+(\mathrm{})=Z_{a,a+2j1}^+(2\mathrm{}1)`$, $`H_{a,a+2j+1}^+(\mathrm{})=Z_{a,a+2j+1}^+(2\mathrm{}1)`$ from Eq.(3.1) and the relation $`\mathrm{\Pi }_{a,a+2j+1}(j+1)=R_{a+2j+1}\mathrm{\Pi }_{a,a+2j1}(j)`$ for maximally occupied segments, we obtain
$$\underset{\mathrm{}=0}{\overset{k}{}}(1)^k\mathrm{}\mathrm{\Pi }_{a,a+2k1}(k\mathrm{})\left(Z_{a,a+2j1}^+(2\mathrm{}1)+R_{a+2j+1}Z_{a,a+2j+1}^+(2\mathrm{}1)\right)=\delta _{k,j}$$
which, upon using
$$Z_{a,a+2j1}^+(2\mathrm{}1)+R_{a+2j+1}Z_{a,a+2j+1}^+(2\mathrm{}1)=Z_{a,a+2j}^+(2\mathrm{})$$
from Eq.(3.1), reduces to the desired inversion relation (3.1).
Note that, as $`𝐃(a)`$ and $`𝐙(a)`$ are lower-triangular, we may decide to truncate them to $`m\times m`$ matrices $`𝐃_m(a)`$ and $`𝐙_m(a)`$ with indices $`i`$ strictly less than $`m`$. These matrices clearly satisfy the inversion relation $`𝐃_m(a)𝐙_m(a)=I_m`$, with $`I_m`$ the $`m\times m`$ identity matrix. For illustration, taking $`m=3`$, we have
$$\begin{array}{cc}\hfill 𝐃_3(a)& =\left(\begin{array}{ccc}1& 0& 0\\ R_{a+1}& 1& 0\\ R_{a+1}R_{a+3}& (R_{a+1}+R_{a+2}+R_{a+3})& 1\end{array}\right)\hfill \\ \hfill 𝐙_3(a)& =\left(\begin{array}{ccc}1& 0& 0\\ R_{a+1}& 1& 0\\ R_{a+1}(R_{a+1}+R_{a+2})& (R_{a+1}+R_{a+2}+R_{a+3})& 1\end{array}\right)\hfill \end{array}$$
which are inverse of one-another.
4. Conserved quantities
4.1. Definition of the conserved quantities
We now present compact expressions for a particular “basis” of conserved quantities of the master equation (2.1). More precisely, we shall define below a set of quantities $`\{\mathrm{\Gamma }_{2i}(n)\}`$ for $`i1`$ and $`n0`$ satisfying $`\mathrm{\Gamma }_{2i}(n)`$= const. independently of $`n`$. Of course, any functions of the $`\mathrm{\Gamma }_{2i}`$’s are conserved as well and the precise choice of basis below has been done in regard of its combinatorial nature. Other, alternative choices will be discussed in Sects.6.1 and 6.2 below.
Using the generating function $`Z_{a,b}^+(k)`$ of previous section for positive walks and the grand-canonical generating function $`V_{a,b}^{}`$ for walks of odd length, we define the quantity $`\mathrm{\Gamma }_{2i}(n)`$ by
$`\mathrm{\Gamma }_{2i}(n)=Z_{n1,n1}^+(2i)\mathrm{\Gamma }_0(n){\displaystyle \underset{j=1}{\overset{i}{}}}Z_{n1,n1+2j}^+(2i)V_{n+2j1,n2}^{}`$
for $`i1`$ and $`n0`$, with $`\mathrm{\Gamma }_0(n)`$ given by
$$\mathrm{\Gamma }_0(n)=R_{n1}V_{n1,n2}^{}+\delta _{n,0}$$
for $`n0`$. The quantity $`\mathrm{\Gamma }_0(n)`$ itself is not stricto sensu a conserved quantity. However, the master equation precisely ensures that $`\mathrm{\Gamma }_0(n)=1`$ for $`n1`$ while $`\mathrm{\Gamma }_0(0)=1`$ by definition (since $`R_1=V_{1,2}^{}=0`$), hence the conservation of $`\mathrm{\Gamma }_0`$ is a tautology. This in turn allows to simplify the expression (4.1) for $`\mathrm{\Gamma }_{2i}(n)`$ by substituting $`\mathrm{\Gamma }_0(n)=1`$. The proof of the conservation of $`\mathrm{\Gamma }_{2i}(n)`$ however is made much simpler by keeping as such the slightly more involved definition (4.1). Introducing the vectors $`\stackrel{}{\mathrm{\Gamma }}(n)`$ and $`\stackrel{}{V}^{}(n)`$ with components
$$\stackrel{}{\mathrm{\Gamma }}(n)_i=\mathrm{\Gamma }_{2i}(n)\mathrm{and}\stackrel{}{V}^{}(n)_j=(R_{n1}+\delta _{n,0})\delta _{j,0}V_{n+2j1,n2}^{}$$
for $`i,j0`$, the above relations (4.1) and (4.1) read simply
$$\stackrel{}{\mathrm{\Gamma }}(n)=𝐙(n1)\stackrel{}{V}^{}(n)$$
with $`𝐙(a)`$ defined as in (3.1). It will prove useful to invert this relation. This is readily performed by use of the inversion relation (3.1) or (3.1), with the result
$$\stackrel{}{V}^{}(n)=𝐃(n1)\stackrel{}{\mathrm{\Gamma }}(n)$$
with $`𝐃(a)`$ defined in (3.1), namely
$$V_{n+2j1,n2}^{}=\underset{i=0}{\overset{j}{}}(1)^{i1}\mathrm{\Pi }_{n1,n+2j2}(i)\mathrm{\Gamma }_{2j2i}(n)+\delta _{j,0}(R_{n1}+\delta _{n,0})$$
with again the convention that $`\mathrm{\Pi }_{n1,n2}(0)=1`$.
4.2. Proof of the conservation
We are now ready to prove that the quantities $`\mathrm{\Gamma }_{2i}(n)`$ of Eq.(4.1) are conserved quantities of the master equation (2.1). To this end, we shall now use the fundamental equation for $`V^{}`$
$$V_{a+1,b}^{}+R_aV_{a1,b}^{}=V_{a,b1}^{}+R_{b+1}V_{a,b+1}^{}$$
directly inherited from the fundamental equation (3.1) for $`Z_{a,b}(k)`$. Taking $`a=n+2j1`$ and $`b=n1`$, we get the identity
$$\begin{array}{cc}\hfill V_{(n+1)+2j1,(n+1)2}^{}+R_{n+2j1}& V_{(n+1)+2(j1)1,(n+1)2}^{}\hfill \\ & =V_{n+2j1,n2}^{}+R_nV_{(n+2)+2(j1)1,(n+2)2}^{}\hfill \end{array}$$
For $`j1`$ and $`n0`$, we may substitute Eq.(4.1) into Eq.(4.1), leading to
$$\begin{array}{cc}\hfill (\mathrm{\Gamma }_{2j}(n+1)& +\underset{i=1}{\overset{j}{}}(1)^{i1}\mathrm{\Pi }_{n,(n+1)+2j2}(i)\mathrm{\Gamma }_{2j2i}(n+1))\hfill \\ \hfill +R_{n+2j1}& \left(\underset{i=0}{\overset{j1}{}}(1)^{i1}\mathrm{\Pi }_{n,(n+1)+2(j1)2}(i)\mathrm{\Gamma }_{2(j1)2i}(n+1)+\delta _{j1,0}R_n\right)\hfill \\ \hfill =(\mathrm{\Gamma }_{2j}(n)& +\underset{i=1}{\overset{j}{}}(1)^{i1}\mathrm{\Pi }_{n1,n+2j2}(i)\mathrm{\Gamma }_{2j2i}(n))\hfill \\ \hfill +R_n& \left(\underset{i=0}{\overset{j1}{}}(1)^{i1}\mathrm{\Pi }_{n+1,(n+2)+2(j1)2}(i)\mathrm{\Gamma }_{2(j1)2i}(n+2)+\delta _{j1,0}R_{n+1}\right)\hfill \end{array}$$
where we have used $`\mathrm{\Pi }_{n1,n+2j2}(0)=\mathrm{\Pi }_{n,n+2j1}(0)=1`$ for all $`j0`$ (unique empty dimer configuration). Noting that the boundary terms on both sides cancel, we now change $`i(i1)`$ in the last sum of each side and rearrange the factors of $`\mathrm{\Gamma }`$, leading us to
$$\begin{array}{cc}\hfill \mathrm{\Gamma }_{2j}(n+1)& \mathrm{\Gamma }_{2j}(n)\hfill \\ & =\underset{i=1}{\overset{j}{}}(1)^{i1}\left\{\mathrm{\Pi }_{n,n+2j1}(i)R_{n+2j1}\mathrm{\Pi }_{n,n+2j3}(i1)\right\}\mathrm{\Gamma }_{2j2i}(n+1)\hfill \\ & \underset{i=1}{\overset{j}{}}(1)^{i1}\left\{\mathrm{\Pi }_{n1,n+2j2}(i)R_n\mathrm{\Pi }_{n+1,n+2j2}(i1)\right\}\mathrm{\Gamma }_{2j2i}(n)\hfill \\ & \underset{i=1}{\overset{j}{}}(1)^iR_n\mathrm{\Pi }_{n+1,n+2j2}(i1)\left(\mathrm{\Gamma }_{2j2i}(n+2)\mathrm{\Gamma }_{2j2i}(n)\right)\hfill \\ & =\underset{i=1}{\overset{j}{}}(1)^{i1}\mathrm{\Pi }_{n,n+2j2}(i)\left(\mathrm{\Gamma }_{2j2i}(n+1)\mathrm{\Gamma }_{2j2i}(n)\right)\hfill \\ & \underset{i=1}{\overset{j}{}}(1)^iR_n\mathrm{\Pi }_{n+1,n+2j2}(i1)\left(\mathrm{\Gamma }_{2j2i}(n+2)\mathrm{\Gamma }_{2j2i}(n)\right)\hfill \end{array}$$
Fig. 11: A schematic representation of the fundamental identity (4.12). A configuration of hard dimers on the segment $`[a,b]`$ may be viewed either as a configuration on the segment $`[a1,b]`$ where the unit segment $`[a1,a]`$ is empty (hence the subtraction) or as a configuration on the segment $`[a,b+1]`$ where the unit segment $`[b,b+1]`$ is empty.
In the last equality, we have used the property
$$\begin{array}{cc}\hfill \mathrm{\Pi }_{n,n+2j1}(i)& R_{n+2j1}\mathrm{\Pi }_{n,n+2j3}(i1)\hfill \\ & =\mathrm{\Pi }_{n1,n+2j2}(i)R_n\mathrm{\Pi }_{n+1,n+2j2}(i1)=\mathrm{\Pi }_{n,n+2j2}(i)\hfill \end{array}$$
which is a particular instance ($`a=n`$, $`b=n+2j2`$) of the fundamental identity
$$\mathrm{\Pi }_{a,b}(i)=\mathrm{\Pi }_{a1,b}(i)R_a\mathrm{\Pi }_{a+1,b}(i1)=\mathrm{\Pi }_{a,b+1}(i)R_{b+1}\mathrm{\Pi }_{a,b1}(i1)$$
satisfied by generating functions for hard dimers. This identity is illustrated in Fig.11 and may be viewed as the hard dimer counterpart of Eq.(3.1) for random walks.
From Eq.(4.1), we immediately deduce by induction on $`j`$ that
$$\mathrm{\Gamma }_{2j}(n+1)=\mathrm{\Gamma }_{2j}(n)$$
for all $`j1`$ and $`n0`$ provided that $`\mathrm{\Gamma }_0(n+1)=\mathrm{\Gamma }_0(n)`$ for all $`n0`$. As already mentioned, this last requirement is precisely guaranteed by the master equation (2.1). The $`\mathrm{\Gamma }_{2i}`$’s therefore form a set of conserved quantities for this equation.
For illustration, we list below the first two conserved quantities for the truncated case of graphs with valences up to $`2m=6`$, corresponding to the truncated master equation (2.1):
$$\begin{array}{cc}\hfill \mathrm{\Gamma }_2(n)& =R_nV_{n+1,n2}^{}\hfill \\ \hfill \mathrm{\Gamma }_4(n)& =R_n(R_n+R_{n+1})(R_n+R_{n+1}+R_{n+2})V_{n+1,n2}^{}V_{n+3,n2}^{}\hfill \end{array}$$
with
$$\begin{array}{cc}\hfill V_{n+1,n2}^{}& =R_{n+1}R_nR_{n1}\left(g_2+g_3(R_{n+2}+R_{n+1}+R_n+R_{n1}+R_{n2})\right)\hfill \\ \hfill V_{n+3,n2}^{}& =g_3R_{n+3}R_{n+2}R_{n+1}R_nR_{n1}\hfill \end{array}$$
where we explicitly substituted $`\mathrm{\Gamma }_0(n)=1`$ into Eq.(4.1).
5. Graph interpretation
5.1. Conserved quantities as multi-point correlation functions
The constant value of the conserved quantities has a nice combinatorial interpretation as a multi-point correlation function. More precisely, let us define by $`G_{2i}(\{g_k\})`$ the so-called (disconnected) $`2i`$-point function, i.e. the generating function for possibly disconnected $`2i`$-leg diagrams, namely graphs with inner vertices of even valences (weighted $`g_k`$ per $`2k`$-valent vertex) and with $`2i`$ legs, i.e. univalent vertices, adjacent to the same (external) face. These legs are distinguished and labeled $`1,2,\mathrm{},2i`$ counterclockwise. We have the identification
$$\mathrm{\Gamma }_{2i}(n)=G_{2i}$$
for all $`n0`$ and all $`i0`$ with the convention that $`G_0=1`$. To prove this, we simply evaluate $`\mathrm{\Gamma }_{2i}(n)`$ at $`n=0`$. Noting that $`V_{2j1,2}^{}=0`$ as it is proportional to $`R_1=0`$, we are left with
$$\mathrm{\Gamma }_{2i}(0)=Z_{1,1}^+(2i)=Z_{1,1}(2i)=Z_{0,1}(2i1)$$
where we first note that $`R_1=0`$ automatically selects positive walks and we then remove the first (ascending) step with weight $`1`$.
Fig. 12: Bijection between (a) two-leg diagrams with both legs in the same face and whose outcoming leg is adjacent to a $`2i`$-valent vertex and (c) $`2i`$-leg diagrams. In the intermediate step (b), we have glued the two legs into a rooted edge and cut out the $`2i`$-valent vertex. The legs are labeled counterclockwise by $`1,2,\mathrm{}2i`$ with the former incoming leg labeled $`1`$. Note that the $`2i`$-leg diagrams need not be connected in general.
As explained in Sect.2.2, the quantity $`Z_{0,1}(2i1)`$ is the generating function for blossom trees with a $`2i`$-valent root vertex and a contour walk with $`0`$ depth. Returning to the graph interpretation, it is identified via the bijection of Sect.2.1 as the generating function for two-leg diagrams with an outcoming leg attached to a $`2i`$-valent vertex and adjacent to the same (external) face as the incoming leg. The $`2i`$-valent vertex receives a weight $`1`$ as opposed to all other inner vertices weighted by the usual factors $`g_k`$ if they are $`2k`$-valent. Such two-leg diagrams are in bijection with the $`2i`$-leg diagrams defined above. Indeed, starting from these two-leg diagrams and gluing their two legs into a marked edge, we simply erase the unweighted $`2i`$-valent vertex and obtain the desired $`2i`$-leg diagrams with a marked leg which receives the label $`1`$ (see Fig.12 for an example). Note that $`2i`$-leg diagrams need not be connected except for $`i=1`$ and may split into connected $`2j`$-leg diagrams with some $`j`$’s summing to $`i`$. Eq.(5.1) follows.
An outcome of this result is that we obtain a host of explicit expressions for $`G_{2i}`$ in terms of the generating functions $`R_i`$ of blossom trees corresponding to various choices of $`n`$ in Eq.(5.1). The first expression is nothing but $`G_{2i}=Z_{0,1}(2i1)`$ corresponding to $`n=0`$. For illustration, for $`i=1,2`$ and $`3`$, we have
$$\begin{array}{cc}\hfill G_2& =R_0\hfill \\ \hfill G_4& =R_0(R_0+R_1)\hfill \\ \hfill G_6& =R_0(R_0^2+2R_0R_1+R_1^2+R_1R_2)\hfill \end{array}$$
Another particularly simple choice consists in using Eq.(5.1) in the limit $`n\mathrm{}`$. This gives explicit formulas for $`G_{2i}`$ in terms of the function $`R`$ solution of Eq.(2.1) (recall that $`R`$ is simply the generating function for two-leg diagrams with arbitrary geodesic distance between the legs). Letting $`R_aR`$ for all $`a`$, we first find that
$$\begin{array}{cc}& Z_{n1,n1+2j}^+(2i)\left(\left(\genfrac{}{}{0pt}{}{2i}{ij}\right)\left(\genfrac{}{}{0pt}{}{2i}{ij1}\right)\right)R^{ij}\hfill \\ & V_{n+2j1,n2}^{}\underset{kj+1}{}g_k\left(\genfrac{}{}{0pt}{}{2k1}{k+j}\right)R^{k+j}\hfill \end{array}$$
which finally yields from Eq.(4.1)
$$\begin{array}{cc}\hfill G_{2i}=& \left(\left(\genfrac{}{}{0pt}{}{2i}{i}\right)\left(\genfrac{}{}{0pt}{}{2i}{i1}\right)\right)R^i\hfill \\ & \underset{k2}{}g_kR^{i+k}\underset{j=1}{\overset{\mathrm{min}(i,k1)}{}}\left(\left(\genfrac{}{}{0pt}{}{2i}{ij}\right)\left(\genfrac{}{}{0pt}{}{2i}{ij1}\right)\right)\left(\genfrac{}{}{0pt}{}{2k1}{k+j}\right)\hfill \end{array}$$
These expressions are determined equivalently as the solutions of the inverse relation (4.1) at large $`n`$, which reads
$$\underset{kj+1}{}g_k\left(\genfrac{}{}{0pt}{}{2k1}{k+j}\right)R^{k+j}=\underset{i=0}{\overset{j}{}}(1)^{i1}\left(\genfrac{}{}{0pt}{}{2ji}{i}\right)R^iG_{2j2i}$$
where $`\left(\genfrac{}{}{0pt}{}{2ji}{i}\right)`$ is the number of hard dimer configurations of $`i`$ dimers on a segment of length $`2j1`$
For illustration in the truncated case with valences up to $`2m=6`$, we find from Eq.(5.1) that
$$\begin{array}{cc}\hfill G_2& =Rg_2R^35g_3R^4\hfill \\ \hfill G_4& =2R^23g_2R^416g_3R^5\hfill \\ \hfill G_6& =5R^39g_2R^550g_3R^6\hfill \end{array}$$
where $`R`$ is now determined by Eq.(2.1).
The explicit expressions (5.1) or (5.1) are to be compared with other known expressions for multi-point correlation functions. The $`2i`$-point function above may indeed be computed alternatively either by use of the planar limit of the one matrix model or by the so-called loop equations. The planar solution of the one-matrix integral may be obtained via saddle point techniques and reads in the so-called one-cut case:
$$\begin{array}{cc}\hfill \underset{\mathrm{}=0}{\overset{i}{}}G_{2i2\mathrm{}}\left(\genfrac{}{}{0pt}{}{2\mathrm{}}{\mathrm{}}\right)R^{\mathrm{}}& =\underset{n\mathrm{}}{lim}n|Q^{2i}(QV^{}(Q))|n1\hfill \\ & =\left(\genfrac{}{}{0pt}{}{2i+1}{i}\right)R^i\underset{k2}{}g_kR^{i+k}\underset{j=1}{\overset{\mathrm{min}(i,k1)}{}}\left(\genfrac{}{}{0pt}{}{2i+1}{ij}\right)\left(\genfrac{}{}{0pt}{}{2k1}{k+j}\right)\hfill \end{array}$$
This expression is readily equivalent to Eq.(5.1) by simply noting that
$$\underset{\mathrm{}=0}{\overset{i}{}}\left(\genfrac{}{}{0pt}{}{2\mathrm{}}{\mathrm{}}\right)\left(\left(\genfrac{}{}{0pt}{}{2i2\mathrm{}}{i\mathrm{}j}\right)\left(\genfrac{}{}{0pt}{}{2i2\mathrm{}}{i\mathrm{}j1}\right)\right)=\left(\genfrac{}{}{0pt}{}{2i+1}{ij}\right)$$
obtained by cutting any walk of length $`2i+1`$ from, say $`0`$ to height $`2j+1`$ at the level of its last $`01`$ step, resulting into a first walk of length, say $`2\mathrm{}`$ from $`0`$ to $`0`$ and a positive walk of length $`2(i\mathrm{})`$ from height $`1`$ to height $`2j+1`$.
On the other hand, the loop equations simply express the (disconnected) $`2i+2`$-point function as a sum of either a $`2i+2k`$-point function if the first leg is connected to a $`2k`$-valent vertex or a product of two lower order correlations if the first leg is connected directly to the $`(2j+2)`$-th leg, namely:
$$G_{2i+2}=\underset{k1}{}g_kG_{2i+2k}+\underset{j=0}{\overset{i}{}}G_{2i2j}G_{2j}$$
That Eq.(5.1) solves this loop equation may be seen as a necessary consistency of the planar limit of the one-matrix model in the one-cut case which implies that Eqs.(5.1) and (5.1) are compatible.
5.2. Relations between conserved quantities for bounded valences
In the truncated case just above of valences up to $`2m=6`$, Eqs.(5.1) imply the relation $`g_1G_2+g_2G_4+g_3G_6+1=G_2`$ provided Eq.(2.1) is satisfied. This may be seen as a particular case of the loop equation (5.1) for $`i=0`$ above in its truncated form. More generally, the truncated loop equations for graphs with valences up to $`2m`$ allow for expressing all $`G_{2i}`$ for $`im`$ as polynomials of the first $`m1`$ values $`G_2,G_4,\mathrm{},G_{2(m1)}`$. This gives a set of polynomial relations between the values of the conserved quantities $`\mathrm{\Gamma }_{2i}(n)`$.
Using the results of previous Section, this interdependence may be rephrased into linear relations between the $`G_{2i}`$’s by writing Eq.(5.1) for $`jm`$ in which case the l.h.s. vanishes. This results in
$$0=\underset{i=0}{\overset{j}{}}(1)^{i1}\left(\genfrac{}{}{0pt}{}{2ji}{i}\right)R^iG_{2j2i}$$
where $`R`$ is determined by the polynomial truncation of Eq.(2.1) with $`g_k=0`$ for $`k>m`$. For illustration, for $`m=3`$, we have the first two relations
$$\begin{array}{cc}\hfill G_6& =5RG_46R^2G_2+R^3\hfill \\ \hfill G_8& =7RG_615R^2G_4+10R^3G_2R^4\hfill \end{array}$$
and similar relations for $`G_{2i}`$’s with higher indices.
5.3. Combinatorial interpretation of $`\mathrm{\Gamma }_2(n)`$
Fig. 13: Any blossom tree (a) whose root is encircled by $`i1`$ edges in the closing process of Sect.2.1 may be rerooted at the first excess bud clockwise from the root, while this original root is replaced by a leaf (b). This results in a rooted tree with a, say $`2k`$-valent root vertex adjacent to $`k2`$ buds and $`k+1`$ blossom subtrees (with necessarily $`k2`$). The root of this new tree is encircled in the closing process by $`i1`$ edges. In (c) and (d) we focus on the rearrangement of bud-leaf pairs in a general case. The excess buds in (d) are those of (c) except the first one, now promoted to root. This leaves three unmatched leaves.
The conservation of $`\mathrm{\Gamma }_2(n)`$ has a nice combinatorial explanation in the language of blossom trees. Indeed, assuming $`\mathrm{\Gamma }_0(n)=1`$ for all $`n0`$, the equality $`\mathrm{\Gamma }_2(n)=\mathrm{\Gamma }_2(0)`$ for all $`n1`$, with $`\mathrm{\Gamma }_2(n)`$ defined by Eq.(4.1), may be rewritten as
$$(R_nR_0)=V_{n+1,n2}^{}$$
The quantity $`R_nR_0`$ is the generating function for blossom trees with contour walk of depth $`i`$ with $`0<in`$. In other words, in the closing process of these trees into two-leg diagrams, their root is encircled by at least one and at most $`n`$ edges separating it from the external face. Picking the bud from which the deepest encircling edge originates, we may reroot the tree at this bud and replace the original root by a leaf (see Fig.13). The resulting object is a tree satisfying (B1), (B2) and (B3) except for the root vertex which now has exactly $`(k2)`$ buds if it is $`2k`$-valent (note that by construction, this vertex has valence at least $`4`$ as it originally carried a bud). In particular, all the proper subtrees of this tree are ordinary blossom trees. Finally, the contour walk of this new tree (now stepping from height $`0`$ to height $`+3`$) has clearly depth $`(i1)(n1)`$. The desired generating function may again be obtained by reading the sequence of buds and subtrees counterclockwise around the root vertex and reads $`Z_{n+1,n2}(2k1)`$ if the new root vertex is $`2k`$-valent. Summing over all possible valences, we get the generating function $`V_{n+1,n2}^{}`$. As the above re-rooting procedure clearly establishes a bijection between the two types of trees at hand, this provides a bijective proof of the equality (5.1). As this equality is valid for all $`n1`$, this in turn proves the conservation of $`\mathrm{\Gamma }_2(n)`$.
It would be nice to have similar bijective proofs for the conservation of $`\mathrm{\Gamma }_{2i}(n)`$ for $`i2`$ as well. For $`2i=4`$, it is easily seen that, upon using $`\mathrm{\Gamma }_0(n)=1`$ and $`\mathrm{\Gamma }_2(n+2)=\mathrm{\Gamma }_2(n)=R_0`$, the equality $`\mathrm{\Gamma }_4(n)=\mathrm{\Gamma }_4(0)`$ may be rewritten as
$$R_0(R_{n+1}R_1)=V_{n+3,n2}^{}+V_{n+3,n}^{}V_{n+1,n2}^{}$$
The l.h.s. of this identity is nothing but the generating function for pairs made of a blossom tree with contour walk of depth $`0`$ and a blossom tree with contour walk of depth $`i`$ with $`1<in+1`$. The r.h.s. may be interpreted as the generating function for the collection of two types of objects
(1) rooted trees satisfying (B1), (B2) and (B3) except at the root vertex which carries exactly $`(k3)`$ buds if it is $`2k`$-valent (with necessarily $`k3`$). The contour walk of this tree moreover has depth at most $`n1`$;
(2) pairs of rooted trees satisfying (B1), (B2) and (B3) except at their root vertex which carries exactly $`(k2)`$ buds if it is $`2k`$-valent (with necessarily $`k2`$). The first of these trees has a contour walk of depth at most $`n+1`$ and the other a contour walk of depth at most $`n1`$.
This suggests the existence of a bijection between the two collections of objects above. If such a bijection could be exhibited, it would provide an alternative bijective proof of the conservation of $`\mathrm{\Gamma }_4(n)`$.
For larger $`i`$, the conservation of $`\mathrm{\Gamma }_{2i}(n)`$ yields more involved relations of the type above which may suggest higher order bijections as well. For instance for $`2i=6`$, we get the relation
$$\begin{array}{cc}\hfill R_0R_1(R_{n+2}& R_2)R_0(R_{n+1}R_1)(R_{n+3}R_1)=\hfill \\ & V_{n+5,n2}^{}+V_{n+5,n}^{}V_{n+1,n2}^{}+V_{n+5,n+2}^{}V_{n+3,n2}^{}+V_{n+5,n+2}^{}V_{n+3,n}^{}V_{n+1,n2}^{}\hfill \end{array}$$
while for $`2i=8`$, we have
$$\begin{array}{cc}& R_0R_1R_2(R_{n+3}R_3)R_0R_1((R_{n+1}R_1)(R_{n+4}R_2)+(R_{n+2}R_2)(R_{n+4}R_1)\hfill \\ & +(R_{n+2}R_2)(R_{n+5}R_2))+R_0((R_{n+1}R_1)(R_{n+3}R_1)(R_{n+5}R_1))\hfill \\ & =V_{n+7,n2}^{}+V_{n+7,n}^{}V_{n+1,n2}^{}+V_{n+7,n+4}^{}V_{n+5,n2}^{}\hfill \\ & +V_{n+7,n+2}^{}V_{n+3,n}^{}V_{n+1,n2}^{}+V_{n+7,n+4}^{}V_{n+5,n}^{}V_{n+1,n2}^{}+V_{n+7,n+4}^{}V_{n+5,n+2}^{}V_{n+3,n2}^{}\hfill \\ & +V_{n+7,n+2}^{}V_{n+3,n2}^{}+V_{n+7,n+4}^{}V_{n+5,n+2}^{}V_{n+3,n}^{}V_{n+1,n2}^{}\hfill \end{array}$$
These relations still await a bijective proof.
6. Discussion
6.1. Alternative expressions for the conserved quantities
As already mentioned, the definition (4.1) corresponds to a particular choice of conserved quantities involving positive walks. This choice however does not respect the “time reversal” symmetry of the master equation, namely under $`R_{nj}R_{n+j}`$ for all $`j`$ (see Eq.(2.1) for illustration). In this Section, we present another choice of conserved quantities that respect this symmetry. More precisely, we define
$`\begin{array}{cc}\hfill \stackrel{~}{\mathrm{\Gamma }}_{2i}(n)& =\left\{Z_{n,n1}(2i1)R_{n1}R_{n+1}Z_{n2,n+1}(2i1)\right\}\stackrel{~}{\mathrm{\Gamma }}_0(n)\hfill \\ & {\displaystyle \underset{j=1}{\overset{i}{}}}\left\{Z_{nj,n+j1}(2i1)R_{nj1}R_{n+j+1}Z_{nj2,n+j+1}(2i1)\right\}V_{n+j,nj1}^{}\hfill \end{array}`$
for $`i1`$ and $`n0`$, with $`\stackrel{~}{\mathrm{\Gamma }}_0(n)`$ given by
$$\stackrel{~}{\mathrm{\Gamma }}_0(n)=R_nV_{n,n1}^{}$$
for $`n0`$. Note that in both $`Z`$’s and $`V^{}`$’s, the indices lead to symmetric expressions.
By techniques similar to those of Sect.4.1, one can invert Eqs. (6.1) and (6.1) into
$$V_{n+j,nj1}^{}=\underset{i=0}{\overset{j}{}}(1)^{i1}\mathrm{\Pi }_{nj,n+j1}(i)\stackrel{~}{\mathrm{\Gamma }}_{2j2i}(n)+\delta _{j,0}R_n$$
with the convention $`\mathrm{\Pi }_{n,n1}(0)=1`$. The passage from Eq.(6.1) to Eq.(6.1) makes use of an inversion relation similar to Eq.(3.1). This relation is explicited in Appendix B.
The conservation of $`\stackrel{~}{\mathrm{\Gamma }}_{2i}(n)`$ may be derived along the same lines as in Sect.4.2. Using again Eq.(4.1) now with $`a=n+j`$ and $`b=nj`$, we have the relation
$$\begin{array}{cc}\hfill V_{(n+1)+j,(n+1)j1}^{}+R_{n+j}& V_{n+(j1),n(j1)1}^{}\hfill \\ & =V_{n+j,nj1}^{}+R_{nj+1}V_{(n+1)+(j1),(n+1)(j1)1}^{}\hfill \end{array}$$
For $`j1`$ and $`n0`$, substituting Eq.(6.1) into this relation yields
$$\begin{array}{cc}\hfill (\stackrel{~}{\mathrm{\Gamma }}_{2j}(n+1)& +\underset{i=1}{\overset{j}{}}(1)^{i1}\mathrm{\Pi }_{(n+1)j,(n+1)+j1}(i)\stackrel{~}{\mathrm{\Gamma }}_{2j2i}(n+1))\hfill \\ \hfill +R_{n+j}& \left(\underset{i=0}{\overset{j1}{}}(1)^{i1}\mathrm{\Pi }_{n(j1),n+(j1)1}(i)\stackrel{~}{\mathrm{\Gamma }}_{2(j1)2i}(n)+\delta _{(j1),0}R_n\right)\hfill \\ \hfill =(\stackrel{~}{\mathrm{\Gamma }}_{2j}(n)& +\underset{i=1}{\overset{j}{}}(1)^{i1}\mathrm{\Pi }_{nj,n+j1}(i)\stackrel{~}{\mathrm{\Gamma }}_{2j2i}(n))\hfill \\ \hfill +R_{nj+1}(\underset{i=0}{\overset{j1}{}}& (1)^{i1}\mathrm{\Pi }_{(n+1)(j1),(n+1)+(j1)1}(i)\stackrel{~}{\mathrm{\Gamma }}_{2(j1)2i}(n+1)+\delta _{(j1),0}R_{n+1})\hfill \end{array}$$
As the boundary terms on both sides cancel, upon setting $`i(i1)`$ in the last sum of each side, we may rewrite this identity as
$$\begin{array}{cc}\hfill \stackrel{~}{\mathrm{\Gamma }}_{2j}(n+1)& \stackrel{~}{\mathrm{\Gamma }}_{2j}(n)\hfill \\ & =\underset{i=1}{\overset{j}{}}(1)^{i1}\left\{\mathrm{\Pi }_{nj+1,n+j}(i)+R_{nj+1}\mathrm{\Pi }_{nj+2,n+j1}(i1)\right\}\stackrel{~}{\mathrm{\Gamma }}_{2j2i}(n+1)\hfill \\ & \underset{i=1}{\overset{j}{}}(1)^{i1}\left\{\mathrm{\Pi }_{nj,n+j1}(i)+R_{n+j}\mathrm{\Pi }_{nj+1,n+j2}(i1)\right\}\stackrel{~}{\mathrm{\Gamma }}_{2j2i}(n)\hfill \\ & =\underset{i=1}{\overset{j}{}}(1)^{i1}\left\{\mathrm{\Sigma }_{nj+1,n+j1}(i)\right\}\left(\stackrel{~}{\mathrm{\Gamma }}_{2j2i}(n+1)\stackrel{~}{\mathrm{\Gamma }}_{2j2i}(n)\right)\hfill \end{array}$$
where we have defined
$$\mathrm{\Sigma }_{a,b}(i)\mathrm{\Pi }_{a,b+1}(i)+\mathrm{\Pi }_{a1,b}(i)\mathrm{\Pi }_{a,b}(i)$$
In the last line of Eq.(6.1) we have used the relation $`\mathrm{\Pi }_{a,b+1}(i)+R_a\mathrm{\Pi }_{a+1,b}(i1)=\mathrm{\Pi }_{a1,b}(i)+R_{b+1}\mathrm{\Pi }_{a,b1}(i1)=\mathrm{\Sigma }_{a,b}(i)`$ which is a direct consequence of Eq.(4.1). We immediately deduce from Eq.(6.1) that all $`\stackrel{~}{\mathrm{\Gamma }}_{2j}(n)`$ for $`j1`$ are conserved provided $`\stackrel{~}{\mathrm{\Gamma }}_0(n)`$ is independent of $`n`$. This is the case when the master equation (2.1) is satisfied, as it reads precisely $`\stackrel{~}{\mathrm{\Gamma }}_0(n)=1`$ for all $`n0`$. This completes the proof of conservation of $`\stackrel{~}{\mathrm{\Gamma }}_{2j}`$.
For illustration, we list below the two independent conserved quantities for the truncated case with up to $`2m=6`$-valent graph corresponding to the truncated master equation (2.1):
$$\begin{array}{cc}\hfill \stackrel{~}{\mathrm{\Gamma }}_2(n)& =R_nV_{n+1,n2}^{}\hfill \\ \hfill \stackrel{~}{\mathrm{\Gamma }}_4(n)& =R_n(R_{n1}+R_n+R_{n+1})R_{n1}R_{n+1}\hfill \\ & (R_{n1}+R_n+R_{n+1})V_{n+1,n2}^{}V_{n+2,n3}^{}\hfill \end{array}$$
with
$$\begin{array}{cc}\hfill V_{n+1,n2}^{}& =R_{n+1}R_nR_{n1}\left(g_2+g_3(R_{n+2}+R_{n+1}+R_n+R_{n1}+R_{n2})\right)\hfill \\ \hfill V_{n+2,n3}^{}& =g_3R_{n+2}R_{n+1}R_nR_{n1}R_{n2}\hfill \end{array}$$
where we explicitly substituted $`\stackrel{~}{\mathrm{\Gamma }}_0(n)=1`$ into Eq.(6.1).
When compared with Eq.(4.1), we see that $`\stackrel{~}{\mathrm{\Gamma }}_2(n)=\mathrm{\Gamma }_2(n)`$ and $`\stackrel{~}{\mathrm{\Gamma }}_4(n)=\mathrm{\Gamma }_4(n1)+(R_{n+1}+R_n+R_{n1})(\mathrm{\Gamma }_2(n)\mathrm{\Gamma }_2(n1))`$. This shows that the new set of conserved quantities is not independent from that of Section 4, as expected.
This interdependence can be made even more explicit by noting that, for $`n=0`$, we again have $`\stackrel{~}{\mathrm{\Gamma }}_{2i}(0)=Z_{0,1}(2i1)=\mathrm{\Gamma }_{2i}(0)`$. When Eq.(2.1) is satisfied, the constant value of $`\stackrel{~}{\mathrm{\Gamma }}_{2i}(n)`$ is therefore equal to $`G_{2i}`$. This is also apparent from the large $`n`$ limit of Eq.(6.1) which is identical to the large $`n`$ limit of Eq.(4.1). As a last remark, we note that the conservation of the $`\mathrm{\Gamma }`$’s is granted by that of the $`\stackrel{~}{\mathrm{\Gamma }}`$’s as we may then replace $`\stackrel{~}{\mathrm{\Gamma }}_{2j2i}(n)`$ in the r.h.s. of Eq.(6.1) by a shifted value $`\stackrel{~}{\mathrm{\Gamma }}_{2j2i}(nj+1)`$ which, upon a global shift of $`nn+j1`$, show that the $`\stackrel{~}{\mathrm{\Gamma }}_{2i}`$’s also obey Eq.(4.1) for $`j1`$. Together with the master equation which guarantees that $`\stackrel{~}{\mathrm{\Gamma }}_0(n)=\mathrm{\Gamma }_0(n)=1`$, this implies that, when the $`\stackrel{~}{\mathrm{\Gamma }}`$’s are conserved, we necessarily have $`\mathrm{\Gamma }_{2i}(n)=\stackrel{~}{\mathrm{\Gamma }}_{2i}(n)`$ for all $`n`$, hence these are conserved as well and take the same constant values.
6.2. Compacted conserved quantities
Let us again restrict ourselves to the truncated case of valences up to $`2m`$ and concentrate on the $`(m1)`$ first conserved quantities $`\stackrel{~}{\mathrm{\Gamma }}_{2i}(n)`$ for $`i=1,\mathrm{},m1`$. From the definition (6.1), we note that the largest index carried by an $`R_a`$ comes from the contributions $`g_mZ_{n+j,nj1}(2m1)`$ in the term $`V_{n+j,nj1}^{}`$ appearing in the r.h.s.. This index is equal to $`a=n+m1`$ and reached for each $`j`$ by a unique path of length $`2m1`$ starting with $`mj1`$ up steps followed by $`m+j`$ down steps. This term reads precisely $`g_m\underset{\mathrm{}=1m}{\overset{j}{}}R_n\mathrm{}`$. Similarly, the largest index carried by an $`R_a`$ in $`\stackrel{~}{\mathrm{\Gamma }}_0(n)`$ comes from the contribution $`g_mZ_{n,n1}(2m1)`$ in the term $`V_{n,n1}^{}`$ and reads $`g_m\underset{\mathrm{}=1m}{\overset{0}{}}R_n\mathrm{}`$.
We have the remarkable identity
$$\begin{array}{cc}\hfill Z_{n1,n1}& (2i)=\left\{Z_{n,n1}(2i1)R_{n1}R_{n+1}Z_{n2,n+1}(2i1)\right\}\hfill \\ & +\underset{j=1}{\overset{i}{}}\left\{Z_{nj,n+j1}(2i1)R_{nj1}R_{n+j+1}Z_{nj2,n+j+1}(2i1)\right\}\underset{\mathrm{}=1}{\overset{j}{}}R_n\mathrm{}\hfill \end{array}$$
which may be inverted into
$$\underset{\mathrm{}=1}{\overset{j}{}}R_n\mathrm{}=\underset{i=0}{\overset{j}{}}(1)^{i1}\mathrm{\Pi }_{nj,n+j1}(i)Z_{n1,n1}(2j2i)$$
These identities may be proved with arguments similar to that of Sect.3.2 and follow from general inversion formulas listed in Appendix B. We deduce that in the combination
$`\mathrm{\Theta }_{2i}(n)=\stackrel{~}{\mathrm{\Gamma }}_{2i}(n)+\left(1\stackrel{~}{\mathrm{\Gamma }}_0(n)\right)Z_{n1,n1}(2i)`$
all terms containing $`R_{n+m1}`$ are cancelled. Note that the definition (6.1) does not involve the particular value of $`m`$ and therefore gives a universal recipe for compacting the conserved quantities, irrespectively of the precise (finite) order of truncation. The quantities $`\mathrm{\Theta }_{2i}(n)`$ form clearly an alternative set of conserved quantities that take the same constant values $`G_{2i}`$, but that involve one less $`R_a`$ than the original $`\mathrm{\Gamma }_{2i}`$ or $`\stackrel{~}{\mathrm{\Gamma }}_{2i}`$.
For illustration, for $`m=3`$, we have
$$\begin{array}{cc}\hfill \mathrm{\Theta }_2(n)& =R_n+R_{n1}R_nR_{n1}[1g_1g_2(R_n+R_{n1})\hfill \\ & g_3(R_{n+1}R_n+(R_n+R_{n1})^2+R_{n1}R_{n2}R_{n+1}R_{n2})]\hfill \\ \hfill \mathrm{\Theta }_4(n)& =R_{n+1}R_n+R_{n1}R_{n2}+(R_n+R_{n1})^2\hfill \\ & R_nR_{n1}[(1g_1)(R_{n+1}+R_n+R_{n1}+R_{n2})\hfill \\ & g_2(R_{n+1}+R_n+R_{n1})(R_n+R_{n1}+R_{n2})\hfill \\ & g_3(R_{n+1}+R_n+R_{n1}+R_{n2})(R_{n+1}R_n+(R_n+R_{n1})^2+R_{n1}R_{n2})]\hfill \end{array}$$
The value of $`\mathrm{\Theta }_2(n)`$ at $`g_1=g_3=0`$ was already obtained in Ref..
6.3. Application to tetra- and hexavalent graphs
In the truncated case of graphs with valences up to $`2m`$, the truncated master equation is a recursion relation on $`n`$ allowing for the recursive determination of all $`R_n`$’s with $`nm1`$ from the knowledge of $`R_0,R_1,\mathrm{},R_{m2}`$. These initial values are completely determined by using the $`m1`$ first conserved quantities in the form $`\mathrm{\Gamma }_{2i}(0)=\mathrm{\Gamma }_{2i}(\mathrm{})=G_{2i}`$ (or similar equalities for $`\stackrel{~}{\mathrm{\Gamma }}_{2i}`$ or $`\mathrm{\Theta }_{2i}`$) with $`i=1,\mathrm{},m1`$. For instance, in the case $`m=3`$ of graphs with bi-, tetra- and hexavalent vertices only, this yields, by use of the first two lines of (5.1) and (5.1)
$$\begin{array}{cc}\hfill R_0& =Rg_2R^35g_3R^4\hfill \\ \hfill R_1& =\frac{R(16g_3R^3g_2^2R^425g_3^2R^6g_2(R^2+10g_3R^5))}{1g_2R^25g_3R^3}\hfill \end{array}$$
with $`R`$ solution of Eq.(2.1). This result is in agreement with the expressions obtained in Ref. via the exact solution (2.1).
Another possible application concerns local environments within graphs. For instance, we have access to the distribution of the number of faces adjacent to the external face in rooted planar graphs, namely graphs with a marked oriented edge, with the external face adjacent to this rooted edge and lying on its right. These rooted graphs are in bijection with two-leg diagrams with both legs in the same (external) face, as readily seen by gluing the two legs counterclockwise around the diagram into a rooted edge oriented from the outcoming leg to the incoming one. The rooted graphs are therefore enumerated by $`R_0`$. We may then assign an extra weight $`x`$ to each face adjacent to the external one, i.e. a weight $`x^p`$ whenever the external face has $`p`$ adjacent faces. The corresponding generating function, still denoted $`R_0R_0(x)`$, is determined by the master equation (2.1) for all $`n1`$ and by a modified equation at $`n=0`$, namely
$$R_0=x+V_{0,1}^{}$$
In particular, all conserved quantities are valid for $`n1`$ and their constant value remains unchanged, given by $`G_{2i}`$.
In the truncated case of valences up to $`2m`$, we obtain a closed algebraic system involving $`R_0,R_1,\mathrm{},R_{m1}`$ by supplementing Eq.(6.1) by the $`m1`$ first conservation laws $`\mathrm{\Theta }_{2i}(1)=G_{2i}`$ for $`i=1,\mathrm{}m1`$. Upon eliminating all $`R_n`$’s with $`n>0`$, we are left with a single algebraic equation determining $`R_0(x)`$. Let us now illustrate this in the cases of pure tetravalent and pure hexavalent graphs.
Fig. 14: The twelve rooted tetravalent planar graphs with up to $`2`$ vertices, hence with weights $`g_2^p`$, $`p=0,1,2`$. Assigning a weight $`x`$ to each face adjacent to the external one, we reproduce the first three terms of the expansion (6.17)
For pure tetravalent graphs, we obtain the equation
$$x(x1)+(1x+g_2R(R2)2g_2^2R^3)R_0+xg_2R_0^2=0$$
with $`R`$ solution of $`R=1+3g_2R^2`$. This leads to the expansion
$$R_0=x+g_2(x+x^2)+g_2^2(3x+4x^2+2x^3)+g_2^3(14x+20x^2+15x^3+5x^4)+\mathrm{}$$
The corresponding rooted graphs up to order $`g_2^2`$ are displayed in Fig.14 for illustration. From the knowledge of $`R_0(x)`$, we may extract the distribution of the number of faces adjacent to the external face in large tetravalent rooted graphs by considering the limit
$$\mathrm{\Delta }(x)\underset{p=1}{\overset{\mathrm{}}{}}x^p𝒫(p)=\underset{N\mathrm{}}{lim}\frac{R_0(x)|_{g_2^N}}{R_0(1)|_{g_2^N}}$$
Here $`𝒫(p)`$ denotes the probability for the external face of infinitely large tetravalent rooted graphs to have $`p`$ adjacent faces. The large $`N`$ limits above are governed by the approach to the critical point $`g_2g_2^{}=1/12`$ where $`R`$ and $`R_0`$ become singular. Expanding in powers of $`ϵ=\sqrt{(g_2^{}g_2)/g_2^{}}`$, the above ratio giving $`\mathrm{\Delta }(x)`$ is obtained as the ratio of the coefficients of $`ϵ^3`$ in the expansions of $`R_0(x)`$ and $`R_0(1)`$ respectively. Using Eq.(6.1), we get the following algebraic equation for $`\mathrm{\Delta }(x)`$
$$27x(x1)+4(43x)^3\mathrm{\Delta }(1x\mathrm{\Delta })=0$$
hence
$$\mathrm{\Delta }(x)=\frac{1}{2x}\left(1\frac{89x}{(43x)^{\frac{3}{2}}}\right)$$
and
$$𝒫(p)=\left(\frac{3}{16}\right)^{p+1}\frac{(2p+1)!}{(p+1)!(p1)!}$$
For pure hexavalent graphs, we obtain the equation
$$x(x1)^2(x1)(x1+2g_3R^2(32R)+24g_3^2R^5)R_0+Rxg_3(R25g_3R^3)R_0^2+x^2g_3R_0^3=0$$
with $`R`$ solution of $`R=1+10g_3R^3`$. This now leads to the expansion
$$R_0=x+g_3(2x+2x^2+x^3)+g_3^2(28x+32x^2+25x^3+12x^4+3x^5)+\mathrm{}$$
From $`R_0(x)`$, we may again extract the distribution of the number of faces adjacent to the external face in large hexavalent rooted graphs by considering the same limiting ratio as in (6.1) with $`g_2g_3`$ and where $`𝒫(p)`$ now denotes the probability for the external face of infinitely large hexavalent rooted graphs to have $`p`$ adjacent faces. The corresponding large $`N`$ limits are now governed by the approach to the critical point $`g_3g_3^{}=2/135`$ where $`R`$ and $`R_0`$ become singular. Performing a similar expansion as before, we arrive at the following algebraic equation for $`\mathrm{\Delta }(x)`$
$$125x(x1)^29(x1)(125x^3+475x^2+200x96)\mathrm{\Delta }27x(x+4)(75x)^3\mathrm{\Delta }^2(1x\mathrm{\Delta })=0$$
from which we deduce the probabilities
$$𝒫(1)=\frac{125}{864},𝒫(2)=\frac{1625}{10368},𝒫(3)=\frac{865625}{5971968}$$
for the external face to have $`1,2`$ or $`3`$ neighboring faces.
7. Conclusion
In this paper, we have explicitly constructed various sets of conserved quantities for the (integrable) master equation which determines the generating function $`R_n`$ for two-leg diagrams with vertices of even valences. These conserved quantities are constructed in terms of generating functions for random walks, whose relations with heaps of dimers and hard dimers were instrumental to grant the conservation.
The precise properties that we used, namely some boson/fermion correspondences and their associated inversion relations, as well as some fundamental relations satisfied by either walks or hard dimers, are very general and presumably can be adapted to other cases of interest. Indeed, in the planar graph enumeration framework, it was already recognized that many other classes of (decorated) graphs have generating functions which satisfy integrable systems of algebraic master equations. These include graphs with arbitrary even or odd valences, as well as the so-called $`p`$-constellations considered in Ref.. In both cases, the two-leg diagram generating functions display a form similar to that of Eq.(2.1) . In a particular case of $`3`$-constellations, namely that of graphs with bicolored trivalent vertices (so-called bicubic maps), a conserved quantity was already identified in Ref.. We suspect that the above methods can be extended to the even more general case of planar bipartite graphs, also enumerated (without geodesic distance) by two-matrix models.
Beside the dependence on the geodesic distance, we know from matrix-model analysis that another integrable structure underlies the topological expansion for the generating functions of graphs with arbitrary genus. It would be desirable to relate the two directions of integrability. A general theory of discrete integrable equations was built in Ref. with (Hirota bilinear) master equations involving more than one integer index $`n`$ (see also for the theory of discrete Painlevé equations). This suggests to look for a larger integrable structure that would include both the dependence on (possibly several) geodesic distances as well as that on topology.
As explained in Sect.5.3, the conservation properties have a strong “bijective flavor” in the language of blossom trees. A bijective explanation for the conservation would be very instructive and seems to be within reach in view if the relations (5.1) and (5.1). If such a proof exists, it must involve only generic properties of blossom trees, or equivalently of planar graphs. This could allow to trace back the somewhat mysterious generic appearance of conserved quantities in the context of planar graph enumeration problems to these generic properties.
Finally, we may hope that the conservation properties have a natural interpretation of their own both in the language of labeled mobiles and in that of spatially extended branching processes. This is yet to be found. Note in this context the recent appearance of yet another integrable master equation governing the so-called “embedded binary trees” studied in Ref..
To conclude, the general case of graphs with vertices of even valence is known to give access to multicritical transitions points of 2D Quantum Gravity by proper fine tuning of the parameters $`g_i`$. As we have now a complete picture of the conserved quantities, we may use them to describe these multicritical points. In the scaling limit of large graphs, the master equation is known to reduce at such points to an integrable differential equation related to the KdV hierarchy . Our conserved quantities provide explicit discrete counterparts of the integrals of motion for these equations. These integrals are indeed recovered from our expressions in the scaling limit.
Acknowledgments: We acknowledge the support of the EU network on “Discrete Random Geometry”, grant HPRN-CT-1999-00161. We thank Jérémie Bouttier for useful discussions and active interactions at an early stage of this work. We thank Mireille Bousquet-Mélou for enlightening discussions on the combinatorics of heaps of pieces.
Appendix A. Relation between heaps and walks
We consider the generating functions $`Z_{a,b}(2k1)`$ for random walks of $`2k1`$ steps from height $`a`$ to height $`b`$ with weight $`R_i`$ per descent $`i(i1)`$, and its truncated version $`Z_{a,b}^+(2k1)`$ restricted to “positive” walks with heights larger or equal to $`a`$. Similarly, we consider the generating function $`H_{a,b}(k)`$ for pyramids of $`k`$ dimers with base $`[a,b]`$ (with $`(ba)`$ odd and positive) with weight $`R_i`$ per dimer in the stripe $`i`$ except for those in the right projection. We also consider its truncated half-pyramid version $`H_{a,b}^+(k)`$ with dimers only in stripes $`i>a`$. In both cases, the truncation simply amounts to taking $`R_i0`$ for all $`ia`$.
We wish to prove that
$$Z_{a,b}(2k1)=H_{a,b}(k),Z_{a,b}^+(2k1)=H_{a,b}^+(k)$$
for all $`k1`$ and $`(ba)`$ odd and positive.
Fig. 15: Bijection between walks from $`a`$ to $`b>a`$ with $`2k1`$ steps and pyramids of base $`[a,b]`$ with $`k`$ dimers, as explained in the text. In (a), the times/locations of pebble droppings are indicated by filled black ellipses. Each pebble stays in place during some time (indicated by a black horizontal segment) until it is possibly picked up. In (b), we put a dimer at each pick-up time/location. This configuration is completed into a pyramid of base \[a,b\] in (c). The inverse mapping makes use of the splitting of dimers into up- or down-pointing triangles (d), whose sequence along each line determines the steps of the walk.
These relations are a consequence of the following bijection, borrowed from Ref., between on the one hand random walks of $`2k1`$ steps from height $`a`$ to height $`b>a`$ and on the other hand pyramids with base $`[a,b]`$ and a total of $`k`$ dimers. Starting from a walk, as illustrated in Fig.15-(a), viewed as the time (horizontal) evolution of a walker on the integer (vertical) line, the walker acts as the Little Thumb by dropping small pebbles at each passage in a unit segment unless a pebble is already present in which case he picks it up. As shown in Fig.15-(b), putting dimers at the times/locations of the pick-up events results in a heap of dimers. Moreover, if the walker goes from $`a`$ to $`b>a`$, there are $`ba`$ pebbles left behind. By completing the heap on the right by the maximally occupied hard dimer configuration on the segment $`[a,b]`$, we obtain a pyramid of base $`[a,b]`$ and with a total of $`k`$ dimers if the walk has $`2k1`$ steps, as illustrated in Fig.15-(c). This establishes a bijection whose inverse goes as follows. We split each dimer of the pyramid not in the right projection into a pair of up- and down-pointing triangles, as shown in Fig.15-(d). Each such triangle has its horizontal edge on a line separating stripes. We also add on the right a succession of $`ba`$ up-pointing triangles adjacent to the lines $`y=a,a+1,\mathrm{},b1`$. For each line $`y=i`$, we record the sequence of up- and down-pointing triangles. The walk is reconstructed by starting from height $`a`$ and using successively the triangles in the above sequences. More precisely, at each position $`y=i`$, the walker makes an up- or down-step according to the up-or down-pointing nature of the first unused triangle along the line $`y=i`$.
In the above bijection, the descending steps of the walk from height $`i`$ to height $`i1`$ are in one-to-one correspondence with the dimers in the stripe $`i`$, except for those of the right projection. Note that these descending steps may correspond either to the dropping or to the pick-up of a pebble. In particular, weighting each descent of the walk from height $`i`$ to height $`i1`$ by $`R_i`$ (resp. $`0`$ if $`ia`$) amounts to weight each dimer in the stripe $`i`$ by $`R_i`$ (resp. $`0`$ if $`ia`$), except for those dimers in the right projection of the pyramid (resp. half-pyramid). This immediately yields the relations (A.1).
Appendix B. Useful identities
We have the following useful identities, proved by arguments similar to that of Sect.3.2.
For $`j1`$ and $`k0`$:
$$\begin{array}{cc}\hfill \underset{i=0}{\overset{j}{}}(1)^{ji}\mathrm{\Pi }_{nj,n+j1}(ji)& \left(\underset{\mathrm{}=1}{\overset{k}{}}R_{n+k+12\mathrm{}}\right)Z_{nk,n+k1}(2i1)\hfill \\ & =\mathrm{\Pi }_{nj,n+j1}(j)\times \delta _{jk}\times \delta _{j=k\mathrm{mod}2}\hfill \end{array}$$
We immediately deduce from this identity the inversion relation
$$\begin{array}{cc}\hfill \underset{i=0}{\overset{j}{}}(1)^{ji}\mathrm{\Pi }_{nj,n+j1}(ji)(& Z_{nk,n+k1}(2i1)\hfill \\ & R_{nk1}R_{n+k+1}Z_{nk2,n+k+1}(2i1))=\delta _{j,k}\hfill \end{array}$$
The inversion of Eq.(6.1) into Eq.(6.1) follows from this identity.
For $`j1`$:
$$\begin{array}{cc}\hfill \underset{i=0}{\overset{j}{}}(1)^{ji}\mathrm{\Pi }_{nj,n+j1}(ji)R_{n1}& Z_{n2,n1}(2i1)\hfill \\ & =\mathrm{\Pi }_{nj,n+j1}(j)\times \delta _{j=0\mathrm{mod}2}+\underset{\mathrm{}=1}{\overset{j}{}}R_n\mathrm{}\hfill \end{array}$$
This relation, together with Eq.(B.1) for $`k=0`$, allows to prove Eqs. (6.1) and (6.1) by noting that $`Z_{n1,n1}(2i)=Z_{n,n1}(2i1)+R_{n1}Z_{n2,n1}(2i1)`$.
References
relax E. Brézin, C. Itzykson, G. Parisi and J.-B. Zuber, Planar Diagrams, Comm. Math. Phys. 59 (1978) 35-51. relax A. Fokas, A. Its and A. Kitaev, Discrete Painlevé equations and their appearance in quantum gravity, Comm. Math. Phys. 142 (1991) 313-344. relax V. Kazakov, Bilocal regularization of models of random surfaces, Phys. Lett. B150 (1985) 282-284; F. David, Planar diagrams, two-dimensional lattice gravity and surface models, Nucl. Phys. B257 (1985) 45-58; J. Ambjorn, B. Durhuus and J. Fröhlich, Diseases of triangulated random surface models and possible cures, Nucl. Phys. B257(1985) 433-449; V. Kazakov, I. Kostov and A. Migdal Critical properties of randomly triangulated planar random surfaces, Phys. Lett. B157 (1985) 295-300. relax See for instance: P. Di Francesco, P. Ginsparg and J. Zinn–Justin, 2D Gravity and Random Matrices, Physics Reports 254 (1995) 1-131, and references therein; see also B. Eynard, Random Matrices, Saclay Lecture Notes (2000), available at http://www-spht.cea.fr/lectures\_notes.shtml relax G. Schaeffer, Conjugaison d’arbres et cartes combinatoires aléatoires, PhD Thesis, Université Bordeaux I (1998); available at http://www.lix.polytechnique.fr/~schaeffe/Biblio. relax G. Schaeffer, Bijective census and random generation of Eulerian planar maps, Electronic Journal of Combinatorics, vol. 4(1) (1997) R20. relax J. Bouttier, P. Di Francesco and E. Guitter, Census of planar maps: from the one-matrix model solution to a combinatorial proof, Nucl. Phys. B645\[PM\] (2002) 477-499, arXiv:cond-mat/0207682. relax J. Bouttier, P. Di Francesco and E. Guitter, Geodesic distance in planar graphs, Nucl. Phys. B663\[FS\] (2003) 535-567, arXiv:cond-mat/0303272. relax X. Viennot, Heaps of pieces 1: basic definitions and combinatorial lemmas, in G. Labelle and P. Leroux Eds. Combinatoire énumérative, Lect. Notes in Math. vol 1234 (1986) 321-350. relax M. Bousquet-Mélou and A. Rechnitzer, Lattice animals and heaps of dimers, Discrete Math. 258 (2002) 235-274. relax P. Di Francesco and E. Guitter, Critical and multicritical semi-random $`(1+d)`$-dimensional lattices and hard objects in $`d`$ dimensions, J. Phys. A Math. Gen, 35 (2002) 897-927. relax J. Bouttier, P. Di Francesco and E. Guitter, Planar Maps as Labeled Mobiles, Electronic Journal of Combinatorics, vol. 11(1) (2004) R69. relax R. Cori and B. Vauquelin, Planar maps are well labeled trees, Canad. J. Math. 33 (5) (1981) 1023-1042; D. Arquès, Les hypercartes planaires sont des arbres très bien étiquetés, Discr. Math. 58(1) (1986) 11-24; M. Marcus and G. Schaeffer, Une bijection simple pour les cartes orientables, http://www.lix.polytechnique.fr/~schaeffe/Biblio/. relax J. Bouttier, P. Di Francesco and E. Guitter, Statistics of planar maps viewed from a vertex: a study via labeled trees, Nucl. Phys. B675\[FS\] (2003) 631-660, arXiv:cond-mat/0307606. relax M. Bousquet-Mélou and G. Schaeffer, Enumeration of planar constellations, Adv. in Applied Math., 24 (2000) 337-368. relax P. Di Francesco, Geodesic distance in planar graphs: An integrable approach, preprint (2003), to appear in the Ramanujan Journal relax M. Jimbo and T. Miwa, Solitons and infinite dimensional Lie algebras, Publ. RIMS, Kyoto Univ. 19 No. 3 (1983) 943-1001, eq.(2.12). relax B. Grammaticos, F. Nijhoff and A. Ramani, Discrete Painlevé equations, in The Painlevé property, one century later, R. Conte Ed., CRM series in Math. Phys. (1999) 413-516. relax M. Bousquet-Mélou, Limit laws for embedded trees. Applications to the integrated superBrownian excursion preprint arXiv:math.CO/0501266.
|
warning/0506/hep-ph0506235.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
The leading order BFKL kernel , derived from Feynman diagrams in momentum space, has been investigated in much detail. As a function of the transverse momenta, it is a meromorphic function. In the color singlet exchange channel, it describes the Pomeron contribution in pQCD, consisting of ladder diagrams with reggeized gluons. This BFKL Pomeron couples to the impact factors of colourless particles which, because of gauge invariance, vanish as one of the two reggeized gluons carries a zero transverse momentum. This property allows us to modify the space of functions to which the Pomeron wave function belongs. On the other hand, in the color octet exchange channel the BFKL equation has the bootstrap solution, which represents a fundamental consistency property derived from the s-channel unitarity. A manifestation of this bootstrap property also takes place for the color singlet state of two gluons and inside the coupling of three or more gluons to colorless particle impact factors. In particular, it plays an important role in the Odderon solution which appears as a bound state of three reggeized gluons.
An important virtue of the color singlet kernel is its invariance under the Möbius transformations in the space of transverse coordinates. Exploiting the gauge invariance of the colorless impact factor to which the BFKL Pomeron couples, the Möbius symmetry allows to search solutions of the BFKL eigenvalues equation in the space of Möbius functions which vanish as the coordinates of the two reggeized gluons coincide, and to define a spectral representation of the BFKL kernel in terms of conformal eigenfunctions. In this space of functions, the BFKL kernel also enjoys the property of holomorphic separability. Translating back to the momentum representation, the kernel in the corresponding spectral form acquires $`\delta `$ function - like pieces which, because of the special properties of the impact factors, do not contribute to physical amplitudes.
As mentioned above, the BFKL kernel, when applied to the color octet wave function in momentum space, satisfies the bootstrap condition, which reflects the s-channel consistency of the BFKL calculation. When transforming to transverse coordinates one finds that, in the space of Möbius functions, this bootstrap condition cannot be satisfied, i.e. the colour octet wave function lies outside this space of functions. In this paper we will define a similarity transform, $`\mathrm{\Phi }`$, which takes us from the Möbius space of functions (M space) to another space of functions (named ‘analytic Feynman diagram’ (AF) space) in which the bootstrap holds. With the same transformation we can also define ‘deformed’ Möbius transformations and a ‘deformed’ spectral representation of the BFKL kernel.
The paper is organized as follows. In the following section we review the properties of the BFKL kernel, both in momentum space and in the space of transverse coordinates. In section 3 we introduce the deformed representation, and we define the similarity transform which takes us from the Möbius space ($`M`$ space) to the new deformed space ($`AF`$ space) of functions. In section 4 we explicitly show that, in the $`AF`$ representation, the bootstrap properties are fulfilled. Section 5 is devoted to the deformed representation of the conformal algebra which follows from the particular choice of the scalar product; in particular we compute the transformed generators of the Möbius group $`𝔰𝔩(2,)`$. Some details of our calculations are collected in two appendices.
## 2 The BFKL equation and bootstrap relation in LLA
In this section we give a brief review of the BFKL approach describing the dynamics of the reggeized gluons in LLA of perturbative QCD. Let us start from the Schrödinger-like BFKL equation describing the compound state of two reggeized gluons,
$$K_2^{(R)}\psi _E=E\psi _E,K_2^{(R)}=\frac{N_c}{2}\left(\stackrel{~}{\omega }_1+\stackrel{~}{\omega }_2\right)\lambda _R\overline{V}_{12}.$$
(1)
Here $`R`$ labels the colour representation of the two gluon state and in the singlet and octet channel one has respectively $`\lambda _1=N_c`$ and $`\lambda _8=N_c/2`$. The symbol $``$ denotes an integration in the transverse space in the case of the integral operator $`\overline{V}_{12}`$, while $`E=\omega =1j`$ where $`j`$ is the $`t`$-channel angular momentum. The eigenvalues of $`K_2^R`$ give the positions of singularities of the $`t`$ channel partial waves, related to the scattering amplitude by the Mellin transformation in the variable $`\mathrm{ln}s`$, where $`s`$ is the squared total energy of colliding particles.
In LLA, in the momentum representation, the gluon trajectory (scaled by $`N_c/2`$) is given by the well known expression
$$\stackrel{~}{\omega }_i=\stackrel{~}{\omega }(𝒌_i)=cd^2𝒍\frac{𝒌_i^2}{𝒍^2(𝒌_i𝒍)^2},c=\frac{g^2}{(2\pi )^3}$$
(2)
and the interaction term is defined by its action on the wave functions
$`\overline{V}_{12}\varphi (𝒌_1,𝒒𝒌_1)={\displaystyle d^2𝒌_1^{}\overline{V}(𝒌_1,𝒒𝒌_1|𝒌_1^{},𝒒𝒌_1^{})\varphi (𝒌_1^{},𝒒𝒌_1^{})}`$
$`=`$ $`c{\displaystyle d^2𝒌_1^{}\left[\frac{𝒌_1^2}{(𝒌_1^{})^2(𝒌_1𝒌_1^{})^2}+\frac{(𝒒𝒌_1)^2}{(𝒒𝒌_1^{})^2(𝒌_1𝒌_1^{})^2}\frac{𝒒^2}{(𝒌_1^{})^2(𝒒𝒌_1^{})^2}\right]\varphi (𝒌_1^{},𝒒𝒌_1^{})}`$
and $`𝒒=𝒌_1+𝒌_2`$. Let us note that this form is proper when the kernel acts on amputated (without gluon propagators) impact factors, otherwise the propagators should be removed from the kernel, as for example is the case for the expression in eq. (9).
We stress that this form of the LL BFKL kernel is obtained directly from the Feynman diagrams analysis in perturbative QCD and has a simple analytic behavior in the transverse momentum space. In particular it does not contain $`\delta `$-functions $`\delta ^2(k_1^{})`$ and $`\delta ^2(qk_1^{})`$.
In the construction of the BFKL kernels a very important property is the gluon reggeization which can be verified with the bootstrap relation for the amplitude with the octet quantum numbers . The bootstrap relation is a consequence of the $`s`$-channel unitarity. It claims, that the scattering amplitude with octet quantum numbers obtained as a solution of the corresponding BFKL equation should coincide with the Born term multiplied by the Regge factor $`s^{\stackrel{~}{\omega }(𝒒)}`$.
Because the BFKL equation was obtained by summing contributions from the multi-particle production described by the multi-Regge amplitudes, the bootstrap relation valid in the leading and next-to-leading orders allows to connect the gluon trajectory and interaction terms. In LLA the bootstrap condition for the BFKL kernel can be written as
$$\omega (𝒒)\omega (𝒌_1)\omega (𝒌_2)=\overline{V}_{12}1\mathrm{or}\overline{\mathrm{K}}_{12}^{(8)}1=\omega (𝒒),$$
(4)
where the constant $`1`$ is the wave function which can be conveniently obtained after rescaling any function depending only on $`𝒒`$, and the kernel $`K_{12}^{(8)}`$ acts on the amputated amplitudes. Let us note that it is crucial in the bootstrap relation in LLA that the two gluons are located at the same point in the transverse coordinate plane. It has played a crucial role in the discovery of the leading Odderon solution in the LLA of perturbative QCD and also some general relation between bound states of $`n`$ and $`n+1`$ reggeized gluons.
In the following we shall use its equivalent form , which has the virtue of being infrared finite and which uses the standard BFKL singlet kernel:
$$\overline{K}_{12}^{(1)}1=2\omega (𝒒)+\omega (𝒌_1)+\omega (𝒌_2).$$
(5)
On using an infrared mass regularization ($`m0`$) one may write
$$\omega (𝒌)=\frac{1}{2}\overline{\alpha }_s\mathrm{log}\left(\frac{𝒌^2}{m^2}\right),$$
(6)
where $`\overline{\alpha }_s=\alpha _sN_c/\pi `$ and $`\alpha _s=g^2/(4\pi )`$, so that a relation equivalent to the bootstrap, but explicitely infraredly finite, is given by
$$\overline{K}_{12}^{(1)}1=\frac{1}{2}\overline{\alpha }_s\mathrm{log}\left(\frac{𝒒^4}{𝒌_1^2𝒌_2^2}\right).$$
(7)
The 2-gluon kernel (1) in the singlet channel has been investigated in great details in the coordinate representation . The amplitude for the scattering of colorless objects is factorized, in the high energy limit, in the product of the Green’s function (which exponentiates the BFKL kernel) and two impact factors. The impact factors vanish as the momentum of one of the attached reggeized gluons goes to zero. This permits to choose a special representation for the two-gluon propagator and for the full BFKL kernel acting in the space of the so called Möbius functions (for two gluon states these are functions $`f(\rho _1,\rho _2)`$ such that $`f(\rho ,\rho )=0`$. This is a special choice among infinite others compatible with the gauge freedom of integrating a colorless impact factor with any function belonging to the set defined by the equivalence relation
$$f(𝝆_1,𝝆_2)\stackrel{~}{f}(𝝆_1,𝝆_2)=f(𝝆_1,𝝆_2)+f^{(1)}(𝝆_1)+f^{(2)}(𝝆_2),$$
(8)
since the corresponding shift in the momentum representation is proportional to $`\delta ^{(2)}(𝒑_i)`$.
In such a representation the operators are Möbius (conformal) invariant . Using complex coordinates, the BFKL hamiltonian $`H_{12}=K_{12}^{(1)}\mathrm{\hspace{0.17em}2}/\overline{\alpha }_s`$ acting on functions with propagators included can be written in the operator form
$$H_{12}=\mathrm{ln}\left|p_1\right|^2+\mathrm{ln}\left|p_2\right|^2+\frac{1}{p_1p_2^{}}\mathrm{ln}\left|\rho _{12}\right|^2p_1p_2^{}+\frac{1}{p_1^{}p_2}\mathrm{ln}\left|\rho _{12}\right|^2p_1^{}p_24\mathrm{\Psi }(1),$$
(9)
where $`\mathrm{\Psi }(x)=d\mathrm{ln}\mathrm{\Gamma }(x)/dx`$, and we introduced the gluon holomorphic momenta $`p_r=i/\rho _r`$. On the Möbius space of functions the holomorphic separability applies:
$$H_{12}=h_{12}+h_{12}^{},h_{12}=\underset{r=1}{\overset{2}{}}\left(\mathrm{ln}p_r+\frac{1}{p_r}\mathrm{ln}(\rho _{12})p_r\mathrm{\Psi }(1)\right),$$
(10)
and it is possible to find more easily the solutions of the homogeneous BFKL pomeron equation which belong to irreducible unitary representations of the Möbius group. In particular the symmetry generators in this representation are
$$M_r^3=\rho _r_r,M_r^+=_r,M_r^{}=\rho _r^2_r.$$
(11)
For two reggeized gluons one has $`M^k=_{r=1}^2M_r^k`$ and the Casimir operator is defined as follows
$$M^2=|\stackrel{}{M}|^2=\rho _{12}^2_1_2,$$
(12)
where $`\stackrel{}{M}=_{r=1}^2\stackrel{}{M}_r`$ and $`\stackrel{}{M}_r(M_r^+,M_r^{},M_r^3)`$.
The eigenfunctions of the BFKL kernel are also eigenstates of two Casimir operators and given by
$$E_{h,\overline{h}}(𝝆_{10},𝝆_{20})\rho |h=\left(\frac{\rho _{12}}{\rho _{10}\rho _{20}}\right)^h\left(\frac{\rho _{12}^{}}{\rho _{10}^{}\rho _{20}^{}}\right)^{\overline{h}},$$
(13)
where $`h=\frac{1+n}{2}+i\nu `$ , $`\overline{h}=\frac{1n}{2}+i\nu `$ are conformal weights for the principal series of the unitary representations of the Möbius group, $`n`$ is the conformal spin and $`d=12i\nu `$ is the anomalous dimension of the operator $`O_{h,\overline{h}}(𝝆_0)`$ describing the compound state of two reggeized gluons (note, that here and below we use other notations for conformal weights $`m,\stackrel{~}{m}`$ in Refs. ). The corresponding eigenvalues of the BFKL kernel are given by
$$\chi _h\chi (\nu ,n)=\overline{\alpha }_s\left(\psi (\frac{1+|n|}{2}+i\nu )+\psi (\frac{1+|n|}{2}i\nu )2\psi (1)\right)=\overline{\alpha }_sϵ_h.$$
(14)
The action of the BFKL kernel can also be written, again on the space of Möbius functions, after a duality transformation , in terms of integral operator appearing in the dipole picture
$$H_{12}f_\omega (𝝆_1,𝝆_2)=\frac{d^2𝝆_3}{\pi }\frac{\left|𝝆_{12}\right|^2}{\left|𝝆_{13}\right|^2\left|𝝆_{23}\right|^2}\left(f_\omega (𝝆_1,𝝆_2)f_\omega (𝝆_1,𝝆_3)f_\omega (𝝆_2,𝝆_3)\right).$$
(15)
We now consider the pomeron eigenstates in the momentum representation. Different forms of the Fourier transform of the function in eq. (13) have been given in the most general case. One finds a sum of an analytic term $`\stackrel{~}{E}_{h,\overline{h}}^A`$, which has been given in an explicit way or in a integral form, and a $`\delta `$-like one $`\stackrel{~}{E}_{h,\overline{h}}^\delta `$. More precisely we obtain
$`\stackrel{~}{E}_{h,\overline{h}}(𝒌_1,𝒌_2)`$ $`=`$ $`{\displaystyle \frac{d^2𝒓_1}{(2\pi )^2}\frac{d^2𝒓_2}{(2\pi )^2}\left(\frac{r_{12}}{r_1r_2}\right)^h\left(\frac{r_{12}^{}}{r_1^{}r_2^{}}\right)^{\overline{h}}e^{i(𝒌_1𝒓_1+𝒌_2𝒓_2)}}=\stackrel{~}{E}_{h,\overline{h}}^A(𝒌_1,𝒌_2)+\stackrel{~}{E}_{h,\overline{h}}^\delta (𝒌_1,𝒌_2)`$ (16)
$`=`$ $`k|h=k|h^A+k|h^\delta ,`$
where a bra-ket compact notation is also introduced. By explicit computation one finds
$$\stackrel{~}{E}_{h,\overline{h}}^\delta (𝒌_1,𝒌_2)=\left[\delta ^{(2)}(𝒌_1)+(1)^n\delta ^{(2)}(𝒌_2)\right]\frac{i^n}{2\pi }2^{1h\overline{h}}\frac{\mathrm{\Gamma }(1\overline{h})}{\mathrm{\Gamma }(h)}q^{\overline{h}1}q^{h1},$$
(17)
where the complex form $`v=v_x+iv_y`$ of any bidimensional transverse vector $`𝒗=(v_x,v_y)`$ (coordinate or momentum) is used. In the calculation of a physical cross section one performs integrations with two-gluon colourless impact factors and this $`\delta `$-like term gives zero contribution, so that all the physics is related to the analytic term. In fact the $`\delta `$-terms have been introduced when one has chosen to move to the Möbius representation which is characterized by the simple and beautiful form in the coordinate representation.
The analytic part can be written, for example, as given in wherein one finds
$$\stackrel{~}{E}_{h\overline{h}}^A(𝒌_1,𝒌_2)=C_v\left(X(𝒌_1,𝒌_2)+(1)^nX(𝒌_2,𝒌_1)\right),$$
(18)
with the coefficient $`C_v`$ given by
$$C_v=\frac{(i)^n}{(4\pi )^2}h\overline{h}(1h)(1\overline{h})\mathrm{\Gamma }(1h)\mathrm{\Gamma }(1\overline{h}).$$
(19)
The expression for $`X`$ in complex notation was written in terms of the hypergeometric functions
$`X(𝒌_1,𝒌_2)=\left({\displaystyle \frac{k_1}{2}}\right)^{\overline{h}2}\left({\displaystyle \frac{k_2^{}}{2}}\right)^{h2}{}_{2}{}^{}F_{1}^{}\left(\begin{array}{cc}1h,2h& \\ 2& \end{array}|{\displaystyle \frac{k_1^{}}{k_2^{}}}\right){}_{2}{}^{}F_{1}^{}\left(\begin{array}{cc}1\overline{h},2\overline{h}& \\ 2& \end{array}|{\displaystyle \frac{k_2}{k_1}}\right).`$ (24)
Another expression for $`\stackrel{~}{E}_{h\overline{h}}^A(𝒌_1,𝒌_2)`$ which will be very useful for our purposes has been given in in an integral form:
$$\stackrel{~}{E}_{h\overline{h}}^A(𝒌_1,𝒌_2)=C_l\frac{1}{𝒌_1^2𝒌_2^2}d^2𝒑\left[\frac{p(qp)}{k_1p}\right]^{\overline{h}1}\left[\frac{p^{}(q^{}p^{})}{k_1^{}p^{}}\right]^{h1},$$
(25)
where
$$C_l=(1)^n\frac{i^{\overline{h}h}h\overline{h}}{2^{h+\overline{h}}\pi ^3}\frac{\mathrm{\Gamma }(1h)}{\mathrm{\Gamma }(\overline{h})}.$$
(26)
Below we shall consider the completeness relation and the spectral representation for the BFKL kernel (cf. Ref. ).
Let us before discuss the scalar product in the space of functions where the kernel is acting on. For the kernel in the momentum space associated to the Feynman diagram derivation there is only one scalar product available. If we consider non amputated functions of two momenta $`𝒌_1,𝒌_2`$, i.e. with 2-dimensional propagators $`1/(𝒌_1^2𝒌_2^2)`$ included, the scalar product is defined as
$$f|g𝑑\mu (𝒌)f^{}(𝒌_1,𝒌_2)g(𝒌_1,𝒌_2),$$
(27)
with the integration measure
$$d\mu (𝒌)=𝒌_1^2𝒌_2^2\delta ^2(𝒒𝒌_1𝒌_2)d^2𝒌_1d^2𝒌_2$$
(28)
which kills the extra propagators.
In the Möbius space of functions (in coordinate space) for the principal series of the Möbius group representation, whenever functions with conformal weight $`h=0`$ or $`h=1`$ are not considered, two possible choices are available . Since the Möbius functions contain propagators, one possibility is to remove, as before, the propagators in one function, acting with the operator $`\mathbf{}_1^2\mathbf{}_2^2`$. The other possibility, related to the form of the Casimir operator of the Möbius group, is to use instead the measure $`d^2𝝆_1d^2𝝆_2/|\rho _{12}|^4`$. The latter is not equivalent to the former when we consider the conformal wheights $`h=0`$ or $`h=1`$. Note, that, providing that we go from the Regge kinematics to the deep-inelastic scattering in the Bjorken regime, the additional series of the unitary representations of the Möbius group should be also used (cf. Ref. ).
We now consider the completeness relation in coordinate space , in the space of Möbius functions, where the eigenfunctions of the Casimir operator of the Möbius group, defined in (13), constitute a suitable spectral basis for the operators acting on the space of the colourless impact factors, which, we stress, never “feel” the presence of the terms with a $`\delta `$-distribution behavior in momentum space, i.e. those of eq. (17). For the operators amputated (spatial propagators removed) to the left (with notation $`\widehat{1}_L`$ and $`\widehat{\overline{K}}_{12}^{(1)}`$) one can write
$`\rho |\widehat{1}_L|\rho ^{}`$ $``$ $`(2\pi )^4\delta ^2(𝝆_{11^{}})\delta ^2(𝝆_{22^{}})={\displaystyle d^2𝝆_0\underset{h}{}\frac{N_h}{|𝝆_{12}|^4}E_{h,\overline{h}}(𝝆_{10},𝝆_{20})E_{h,\overline{h}}^{}(𝝆_{1^{}0},𝝆_{2^{}0})},`$
$`\rho |\widehat{\overline{K}}_{12}^{(1)}|\rho ^{}`$ $`=`$ $`{\displaystyle d^2𝝆_0\underset{h}{}\frac{N_h}{|𝝆_{12}|^4}E_{h,\overline{h}}(𝝆_{10},𝝆_{20})\chi _hE_{h,\overline{h}}^{}(𝝆_{1^{}0},𝝆_{2^{}0})},`$ (29)
where we have defined the weights $`N_h`$, as well as $`\stackrel{~}{N_h}`$ for later use,
$$N_h=16(\nu ^2+n^2/4),\stackrel{~}{N}_h=(2\pi )^2\frac{\nu ^2+n^2/4}{[\nu ^2+(n1)^2/4][\nu ^2+(n+1)^2/4]}$$
(30)
and $`_h_n𝑑\nu `$. If we try to extrapolate these operators to a wider domain which contains functions which no longer vanish for zero momenta, contrary to the colorless impact factor case, one needs some care. For example, for the function, which in momentum space depends only on the total momentum, just the case we meet in the bootstrap relation in LLA, one obtains the the result $`\delta ^2(\rho _{12})`$ in coordinate space (for the amputated impact factor). But such generalized function is orthogonal to the basis constituted by the functions in eq. (13) since $`d^2𝝆_1^{}E_{h,\overline{h}}^{}(\rho _{1^{}0},\rho _{2^{}0})\delta ^2(\rho _{1^{}2^{}})=0`$ and therefore the application of the operators in eq. (29) on this function will give zero. In other words the bootstrap relation cannot be fullfilled in such a framework with the spectral representations given in eq. (29).
But this should not be surprising, since the choice done thanks to the equivalence relation in eq. (8), is based on a restriction of the action of the kernel to the functions similar to colorless impact factors.
## 3 Deformed representation
Moving to the momentum representation, one may write the Fourier transform of the relations in (29) and the same considerations done above may be repeated in this case, leading to the same conclusions. In particular one cannot write a bootstrap relation in a spectral basis built upon the states of eq. (16). On the other hand one notices that the BFKL kernel $`K_{12}^{(1)}`$ in momentum space, as constructed from the Feynman diagram analysis, has an analytic behavior which must be reflected also in its spectral decomposition. It can also be noted that the behavior in $`|𝒒|`$ of the $`\delta `$-like term (17) is $`|𝒒|^{1+2i\nu }`$, which is singular in the forward limit where analytic behavior is expected for fixed $`\stackrel{}{k}_1=\stackrel{}{k}_2`$.
Therefore it is natural to suggest the modified relations for the eigenfunction completeness and for the spectral representation of the BFKL kernel:
$`k|\widehat{1}_L|k^{}`$ $``$ $`1={\displaystyle \underset{h}{}}\stackrel{~}{N}_h\stackrel{~}{E}_{h\overline{h}}^A(𝒌_1,𝒌_2)\stackrel{~}{E}_{h\overline{h}}^A(𝒌_1^{},𝒌_2^{}),`$
$`k|\widehat{\overline{K}}_{12}^{(1)}|k^{}`$ $`=`$ $`{\displaystyle \underset{h}{}}\stackrel{~}{N}_h\stackrel{~}{E}_{h\overline{h}}^A(𝒌_1,𝒌_2)\chi _h\stackrel{~}{E}_{h\overline{h}}^A(𝒌_1^{},𝒌_2^{}),`$ (31)
$`k|\widehat{G}_{12}^{(1)}(y)|k^{}`$ $`=`$ $`{\displaystyle \underset{h}{}}\stackrel{~}{N}_h\stackrel{~}{E}_{h\overline{h}}^A(𝒌_1,𝒌_2)e^{y\chi _h}\stackrel{~}{E}_{h\overline{h}}^A(𝒌_1^{},𝒌_2^{}),`$
where only the analytic contribution from each state of the spectral basis is used and the measure for integration is defined in eq. (28), according to our choice of considering functions with propagators included. Below we shall verify this anzatz.
Firstly we consider a space of functions (with removed propagators) which includes functions corresponding to the colorless impact factors and at least one function depending only on the total momentum, which is associated to a particular colored impact factor. Secondly we modify the spectral representation of the LL BFKL kernel, wherein the eigenvalues are assumed to be the same expressions of the conformal weights as for the restriction on the Möbius space of function, and the eigenfunctions are deformed in order to be analytic in the momentum space. A similar idea was already considered in the literature in an attempt to couple the LL BFKL pomeron to a quark .
We proceed now with the definition of a transformation between the Möbius (M) basis and the analytic Feynman (AF) basis. Let us observe that one can go from one basis to the other by a simple transformation, which reads in coordinate representation as follows
$`\mathrm{\Phi }^1:\mathrm{M}\mathrm{AF}`$ , $`E_h^A(𝝆_{10},𝝆_{20})=E_h^M(𝝆_{10},𝝆_{20})\underset{\rho _1\mathrm{}}{lim}E_h^M(𝝆_{10},𝝆_{20})\underset{\rho _2\mathrm{}}{lim}E_h^M(𝝆_{10},𝝆_{20})`$ (32)
$`=E_h^M(𝝆_{10},𝝆_{20})2^{h\overline{h}}\left(E_h^M(𝝆_{20},𝝆_{20})+(1)^nE_h^M(𝝆_{10},𝝆_{10})\right),`$
$`\mathrm{\Phi }:\mathrm{AF}\mathrm{M}:`$ , $`E_h^M(𝝆_{10},𝝆_{20})=E_h^A(𝝆_{10},𝝆_{20})+{\displaystyle \frac{1}{2^{h+\overline{h}}2}}\left(E_h^A(𝝆_{20},𝝆_{20})+(1)^nE_h^A(𝝆_{10},𝝆_{10})\right).`$
These relations are clearly non local and have been obtained by noting that
$$E_h^M(𝝆,𝝆)=2^{h+\overline{h}}\left(\frac{1}{\rho }\right)^h\left(\frac{1}{\rho ^{}}\right)^{\overline{h}},E_h^A(𝝆,𝝆)=(2^{h+\overline{h}}2)\left(\frac{1}{\rho }\right)^h\left(\frac{1}{\rho ^{}}\right)^{\overline{h}}.$$
(33)
For the case of even conformal spins ($`n`$) the second transformation can be written in a simple local form on observing that
$$\frac{1}{2^{h+\overline{h}}2}\left(E_h^A(𝝆_{20},𝝆_{20})+(1)^nE_h^A(𝝆_{10},𝝆_{10})\right)=\frac{1}{2}\left(E_h^A(𝝆_{10},𝝆_{10})+(1)^nE_h^A(𝝆_{20},𝝆_{20})\right)$$
(34)
while for odd conformal spin such a simple transformation is not possible since $`E_h^A(\rho ,\rho )=0`$. It is related to the fact, that in the last case both functions $`E_h^M(\rho _{10},\rho _{20})`$ and $`E_h^A(\rho _{10},\rho _{20})`$ satisfy the colour transparancy property $`E_h^{M,A}(\rho ,\rho )=0`$, but only $`E_h^M(\rho _{10},\rho _{20})`$ has the simple conformal properties.
It can be easily checked that
$$\mathrm{\Phi }\mathrm{\Phi }^1I_M,\mathrm{\Phi }^1\mathrm{\Phi }I_{AF}$$
(35)
so that $`\mathrm{\Phi }`$ is a 1-1 mapping with inverse really given by $`\mathrm{\Phi }^1`$.
The Green’s function for the evolution in the rapidity $`y`$ based on the deformed analytic spectral basis was also written above (see (31)). In the next section we shall show that with this prescription one is recovering the correct relation compatible with the bootstrap requirement for the gluon reggeization in LLA.
Since the spectrum in both representations is the same it is natural to expect that the BFKL kernel satisfying the bootstrap relation is conformal invariant, in the new deformed analytic Feynman basis obtained by a similarity transformation given by the operator $`\mathrm{\Phi }`$. Namely, the well known generators of the Möbius group $`\stackrel{}{M}_r`$ could be extended to this basis as follows
$$\stackrel{}{M}_r^{AF}=\mathrm{\Phi }^1\stackrel{}{M}_r\mathrm{\Phi }.$$
(36)
Let us summarize the result. If one considers the impact factors derived from Feynman diagrams calculations, they are the functions belonging to the AF-space and not to the M-space. We remind that we have two different completeness relations based on the two spectral basis $`E_h^{AF}`$ and $`E_h^M`$ and in these two spaces there are different definitions for scalar products and normalizations of wave functions. Nevertheless the cardinality of the two basis is the same. The two spaces are related by the 1-1 mapping $`\mathrm{\Phi }:`$ AF-space $``$ M-space as previously discussed (see also Fig. 1) and one space is shifted with respect to the other. Moreover any operator $`O^{AF}`$ in the AF-space which can be decomposed on the spectral basis can be defined on the M-space as $`O^M=\mathrm{\Phi }O^{AF}\mathrm{\Phi }^1`$ and viceversa $`O^{AF}=\mathrm{\Phi }^1O^M\mathrm{\Phi }`$. This can be seen as an isometry between the Möbius space and the analytical Feynman space with the different scalar products in these spaces.
The next interesting question is therefore if it is possible to construct explicitely the new representation defined in eq. (36). We shall study this problem in the section $`5`$, starting from analyzing the Fourier transform of the Möbius algebra in the M-space and later on moving to the AF-space. In the next section we will show that with the meromorphic prescription we indeed recover the bootstrap properties of the BFKL kernel.
## 4 AF representation and the bootstrap relation
For the sake of using a compact notation, let us introduce for some functions of transverse gluon momenta the corresponding wave functions for a color singlet state. These are the power $`|P_\lambda `$, the unity $`|U`$ and the log $`|L`$ states (modulus the propagators), defined by their representations in the momentum space:
$`k|P_\lambda `$ $``$ $`{\displaystyle \frac{1}{𝒌_1^2𝒌_2^2}}\left({\displaystyle \frac{𝒒^4}{𝒌_1^2𝒌_2^2}}\right)^\lambda ,`$
$`k|U`$ $``$ $`{\displaystyle \frac{1}{𝒌_1^2𝒌_2^2}}=\left(k|P_\lambda \right)_{\lambda =0},`$
$`k|L`$ $``$ $`{\displaystyle \frac{1}{𝒌_1^2𝒌_2^2}}\mathrm{log}\left({\displaystyle \frac{𝒒^4}{𝒌_1^2𝒌_2^2}}\right)={\displaystyle \frac{d}{d\lambda }}\left(k|P_\lambda \right)_{\lambda =0}.`$ (37)
Since, as previoulsy discussed, one has for the Möbius-invariant wave function $`h`$ the equality $`h|U=0`$ we obtain
$$h^A|U=h^\delta |U.$$
(38)
Let us recall the completeness relation and the BFKL kernel written as
$`\widehat{1}_L`$ $`=`$ $`{\displaystyle \underset{h}{}}\stackrel{~}{N}_h|h^Ah^A|,`$
$`\widehat{\overline{K}}_{12}^{(1)}`$ $`=`$ $`{\displaystyle \underset{h}{}}\stackrel{~}{N}_h|h^A\chi _hh^A|.`$ (39)
On using the previous relations we write the bootstrap relation in the form of eq. (7) as
$$\overline{\alpha }_s\underset{h}{}\stackrel{~}{N}_h|h^Aϵ_hh^A|U=\frac{\overline{\alpha }_s}{2}\underset{h}{}\stackrel{~}{N}_h|h^Ah^A|L,$$
(40)
where eq. (14) was used. In order to show that our definitions (39) are correct, we should prove that the coefficients of the conformal basis $`|h^A`$ on both the l.h.s. and the r.h.s. do coincide, i.e.
$$ϵ_hh^A|U=\frac{1}{2}h^A|L.$$
(41)
It is therefore enough to calculate the integral given by $`h^A|P_\lambda `$. Verifying that this expression is finite it is sufficient also to calculate the first two terms of the Taylor expansion in $`\lambda `$ to verify the eq. (41), thanks to eq. (37).
We also note that the left hand side of eq. (41) can be calculated easily in an independent way using the relations in eq. (38) and (17), which give:
$$h^A|U=\delta _{n,\mathrm{even}}\frac{i^n}{2\pi }2^{h+\overline{h}}\frac{\mathrm{\Gamma }(h)}{\mathrm{\Gamma }(1\overline{h})}q^hq^{\overline{h}},$$
(42)
where $`\delta _{n,\mathrm{even}}=[1+(1)^n]/2`$ has support (and value $`1`$) for any even integer number.
Let us now find explicitely $`h^A|P_\lambda `$. It is convenient to perform the calculations in the momentum space where we have defined the analytic element $`|h^A`$ of the conformal basis. Taking for it the expression given in eq. (25) we write:
$`h^A|P_\lambda `$ $`=`$ $`{\displaystyle d^2𝒌_1d^2𝒌_2\delta ^{(2)}(𝒒𝒌_1𝒌_2)\stackrel{~}{E}_{h\overline{h}}^A(𝒌_1,𝒌_2)\left(\frac{𝒒^4}{𝒌_1^2𝒌_2^2}\right)^\lambda }`$ (43)
$`=`$ $`C_l^{}|𝒒|^{4\lambda }{\displaystyle }d^2𝒌{\displaystyle }d^2𝒑p^{\overline{h}}(qp)^{\overline{h}}(kp)^{\overline{h}}k^{1\lambda }(qk)^{1\lambda }\times (\mathrm{h}.\mathrm{c}.).`$
Using new integration variables induced by the change $`x=p/q`$ and $`y=q/k`$ with analogous relations for the complex conjugated ones, we obtains
$$h^A|P_\lambda =C_l^{}q^{\overline{h}}q^hI_\lambda (1),$$
(44)
where
$$I_\lambda (z)=d^2𝒙d^2𝒚x^{\overline{h}}(1x)^{\overline{h}}y^{2\lambda \overline{h}}(1y)^{1\lambda }(1xyz)^{\overline{h}}\times (h.c.).$$
(45)
It is convenient to define the function $`I_\lambda (z)`$ because one can find an explicit expression for it in terms of the generalized hypergeometric functions $`{}_{3}{}^{}F_{2}^{}`$. We illustrate some details of the calculation in appendix A.
It is sufficient to represent the result of the integration in power series of $`\lambda `$ for the point $`z=1`$:
$$I_\lambda (1)=\underset{m}{}\frac{1}{m!}I_\lambda ^{(m)}(1)|_{\lambda =0}\lambda ^m.$$
(46)
Recalling the first two terms in the expansion (see eq. (113) in appendix A) one can see, on using the definitions of eq. (37) and (44), which imply $`h^A|U=(h^A|P_\lambda )_{\lambda =0}`$ and $`h^A|L=(dh^A|P_\lambda /d\lambda )_{\lambda =0}`$, that the relation in eq. (41) is verified and therefore also the bootstrap relation. As a check one may evaluate explicitely the quantity
$$(h^A|P_\lambda )_{\lambda =0}=C_l^{}q^{\overline{h}}q^h\frac{2\pi ^2}{(1h)(1\overline{h})}\delta _{n,\mathrm{even}},$$
(47)
using the eq. (26). It is easy to show that it coincides with the expression of eq. (42), computed in a completely different way.
In order to give an alternative check of the representation proposed, we have also calculated explicitly the third term of the expansion of the BFKL Green’s function $`G_{12}^{(1)}(y)=e^{y\overline{K}_{12}^{(1)}}1`$ (the first term is trivial and the second corresponds to the bootstrap relation), which contains the double iteration of the kernel acting on the unity (momentum space). The details are in the appendix B.
Using the spectral representation one can compute an arbitrary number of iterations of the BFKL kernel or the action of the BFKL Green’s function on the impact factor of a quark or gluon line.
## 5 The deformed representation of the Möbius (conformal) algebra
In this section we will examine the representation of the $`𝔰𝔩(2,)`$ algebra in the $`AF`$ space of funtions. To this end we start from the familiar Möbius algebra in the $`M`$ space and transform it into momentum space. We then switch to the $`AF`$ space: in order to retain the correct structure of the algebra we have to construct a new representation of the generator $`M_{}`$.
Let us reconsider the generators of the conformal Möbius group in $`2`$ dimensions in coordinate space given in eq. (11) and the corresponding Casimir operator (12). We will be interested in writing similar relations in momentum space. We start from the conformal eigenfunctions (13) which from now on will carry the superscript ’$`M`$’, and we define their Fourier transforms:
$`E_{h\overline{h}}^M(𝝆_0;𝝆_1,𝝆_2)`$ $`=`$ $`\left({\displaystyle \frac{\rho _{12}}{\rho _{10}\rho _{20}}}\right)^h\left({\displaystyle \frac{\rho _{12}^{}}{\rho _{10}^{}\rho _{20}^{}}}\right)^{\overline{h}},`$ (48)
$`E_{h\overline{h}}^M(𝒒;𝒌_1,𝒌_2)=[E_{h\overline{h}}^M](𝒒;𝒌_1,𝒌_2)`$ $`=`$ $`{\displaystyle 𝑑\mu e^{i𝒒𝝆_0+i𝒌_1𝝆_1+i𝒌_2𝝆_2}E_{h\overline{h}}^M(𝝆_0;𝝆_1,𝝆_2)}=`$ (49)
$`=`$ $`(2\pi )^2\delta ^{(2)}(𝒒𝒌_1𝒌_2)\stackrel{~}{E}_{h\overline{h}}^M(𝒌_1,𝒌_2),`$
$`\stackrel{~}{E}_{h\overline{h}}^M(𝝆_0;𝒌_1,𝒌_2)=\stackrel{~}{}[E_{h\overline{h}}^M](𝝆_0;𝒌_1,𝒌_2)`$ $`=`$ $`{\displaystyle 𝑑\stackrel{~}{\mu }e^{i𝒌_1𝝆_1+i𝒌_2𝝆_2}E_{h\overline{h}}^M(𝝆_0;𝝆_1,𝝆_2)}=`$ (50)
$`=`$ $`e^{i(𝒌_1+𝒌_2)𝝆_0}\stackrel{~}{E}_{h\overline{h}}^M(𝒌_1,𝒌_2),`$
$`\stackrel{~}{E}_{h\overline{h}}^M(𝒌_1,𝒌_2)=\stackrel{~}{}[E_{h\overline{h}}^M](𝝆_0=0;𝒌_1,𝒌_2)`$ $`=`$ $`{\displaystyle 𝑑\stackrel{~}{\mu }e^{i𝒌_1𝝆_1+i𝒌_2𝝆_2}E_{h\overline{h}}^M(0;𝝆_1,𝝆_2)},`$ (51)
where $`d\mu =d^2𝝆_0d^2𝝆_1d^2𝝆_2`$, $`d\stackrel{~}{\mu }=d^2𝝆_1d^2𝝆_2`$ and $``$, $`\stackrel{~}{}`$ are the Fourier transform operators in three and two variables respectively. In coordinate space the action of the generators on the eigenfunctions $`E_{h\overline{h}}^M(𝝆_{10},𝝆_{20})`$ is:
$`M^+E_{h\overline{h}}^M(𝝆_0;𝝆_1,𝝆_2)`$ $`=`$ $`_0E_{h\overline{h}}^M(𝝆_0;𝝆_1,𝝆_2),`$ (52)
$`M^3E_{h\overline{h}}^M(𝝆_0;𝝆_1,𝝆_2)`$ $`=`$ $`(h\rho _0_0)E_{h\overline{h}}^M(𝝆_0;𝝆_1,𝝆_2),`$ (53)
$`M^{}E_{h\overline{h}}^M(𝝆_0;𝝆_1,𝝆_2)`$ $`=`$ $`(2h\rho _0+\rho _0^2_0)E_{h\overline{h}}^M(𝝆_0;𝝆_1,𝝆_2).`$ (54)
Note, that the functions $`E_{h\overline{h}}^M(𝝆_0;𝝆_1,𝝆_2)`$ can be considered as the Clebsch-Gordon coefficients in the expansion of the product of the Reggeon wave functions $`\phi (𝝆_r)`$ in the sum of the irreducible representations
$$\phi (𝝆_1)\phi (𝝆_2)=_{\mathrm{}}^{\mathrm{}}𝑑\nu \underset{n=\mathrm{}}{\overset{\mathrm{}}{}}a_{\nu n}d^2\rho _0E_{h\overline{h}}^M(𝝆_0;𝝆_1,𝝆_2)O_{h\overline{h}}(𝝆_0),$$
(55)
where the conformal weights for the principal series of unitary representations are
$$h=\frac{1}{2}+i\nu +\frac{n}{2},\overline{h}=\frac{1}{2}+i\nu \frac{n}{2}$$
(56)
and the coefficients $`a_{\nu n}`$ are fixed with the use of the normalization conditions for the wave functions $`\phi (𝝆_r)`$ and $`O_{h\overline{h}}(𝝆_0)`$. In the irreducible representations $`O_{h\overline{h}}(𝝆_0)`$ the Möbius group generators are given below
$`M^+O_{h\overline{h}}(𝝆_0)`$ $`=`$ $`_0O_{h\overline{h}}(𝝆_0),`$ (57)
$`M^3O_{h\overline{h}}(𝝆_0)`$ $`=`$ $`(1h+\rho _0_0)O_{h\overline{h}}(𝝆_0),`$ (58)
$`M^{}O_{h\overline{h}}(𝝆_0)`$ $`=`$ $`(2(1h)\rho _0\rho _0^2_0)O_{h\overline{h}}(𝝆_0).`$ (59)
The generators of the Möbius group can be translated into the momentum space, using the relation $`𝝆𝒌=(\rho ^{}k+\rho k^{})/2`$:
$`M^+\stackrel{~}{E}_{h\overline{h}}^M(𝝆_0;𝒌_1,𝒌_2)`$ $`=`$ $`e^{i(𝒌_1+𝒌_2)𝝆_0}\stackrel{~}{M}^+(𝒌)\stackrel{~}{E}_{h\overline{h}}^M(𝒌_1,𝒌_2),`$ (60)
$`M^3\stackrel{~}{E}_{h\overline{h}}^M(𝝆_0;𝒌_1,𝒌_2)`$ $`=`$ $`e^{i(𝒌_1+𝒌_2)𝝆_0}\left(\stackrel{~}{M}^3(𝒌)+\rho _0\stackrel{~}{M}^+(𝒌)\right)\stackrel{~}{E}_{h\overline{h}}^M(𝒌_1,𝒌_2),`$ (61)
$`M^{}\stackrel{~}{E}_{h\overline{h}}^M(𝝆_0;𝒌_1,𝒌_2)`$ $`=`$ $`e^{i(𝒌_1+𝒌_2)𝝆_0}\left(\stackrel{~}{M}^{}(𝒌)2\rho _0\stackrel{~}{M}^3(𝒌)\rho _0^2\stackrel{~}{M}^+(𝒌)\right)\stackrel{~}{E}_{h\overline{h}}^M(𝒌_1,𝒌_2),`$ (62)
where
$`\stackrel{~}{M}^+(𝒌)`$ $`=`$ $`{\displaystyle \frac{i}{2}}(k_1^{}+k_2^{}),`$ (63)
$`\stackrel{~}{M}^3(𝒌)`$ $`=`$ $`(_{k_1^{}}k_1^{}+_{k_2^{}}k_2^{}),`$ (64)
$`\stackrel{~}{M}^{}(𝒌)`$ $`=`$ $`2i(_{k_1^{}}^2k_1^{}+_{k_2^{}}^2k_2^{}).`$ (65)
One can calculate also the action of some bilinear combinations of the generators in the momentum representation
$$M^+M^{}\stackrel{~}{E}_{h\overline{h}}^M(𝝆_0;𝒌_1,𝒌_2)=e^{i(𝒌_1+𝒌_2)𝝆_0}\stackrel{~}{M}^+(𝒌)\left(\stackrel{~}{M}^{}(𝒌)2\rho _0\stackrel{~}{M}^3(𝒌)\rho _0^2\stackrel{~}{M}^+(𝒌)\right)\stackrel{~}{E}_{h\overline{h}}^M(𝒌_1,𝒌_2),$$
$$\left(M^3(M^3+1)+M^+M^{}\right)\stackrel{~}{E}_{h\overline{h}}^M(𝝆_0;𝒌_1,𝒌_2)=e^{i(𝒌_1+𝒌_2)𝝆_0}\stackrel{~}{M}^2\stackrel{~}{E}_{h\overline{h}}^M(𝒌_1,𝒌_2),$$
where the Casimir operator in the ”short” momentum representation is
$$\stackrel{~}{M}^2=\stackrel{~}{M}^3(𝒌)\left(\stackrel{~}{M}^3(𝒌)+1\right)+\stackrel{~}{M}^+(𝒌)\stackrel{~}{M}^{}(𝒌)=(_{k_1^{}}_{k_2^{}})^2k_1^{}k_2^{}.$$
(66)
Note that in our notations $`\stackrel{~}{M}^3`$ is a generator while $`\stackrel{~}{M}^2`$ is the Casimir operator. The algebra of these generators in momentum space is the same as in coordinate space:
$$[\stackrel{~}{M}_r^+,\stackrel{~}{M}_r^3]=\stackrel{~}{M}_r^+,[\stackrel{~}{M}_r^+,\stackrel{~}{M}_r^{}]=2\stackrel{~}{M}_r^3,[\stackrel{~}{M}_r^{},\stackrel{~}{M}_r^3]=\stackrel{~}{M}_r^{}.$$
While the action of the generator $`\stackrel{~}{M}^+`$ does not need special comments, it is interesting to look at others. One finds
$`\stackrel{~}{M}^3\stackrel{~}{E}_h^M(𝒌_1,𝒌_2)`$ $`=`$ $`h\stackrel{~}{E}_h^M(𝒌_1,𝒌_2),`$ (67)
$`\stackrel{~}{M}^{}\stackrel{~}{E}_h^M(𝒌_1,𝒌_2)`$ $`=`$ $`0.`$ (68)
The last relation follows from the action of the Casimir operator
$$\stackrel{~}{M}^2\stackrel{~}{E}_h^M=h(h1)\stackrel{~}{E}_h^M$$
(69)
and from relation (67). In particular, given that
$$\stackrel{~}{M}^2=\frac{1}{2}\left[\stackrel{~}{M}^+\stackrel{~}{M}^{}+\stackrel{~}{M}^{}\stackrel{~}{M}^+\right]+\left(\stackrel{~}{M}^3\right)^2=\stackrel{~}{M}^+\stackrel{~}{M}^{}+\stackrel{~}{M}^3\left(\stackrel{~}{M}^3+1\right),$$
(70)
where the commutation relation $`[\stackrel{~}{M}^+,\stackrel{~}{M}^{}]=2\stackrel{~}{M}^3`$ has been used, one can see that indeed the relation (68 ) is satisfied.
These relations mean that in the M-space in momentum representation the Möbius algebra is projected to
$`[\stackrel{~}{M}^+,\stackrel{~}{M}^3]\stackrel{~}{E}^M=\stackrel{~}{M}^+\stackrel{~}{E}^M,\stackrel{~}{M}^{}\stackrel{~}{M}^+\stackrel{~}{E}^M=2\stackrel{~}{M}^3\stackrel{~}{E}^M,0\stackrel{~}{E}^M=0\stackrel{~}{E}^M.`$ (71)
Let us now move to the $`AF`$ space. We start, in coordinate space, from the definition
$$E_h^A=E_h^ME_h^\delta ,$$
(72)
where, according to eq. (32),
$$E_h^\delta (𝝆_{10},𝝆_{20})=2^{1h\overline{h}}PE_h^M(𝝆_{10},𝝆_{20})=\left(\frac{1}{\rho _{20}}\right)^h\left(\frac{1}{\rho _{20}^{}}\right)^{\overline{h}}+\left(\frac{1}{\rho _{10}}\right)^h\left(\frac{1}{(\rho _{10})^{}}\right)^{\overline{h}}$$
(73)
and $`P`$ is a projector ($`P^2=P`$) defined by its action
$$Pf(x_1,x_2)=\frac{1}{2}\left(f(x_1,x_1)+f(x_2,x_2)\right).$$
(74)
In order to compute the action of the Möbius generators on the analytic part, $`E_h^A`$, of the conformal eigenfunction we first examine the singular piece, $`E_h^\delta `$.
First of all let us observe that the Casimir operator acting on $`E_h^\delta `$ gives zero, as it is trivially checked in the coordinate representation. This means that $`h`$ in such a case can be obtained with the action of a function of the Casimir (solving the eq. (69) for $`h`$) only before applying the projector $`P`$. After the application of $`P`$ one needs to define a new operator to extract the scaling parameter $`h`$. It is related to the fact, that $`E_h^A`$ is not an eigenfunction of the usual Casimir operator. We should construct a new Casimir operator $`\stackrel{~}{M}_{AF}^2`$ for this representation.
In coordinate representation we can deduce the following relations for generators
$`M^+E_h^\delta (𝝆_{10},𝝆_{20})=_0E_h^\delta (𝝆_{10},𝝆_{20}),`$ (75)
$`M^3E_h^\delta (𝝆_{10},𝝆_{20})=(h\rho _0_0)E_h^\delta (𝝆_{10},𝝆_{20}),`$ (76)
which coincides with their action for $`E_h^M`$ while the action of $`M^{}`$ is different. According to this fact the action of Möbius generators on $`E_h^A`$ can be written as
$`M^+E_h^A(𝝆_{10},𝝆_{20})=_0E_h^A(𝝆_{10},𝝆_{20}),`$ (77)
$`M^3E_h^A(𝝆_{10},𝝆_{20})=(h\rho _0_0)E_h^A(𝝆_{10},𝝆_{20})`$ (78)
and
$`M^{}E_h^A(𝝆_{10},𝝆_{20})=(\rho _0^2_0+2h\rho _0)E_h^A(𝝆_{10},𝝆_{20})\left[h\rho _{10}E_h^M(𝝆_{10},𝝆_{10})+h\rho _{20}E_h^M(𝝆_{20},𝝆_{20})\right]2^{h\overline{h}}.`$ (79)
Therefore only the last generator $`M^{}`$, needs to be redefined in the $`AF`$ space.
Let us consider now again the momentum space for these functions. First we analyze the action of the $`\stackrel{~}{M}^3`$ generator on $`\stackrel{~}{E}_h^A(𝒌_1,𝒌_2)`$, which we have written in terms of hypergeometric functions. One can immediately see that
$$\stackrel{~}{M}^3\stackrel{~}{E}_h^A(𝒌_1,𝒌_2)=h\stackrel{~}{E}_h^A(𝒌_1,𝒌_2)$$
(80)
for any $`𝒌_1+𝒌_2`$. This means again that we are not forced to use a non local operator $`\widehat{h}`$ (solving an equation with the corresponding Casimir $`\stackrel{~}{M}_{AF}^2`$) to extract the dimension $`h`$ of an eigenstate, as we had to do in the coordinate representation.
The $`E_h^\delta `$ function is instead very peculiar since we know that the Casimir gives zero. This fact is also verified directly in momentum space, due to the deltas present in $`\stackrel{~}{E}_h^\delta `$. Later we will show it explicitely. Instead we note that the action of the $`\stackrel{~}{M}^3`$ operator is good. Indeed, from the explicit form given in Eq. (17), we obtain
$`\stackrel{~}{M}^3E_h^\delta (𝒌_1,𝒌_2)=hE_h^\delta (𝒌_1,𝒌_2).`$ (81)
This is compatible with the fact that the $`\stackrel{~}{M}^3`$ operator has the same action on the $`\stackrel{~}{E}_h^M(𝒌_1,𝒌_2)`$ and $`\stackrel{~}{E}_h^A(𝒌_1,𝒌_2)`$ functions.
The fact that we may choose to extract the dimension $`h`$ in momentum space using the $`\stackrel{~}{M}^3`$ operator may be useful to simplify the construction of the mappings $`\mathrm{\Phi }`$ and $`\mathrm{\Phi }^1`$. On defining
$$F=2^{1\widehat{h}\widehat{\overline{h}}}$$
(82)
one may write the mapping introduced in eq. (32) as
$`\mathrm{\Phi }^1:MAF,\mathrm{\Phi }^1=1F^\delta P=1PF^M,`$ (83)
$`\mathrm{\Phi }:AFM,\mathrm{\Phi }=1+{\displaystyle \frac{F^\delta }{1F^\delta }}P=1+P{\displaystyle \frac{F^{AF}}{1F^{AF}}},`$ (84)
where in general $`PFP=FP`$ and which imply $`\mathrm{\Phi }\mathrm{\Phi }^1=\mathrm{\Phi }^1\mathrm{\Phi }=1`$ and we have denoted $`F^\delta =F(\widehat{h}(\stackrel{~}{M}^3))`$, $`F^M=F(\widehat{h}(\stackrel{~}{M}^2))`$ and similarly $`F^{AF}=F(\widehat{h}(\stackrel{~}{M}_{AF}^2))`$. Clearly in the relation for the mappings one can use different forms for the operator $`F`$ depending on which space it acts.
In momentum space $`\widehat{h}=\stackrel{~}{M}^3`$ then one can write $`[P,F]=0`$ (with some abuse of notation) and things are further simplified.
Proceeding along this line one can give a deformed representation $`\stackrel{~}{M}_{AF}^{}`$ which lives on the AF-space. In order to do this, let us first study
$$\stackrel{~}{M}^{}\stackrel{~}{E}_h^A=\stackrel{~}{M}^{}(\stackrel{~}{E}_h^M\stackrel{~}{E}_h^\delta )=\stackrel{~}{M}^{}\stackrel{~}{E}_h^\delta $$
(85)
and
$`\stackrel{~}{M}^{}\stackrel{~}{E}_h^\delta `$ $`=`$ $`2i(_{k_1^{}}^2k_1^{}+_{k_2^{}}^2k_2^{})\left[\delta ^{(2)}(𝒌_1)+(1)^n\delta ^{(2)}(𝒌_2)\right]c_h(k_1+k_2)^{\overline{h}1}(k_1^{}+k_2^{})^{h1}`$ (86)
$`=`$ $`2ih(h1){\displaystyle \frac{1}{k_1^{}+k_2^{}}}\stackrel{~}{E}_h^\delta =h(h1)\left(\stackrel{~}{M}^+\right)^1\stackrel{~}{E}_h^\delta `$
$`=`$ $`\left(\stackrel{~}{M}^+\right)^1h(h1){\displaystyle \frac{F}{1F}}P\stackrel{~}{E}_h^A.`$
Note that $`\stackrel{~}{M}^+`$ has the same form on any of the spaces of functions we consider. The resulting action of $`\stackrel{~}{M}^{}`$ on $`\stackrel{~}{E}_h^A`$ has to be studied in order to see if it is possible to define a new representation for $`\stackrel{~}{M}^{}`$ on AF-space which satisfies the conformal algebra. But before let us note that from the previous relation we have
$`\stackrel{~}{M}^+\stackrel{~}{M}^{}\stackrel{~}{E}_h^\delta =h(h1)\stackrel{~}{E}_h^\delta ,`$
$`\stackrel{~}{M}^{}\stackrel{~}{M}^+\stackrel{~}{E}_h^\delta =h(h+1)\stackrel{~}{E}_h^\delta ,`$ (87)
so that we reobtain, recalling eq. (70),
$$\stackrel{~}{M}^2\stackrel{~}{E}_h^\delta =0.$$
(88)
In general we have defined the Möbius generators on the AF space (spanned by the Basis functions $`\stackrel{~}{E}_h^A`$) using the isometry with the M-space:
$$\stackrel{~}{M}_{AF}=\mathrm{\Phi }^1\stackrel{~}{M}\mathrm{\Phi }.$$
(89)
Let us look at each of the three generators. From the previous discussion, also in the coordinate representation, we observe that
$$\stackrel{~}{M}_{AF}^+=\stackrel{~}{M}^+,\stackrel{~}{M}_{AF}^3=\stackrel{~}{M}^3,$$
(90)
which means that in momentum space the actions of these two operators is the same in both M-space and AF-space.
We therefore define for the AF-space a new generator in momentum representation such that
$$\stackrel{~}{M}_{AF}^{}\stackrel{~}{E}_h^A=0,$$
(91)
which puts into relation, as before, the action of the Casimir operator such that
$$\stackrel{~}{M}_{AF}^2\stackrel{~}{E}_h^A=h(h1)\stackrel{~}{E}_h^A,$$
(92)
together with the piece of the Möbius algebra $`[\stackrel{~}{M}_{AF}^+,\stackrel{~}{M}_{AF}^{}]=2\stackrel{~}{M}_{AF}^3`$ and the action of $`\stackrel{~}{M}_{AF}^3`$. According to the previous result (see eqs. (85) and (86)), by construction the choice is
$$\stackrel{~}{M}_{AF}^{}=\stackrel{~}{M}^{}\left(\stackrel{~}{M}^+\right)^1h(h1)\frac{F}{1F}P=\stackrel{~}{M}^{}\left(\stackrel{~}{M}^+\right)^1PG(\stackrel{~}{M}_{AF}^2),$$
(93)
where, in the last form, G is a non local operator such that
$$G\stackrel{~}{E}_h^A=h(h1)\frac{F(h)}{1F(h)}\stackrel{~}{E}_h^A.$$
(94)
Let us finally verify that, with this new representation of $`\stackrel{~}{M}_{AF}^{}`$, the algebra of the Möbius generators is correct. It is easy to see that
$$[\stackrel{~}{M}_{AF}^{},\stackrel{~}{M}_{AF}^3]=\stackrel{~}{M}_{AF}^{},$$
(95)
which is a relation that is projected to zero when applied to the function $`\stackrel{~}{E}_h^A`$. In order to check that the Möbius algebra is closed we need to examine the relation
$$[\stackrel{~}{M}_{AF}^{},\stackrel{~}{M}_{AF}^+]=+2\stackrel{~}{M}_{AF}^3$$
(96)
understood to act on $`\stackrel{~}{E}_h^A`$. One then has to verify that
$$\stackrel{~}{M}_{AF}^{}\stackrel{~}{M}_{AF}^+\stackrel{~}{E}_h^A=2h\stackrel{~}{E}_h^A.$$
(97)
Using eq. (87) we derive also
$$\stackrel{~}{M}^{}\stackrel{~}{M}^+\stackrel{~}{E}_h^A=\left(\stackrel{~}{M}^+\stackrel{~}{M}^{}+2\stackrel{~}{M}^3\right)\left(\stackrel{~}{E}_h^M\stackrel{~}{E}_h^\delta \right)=2h\stackrel{~}{E}_h^A+h(h1)\stackrel{~}{E}_h^\delta $$
(98)
and making the difference between eqs. (97) and (98) we therefore obtain
$$\left(\stackrel{~}{M}_{AF}^{}\stackrel{~}{M}^{}\right)\stackrel{~}{M}^+\stackrel{~}{E}_h^A=h(h1)\stackrel{~}{E}_h^\delta .$$
(99)
If $`\stackrel{~}{M}^+`$ commutes with $`(\stackrel{~}{M}_{AF}^{}\stackrel{~}{M}^{})`$ this is equivalent to
$$\left(\stackrel{~}{M}_{AF}^{}\stackrel{~}{M}^{}\right)\stackrel{~}{E}_h^A=h(h1)\left(\stackrel{~}{M}^+\right)^1\stackrel{~}{E}_h^\delta =\left(\stackrel{~}{M}^+\right)^1PG(\stackrel{~}{M}_{AF}^2)\stackrel{~}{E}_h^A,$$
(100)
which coincides with the starting definition of the generator. The commutation relation is fullfilled since the operator $`\stackrel{~}{M}^+`$ is just a multiplicative operator associated to the total momentum and therefore commutes with the action of the projector $`P`$ and moreover it commutes with the function of the Casimir $`G(\stackrel{~}{M}_{AF}^2)`$.
Finally let us look at the relation between the two Casimir operators in M-space and AF-space: using $`\stackrel{~}{M}_{AF}^2=\frac{1}{2}\left[\stackrel{~}{M}_{AF}^+\stackrel{~}{M}_{AF}^{}+\stackrel{~}{M}_{AF}^{}\stackrel{~}{M}_{AF}^+\right]+\left(\stackrel{~}{M}_{AF}^3\right)^2`$ we find from the previous relations
$$\stackrel{~}{M}_{AF}^2+PG(\stackrel{~}{M}_{AF}^2)=\stackrel{~}{M}^2.$$
(101)
This last relation, applied to $`\stackrel{~}{E}_h^A`$, gives
$$h(h1)\stackrel{~}{E}_h^A+h(h1)\stackrel{~}{E}_h^\delta =\stackrel{~}{M}^2\stackrel{~}{E}_h^A,$$
(102)
which indeed is compatible with $`\stackrel{~}{M}^2\stackrel{~}{E}_h^M=h(h1)\stackrel{~}{E}_h^M`$.
## 6 Conclusions
Among the properties of the LL BFKL kernel the bootstrap relation connected to the gluon reggeization is a fundamental consistency condition. This bootstrap property leads to an important feature of the BFKL kernel also in the color singlet state: this can be demonstrated most easily in momentum space where the BFKL kernel is meromorphic. Many previous investigations, exploiting the conformal symmetry of the kernel, have been carried out in the space of Möbius functions ($`M`$). In this space of functions, however, the bootstrap relation for the BFKL kernel is not satisfied.
In this paper we have defined a modified space of functions ($`AF`$), in which the bootstrap property is valid. In particular, we have derived a spectral representation of the kernel which is also useful in evaluating the coupling of the BFKL Green’s function to colored impact factors, and we have verified explicitly that the bootstrap relation is fulfilled. We also have derived the corresponding represention of the Möbius algebra; in order to act in this modified space of functions, one of the Möbius generators has to be deformed.
Our discussion has been limited to the the case of two-gluon Green’s functions. Since in both spaces, $`M`$ and $`AF`$, the eigenvalues of the Casimir operator, $`\stackrel{~}{M}^2`$ and $`\stackrel{~}{M}_{AF}^2`$ , resp., are the same, and since the hamiltonian for the pairwise interaction of two reggeized gluons can be expressed in terms of this Casimir operator, one can expect that, for states consisting of more than two gluons all remarkable properties of the multi-colour BFKL dynamics derived in the Möbius picture can be generalized to the AF picture. This includes, in particular, the holomorphic separability and integrability . We are going to return to these interesting problems in future publications.
## Acknowledgements
Part of this work has been done while one of us (L.N.L) has been visiting the II.Institut f.Theoretische Physik, University Hamburg, and the University of Bologna, Bologna. Two of us (J.B. and G.P.Vacca) gratefully acknowledge the hospitality of the Nuclear Physics Institute in Gatchina, St.Petersburg.
## Appendix A
The calculation of the integral in eq. (45) is carried on noting that one can find a third order differential operator in $`z`$ and another in $`z^{}`$, related to the $`{}_{3}{}^{}F_{2}^{}`$ functions, which, when applied to the integrand, gives a total derivative in an integration variable so that, on applying the Stokes theorem, one finds that $`I_\lambda (z)`$ satisfies the related differential equation in both the holomorphic and antiholomorphic sectors. This allows to write it as a sum of products of the linearly independent solutions of the two differential equations and on imposing the single valueness and the correct normalization (at some convenient point) the full expression is found . In particular this has the following, so called, conformal block structure
$$I_\lambda (z)=\underset{i=0}{\overset{2}{}}\lambda _iu_i(z)\overline{u}_i(z^{}),$$
(103)
where $`u_i`$ and $`\overline{u}_i`$ are three independent solutions of the generalized hypergeometric differential equations in $`z`$ and $`z^{}`$. The coefficients $`\lambda _i`$ depend through $`\mathrm{\Gamma }`$ functions on the conformal weights. The general form of $`I_\lambda (z)`$ is pretty complicated but it simplifies considerably for $`z=1`$ using the relations
$`{}_{3}{}^{}F_{2}^{}\left(\begin{array}{c}a,b,c\\ \frac{a+b+1}{2},2c\end{array}|1\right)`$ $`=`$ $`\pi ^{1/2}{\displaystyle \frac{\mathrm{\Gamma }\left(c+\frac{1}{2}\right)\mathrm{\Gamma }\left(\frac{1+a+b}{2}\right)\mathrm{\Gamma }\left(c+\frac{1ab}{2}\right)}{\mathrm{\Gamma }\left(\frac{1+a}{2}\right)\mathrm{\Gamma }\left(\frac{1+b}{2}\right)\mathrm{\Gamma }\left(c+\frac{1a}{2}\right)\mathrm{\Gamma }\left(c+\frac{1b}{2}\right)}},`$ (106)
$`{}_{3}{}^{}F_{2}^{}\left(\begin{array}{c}a,1a,c\\ f,2c+1f\end{array}|1\right)`$ $`=`$ $`\pi {\displaystyle \frac{\mathrm{\Gamma }\left(f\right)\mathrm{\Gamma }\left(2c+1f\right)2^{12c}}{\mathrm{\Gamma }\left(\frac{a+f}{2}\right)\mathrm{\Gamma }\left(\frac{1a+f}{2}\right)\mathrm{\Gamma }\left(c+\frac{1+af}{2}\right)\mathrm{\Gamma }\left(1+c\frac{a+f}{2}\right)}}.`$ (109)
On using the above relation, after algebraic simplifications, we find
$$I_\lambda (1)=I^{(0)}+I^{(1)}+I^{(2)}$$
(110)
with
$`I^{(0)}`$ $`=`$ $`{\displaystyle \frac{2^{3+4\lambda }\mathrm{\Gamma }\left(1\frac{h}{2}\right)\mathrm{\Gamma }\left(1\frac{\overline{h}}{2}\right)\mathrm{\Gamma }^2\left(1\lambda \right)\mathrm{\Gamma }^2\left(\lambda \right)\mathrm{\Gamma }\left(\frac{1h}{2}+\lambda \right)\mathrm{\Gamma }\left(\frac{1\overline{h}}{2}+\lambda \right)}{\mathrm{\Gamma }\left(\frac{3h}{2}\right)\mathrm{\Gamma }\left(\frac{3\overline{h}}{2}\right)\mathrm{\Gamma }\left(1\frac{h}{2}\lambda \right)\mathrm{\Gamma }\left(1\frac{\overline{h}}{2}\lambda \right)}}\times `$
$`\times {\displaystyle \frac{\mathrm{sin}\pi (\overline{h}2\lambda )\mathrm{sin}\pi \lambda \mathrm{tan}\pi \overline{h}}{\mathrm{sin}\pi (\overline{h}\lambda )}},`$
$`I^{(1)}`$ $`=`$ $`{\displaystyle \frac{2^{3+4\lambda }\mathrm{\Gamma }\left(\frac{h1}{2}\right)\mathrm{\Gamma }\left(\frac{\overline{h}1}{2}\right)\mathrm{\Gamma }^2\left(1\lambda \right)\mathrm{\Gamma }^2\left(\lambda \right)\mathrm{\Gamma }\left(\frac{h}{2}+\lambda \right)\mathrm{\Gamma }\left(\frac{\overline{h}}{2}+\lambda \right)}{\mathrm{\Gamma }\left(\frac{h}{2}\right)\mathrm{\Gamma }\left(\frac{\overline{h}}{2}\right)\mathrm{\Gamma }\left(\frac{1+h}{2}\lambda \right)\mathrm{\Gamma }\left(\frac{1+\overline{h}}{2}\lambda \right)}}\times `$
$`\times {\displaystyle \frac{\mathrm{sin}\pi (\overline{h}+2\lambda )\mathrm{sin}\pi \lambda \mathrm{tan}\pi \overline{h}}{\mathrm{sin}\pi (\overline{h}+\lambda )}},`$
$`I^{(2)}`$ $`=`$ $`{\displaystyle \frac{2^{4\lambda }\pi ^6\mathrm{\Gamma }^2\left(1\lambda \right)}{\mathrm{\Gamma }\left(\frac{3h}{2}\right)\mathrm{\Gamma }\left(\frac{h}{2}\right)\mathrm{\Gamma }\left(\frac{3\overline{h}}{2}\right)\mathrm{\Gamma }\left(\frac{\overline{h}}{2}\right)\mathrm{\Gamma }\left(\frac{1+h}{2}\lambda \right)\mathrm{\Gamma }\left(\frac{1+\overline{h}}{2}\lambda \right)}}\times `$ (111)
$`\times {\displaystyle \frac{1}{\mathrm{\Gamma }\left(1\frac{h}{2}\lambda \right)\mathrm{\Gamma }\left(1\frac{\overline{h}}{2}\lambda \right)\mathrm{\Gamma }^2\left(1+\lambda \right)}}{\displaystyle \frac{1}{\mathrm{sin}\pi (\overline{h}\lambda )\mathrm{sin}\pi (\overline{h}+\lambda )}}.`$
The expressions of $`I^{(0)}`$ and $`I^{(1)}`$ have a simple pole at $`\lambda =0`$ and opposite residues for this pole, while $`I^{(2)}`$ is holomorphic at $`\lambda =0`$, therefore the full function in eq. (110) is not singular at $`\lambda =0`$. Using the transformation properties of the Gamma function, each of three terms can be reduced to a common multiplicative factor and expressions involving only trigonometric functions. These terms are combined together to give the following compact form for $`I_\lambda (1)`$
$`I_\lambda (1)`$ $`=`$ $`{\displaystyle \frac{2}{(1h)(1\overline{h})}}\delta _{n,\mathrm{even}}\times `$ (112)
$`{\displaystyle \frac{\mathrm{\Gamma }\left(\frac{h}{2}+\lambda \right)\mathrm{\Gamma }\left(\frac{\overline{h}}{2}+\lambda \right)\mathrm{\Gamma }\left(\frac{1h}{2}+\lambda \right)\mathrm{\Gamma }\left(\frac{1\overline{h}}{2}+\lambda \right)}{\mathrm{\Gamma }\left(\frac{h}{2}\right)\mathrm{\Gamma }\left(\frac{\overline{h}}{2}\right)\mathrm{\Gamma }\left(\frac{1h}{2}\right)\mathrm{\Gamma }\left(\frac{1\overline{h}}{2}\right)}}\times `$
$`16^\lambda \left(1+i^n{\displaystyle \frac{\mathrm{sin}2\pi \lambda }{\mathrm{sin}\pi \left(\frac{h+\overline{h}}{2}\right)}}\right)\mathrm{\Gamma }^2\left(1\lambda \right)\mathrm{\Gamma }^2\left(\lambda \right)\mathrm{sin}^2\pi \lambda .`$
Now the first few derivatives with respect to $`\lambda `$ can be easily calculated. In particular we obtain, after some manipulations,
$`c_0I_\lambda (1)|_{\lambda =0}`$ $`=`$ $`{\displaystyle \frac{2\pi ^2}{(1h)(1\overline{h})}}\delta _{n,\mathrm{even}},`$
$`I_\lambda ^{}(1)|_{\lambda =0}`$ $`=`$ $`2ϵ_hc_0,`$
$`I_\lambda ^{\prime \prime }(1)|_{\lambda =0}`$ $`=`$ $`[\left(2ϵ_h\right)^2(1)^{h\overline{h}}\mathrm{\hspace{0.17em}4}\pi ^2\mathrm{csc}^2{\displaystyle \frac{h+\overline{h}}{2}}+`$ (113)
$`\psi ^{}\left({\displaystyle \frac{h}{2}}\right)+\psi ^{}\left({\displaystyle \frac{\overline{h}}{2}}\right)+\psi ^{}\left({\displaystyle \frac{1h}{2}}\right)+\psi ^{}\left({\displaystyle \frac{1\overline{h}}{2}}\right)]c_0.`$
## Appendix B
We recompute here the double iteration of the BFKL kernel in momentum space acting on a constant function and compare it with the expressions obtained with the use of the above spectral representation of the kernel (and therefore of the Green’s function). Let us write before the result of the direct integration in momentum space:
$`(\overline{K}_{12}^{(1)})^21`$ $`=`$ $`{\displaystyle \frac{1}{2}}\overline{\alpha }_s\overline{K}_{12}^{(1)}\mathrm{log}\left({\displaystyle \frac{𝒒^4}{𝒌_1^2𝒌_2^2}}\right)=`$ (114)
$`=`$ $`\left({\displaystyle \frac{1}{2}}\overline{\alpha }_s\right)^2\left[\mathrm{log}^2\left({\displaystyle \frac{𝒒^2}{𝒌_1^2}}\right)+\mathrm{log}^2\left({\displaystyle \frac{𝒒^2}{𝒌_2^2}}\right)\right].`$
Before giving some details of the computation let us consider what we obtain from the spectral approach. On using the spectral representation for the kernel, the relation (114) can be written as
$$k|\left(\widehat{\overline{K}}_{12}^{(1)}\right)^2|U=\left(\overline{\alpha }_s\right)^2\underset{h}{}\stackrel{~}{N}_hk|h^Aϵ_h^2h^A|U,$$
(115)
a relation which we have checked numerically. This check has been also performed in the case of the single iteration (bootstrap relation). The argument of the integration over $`\nu `$ oscillates with a very slow decay for fixed conformal spin, but summing over some tens of $`n`$ gives a good suppression of the tails and the integral can be computed.
Finally we sketch the derivation of eq. (114). There are three kinds of integral involved; they can be calculated with the dimensional regularization and the Feynman parameterization, in the case $`𝒌_{1,2}0,𝒌_1𝒌_2`$.
The needed integrals are:
$`\omega (𝒌_1)`$ $`=`$ $`{\displaystyle d^d𝒌^{}\frac{𝒌_1^2}{(𝒌^{})^{\mathrm{\hspace{0.17em}2}}(𝒌_1𝒌^{})^2}},`$
$`J_1(𝒌_1,𝒌_2)`$ $`=`$ $`{\displaystyle d^d𝒌^{}\frac{𝒌_1^2}{(𝒌^{})^{\mathrm{\hspace{0.17em}2}}(𝒌_1𝒌^{})^2}\mathrm{log}\frac{𝒒^2}{(𝒌^{})^{\mathrm{\hspace{0.17em}2}}}},`$ (116)
$`J_2(𝒌_1,𝒌_2)`$ $`=`$ $`{\displaystyle d^d𝒌^{}\frac{𝒌_1^2}{(𝒌^{})^{\mathrm{\hspace{0.17em}2}}(𝒌_1𝒌^{})^2}\mathrm{log}\frac{𝒒^2}{(𝒒𝒌^{})^{\mathrm{\hspace{0.17em}2}}}},`$
where $`d=2+2ϵ`$, $`𝒒=𝒌_1+𝒌_2`$ and $`\omega (𝒌_1)`$ is the gluon trajectory (2) rescaled by $`N_c/2c`$.
The first two integral can be easily calculated exactly, giving:
$`\omega (𝒌_1)`$ $`=`$ $`\pi ^{1+ϵ}{\displaystyle \frac{\mathrm{\Gamma }(1ϵ)\mathrm{\Gamma }^2(ϵ)}{\mathrm{\Gamma }(2ϵ)}}(𝒌_1^2)^ϵ=`$
$`=`$ $`\pi ^{1+ϵ}\mathrm{\Gamma }(1ϵ)(𝒒^2)^ϵ\left[{\displaystyle \frac{2}{ϵ}}2\mathrm{log}{\displaystyle \frac{𝒒^2}{𝒌_1^2}}+𝒪(ϵ)\right],`$
$`J_1(𝒌_1,𝒌_2)`$ $`=`$ $`\pi ^{1+ϵ}{\displaystyle \frac{\mathrm{\Gamma }(1ϵ)\mathrm{\Gamma }^2(ϵ)}{\mathrm{\Gamma }(2ϵ)}}\left[\pi \mathrm{cot}\pi ϵ+\mathrm{log}{\displaystyle \frac{𝒒^2}{𝒌_1^2}}+\psi (2ϵ)\psi (1)\right](𝒌_1^2)^ϵ=`$ (117)
$`=`$ $`\pi ^{1+ϵ}\mathrm{\Gamma }(1ϵ)(𝒒^2)^ϵ\left[{\displaystyle \frac{1}{ϵ^2}}+{\displaystyle \frac{1}{ϵ}}\mathrm{log}{\displaystyle \frac{𝒒^2}{𝒌_1^2}}{\displaystyle \frac{\pi ^2}{6}}{\displaystyle \frac{3}{2}}\mathrm{log}^2{\displaystyle \frac{𝒒^2}{𝒌_1^2}}+𝒪(ϵ)\right].`$
The third integral is calculated to the order $`𝒪(ϵ)`$:
$`J_2(𝒌_1,𝒌_2)`$ $`=`$ $`\pi ^{1+ϵ}\mathrm{\Gamma }(1ϵ)(𝒒^2)^ϵ\left[{\displaystyle \frac{1}{ϵ}}\mathrm{log}{\displaystyle \frac{𝒒^2}{𝒌_2^2}}{\displaystyle \frac{1}{2}}\mathrm{log}{\displaystyle \frac{𝒒^2}{𝒌_2^2}}\mathrm{log}{\displaystyle \frac{𝒒^2}{𝒌_1^2}}\mathrm{log}{\displaystyle \frac{𝒒^2}{𝒌_2^2}}+𝒪(ϵ)\right].`$ (118)
Putting together all the pieces to construct the complete action of the kernel, all the singulatities cancel and the result, which is finite in the limit $`ϵ0`$, correspond to the expression given in eq. (114).
|
warning/0506/quant-ph0506239.html
|
ar5iv
|
text
|
# Adventures of the Coupled Yang-Mills Oscillators: II. YM-Higgs Quantum Mechanics
## 1 Introduction
We here continue our study of the quantum mechanical motion in the $`x^2y^2`$ potentials of phase space dimensions $`2n`$ with $`n=2,3`$, which arise in the spatially homogeneous limit of the Yang-Mills (YM) equations. As is well known (see for a review), these systems exhibit a rich chaotic behavior despite their extreme simplicity. Especially the $`n=2`$ model, the central object of this and also our previous investigation (see \- we will henceforth refer to this work as I), has been widely studied.
Classically, this model possesses a logarithmically divergent volume of energetically accessible phase space $`\mathrm{\Gamma }_E`$ <sup>1</sup><sup>1</sup>1This is in violation of Weil’s famous theorem , which states that the average energy level density $`dN/dE`$ is asymptotically proportional to $`\mathrm{\Gamma }_E`$., but its quantum mechanical version (YM quantum mechanics, YMQM) has a discrete spectrum . Physically, the explanation is obvious: Quantum fluctuations, e.g. zero-point fluctuations, forbid that the trajector escapes along the $`x`$ or $`y`$ axis where the potential energy vanishes. The system is thus confined to a finite volume, and this implies the discreteness of the energy levels. Classically this escape is always possible without increasing energy. As we shall see below, these classically allowed configurations result in singularities of the quasiclassical partition function.
In I we calculated the higher-order quantum corrections to the partition function (heat kernel) $`Z(t)`$ for the $`x^2y^2`$ potential using the approximation based on the adiabatic separation of the motion in $`x`$ and $`y`$ in the hyperbolic channels of the equipotential surface $`xy=\mathrm{const}`$. The main assumption of this method is that the final results of the calculations do not depend on the artificial boundary $`Q`$ dividing the central region $`x,y[Q,Q]`$ from the channels $`x,y[Q,\mathrm{}]`$. We showed in I that this assumption, after improvement of the quantum mechanical treatment of the motion in the channels, is correct not only for the Thomas-Fermi (TF) term but also for the leading (in powers of $`tQ^41`$) higher-order quantum corrections, and we derived $`Q`$-independent asymptotic series in the parameter $`\lambda ^2=g^2\mathrm{}^4t^3`$ for contribution of each region to $`Z(t)`$.
In the present paper, like in I, we explore the properties of the $`x^2y^2`$ model beyond the TF approximation, but we pursue a different approach. We calculate the higher-order corrections to $`Z(t)`$ as in by starting from the Yang-Mills-Higgs quantum mechanics (YMHQM) and taking the limit $`v0`$, where $`v`$ is the vacuum expectation value of the Higgs field. In the $`n=2,3`$ $`x^2y^2`$ models, $`v`$ determines the strength of the harmonic potential
$`V(x,y)`$ $`={\displaystyle \frac{1}{2}}v^2(x^2+y^2)`$ $`(n=2),`$
$`V(x,y,z)`$ $`={\displaystyle \frac{1}{2}}v^2(x^2+y^2+z^2)`$ $`(n=3).`$
Due to the above mentioned logarithmic infinity of the classical phase space volume of the $`x^2y^2`$ model it is impossible to completely disentangle the nonlinear coupled oscillators from the harmonic oscillations generated by the Higgs potential leading to a term in $`Z(t)`$ proportional to $`\mathrm{ln}v`$. In the $`n=3`$ case, which has a finite phase space volume at fixed energy, this method yields an expression for $`Z(t)`$ that coincides in the limit $`v0`$ with the one obtained adiabatic separation method . This is due to the negligible time spent by the classical trajectory in the depth of the hyperbolic channels . Higher-order corrections change this situation essentially, as we shall see below.
Here we use the approach of ref. with the limiting procedure $`v0`$ for the calculation of $`Z(t)`$ beyond the TF approximation by applying the Wigner-Kirkwood (WK) method (see for a review of the WK method). The higher-order corrections $`𝒪(\mathrm{}^k)`$ in the WK approach lead to a new phenomenon in the limit $`v0`$: for $`k2`$ they yield power-like singularities of the form $`v^k`$. These singularities are not cancelled at a given power $`k`$ as one might expect. The situation is completely different when one includes the quantum fluctuations in the channels of the $`x^2y^2`$ potential to all orders. These generate a confining potential, which does not disappear in the limit $`v0`$, closes the flat direction, and eliminates the mentioned singularities, which are essentially classical. Taking these fluctuations into account, we are able to compare the expression for $`Z(t)`$ obtained by our method with the result for $`Z(t)`$ obtained by the method of in the TF approximation.
Concerning the $`n=3`$ case with its finite phase-space volume, the singularities appear also at the higher-order corrections in contrast to the TF approximation and again are eliminated by the quantum fluctuations corresponding to the specific quartic form of the potential characteristic of the YM quantum mechanics.<sup>2</sup><sup>2</sup>2In the corresponding supersymmetric Yang-Mills quantum mechanical system, the wave function is not confined due to the cancellations between bosonic and fermionic degrees of freedom, and the spectrum is continuous .
Finally, we develop a novel approximation scheme, which relies on a resummation of certain terms in the KW expansion and thus introduces a nonvanishing Higgs potential, which avoids the divergences of the TF approximation. This approach is motivated by the need to take the quantum fluctuations inside the hyperbolic channels into account even in the lowest order approximation. We show that the new approach reproduces the TF result obtained with the method of without requiring an artificial subdivision of the phase space into different regions. A modified WK expansion can be derived to systematically improve on this result.
In the next two sections we present the YMH system and the WK method of calculating $`Z(t)`$ beyond the TF approximation.
## 2 Yang-Mills-Higgs classical and quantum mechanics: The Thomas-Fermi approximation
There are several mechanisms that can suppress and even eliminate the classical chaos of the YM system (see ). One is the Higgs mechanism . For spatially homogeneous fields (long wave length limit of YM system), if only the interaction of the gauge fields with the Higgs vacuum is considered, the classical Hamiltonian for $`n=2`$ is given by the expression
$$H=\frac{1}{2}(p_x^2+p_y^2)+\frac{g^2}{2}x^2y^2+\frac{v^2}{2}(x^2+y^2),$$
(1)
where $`v=\varphi `$ is the vacuum expectation value of the Higgs field $`\varphi `$. It is known that there is a classical transition from chaos to regular motion as $`v`$ gets large enough. More precisely, the chaos disappears when $`g^2v^4/E>0.6`$ , where $`E=H`$ is the energy. The analogous transition in the adjacent energy-level spacing distribution was predicted and established in several papers . The quantized counterpart of (1) is
$$\widehat{H}=\frac{\mathrm{}^2}{2}_{x,y}^2+\frac{g^2}{2}x^2y^2+\frac{v^2}{2}(x^2+y^2).$$
(2)
As in I, we measure all quantities in units of the energy $`E`$ with dimensions $`[H]=1,[t]=1,[x],[y]=1/4,[g]=0,[v]=1/4,[\mathrm{}]=3/4`$.
It is obvious that the operator (2) has a discrete spectrum as it has for $`v=0`$. The TF approximation to the heat kernel or partition function $`Z(t)=\mathrm{Tr}[\mathrm{exp}(t\widehat{H})]`$ is the standard lowest-order semiclassical approximation valid for small $`\mathrm{}t^{3/4}1`$. It is obtained by substituting the classical Hamiltonian for its quantum counterpart and replacing the trace of the heat kernel by the integral over the phase-space volume normalized by ($`2\pi \mathrm{})^n`$, where $`2n`$ is the phase-space dimension. In other words, the TF approximation takes into account only the discreteness of the quantum mechanical phase space, but considers momenta and coordinates (in our case, the field amplitudes $`x`$ and $`y`$) as commuting variables. This method was used in numerous papers (see e.g. ). For the calculation of the energy level density $`\rho (E)=dN(E)/dE`$ at asymptotic energies, the TF approximation is a consistent approach since, as we shall see below, all corrections to the TF term are structures with factors $`\mathrm{}^kt^{\mathrm{}}`$ with $`k,\mathrm{}`$ positive integers. For the asymptotic energy level density $`\rho (E)`$ or $`N(E)`$ these corrections are negligible according the Karamata-Tauberian theorems relating the most singular part of $`Z(t)`$ to the asymptotic level density, $`N(E)=𝑑E\rho (E)=^1[Z(t)/t]`$, where $`^1`$ denotes the inverse Laplace transform.
In $`Z(t)`$ and $`N(E)`$ were calculated for the Hamiltonian (1). We give the precise expression for $`Z(t)`$ of the YMHQM system in the TF approximation:
$$Z(t)=\frac{1}{\sqrt{2\pi }g\mathrm{}^2t^{3/2}}\mathrm{exp}\left(\frac{tv^4}{4g^2}\right)K_0\left(\frac{tv^4}{4g^2}\right),$$
(3)
where $`K_0(z)`$ is the modified Bessel function of the third kind. For the most interesting limit $`v0`$ we get:
$$Z(t)\frac{1}{\sqrt{2\pi }g\mathrm{}^2t^{3/2}}\left(\mathrm{ln}\frac{8g^2}{tv^4}C\right),$$
(4)
where $`C`$ is the Euler constant. The impossibility of disentangling the coupled oscillators from the uncoupled ones is expressed by the logarithmic divergence of the phase space volume for $`n=2`$ as we already mentioned in the Introduction. Below we compare (4) with the corresponding expression for $`Z(t)`$ obtained in for the pure $`x^2y^2`$ model. Because we shall often encounter the pre-factor appearing in (3) and (4) in the following, we introduce the special symbol $`K`$ for it:
$$K=(2\pi g^2\mathrm{}^4t^3)^{1/2}(2\pi \lambda ^2)^{1/2}.$$
(5)
## 3 Beyond the TF approximation: The Kirkwood-Wigner expansion
In the present paper, unlike in I, we apply the WK expansion in all of phase space, avoiding the division of the phase space into a central region and hyperbolic channels. As we will see below, this poses no problems as long as $`v0`$. However, singularities appear in the limit $`v0`$, unless the WK expansion is modified to include quantum fluctuations in a nonperturbative way.
Since we described the WK method in detail in I, we only give a very brief outline here. We start from the equation (I–12), a set of recurrent differential equations for the kernels $`W_k`$ of the partition function $`Z_k(t)`$ at the $`k`$-th order in $`\mathrm{}`$:
$$Z_k(t)=\frac{\mathrm{}^k}{(2\pi \mathrm{})^n}𝑑\mathrm{\Gamma }W_k(\stackrel{}{r},\stackrel{}{p};t)e^{tH},$$
(6)
where $`d\mathrm{\Gamma }=dxdydp_xdp_y`$ and the classical Hamiltonian $`H`$ given by (1). We begin with the partition function at second order in $`\mathrm{}`$ ($`k=2`$). Integrating over $`p_x`$ and $`p_y`$ and making use of the symmetry of the Hamiltonian (1) with respect to the interchange $`xy`$, we obtain:
$$𝑑\mathrm{\Gamma }W_2e^{tH}=\frac{\pi t}{3}\left[\left(g^2+\frac{tv^4}{2}\right)I_{10}+\frac{tg^4}{2}I_{21}v^2I_{00}+tg^2v^2I_{11}\right],$$
(7)
where we have used the notation (for $`mn`$): <sup>3</sup><sup>3</sup>3Note that the case $`m=n`$ needs to be calculated separately from the case $`m>n`$; see below.
$$I_{mn}=4_0^{\mathrm{}}𝑑x_0^{\mathrm{}}𝑑yx^{2m}y^{2n}\mathrm{exp}\left[\frac{t}{2}\left(v^2(x^2+y^2)+g^2x^2y^2\right)\right].$$
(8)
Note that the factors $`\mathrm{}^2`$ from the KW expansion and from the normalization of the phase space volume element have canceled, making $`Z_2`$ independent of $`\mathrm{}`$. Integrating over $`y`$ and introducing the new variable $`u=g^2x^2/v^2`$, we obtain:
$$I_{mn}=\frac{\sqrt{2\pi }(2n1)!!(v^2)^{mn}}{t^{n+\frac{1}{2}}g^{2m+1}}_0^{\mathrm{}}𝑑uu^{m\frac{1}{2}}(1+u)^{n\frac{1}{2}}e^{uz}$$
(9)
with $`z=tv^4/2g^2`$. The integral over $`u`$ is related to the Whittaker function $`W_{\lambda ,\mu }(z)`$ (see , equation 9.222.1):
$$I_{mn}=\frac{\sqrt{2\pi }(2n1)!!(v^2)^{mn}}{t^{n+\frac{1}{2}}g^{2m+1}}e^{z/2}z^{\frac{mn+1}{2}}\mathrm{\Gamma }\left(\frac{2m+1}{2}\right)W_{\frac{m+n}{2},\frac{mn}{2}}(z)$$
(10)
For $`m=n`$ the Whittaker function has only a logarithmic divergence at $`z=0`$. For $`mn1`$ power-like singularities appear. For completeness we give $`Z_2(t)`$ explicitly
$`Z_2(t)`$ $`=`$ $`{\displaystyle \frac{tv^2}{12\sqrt{2\pi }gt^{1/2}}}[2(1+{\displaystyle \frac{tv^4}{2g^2}})z^1\mathrm{\Gamma }\left({\displaystyle \frac{3}{2}}\right)W_{\frac{1}{2},\frac{1}{2}}(z)`$ (11)
$`+z^1\mathrm{\Gamma }\left({\displaystyle \frac{5}{2}}\right)W_{\frac{3}{2},\frac{1}{2}}(z)2z^{1/2}\mathrm{\Gamma }\left({\displaystyle \frac{1}{2}}\right)W_{0,0}(z)`$
$`+2z^{1/2}\mathrm{\Gamma }\left({\displaystyle \frac{3}{2}}\right)W_{1,0}(z)]`$
As is easily seen from (11), $`Z_2(t)`$ has a singularity of the form $`v^2`$ at $`v=0`$. Using the limit of the Whittaker function $`W_{\lambda ,\mu }(z)`$ for small $`z`$, we find:
$$Z_2(t)K\frac{\mathrm{}^2g^2t^{3/2}}{6(tv^4)^{1/2}}(v0)$$
(12)
For completeness we give here the $`Z_2`$ from the KW method in the limit $`g=0`$, i. e. for two free harmonic oscillators using the asymptotic form of the Whittaker function:
$$Z_2(t)=\frac{1}{12}.$$
(13)
Together with the expression for the TF term we have
$$Z_0(t)+Z_2(t)=\frac{1}{\mathrm{}^2v^2t^2}\left(1\frac{1}{12}\mathrm{}^2v^2t^2\right),$$
(14)
which are the first two terms in the Taylor expansion of the exact partition function for the two-dimensional harmonic oscillator at small $`\mathrm{}vt`$:
$$Z(t)=\left[2\mathrm{sinh}\frac{1}{2}\mathrm{}vt\right]^2.$$
(15)
## 4 Higher-order corrections and the limit $`v0`$ for YMHQM
At higher order $`\mathrm{}^k`$ ($`k2`$) the mathematical structure of terms $`W_k`$ and $`Z_k`$ changes essentially. In the expression (8) higher powers $`m`$ and $`n`$ appear and the difference between them increases ($`mn1`$), causing the second index of the Whittaker function to exceed $`\mu =1/2`$. As a result, in the limit $`z=tv^4/2g^20`$ the finite sum of the power-like singular terms of $`W_{\lambda ,\mu }(z)`$ begins to play a crucial role.
A systematic analysis of the higher-order corrections using Mathematica leads to the conclusion that there is a correlation between the power of $`\mathrm{}`$ (for $`k2`$) and the most singular terms in (10): $`mn=k/2`$ (due to the symmetry against exchange $`xy`$ it is always possible to put $`m>n`$). The case $`m=n`$ requires special consideration and, as found, leads only to logarithmic singularities.
It is easy to see that the next, less singular terms correspond to the case $`mn=(k/2)2\mathrm{}`$ ($`\mathrm{}=1,2,\mathrm{};\mathrm{}<k/4`$). For the general expression of $`Z_k^{(m,n)}(t)`$ we need to determine the powers of $`g^2`$ and $`t`$. For $`g^2`$ it is simply $`g^{2m}`$. To determine the power of $`t`$ we note that for each $`k`$ there are always terms without factors of $`p_x`$ and $`p_y`$ and minimal power of $`t`$ at given $`k`$. For such terms the factor is $`t^{2mn1}`$. For the terms containing factors of $`p_x`$ and $`p_y`$ the power of $`t`$ does not change after integration over the momenta. For the less singular terms with $`mn=\frac{1}{2}k2\mathrm{}`$ the corresponding factor is $`t^{2mn13\mathrm{}}`$.
Now we are in position to write the general expression for the most singular terms contributing to the partition function at the order of $`\mathrm{}^k(k=2,4,6,\mathrm{}`$). After integration over $`p_x,p_y,x`$ and $`y`$, keeping only terms having $`mn=k/2`$ ($`k/2mk,0nk/2`$) we obtain:
$$Z_k^{(m,n)}(t)=K\left(\frac{4g^4\mathrm{}^4t^3}{v^4t}\right)^{k/4}\mathrm{\Gamma }\left(\frac{k}{2}\right)(2mk1)!!.$$
(16)
We see that there are strong singularities at $`v=0`$, and further analysis shows that these are not cancelled by the summation of all most singular terms at a given $`k`$. For the less singular terms with $`mn=k/22\mathrm{}`$ with $`\mathrm{}<k/4`$ we have:
$$Z_k^{(m,n,\mathrm{}}(t)=K\left(\frac{4g^4\mathrm{}^4t^3}{v^4t}\right)^{k/4}\left(\frac{v^4t}{4g^4}\right)^{\mathrm{}}\mathrm{\Gamma }\left(\frac{k}{2}\mathrm{}\right)(2mk1)!!.$$
(17)
Due to the appearance of $`(mn)`$ in the argument of the gamma function in (17), the case with $`m=n`$ must be considered separately. It is obvious that these terms have no power-like divergences, but only logarithmic singularities like the TF term. We obtain
$$Z_k^{(m=n)}(t)=K(g^2\mathrm{}^4t^3)^{k/4}\left[\mathrm{ln}\frac{2g^2}{v^4t}2C\psi \left(m+\frac{1}{2}\right)\right],$$
(18)
where $`\psi (x)`$ is the logarithmic derivative of the Gamma function. Let us briefly discuss these results. The power-like singularities in the KW approach for $`k2`$ are related to the possibility of escaping classically along the axis $`x=0`$ or $`y=0`$ where, in the limit $`v0`$, the potential energy vanishes. Non-zero $`v`$ forbids such escape to infinity. These singularities affect any classically calculated distribution function, in particular, the heat kernel $`Z(t)`$. They also show that the trajectories lie deep inside the channels most of the time. Quantum mechanical fluctuations forbid any escapes along the axes.
The $`v^k`$ singularities have more resemblance with the infrared singularities, they are related to the behavior at long distances and different from the usual ultraviolet divergences connected with the asymptotic expansion in powers of $`h^k`$. The absence of power-like singularities for $`m=n`$ is easily explained: for $`m=n`$ the configurations dominate along the diagonals ($`|x|=|y|`$), whereas for $`mn`$ the configurations populate the channels and have a trend to escape if quantum fluctuations do not forbid this. It is thus clear that we have to take into account the effect of the quantum fluctuations on the motion inside the channels, which the perturbative expansion of the WK method fails to do.
## 5 Quantum fluctuations and power-like singularities
In this Section we will attempt to include quantum fluctuations created by the form of $`x^2y^2`$ potential in a framework, which does not rely directly on the adiabatic separation of the motion inside the hyperbolic channels as it was elaborated in detail in I. Let us consider the motion along the $`x`$-axis. The heat kernel for the $`x^2y^2`$ potential generates a mean spread in $`y`$ at the position $`x`$ of the order of $`\delta y(g^2x^2t/2)^{1/2})`$. Quantum mechanics dictates that the spread of the conjugate momentum $`p_y`$ is at least $`\delta p_y\mathrm{}g|x|(t/2)^{1/2}`$. Analogous relations hold between $`y`$ and $`p_x`$: $`\delta p_x\mathrm{}g|y|(t/2)^{1/2}`$ for the motion along the $`y`$-axis. We now propose a modification of the WK formalism, which incorporates these relationships into the generating functional for the expansion in powers of $`\mathrm{}`$.
To achieve this, we resum the term $`\frac{1}{2}\mathrm{}^2t(\mathrm{\Delta }V)`$ in the WK operator (I–10) to all orders by writing
$`W(\stackrel{}{r},\stackrel{}{p};t)=\stackrel{~}{W}(\stackrel{}{r},\stackrel{}{p};t)\mathrm{exp}\left({\displaystyle \frac{\mathrm{}^2}{4}}t^2\mathrm{\Delta }V\right).`$ (19)
Inserting this definition into the differential equation (I–10) for the WK kernel $`W`$ yields an equation for the phase space function $`\stackrel{~}{W}`$:
$`{\displaystyle \frac{\stackrel{~}{W}}{t}}`$ $`=`$ $`{\displaystyle \frac{\mathrm{}^2}{2}}[\mathrm{\Delta }+t^2(V)^2{\displaystyle \frac{2it}{\mathrm{}}}(\stackrel{}{p}V){\displaystyle \frac{\mathrm{}^2}{4}}t^2(\mathrm{\Delta }\mathrm{\Delta }V)`$ (20)
$`+{\displaystyle \frac{2}{\mathrm{}}}(i\stackrel{}{p}\mathrm{}tV){\displaystyle \frac{\mathrm{}^2}{2}}t^2(\mathrm{\Delta }V)`$
$`{\displaystyle \frac{\mathrm{}}{2}}t^2(i\stackrel{}{p}\mathrm{}tV)(\mathrm{\Delta }V)]\stackrel{~}{W}.`$
The term in the exponent of (19):
$$\frac{\mathrm{}^2}{4}t^2\mathrm{\Delta }V=\frac{\mathrm{}^2}{4}g^2t^2(x^2+y^2)$$
(21)
acts like a Higgs potential with $`v_{\mathrm{eff}}^2=\mathrm{}^2g^2t/2`$. What is unusual about this term is that the effective potential itself is time-dependent. The connection to the argument given at the beginning of this section becomes evident, when one recognizes that the exponent represents the lower bound associated with the kinetic energy demanded by the uncertainty relation:
$$\frac{1}{2}(\delta p_x^2+\delta p_y^2)\frac{\mathrm{}^2}{4}t\mathrm{\Delta }V.$$
(22)
It is straightforward to derive a recursion relation analogous to (I–12) for the coefficients of the expansion of $`\stackrel{~}{W}`$ in powers of $`\mathrm{}`$: <sup>4</sup><sup>4</sup>4Note that the expansion of $`\stackrel{~}{W}`$ in powers of $`\mathrm{}`$ does not yield an expansion of $`Z(t)`$ strictly in powers of $`\mathrm{}`$ because of the nonpolynomial factor containing $`\mathrm{}`$ in (19).
$`{\displaystyle \frac{\stackrel{~}{W}_k}{t}}`$ $`=`$ $`i\stackrel{}{p}\left[t(V)\right]\stackrel{~}{W}_{k1}+{\displaystyle \frac{1}{2}}\left[\mathrm{\Delta }+t^2(V)^22tV\right]\stackrel{~}{W}_{k2}`$ (23)
$`\left[{\displaystyle \frac{it^2}{4}}\stackrel{}{p}(\mathrm{\Delta }V)\right]\stackrel{~}{W}_{k3}`$
$`+\left[{\displaystyle \frac{t^3}{4}}V(\mathrm{\Delta }V){\displaystyle \frac{t^2}{4}}(\mathrm{\Delta }V)+{\displaystyle \frac{t^2}{8}}\mathrm{\Delta }\mathrm{\Delta }V\right]\stackrel{~}{W}_{k4}.`$
Before we apply this trick to the quartic YM oscillator, it is useful to briefly explore how it works for the harmonic oscillator. It is easy to see that (23) yields the correct expansion for the partition function for the potential $`V=\frac{1}{2}v^2x^2`$. Indeed, by calculating $`\stackrel{~}{W}_0,\stackrel{~}{W}_2,\stackrel{~}{W}_4`$ and integrating over $`p`$ and $`x`$, we obtain
$`\stackrel{~}{Z}_0(t)`$ $`=`$ $`{\displaystyle \frac{1}{\mathrm{}vt}}e^{(\mathrm{}vt)^2/4},`$
$`\stackrel{~}{Z}_2(t)`$ $`=`$ $`{\displaystyle \frac{5}{24}}\mathrm{}vte^{(\mathrm{}vt)^2/4},`$
$`\stackrel{~}{Z}_4(t)`$ $`=`$ $`{\displaystyle \frac{127}{5760}}(\mathrm{}vt)^3e^{(\mathrm{}vt)^2/4},`$ (24)
giving the expansion in powers of $`(\mathrm{}vt)`$ (up to $`\mathrm{}^4`$) of the exact partition function of the harmonic oscillator, $`Z(t)=(2\mathrm{sinh}\mathrm{}vt/2)^1`$. Note that, in this case, $`\stackrel{~}{Z}_k`$ contributes to all powers in the $`\mathrm{}`$ expansion of $`Z(t)`$ beginning with $`\mathrm{}^{k1}`$.
Now consider the non-linear potential $`\frac{1}{2}g^2x^2y^2`$. The lowest term is easily obtained from (3) by substituting $`vv_{\mathrm{eff}}`$:
$$\stackrel{~}{Z}_0=K\mathrm{exp}\left(\frac{\mathrm{}^4}{16}g^2t^3\right)K_0\left(\frac{\mathrm{}^4}{16}g^2t^3\right)K\left[\mathrm{ln}\frac{1}{g^2\mathrm{}^4t^3}+5\mathrm{ln}2C\right],$$
(25)
which coincides (with logarithmic precision) with the result obtained in for $`Z(t)`$. The second-order term is
$$\stackrel{~}{Z}_2=\frac{g^2t}{24\pi }(4\stackrel{~}{I}_{10}+g^2t\stackrel{~}{I}_{21})$$
(26)
with
$$\stackrel{~}{I}_{mn}=4_0^{\mathrm{}}𝑑x_0^{\mathrm{}}𝑑yx^{2m}y^{2n}\mathrm{exp}\left(\frac{1}{2}g^2x^2y^2t\frac{\mathrm{}^2}{4}g^2(x^2+y^2)t^2\right).$$
(27)
These integrals correspond to those defined in (8) with the substitution $`v^2\mathrm{}^2g^2t/2`$. The exact analytical expression for these integrals in terms of Whittaker functions was given in (10). In the following we retain only the first term in the (finite) power series expansion of the Whittaker function; later we shall correct for this simplification. Using the condition $`z=g^2\mathrm{}^4t^3/8\lambda ^2/81`$, allowing us to neglect terms involving $`\mathrm{ln}\lambda ^2`$ compared with terms of the form $`\lambda ^2`$, we obtain
$`\stackrel{~}{I}_{10}={\displaystyle \frac{\sqrt{2\pi }}{(g^2t)^{1/2}}}{\displaystyle \frac{4}{(\mathrm{}gt)^2}},`$
$`\stackrel{~}{I}_{21}={\displaystyle \frac{\sqrt{2\pi }}{(g^2t)^{3/2}}}{\displaystyle \frac{4}{(\mathrm{}gt)^2}},`$ (28)
yielding the result
$$\stackrel{~}{Z}_2(t)=\frac{5}{3}K.$$
(29)
For $`\stackrel{~}{Z}_4`$ we retain only terms with $`mn=2`$ as the most singular ones in the limit $`v0`$, based on the usual arguments. This gives:
$$\stackrel{~}{Z}_4(t)=\frac{(\mathrm{}gt)^4}{2\pi \mathrm{}^2t}\left[\frac{1}{30}\stackrel{~}{I}_{20}+\frac{g^2t}{180}\stackrel{~}{I}_{31}+\frac{(g^2t)^2}{576}\stackrel{~}{I}_{42}\right].$$
(30)
For the needed integrals we obtain in the same approximation ($`\lambda ^21`$):
$`\stackrel{~}{I}_{20}={\displaystyle \frac{\sqrt{2\pi }}{(g^2t)^{1/2}}}{\displaystyle \frac{16}{(\mathrm{}gt)^4}},`$
$`\stackrel{~}{I}_{31}={\displaystyle \frac{\sqrt{2\pi }}{(g^2t)^{3/2}}}{\displaystyle \frac{16}{(\mathrm{}gt)^4}},`$
$`\stackrel{~}{I}_{42}={\displaystyle \frac{\sqrt{2\pi }}{(g^2t)^{5/2}}}{\displaystyle \frac{48}{(\mathrm{}gt)^4}}.`$ (31)
Substituting (31) into (30) we obtain:
$$\stackrel{~}{Z}_4(t)=\frac{127}{180}K.$$
(32)
Collecting all results, we have:
$$\stackrel{~}{Z}_{0+2+4}(t)=K\left[\mathrm{ln}\frac{1}{g^2\mathrm{}^4t^3}+5\mathrm{ln}2C+\frac{427}{180}\right],$$
(33)
We improve upon the approximation made above to the Whittaker function and consider the contribution of all singular terms in the asymptotic expansion of $`W_{\kappa ,\mu }(z)`$ for small $`z`$:
$$\underset{p=0}{\overset{mn1}{}}\mathrm{\Gamma }(mnp)\mathrm{\Gamma }\left(n+\frac{1}{2}+p\right)\left(\frac{\lambda ^2}{8}\right)^p,$$
(34)
where again $`\lambda ^2=g^2\mathrm{}^4t^3`$. The complete expression for the contribution to $`\stackrel{~}{Z}_k`$ ($`k=2,4,6,\mathrm{}`$) from the most singular terms is
$$\stackrel{~}{Z}_k^{(m,n)}=K\frac{2^k(2n1)!!}{\mathrm{\Gamma }\left(n+\frac{1}{2}\right)}\underset{p=0}{\overset{\frac{1}{2}k1}{}}\mathrm{\Gamma }\left(\frac{1}{2}kp\right)\mathrm{\Gamma }\left(n+\frac{1}{2}+p\right)\left(\frac{\lambda ^2}{8}\right)^p.$$
(35)
For the less singular terms ($`mn=\frac{1}{2}k2\mathrm{}`$, $`\mathrm{}<\frac{1}{4}k`$) we get
$`\stackrel{~}{Z}_k^{(m,n,\mathrm{})}`$ $`=`$ $`K\lambda ^2\mathrm{}{\displaystyle \frac{2^{k4\mathrm{}}(2n1)!!}{\mathrm{\Gamma }\left(n+\frac{1}{2}\right)}}`$ (36)
$`{\displaystyle \underset{p=0}{\overset{\frac{1}{2}k2\mathrm{}1}{}}}\mathrm{\Gamma }\left({\displaystyle \frac{1}{2}}k2\mathrm{}p\right)\mathrm{\Gamma }\left(n+{\displaystyle \frac{1}{2}}+p\right)\left({\displaystyle \frac{\lambda ^2}{8}}\right)^p,`$
and for the logarithmic term ($`m=n`$) we obtain
$$\stackrel{~}{Z}_k^{(m=n)}=K\lambda ^{\frac{1}{2}k}\left[\mathrm{ln}\lambda ^2+3\mathrm{ln}22C\psi \left(m+\frac{1}{2}\right)\right].$$
(37)
It is clear that these contributions again generate an asymptotic series in the small parameter $`\lambda ^2=g^2\mathrm{}^4t^3`$ of the following form:
$`Z(t)`$ $`=`$ $`K[\mathrm{ln}\lambda ^2+5\mathrm{ln}2C+{\displaystyle \underset{k=2,4\mathrm{}}{}}2^k{\displaystyle \underset{n=0}{\overset{k/2}{}}}a_n^{(k/2)}{\displaystyle \frac{2^k(2n1)!!}{\mathrm{\Gamma }\left(n+\frac{1}{2}\right)}}`$ (38)
$`{\displaystyle \underset{p=0}{\overset{\frac{1}{2}k1}{}}}\mathrm{\Gamma }({\displaystyle \frac{1}{2}}kp)\mathrm{\Gamma }(n+{\displaystyle \frac{1}{2}}+p)({\displaystyle \frac{\lambda ^2}{8}})^p],`$
where the $`a_n^{(k/2)}`$ ($`n=0,1,\mathrm{},\frac{k}{2}`$) are the coefficients of the structures $`(g^2t)^n\stackrel{~}{I}_{mn}`$ ($`mn=\frac{k}{2}`$) in the expression $`𝑑\mathrm{\Gamma }\stackrel{~}{W}_ke^{Ht}`$ ($`k=2,4,6,\mathrm{}`$), analogous to the coefficients $`a_m^{(n)}`$ ($`m=0,1,\mathrm{},3n`$) in (I–71). For $`k=4`$, e.g., these numbers are:
$$a_0^{(2)}=\frac{1}{30},a_1^{(2)}=\frac{1}{180},a_2^{(2)}=\frac{1}{576}.$$
(39)
We note that this asymptotic series closely resembles the one derived in I using a completely different approach, splitting the $`xy`$ plane into two integration region and treating the quantum fluctuations exactly in the region containing the hyperbolic channels. The leading logarithmic term is identical in both cases, but it is not clear that the constant coincides. The expansion parameter $`\lambda ^2`$ is the same for both series. Unfortunately, our inability to find a simple general algorithm for the coefficients, $`a_n^{(k/2)}`$ here and $`a_m^{(n)}`$ in I, has prevented us to compare the two results in detail. The less singular terms also lead, after summation over $`k`$, to an asymptotic series in the parameter $`\lambda ^2`$. The same is true for the terms $`Z_k^{(m,m)}`$ with $`m=n`$.
At the end of Section 4 we commented on the analogy between the limit $`v0`$ and the infrared problem. We now can make this analogy more precise. The infrared limit corresponds to the behavior of the system at large distances or deep inside the channels, in the terminology of the paper I. When we compare the limiting forms for $`v0`$ of the quantities $`I_{mn}`$ derived here with the expression (I–25) for the analogous quantities derived for large distances ($`tQ^41`$) in I, we obtain the correspondence:
$$Q^{2(mn)}\frac{mn}{\mathrm{\Gamma }\left(n+\frac{1}{2}\right)}\frac{1}{(tv^2)^{mn}}.$$
(40)
Evidently the limit $`v0`$ corresponds to the limit $`Q\mathrm{}`$, supporting our claim that it represents the infrared limit.
## 6 The three dimensional YMHQM
Finally, we briefly consider the $`n=3`$ case of the YMHQM model. The Hamiltonian for $`n=3`$ YMH classical mechanics is:
$$H=\frac{1}{2}(p_x^2+p_y^2+p_z^2)+\frac{g^2}{2}(x^2y^2+y^2z^2+z^2x^2).$$
(41)
In its quantum counterpart, $`\stackrel{}{p}^2`$ is replaced with $`\mathrm{}^2^2`$. As we know the contribution to $`Z(t)`$ from the channels is negligible in the TF approximation if we apply the condition $`tQ^41`$. This is due to the fact that deep in a channel, e. g. along the $`x`$-axis, the energetically accessible phase space volume gets pinched as $`x^2`$ and not as $`x^1`$ as it for the $`n=2`$ Hamiltonian. This implies that the limit $`v0`$ is smooth and the expressions for $`Z(t)`$ from and coincide at $`v0`$. Already for the second correction to the TF term this is not true as we shall see below. In the WK approach for $`Z_2`$ using (I–12) and (6) we have
$$Z_2(t)=\frac{1}{(2\pi \mathrm{})^3}𝑑p_x𝑑p_y𝑑p_z𝑑x𝑑y𝑑zW_2(\stackrel{}{p},\stackrel{}{x};t)e^{Ht},$$
(42)
with $`W_2`$ from (I–19). Integrating over the momenta, using the symmetry of the potential energy in (41), we have
$$Z_2(t)=\frac{t^{1/2}}{2(2\pi )^{3/2}\mathrm{}}\left[g^2I_1+\frac{g^4t}{4}I_2\right],$$
(43)
where integrals $`I_1`$ and $`I_2`$ over $`x,y,z`$ are given by
$`I_1`$ $`=`$ $`{\displaystyle _0^{\mathrm{}}}𝑑x𝑑y𝑑zx^2e^{V(x,y,z)t};`$ (44)
$`I_2`$ $`=`$ $`{\displaystyle _0^{\mathrm{}}}𝑑x𝑑y𝑑zx^2(y^2+z^2)e^{V(x,y,z)t};`$ (45)
with
$$V(x,y,z)=\frac{g^2}{2}(x^2y^2+y^2z^2+z^2x^2)+\frac{v^2}{2}(x^2+y^2+z^2).$$
(46)
Integrating over $`x`$ and using polar coordinates $`r,\varphi `$ for the integrations over $`y`$ and $`z`$ we obtain:
$`I_1`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left({\displaystyle \frac{2\pi }{t}}\right)^{3/2}{\displaystyle _0^{\mathrm{}}}r𝑑r{\displaystyle \frac{\mathrm{exp}\left(\frac{1}{16}tg^2r^4\frac{1}{2}tv^2r^2\right)I_0\left(\frac{1}{16}tg^2r^4\right)}{(v^2+g^2r^2)^{3/2}}};`$ (47)
$`I_2`$ $`=`$ $`{\displaystyle \frac{1}{2}}\left({\displaystyle \frac{2\pi }{t}}\right)^{3/2}{\displaystyle _0^{\mathrm{}}}r^5𝑑r{\displaystyle \frac{\mathrm{exp}\left(\frac{1}{16}tg^2r^4\frac{1}{2}tv^2r^2\right)I_0\left(\frac{1}{16}tg^2r^4\right)}{(v^2+g^2r^2)^{3/2}}};`$ (48)
where $`I_0(z)`$ denotes the modified Bessel function of the first kind. We see that $`I_1`$ is divergent at $`r=0`$ if $`v=0`$ in contrast to the TF term.
Again, we obtain a smooth transition in the limit $`v0`$ if we make the substitution used in Section V before taking the limit $`v0`$. As for the $`n=2`$ model, here also the zero-point quantum fluctuations in the channels generate an effective Higgs potential $`\frac{1}{4}\mathrm{}^2g^2t(x^2+y^2+z^2)`$ and render the integral $`I_1`$ convergent. After this substitution, introducing the new variable $`u^2=\frac{1}{16}tg^2r^4`$, we get:
$$Z_2(t)=\frac{\sqrt{2}t^{3/4}}{\mathrm{}g^{1/2}}\left[J_0(\lambda )+4J_2(\lambda )\right]$$
(49)
with $`\lambda =gt^{3/2}\mathrm{}^2(1)`$ and
$$J_b(\lambda )=_0^{\mathrm{}}u^b𝑑u\frac{e^{u^2\lambda u}}{(\lambda +8u)^{3/2}}.$$
(50)
Expanding the exponential function in the small parameter $`\lambda `$, the integral (50) can be expressed as a finite sum of the generalized hypergeometric functions. Retaining the main terms we have
$`J_0(\lambda ){\displaystyle \frac{1}{4\sqrt{\lambda }}},`$
$`J_2(\lambda ){\displaystyle \frac{\mathrm{\Gamma }\left(\frac{3}{4}\right)}{32\sqrt{2}}}.`$ (51)
Thus, the second-order correction $`Z_2(t)`$ for the $`n=3`$ YM model becomes
$$Z_2(t)L\left[1\frac{\mathrm{\Gamma }\left(\frac{3}{4}\right)^3}{2^{5/4}\pi ^{3/2}}(g^2\mathrm{}^4t^3)^{1/4}\right],$$
(52)
where
$$L=\frac{1}{2}\mathrm{\Gamma }\left(\frac{1}{4}\right)^3(2\pi ^2g^2\mathrm{}^4t^3)^{3/4}$$
(53)
is the TF term found in . We see from (52) that the quantum corrections at the order $`\mathrm{}^2`$ are parametrically enhanced due to the quantum fluctuations generated by the effective Higgs potential $`\frac{1}{2}\mathrm{}^2g^2(x^2+y^2+z^2)t`$.
## 7 Conclusions
We have shown in I and here that the richness of the classical YM mechanics with the $`x^2y^2`$ potential translates into, and even gets amplified by, the quantum mechanical properties of the system. The YM quantum mechanics exhibits a confinement property, which strongly influences the quantum mechanical motion in the $`x^2y^2`$ potential. At higher order in $`\mathrm{}`$ (up to $`\mathrm{}^8`$) this results in the vanishing of the leading quantum corrections, when we correctly take into account this property for the motion in the hyperbolic channels. We convinced that this property survives at higher orders, although we did not explicitly demonstrate it.
Here we calculated the quantum corrections to the partition function by adding a Higgs term to the potential, and found that power-like singularities arise in the limit $`v0`$. We associate these essentially classical singularities with the fact that the Wigner-Kirkwood expansion does not take into account the effect of the quantum fluctuations on the motion within the channels. When these fluctuations, which are dictated by the uncertainty relation and the hyperbolic form of the channels, are taken into account, the escape along the coordinate axes, which is classically allowed, is prohibited, and the singularities disappear. As a result, the Thomas-Fermi term for the partition function acquires a renormalization expressed in terms of an asymptotic series in the parameter $`\lambda ^2=g^2\mathrm{}^4t^3`$ within both approaches.
We hope that the lessons we elicited from the present study of the higher-order quantum corrections to the homogeneous limit of the Yang-Mills equations will be useful for an improved understanding of the internal dynamics of the Yang-Mills quantum field theory.
|
warning/0506/hep-th0506151.html
|
ar5iv
|
text
|
# Untitled Document
hep-th/0506151
Counting Dyons in $`𝒩=8`$ String Theory
David Shih,<sup>*</sup><sup>*</sup>Permanent address: Department of Physics, Princeton University, Princeton, NJ 08544, USA. Andrew Strominger<sup>\**</sup><sup>\**</sup>Permanent address: Jefferson Physical Laboratory, Harvard University, Cambridge, MA 02138, USA. and Xi Yin\**
Center of Mathematical Sciences
Zhejiang University, Hangzhou 310027 China
Abstract
A recently discovered relation between 4D and 5D black holes is used to derive exact (weighted) BPS black hole degeneracies for 4D $`𝒩=8`$ string theory from the exactly known 5D degeneracies. A direct 4D microscopic derivation in terms of weighted 4D D-brane bound state degeneracies is sketched and found to agree.
1. Introduction
In this paper, we deduce an exact formula for the modified elliptic genus of string theory in four dimensions with $`𝒩=8`$ supersymmetry. The modified elliptic genus, as we review below, provides a weighted count of BPS states of $`𝒩=8`$ string theory. We derive a formula for it using a recently proposed exact relation between 4D and 5D BPS degeneracies, together with the known degeneracies in 5D. In addition we sketch a direct microscopic counting of D0-D2-D4 bound states which gives the same result. Our hope is that this example will provide a useful laboratory for testing the string theory relations recently proposed in e.g..
Some years ago an explicit formula for the elliptic genus for BPS states in 4D $`𝒩=4`$ theories was presciently conjectured . This formula was recently derived using the 4D-5D connection in . The present work is an extension of to 4D $`𝒩=8`$ theories. Previous work in this direction includes \[6,,7,,8\].
In the next section we review the 5D index defined and computed in . In section 3 we use the 4D-5D connection to derive the 4D index. In section 4 we sketch how this expression should follow (for one element of the U-duality class of black holes) from a microscopic analysis.
2. Review of the 5D modified eliptic genus
In this section, we want to summarize the work of reference on counting the microstates of 1/8 BPS black holes in five dimensions. These can be realized in string theory as the usual D1-D5-momentum system of type IIB on $`T^4\times S^1`$, with $`Q_1`$ D1-branes, $`Q_5`$ D5-branes and integral $`S^1`$ momentum $`n`$. The reason that microstate counting of this system is more difficult than for $`K3`$ compactification is because the usual supersymmetric index that counts these microstates, the orbifold elliptic genus of $`Hilb^k(K3)`$ with $`k=Q_1Q_5`$, vanishes when $`K3`$ is replaced with $`T^4`$. In , this difficulty was overcome by defining (and then computing) a new supersymmetric index $`_2`$, closely related with the elliptic genus, which is nonvanishing for $`T^4`$. We will refer to this new supersymmetric index as the modified elliptic genus of $`Hilb^k(T^4)`$. It is defined to be
$$_2^{(k)}=Tr\left[(1)^{2J_L^32J_R^3}2(J_R^3)^2q^{L_0}\overline{q}^{\overline{L}_0}y^{2J_L^3}\right]$$
where the trace is over states of the sigma model with target space $`Hilb^k(T^4)`$.<sup>1</sup> A free sigma model on $`R^4\times T^4`$ is factored out here, and our definition differs by a factor of 2 from . Here $`J_L^3`$ and $`J_R^3`$ are the left and right half-integral $`U(1)`$ charges of the CFT, and they are identified with generators of SO(4) rotations of the transverse $`R^4`$. The $`S^1`$ momentum is $`n=L_0\overline{L}_0`$. The usual elliptic genus is given by the same formula but without the $`2(J_R^3)^2`$ factor; it is these two insertions of $`J_R^3`$ that make $`_2`$ nonvanishing for $`T^4`$.
As for $`K3`$, here it is convenient to define a generating function for the modified elliptic genus:
$$_2=\underset{k1}{}p^k_2^{(k)}$$
In , this was shown to be given by the following sum
$$_2(p,q,y)=\underset{s,k,n,\mathrm{}}{}s(p^kq^ny^{\mathrm{}})^s\widehat{c}(nk,\mathrm{})$$
with the sum running over $`s,k1`$, $`n0`$, $`\mathrm{}Z`$. Note that the $`\overline{q}`$ dependence has dropped out – only the $`\overline{L}_0=0`$ states contribute to the modified elliptic genus. Of course, the index must have this property in order to count BPS states, since the BPS condition is equivalent to requiring $`\overline{L}_0=0`$.
It was furthermore shown in that the integers $`\widehat{c}(nm,\mathrm{})`$ are the coefficients in the following Fourier expansion
$$Z(q,y)\eta (q)^6\vartheta _1(y|q)^2=\underset{n,\mathrm{}}{}\widehat{c}(n,\mathrm{})q^ny^{\mathrm{}}$$
where $`\eta (q)`$ is the usual Dedekind eta function, and $`\vartheta _1(y|q)`$ is defined by the product formula
$$\vartheta _1(y|q)=i(y^{1/2}y^{1/2})q^{1/8}\underset{n=1}{\overset{\mathrm{}}{}}(1q^n)(1yq^n)(1y^1q^n)$$
Finally, it was observed in that $`\widehat{c}(n,\mathrm{})`$ actually only depends on a single combination of parameters $`4n\mathrm{}^2`$:
$$\widehat{c}(n,\mathrm{})=\widehat{c}(4n\mathrm{}^2)$$
Using (2.1) in (2.1) yields
$$_2(p,q,y)=\underset{s,k,n,\mathrm{}}{}s(p^kq^ny^{\mathrm{}})^s\widehat{c}(4nk\mathrm{}^2)$$
When $`(k,n,\mathrm{})`$ are coprime, $`\widehat{c}(4nk\mathrm{}^2)`$ counts BPS black holes with $`k=Q_1Q_5`$, $`S^1`$ momentum $`n`$ and spin $`J_L^3=\frac{\mathrm{}}{2}`$, multiplied by an overall $`()^{\mathrm{}}`$ and summed over $`J_R^3`$ weighted by $`2(J_R^3)^2()^{2J_R^3}`$:
$$\widehat{c}(4nk\mathrm{}^2)|_{(k,n,\mathrm{})\mathrm{coprime}}=()^{\mathrm{}}\underset{J_R,BPSstates}{}2(J_R^3)^2()^{2J_R^3}$$
When they are not coprime, the black hole can fragment, and the situation is more complicated due to multiple contributions in $`_2`$ . In this paper we will always avoid this complication by choosing coprime charges.
We should note that $`Z(q,y)`$ is also the modified elliptic genus of $`T^4`$, i.e.
$$_2^{(1)}=\underset{n,\mathrm{}}{}\widehat{c}(n,\mathrm{})q^ny^{\mathrm{}}=Z(q,y).$$
This corresponds to the coprime D1-D5 system with $`k=1=Q_1=Q_5`$. By writing
$$Z(q,y)=\underset{m}{}\widehat{c}(4m)q^m\underset{k}{}q^{k^2}y^{2k}+\underset{m}{}\widehat{c}(4m1)q^m\underset{k}{}q^{k^2+k}y^{2k+1}$$
and using (2.1) along with the standard Fourier expansion of the theta function
$$\vartheta _1(y|q)=i\underset{nZ}{}(1)^nq^{(n1/2)^2/2}y^{n1/2}$$
one can reorganize the generating functions for $`\widehat{c}`$ as
$$\begin{array}{cc}& \underset{m}{}\widehat{c}(4m)q^m=q^{\frac{1}{4}}\eta (q)^6\underset{mZ}{}q^{m^2+m},\hfill \\ & \underset{m}{}\widehat{c}(4m1)q^m=q^{\frac{1}{4}}\eta (q)^6\underset{mZ}{}q^{m^2}.\hfill \end{array}$$
These expressions will analyzed microscopically below in section 4.
3. The 4D modified elliptic genus
In this section we use the conjecture of to transform the 5D degeneracies into 4D ones. The fact that the $`\widehat{c}`$ coefficients depend only on the combination $`4nk\mathrm{}^2`$ is very encouraging, for the following reason. We expect the 1/8 BPS 5D degeneracies to be related to degeneracies of 1/8 BPS black holes in 4D, and in 4D U-duality implies that the black hole entropy must depend on the unique quartic invariant of $`E_{7,7}`$, the so-called Cremmer-Julia invariant . In an $`𝒩=4`$ language, this invariant takes the form
$$𝒥=q_e^2q_m^2(q_eq_m)^2$$
where $`q_e`$ and $`q_m`$ are the electric and magnetic charge vectors for $`𝒩=4`$ BPS states. (See e.g. for details on the notation.) This is precisely the dependence of $`\widehat{c}`$ on $`n`$, $`m`$, $`\mathrm{}`$, provided we identify
$$k=\frac{1}{2}q_e^2,n=\frac{1}{2}q_m^2,\mathrm{}=q_eq_m.$$
Note that from the purely 5D point of view, there was no obvious reason that $`\widehat{c}`$ should depend only on the combination $`4nk\mathrm{}^2`$ as there is no 5D U-duality which mixes spins with charges.
Let us now derive the identification (3.1) from the dictionary of , beginning from the IIB spinning 5D D1-D5-n black hole of the previous section. First we T-dual on $`S^1`$ to obtain a black hole with spin $`\frac{\mathrm{}}{2}`$, F-string winding $`n`$, $`Q_1`$ D0-branes, and $`Q_5`$ D4-branes. Now T-dual so that there are $`Q_1+Q_5`$ D2 branes with intersection number $`Q_1Q_5=k`$ on the $`T^4`$. Next we compactify on a single center Taub-NUT, whose asymptotic circle we identify as the the new M-theory circle. The result is three orthogonal sets of $`(n,Q_1,Q_5)`$ D2-branes on $`T^6`$, $`\mathrm{}`$ D0-branes, and one D6-brane. For IIA D-brane configurations with D0, D2, D4, D6 charges $`(q_0,q_{ij},p^{ij},p^0)`$, where $`i=1,\mathrm{}6`$ runs over the $`T^6`$ cycle and $`p^{ij}=p^{ji},q_{ij}=q_{ji}`$ $`𝒥`$ reduces to<sup>2</sup> See e.g. , equation (66), and take $`p^0=p_{87},p_{8i}=0`$, etc. Our definition of $`𝒥`$ differs from that of by a sign.
$$\begin{array}{cc}\hfill 𝒥=\frac{1}{12}& (q_0ϵ_{ijklmn}p^{ij}p^{kl}p^{mn}+p^0ϵ^{ijklmn}q_{ij}q_{kl}q_{mn})\hfill \\ & p^{ij}q_{jk}p^{kl}q_{li}+\frac{1}{4}p^{ij}q_{ij}p^{kl}q_{kl}(p^0q_0)^2+\frac{1}{2}p^0q_0p^{ij}q_{ij}.\hfill \end{array}$$
For our D0-D2-D6 configuration, we can pick a basis of cycles without loss of generality such that the nonzero charges are
$$p^0=0,q_0=\mathrm{},q_{12}=q_{21}=n,q_{34}=q_{43}=Q_1,q_{56}=q_{65}=Q_5$$
Then (3.1) reduces to
$$𝒥=4nk\mathrm{}^2,$$
which, as stated above, is exactly the argument of (2.1).
According to the weighted degeneracy of the 4D black hole resulting from U-duality and Taub-NUT compactification equals that of the original 5D black hole, when $`J_R^3`$ in (2.1) is identified with the generator $`J^3`$ of $`R^3`$ rotations in 4D. Note that, since $`𝒥`$ is odd if and only if $`\mathrm{}`$ is, we may trade $`()^{\mathrm{}}`$ for $`()^𝒥`$ in (2.1). Therefore, for fixed coprime charges, the weighted 4D BPS degeneracy depends only on the the Cremmer-Julia invariant and is given by
$$\underset{J^3,BPSstates}{}2(J^3)^2()^{2J^3}=()^𝒥\widehat{c}(𝒥).$$
Note that, although this formula for the 4D BPS degeneracy was derived assuming a specific D6-D2-D0 configuration, it applies to all D-brane configurations by U-duality.
As a first check on this conjecture, we note that for large charges $`\widehat{c}(𝒥)e^{\pi \sqrt{J}}`$. From the supergravity solutions $`\mathrm{Area}=4\pi \sqrt{J}`$, so there is agreement with the Bekenstein-Hawking entropy.
As an example, let’s consider the modified elliptic genus for the D4-D0 black hole on $`T^6`$, in which we fix the D4 charges and sum over D0 charge $`q_0`$. Consider the $`T^6`$ of the form $`T^2\times T^2\times T^2`$ with $`\alpha _1,\alpha _2,\alpha _3`$ being the three 2-cycles associated with the $`T^2`$’s. Let $`A^1,A^2,A^3`$ be the dual 4-cycles. We shall consider the D4-brane wrapped on the cycle $`[P]=A^1+A^2+A^3`$. Its triple self-intersection number is $`D=PPP=6`$. From (3.1) we have
$$𝒥=4q_0.$$
We then have
$$_2(q)=\underset{q_0Z}{}\widehat{c}(4q_0)q^{q_0}=q^{\frac{1}{4}}\eta (q)^6\underset{mZ}{}q^{m^2+m}.$$
according to (2.1).
A straightforward generalization of this example is the D4-D2-D0 system, where we wrap $`(q_1,q_2,q_3)`$ D2 branes on the 2-cycles $`(\alpha _1,\alpha _2,\alpha _3)`$. In this case, the Cremmer-Julia invariant becomes
$$𝒥=4(q_0+q_1q_2+q_1q_3+q_2q_3)(q_1+q_2+q_3)^2$$
and the sum over $`q_0`$ produces
$$_2(q)=\underset{q_0Z}{}(1)^𝒥\widehat{c}(𝒥)q^{q_0}=\{\begin{array}{cc}q^{\frac{1}{4}}\eta (q)^6_{mZ}q^{m^2+m\frac{1}{4}\stackrel{~}{𝒥}}\hfill & \text{ }q_1+q_2+q_3\text{ even}\hfill \\ q^{\frac{1}{4}}\eta (q)^6_{mZ}q^{m^2\frac{1}{4}\stackrel{~}{𝒥}\frac{1}{4}}\hfill & \text{ }q_1+q_2+q_3\text{ odd}\hfill \end{array}$$
where $`\stackrel{~}{𝒥}=4(q_1q_2+q_1q_3+q_2q_3)(q_1+q_2+q_3)^2`$. Now let us turn to the 4D derivation of (3.1) and (3.1).
4. Microscopic derivation in 4D
In this section we sketch a derivation of (3.1) and (3.1) using a 4D microscopic analysis. The derivation is not complete because, as we will discuss below, we ignore some potential subtleties associated to the fact that $`P`$ is not simply connected. In principle it should be possible to close this gap. A microscopic description of $`T^6`$ black holes using the M-theory picture of wrapped fivebranes has been given in , adapting the description given in for a general Calabi-Yau, in terms of a $`(0,4)`$ 2D CFT living on the M-theory circle. For uniformity and simplicity of presentation, we here will use the IIA description in which fivebrane momenta around the M-theory circle become bound states of D0 branes to D4 branes.
As above (3.1) we examine the special case of the D4-D0 system wrapped on $`[P]=A^1+A^2+A^3`$. The D4-D0 system can be described in terms of the quantum mechanics of $`q_0`$ D0-branes living on the D4-brane world volume $`P`$. The D4-brane world volume $`P`$ is holomorphically embedded in the $`T^6`$. One can compute its Euler character, $`\chi (P)=6`$. It follows from the Riemann-Roch formula that the only modulus of $`P`$ is the overall translation in $`T^6`$.<sup>3</sup> The dual line bundle $`_P`$ of the divisor $`P`$ has only one holomorphic section. However as $`T^6`$ is not simply connected, the line bundle $`_P`$ is not only determined by $`c_1(_P)=[P]`$. In fact the translation of $`T^6`$ takes it to a different line bundle. Since $`\chi (P)=6`$, $`P`$ has $`4+2b_1`$ 2-cycles. By the Lefschetz hyperplane theorem we have $`b_1(P)=b_1(T^6)=6`$, and therefore $`b_2(P)=16`$. All but one of the 2-cycles come from the intersection of $`P`$ with $`\left(\genfrac{}{}{0pt}{}{6}{4}\right)=15`$ 4-cycles in $`T^6`$. We will be mostly interested in 3 of these, denoted by $`\stackrel{~}{\alpha }_i`$, corresponding to intersections of $`A^i`$ with $`P`$. Turning on fluxes along these three 2-cycles corresponds to having charges of D2-branes wrapped on the $`\alpha _i`$’s. Their intersection numbers are
$$\stackrel{~}{\alpha }_i\stackrel{~}{\alpha }_j=\{\begin{array}{cc}\hfill 0,i=j& \\ \hfill 1,ij& \end{array}$$
There is, however, one extra 2-cycle in $`P`$, which we shall denote by $`\beta `$, that does not correspond to any cycle in the $`T^6`$.
One can show from the adjunction formula that $`c_1(P)`$ is Poincaré dual to $`(\stackrel{~}{\alpha }_1+\stackrel{~}{\alpha }_2+\stackrel{~}{\alpha }_3)`$. It then follows from Hirzebruch signature theorem that
$$\sigma (P)=\frac{2}{3}\chi (P)+\frac{1}{3}_Pc_1^2=2.$$
We conclude that the intersection form on $`P`$ is odd (and that $`P`$ is not a spin manifold). Essentially the unique way to extend (4.1) to an odd rank 4 unimodular quadratic form is to have an extra 2-cycle $`\gamma `$ with
$$\gamma \stackrel{~}{\alpha }_i=1,\gamma \gamma =1.$$
Now if we choose $`\beta =2\gamma \stackrel{~}{\alpha }_i`$, we have
$$\beta \stackrel{~}{\alpha }_i=0,\beta \beta =2.$$
Note that $`(\stackrel{~}{\alpha }_i,\beta )`$ is not an integral basis for $`H_2(P,Z)`$, yet $`\beta `$ is the smallest 2-cycle that doesn’t intersect $`\stackrel{~}{\alpha }_i`$. The total intersection form on $`P`$ is the sum of this rank 4 form together with 6 copies of $`\sigma _1`$ coming from the 12 other 2-cycles in $`P`$.
Now one can turn on gauge field flux on the D4-brane world volume along $`\beta `$, which does not correspond to any D2-brane charge. This flux nevertheless induces D0-brane charge. There is a subtlety in the quantization of this flux. As well known, the curvature of the D4-brane world volume induces an anomalous D0-brane charge $`\chi (P)/24=\frac{1}{4}`$. In order that the total D0 charge be integral the flux along the cycle $`\beta `$ on the D4-brane must be half-integer, i.e. of the form $`(m+\frac{1}{2})\beta `$. The total induced D0-brane charge is $`\mathrm{\Delta }q_0=\frac{1}{2}(m+\frac{1}{2})^2\beta \beta \frac{1}{4}=m^2+m`$, which is indeed an integer.<sup>4</sup> In the M-theory picture the anomalous D0 charge is the left-moving zero point energy $`\frac{c_L}{24}=\frac{1}{4}`$, and the 2-cycle fluxes correspond to momentum zero modes of scalars on a Narain lattice.
We ignore here the facts arising form nonzero $`b_1(P)`$ that there is a moduli space of flat connections as well as overall $`T^6`$ translations which must be quantized. These factors are treated in the language of the 2D CFT in . They are found to lead to extra degrees of freedom which are however eliminated by extra gauge constraints. A complete microscopic derivation, not given here, would have to show that a careful accounting of these factors give a trivial correction to our result.
It is now straightforward to reproduce (3.1). Each D0-D4 bound state is in a hypermultiplet which contributes minus one to $`Tr\left[2(J^3)^2()^{2J^3}\right]`$. Counting the number of ways of distributing $`n`$ D0-branes among the $`\chi (P)=6`$ ground states of the supersymmetric quantum mechanics, and then summing over $`n`$, gives the factor of $`q^{1/4}\eta (q)^6=_{k=1}^{\mathrm{}}(1q^k)^6`$ in (3.1). Including finally the sum over fluxes on $`\beta `$, we precisely reproduce the degeneracy (3.1)!
Let us now consider the more general case of D4-D2-D0 system. Again we shall assume $`(p^1,p^2,p^3)=(1,1,1)`$. The D2-brane charges are labelled by $`(q_1,q_2,q_3)`$. The bound state is described by the D4-brane with D2-brane dissolved in its world volume. We end up with the gauge flux
$$F=(m+1/2)\beta +\underset{i=1}{\overset{3}{}}q_i\delta _i,\delta _i\stackrel{~}{\alpha }_j=\delta _{ij}.$$
In above expression $`\delta _i`$ is defined up to a shift of an integer multiple of $`\beta `$. Since we are summing over $`m`$, this ambiguity is irrelevant. We can choose $`\delta _i=\gamma \stackrel{~}{\alpha }_i`$. The total induced D0-brane charge is then
$$\begin{array}{cc}\hfill \mathrm{\Delta }q_0& =\frac{1}{2}F^2\frac{1}{4}\hfill \\ & =(m+1/2)^2+(m+1/2)q_i+\frac{1}{2}q_i^2\frac{1}{4}\hfill \\ & =\left(m+\frac{1}{2}+\frac{1}{2}q_i\right)^2\frac{1}{12}D^{AB}q_Aq_B\frac{1}{4},\hfill \end{array}$$
where $`D^{AB}`$ is the inverse matrix of $`D_{AB}D_{ABC}p^C`$,
$$D^{AB}q_Aq_B=3(2q_1q_2+2q_2q_3+2q_3q_1q_1^2q_2^2q_3^2).$$
Note that $`\frac{1}{3}D^{AB}q_Aq_B=0\mathrm{mod}4`$ if $`q_i`$ is even, and $`\frac{1}{3}D^{AB}q_Aq_B=1\mathrm{mod}4`$ if $`q_i`$ is odd. Therefore $`\mathrm{\Delta }q_0`$ is always an integer, as expected. The Cremmer-Julia invariant is in this case
$$𝒥=4\left(q_0+\frac{1}{12}D^{AB}q_Aq_B\right).$$
The counting of D0-brane states as before gives the generating function
$$\underset{q_0}{}()^𝒥c(𝒥)q^{q_0}=\underset{k=1}{\overset{\mathrm{}}{}}(1q^k)^6\underset{mZ}{}q^{m^2+m\frac{1}{12}D^{AB}q_Aq_B}$$
in the case $`q_i2Z`$ and $`𝒥0\mathrm{mod}4`$, and
$$\underset{q_0}{}()^𝒥c(𝒥)q^{q_0}=\underset{k=1}{\overset{\mathrm{}}{}}(1q^k)^6\underset{mZ}{}q^{m^2\frac{1}{12}D^{AB}q_Aq_B\frac{1}{4}}$$
in the case $`q_i2Z+1`$ and $`𝒥1\mathrm{mod}4`$. These are precisely the degeneracies (3.1) we derived from 5D earlier!
Acknowledgments
This work was supported in part by DOE grant DEFG02-91ER-40654. We are grateful to Greg Moore and Boris Pioline for useful correspondence.
References
relax D. Gaiotto, A. Strominger and X. Yin, “New connections between 4D and 5D black holes,” arXiv:hep-th/0503217. relax J. M. Maldacena, G. W. Moore and A. Strominger, “Counting BPS black holes in toroidal type II string theory,” arXiv:hep-th/9903163. relax G. Lopes Cardoso, B. de Wit and T. Mohaupt, “Corrections to macroscopic supersymmetric black-hole entropy,” Phys. Lett. B 451, 309 (1999) \[arXiv:hep-th/9812082\]; H. Ooguri, A. Strominger and C. Vafa, “Black hole attractors and the topological string,” Phys. Rev. D 70, 106007 (2004) \[arXiv:hep-th/0405146\]; A. Dabholkar, F. Denef, G. W. Moore and B. Pioline, “Exact and asymptotic degeneracies of small black holes,” arXiv:hep-th/0502157; G. L. Cardoso, B. de Wit, J. Kappeli and T. Mohaupt, “Asymptotic degeneracy of dyonic N = 4 string states and black hole entropy,” JHEP 0412, 075 (2004) \[arXiv:hep-th/0412287\]; A. Dabholkar, “Exact counting of black hole microstates,” arXiv:hep-th/0409148; E. Verlinde, “Attractors and the holomorphic anomaly,” arXiv:hep-th/0412139; relax R. Dijkgraaf, E. Verlinde and H. Verlinde, “Counting dyons in N = 4 string theory,” Nucl. Phys. B 484, 543 (1997) \[arXiv:hep-th/9607026\]. relax D. Shih, A. Strominger and X. Yin, “Recounting dyons in N = 4 string theory,” arXiv:hep-th/0505094. relax V. Balasubramanian and F. Larsen, “On D-Branes and Black Holes in Four Dimensions,” Nucl. Phys. B 478, 199 (1996) \[arXiv:hep-th/9604189\]. relax V. Balasubramanian, “How to count the states of extremal black holes in N = 8 supergravity,” arXiv:hep-th/9712215. relax M. Bertolini and M. Trigiante, “Microscopic entropy of the most general four-dimensional BPS black hole,” JHEP 0010, 002 (2000) \[arXiv:hep-th/0008201\]. relax R. Kallosh and B. Kol, “E(7) Symmetric Area of the Black Hole Horizon,” Phys. Rev. D 53, 5344 (1996) \[arXiv:hep-th/9602014\]. relax E. Cremmer and B. Julia, “The SO(8) Supergravity,” Nucl. Phys. B 159, 141 (1979). relax S. Ferrara and R. Kallosh, “Universality of Supersymmetric Attractors,” Phys. Rev. D 54, 1525 (1996) \[arXiv:hep-th/9603090\]. relax J. M. Maldacena, A. Strominger and E. Witten, “Black hole entropy in M-theory,” JHEP 9712, 002 (1997) \[arXiv:hep-th/9711053\].
|
warning/0506/quant-ph0506074.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
Both the matrix eigenvalue problem and the inverse eigenvalue problem are the subjects of intensive research . This paper addresses both problems. A means of producing matrices with any specified eigenvalues and whose eigenvectors are known is presented. Both the elements of the matrix and the eigenvectors of the matrix are given in analytic form; this should facilitate the use of such matrices for analysis in other areas of linear algebra. A reversal of the matrix-generation procedure naturally suggests a matrix-diagonalization algorithm, and we derive the equation which needs to be solved to realize this. The findings we are going to present grew out of research into quantum foundations, but are here presented from a purely mathematical point of view.
The organization of this paper is as follows. In the next section, we present the basic result that makes possible the production of matrices with any desired eigenvalues. In Section 3, we outline the diagonalization routine that arises from it. In Section 4, we give the functions which for the case $`N=5`$ enable the practical realization of both these projects. The results of preliminary application of this theory are reported in Section 5. We conclude with brief comments in Section 6.
## 2 Matrix Generation
Suppose that the $`N^2`$ functions $`\varphi (B_i;C_j)`$ possess the orthonormality property
$$\underset{l=1}{\overset{N}{}}\varphi (B_l;C_i)\varphi ^{}(B_l;C_j)=\delta _{ij},$$
(1)
where $`B`$ and $`C`$ are parameters that take the values $`B_1`$,$`B_2`$,…,$`B_N`$ and $`C_1`$, $`C_2`$,…,$`C_N`$ respectively. The quantities $`\varphi (B_i;C_j)`$ are functions of some variables, collectively denoted by $`x`$. If with the aid of the $`N`$ arbitrary numbers $`\lambda _1,\lambda _2,\mathrm{}\lambda _N`$ we form the $`N\times N`$ matrix $`[M]`$ using the rule
$$M_{ij}=\underset{n=1}{\overset{N}{}}\varphi ^{}(B_i;C_n)\varphi (B_j;C_n)\lambda _n$$
(2)
we find that the eigenvectors of this matrix are
$$[\xi _i]=\left(\begin{array}{c}\varphi ^{}(B_1;C_i)\\ \varphi ^{}(B_2;C_i)\\ :\\ \varphi ^{}(B_N;C_i)\end{array}\right)$$
(3)
with the respective eigenvalues $`\lambda _i.`$ Indeed if we write
$$[V]=[M][\xi _i],$$
(4)
we find that the $`k`$th row of $`[M][\xi _i]`$ is
$`V_k`$ $`=`$ $`{\displaystyle \underset{l=1}{\overset{N}{}}}M_{kl}\varphi ^{}(B_l;C_i)`$
$`=`$ $`{\displaystyle \underset{l=1}{\overset{N}{}}}\left({\displaystyle \underset{n=1}{\overset{N}{}}}\varphi ^{}(B_k;C_n)\varphi (B_l;C_n)\lambda _n\right)\varphi ^{}(B_l;C_i)`$
$`=`$ $`{\displaystyle \underset{n=1}{\overset{N}{}}}\lambda _n\varphi ^{}(B_k;C_n){\displaystyle \underset{l=1}{\overset{N}{}}}\varphi (B_l;C_n)\varphi ^{}(B_l;C_i)`$
$`=`$ $`{\displaystyle \underset{n=1}{\overset{N}{}}}\lambda _n\varphi ^{}(B_k;C_n)\delta _{ni}`$
$`=`$ $`\lambda _i\varphi ^{}(B_k;C_i).`$ (5)
Thus the eigenvalue equation
$$[M][\xi _i]=\lambda _i[\xi _i]$$
(6)
is satisfied. Property Eqn. (1) means that the eigenvectors are orthonormal:
$`[\xi _i]^{}[\xi _j]`$ $`=`$ $`{\displaystyle \underset{l}{}}\varphi (B_l;C_i)\varphi ^{}(B_l;C_j)`$ (7)
$`=`$ $`\delta _{ij}.`$
These results give us a means of generating matrices with specified eigenvalues and known eigenvectors. We only need appropriate functions $`\varphi (B_i;C_j).`$ An example of such functions will be given in Section 4. Given such functions, and a set of eigenvalues $`\{\lambda _i\},`$ we are able generate an infinite number of different matrices by varying the argument $`x`$ of these functions. Each such matrix is generated together with its own orthonormal set of eigenvectors.
On the other hand, given a value of $`x`$, we can generate an infinite number of matrices by using different sets of eigenvalues. In this case, all these matrices share the same set of orthonormal eigenvectors.
## 3 A Diagonalization Algorithm
Since each matrix is generated by means of a specific value of the argument $`x,`$ we should be able, at least in principle, to find a way of deducing this value of $`x`$ for a given matrix. A procedure for doing this would constitute a diagonalization routine. In this section, we derive one such method.
Suppose we are given a matrix $`[M]`$ formed by the prescription Eqn. (2). Its eigenvectors are given by Eqn. (3). If we only knew the value of $`x`$ that was used to form the matrix, then we could fix the eigenvectors and recover the eigenvalues. In order to develop a way of deducing the value of $`x,`$ let us start, for the sake of convenience, by setting
$$\xi (C_i;B_j)=\varphi ^{}(B_j;C_i)$$
(8)
so that
$$[\xi _i]=\left(\begin{array}{c}\varphi ^{}(B_1;C_i)\\ \varphi ^{}(B_2;C_i)\\ :\\ \varphi ^{}(B_N;C_i)\end{array}\right)=\left(\begin{array}{c}\xi (C_i,B_1)\\ \xi (C_i,B_2)\\ :\\ \xi (C_i,B_N)\end{array}\right).$$
(9)
Now, from
$$[M][\xi _i]=\lambda _i[\xi _i]$$
(10)
and the value of $`x`$ that was used to produce the matrix, we can recover the eigenvalue $`\lambda _i`$ from the $`r`$th row by using
$$(\lambda _i)_r=\frac{_lM_{rl}\xi (C_i,B_l)}{\xi (C_i,B_r)}.$$
(11)
The $`s`$th row would give the same value. Hence, comparing the values from the $`r`$th and $`s`$th rows, we can write
$$(\lambda _i)_r(\lambda _i)_s=0.$$
(12)
This must be true for all possible combinations of $`r`$ and $`s:`$
$$\underset{r=1}{\overset{N1}{}}\underset{s=r+1}{\overset{N}{}}\left[(\lambda _i)_r(\lambda _i)_s\right]=0.$$
(13)
In other words:
$$\underset{r=1}{\overset{N1}{}}\underset{s=r+1}{\overset{N}{}}\left[\frac{_lM_{rl}\xi (C_i,B_l)}{\xi (C_i,B_r)}\frac{_lM_{sl}\xi (C_i,B_l)}{\xi (C_i,B_s)}\right]=0.$$
(14)
But if the value of $`x`$ used is not correct, the equation is not satisfied. Thus, this equation is a means of deducing the correct value of $`x`$. A form of this equation better adapted to numerical work is
$$\underset{i=1}{\overset{N}{}}\underset{r=1}{\overset{N1}{}}\underset{s=r+1}{\overset{N}{}}\left[\xi (C_i,B_s)\underset{l}{}M_{rl}\xi (C_i,B_l)\xi (C_i,B_r)\underset{l}{}M_{sl}\xi (C_i,B_l)\right]=0.$$
(15)
We have summed over all the eigenvalues by means of the index $`i`$ because we must ensure that the equation is satisfied for every eigenvalue.
Once the correct value of $`x`$ $`(`$denoted by $`x_0)`$ is found, the eigenvalues are calculated using
$$(\lambda _i)_r=\frac{_lM_{rl}\xi (C_i,B_l)}{\xi (C_i,B_r)}|_{x_0}$$
(16)
and the eigenvectors are
$$[\xi _i]=\left(\begin{array}{c}\xi (C_i,B_1)\\ \xi (C_i,B_2)\\ :\\ \xi (C_i,B_N)\end{array}\right)|_{x_0}.$$
(17)
As is wise in the solution of such equations, it is necessary to establish that the value of $`x`$ obtained is not spurious by putting it back into the equation. In this case, this means using it to calculate the eigenvalue using every line of the eigenvector. Thus, for each eigenvalues, there should be $`N`$ values, which should agree in order for this value of $`x`$ to be acceptable. Of course, in the case where one of the elements of the eigenvectors vanishes, it cannot be used to compute the eigenvalue, and there will be fewer than $`N`$ values to compare.
## 4 Probability Amplitudes
In order to utilise the prescription for generating matrices with specified eigenvalues and to realise the algorithm for diagonalizing them, we need the functions $`\varphi (B_i;C_j).`$ Since the origins of the method presented in this paper lie in quantum theory, we shall call these functions probability amplitudes. A source for such functions is spin theory in quantum mechanics. The treatment of a spin $`s`$ system yields functions suitable for generating matrices of order $`N=2s+1`$. As an example, we take the spin-2 system. In the direction of the unit vector
$$\widehat{𝐜}=(\mathrm{sin}\theta \mathrm{cos}\phi ,\mathrm{sin}\theta \mathrm{sin}\phi ,\mathrm{cos}\theta )$$
(18)
its operator is
$$[\widehat{𝐜}𝐒]=\left(\begin{array}{ccccc}2\mathrm{cos}\theta & \mathrm{sin}\theta e^{i\phi }& 0& 0& 0\\ \mathrm{sin}\theta e^{i\phi }& \mathrm{cos}\theta & \frac{\sqrt{6}}{2}\mathrm{sin}\theta e^{i\phi }& 0& 0\\ 0& \frac{\sqrt{6}}{2}\mathrm{sin}\theta e^{i\phi }& 0& i\frac{\sqrt{6}}{2}\mathrm{sin}\theta e^{i\phi }& 0\\ 0& 0& \frac{\sqrt{6}}{2}\mathrm{sin}\theta e^{i\phi }& \mathrm{cos}\theta & \mathrm{sin}\theta e^{i\phi }\\ 0& 0& 0& \mathrm{sin}\theta e^{i\phi }& 2\mathrm{cos}\theta \end{array}\right).$$
(19)
The eigenvalues of this matrix are $`2,1,0,1`$ and $`2`$, with the respective normalized eigenvectors
$$[\chi _{m=2}^{(\widehat{𝐜})}]=\left(\begin{array}{c}\mathrm{cos}^4\frac{\theta }{2}e^{i2\phi }\\ \mathrm{sin}\theta \mathrm{cos}^2\frac{\theta }{2}e^{i\phi }\\ \frac{\sqrt{6}}{4}\mathrm{sin}^2\theta \\ \mathrm{sin}\theta \mathrm{sin}^2\frac{\theta }{2}e^{i\phi }\\ \mathrm{sin}^4\frac{\theta }{2}e^{i2\phi }\end{array}\right),$$
(20)
$$[\chi _{m=1}^{(\widehat{𝐜})}]=\left(\begin{array}{c}\mathrm{sin}\theta \mathrm{cos}^2\frac{\theta }{2}e^{i2\phi }\\ (3\mathrm{sin}^2\frac{\theta }{2}\mathrm{cos}^2\frac{\theta }{2})\mathrm{cos}^2\frac{\theta }{2}e^{i\phi }\\ \frac{\sqrt{6}}{2}\mathrm{sin}\theta \mathrm{cos}\theta \\ (3\mathrm{cos}^2\frac{\theta }{2}\mathrm{sin}^2\frac{\theta }{2})\mathrm{sin}^2\frac{\theta }{2}e^{i\phi }\\ \mathrm{sin}\theta \mathrm{sin}^2\frac{\theta }{2}e^{i2\phi }\end{array}\right),$$
(21)
$$[\chi _{m=0}^{(\widehat{𝐜})}]=\left(\begin{array}{c}\frac{\sqrt{6}}{4}\mathrm{sin}^2\theta e^{i2\phi }\\ \frac{\sqrt{6}}{2}\mathrm{sin}\theta \mathrm{cos}\theta e^{i\phi }\\ \frac{1}{2}(2\mathrm{cos}^2\theta \mathrm{sin}^2\theta )\\ \frac{\sqrt{6}}{2}\mathrm{sin}\theta \mathrm{cos}\theta e^{i\phi }\\ \frac{\sqrt{6}}{4}\mathrm{sin}^2\theta e^{i2\phi }\end{array}\right),$$
(22)
$$[\chi _{m=1}^{(\widehat{𝐜})}]=\left(\begin{array}{c}\mathrm{sin}\theta \mathrm{sin}^2\frac{\theta }{2}e^{i2\phi }\\ (3\mathrm{cos}^2\frac{\theta }{2}\mathrm{sin}^2\frac{\theta }{2})\mathrm{sin}^2\frac{\theta }{2}e^{i\phi }\\ \frac{\sqrt{6}}{2}\mathrm{sin}\theta \mathrm{cos}\theta \\ (3\mathrm{sin}^2\frac{\theta }{2}\mathrm{cos}^2\frac{\theta }{2})\mathrm{cos}^2\frac{\theta }{2}e^{i\phi }\\ \mathrm{sin}\theta \mathrm{cos}^2\frac{\theta }{2}e^{i2\phi }\end{array}\right)$$
(23)
and
$$[\chi _{m=2}^{(\widehat{𝐜})}]=\left(\begin{array}{c}\mathrm{sin}^4\frac{\theta }{2}e^{i2\phi }\\ \mathrm{sin}\theta \mathrm{sin}^2\frac{\theta }{2}e^{i\phi }\\ \frac{\sqrt{6}}{4}\mathrm{sin}^2\theta \\ \mathrm{sin}\theta \mathrm{cos}^2\frac{\theta }{2}e^{i\phi }\\ \mathrm{cos}^4\frac{\theta }{2}e^{i2\phi }\end{array}\right).$$
(24)
We have used the index $`m`$ to label the eigenvalues of the spin matrix because that is the convention. Also, we need to keep the distinction between the eigenvalues of this matrix and the eigenvalues $`\{\lambda _i\}`$ belonging to the matrix we wish to generate once we have obtained the probability amplitudes.
For the direction
$$\widehat{𝐛}=(\mathrm{sin}\theta ^{}\mathrm{cos}\phi ^{},\mathrm{sin}\theta ^{}\mathrm{sin}\phi ^{},\mathrm{cos}\theta ^{})$$
(25)
the eigenvectors of the operator are identical, except that $`(\theta ,\phi )`$ is replaced by $`(\theta ^{},\phi ^{}).`$ We now associate the parameter $`B`$ with the direction $`\widehat{𝐛}`$, and the parameter $`C`$ with the direction $`\widehat{𝐜}`$. The notation is such that the values $`B_1,B_2,\mathrm{},B_5`$ correspond respectively to the eigenvalues $`2,1,\mathrm{},2`$ . The same holds for $`C`$.
Some familiarity with spin theory in quantum mechanics is perhaps necessary to fully appreciate these functions. Each element in each eigenvector is a probability amplitude, and its modulus square is a probability for a certain measurement. However, from a mathematical point of view, all that is needed is the fact that these elements possess the property Eqn. (1). The scalar products of the eigenvectors corresponding to $`\widehat{𝐛}`$ with those corresponding to $`\widehat{𝐜}`$ give the most generalized forms of these functions. Thus the required quantities are
$$\varphi (B_i;C_j)=[\chi _{m_j}^{(\widehat{𝐜})}]^{}[\chi _{m_i}^{(\widehat{𝐛})}].$$
(26)
Evidently, in this case, $`x=(\theta ,\phi ,\theta ^{},\phi ^{}).`$
To illustrate the notation, we give a few of these functions:
$`\varphi (B_1;C_1)`$ $`=`$ $`[\chi _{m=2}^{(\widehat{𝐜})}]^{}[\chi _{m=2}^{(\widehat{𝐛})}]`$
$`=`$ $`\mathrm{cos}^4{\displaystyle \frac{\theta }{2}}\mathrm{cos}^4{\displaystyle \frac{\theta ^{}}{2}}e^{i2(\phi \phi ^{})}`$ (27)
$`+\mathrm{sin}\theta ^{}\mathrm{sin}\theta \mathrm{cos}^2{\displaystyle \frac{\theta ^{}}{2}}\mathrm{cos}^2{\displaystyle \frac{\theta }{2}}e^{i(\phi \phi ^{})}+{\displaystyle \frac{3}{8}}\mathrm{sin}^2\theta ^{}\mathrm{sin}^2\theta `$
$`+\mathrm{sin}\theta ^{}\mathrm{sin}\theta \mathrm{sin}^2{\displaystyle \frac{\theta ^{}}{2}}\mathrm{sin}^2{\displaystyle \frac{\theta }{2}}e^{i(\phi \phi ^{})}`$
$`+\mathrm{sin}^4{\displaystyle \frac{\theta }{2}}\mathrm{sin}^4{\displaystyle \frac{\theta ^{}}{2}}e^{i2(\phi \phi ^{})},`$
$`\varphi (B_1;C_2)`$ $`=`$ $`[\chi _{m=1}^{(\widehat{𝐜})}]^{}[\chi _{m=2}^{(\widehat{𝐛})}]`$
$`=`$ $`\mathrm{sin}\theta \mathrm{cos}^4{\displaystyle \frac{\theta ^{}}{2}}\mathrm{cos}^2{\displaystyle \frac{\theta }{2}}e^{i2(\phi \phi ^{})}`$ (28)
$`+(3\mathrm{sin}^2{\displaystyle \frac{\theta }{2}}\mathrm{cos}^2{\displaystyle \frac{\theta }{2}})\mathrm{sin}\theta ^{}\mathrm{cos}^2{\displaystyle \frac{\theta ^{}}{2}}\mathrm{cos}^2{\displaystyle \frac{\theta }{2}}e^{i(\phi \phi ^{})}`$
$`{\displaystyle \frac{3}{4}}\mathrm{sin}^2\theta ^{}\mathrm{sin}\theta \mathrm{cos}\theta `$
$`(3\mathrm{cos}^2{\displaystyle \frac{\theta }{2}}\mathrm{sin}^2{\displaystyle \frac{\theta }{2}})\mathrm{sin}\theta ^{}\mathrm{sin}^2{\displaystyle \frac{\theta ^{}}{2}}\mathrm{sin}^2{\displaystyle \frac{\theta }{2}}e^{i(\phi \phi ^{})}`$
$`\mathrm{sin}\theta \mathrm{sin}^4{\displaystyle \frac{\theta ^{}}{2}}\mathrm{sin}^2{\displaystyle \frac{\theta }{2}}e^{i2(\phi \phi ^{})}`$
and
$`\varphi (B_5;C_5)`$ $`=`$ $`[\chi _{m=2}^{(\widehat{𝐜})}]^{}[\chi _{m=2}^{(\widehat{𝐛})}]`$
$`=`$ $`\mathrm{cos}^4{\displaystyle \frac{\theta }{2}}\mathrm{cos}^4{\displaystyle \frac{\theta ^{}}{2}}e^{i2(\phi \phi ^{})}`$ (29)
$`+\mathrm{sin}\theta ^{}\mathrm{sin}\theta \mathrm{sin}^2{\displaystyle \frac{\theta ^{}}{2}}\mathrm{sin}^2{\displaystyle \frac{\theta }{2}}e^{i(\phi \phi ^{})}+{\displaystyle \frac{3}{8}}\mathrm{sin}^2\theta ^{}\mathrm{sin}^2\theta `$
$`+\mathrm{sin}\theta ^{}\mathrm{sin}\theta \mathrm{cos}^2{\displaystyle \frac{\theta ^{}}{2}}\mathrm{cos}^2{\displaystyle \frac{\theta }{2}}e^{i(\phi \phi ^{})}`$
$`+\mathrm{cos}^4{\displaystyle \frac{\theta }{2}}\mathrm{cos}^4{\displaystyle \frac{\theta ^{}}{2}}e^{i2(\phi \phi ^{})}.`$
There are 22 other functions. We remind ourselves that since
$$\xi (C_i,B_j)=\varphi ^{}(B_j;C_i),$$
(30)
then
$$\xi (C_i,B_j)=[\chi _{m_i}^{(\widehat{𝐛})}]^{}[\chi _{m_j}^{(\widehat{𝐜})}].$$
(31)
These are the functions which enter into Eqn. (15).
In general, the probability amplitudes satisfy the inter-dependence law
$$\varphi (B_i;C_j)=\underset{l=1}{\overset{5}{}}\varphi (B_i;D_l)\varphi ^{}(D_l;C_j),$$
(32)
where $`\widehat{𝐝}`$ is a third direction to which corresponds the new parameter $`D`$. This is a particular form of a fundamental quantum property of three sets of probability amplitudes belonging to one quantum system ,
$$\psi (A_i;C_j)=\underset{l=1}{\overset{N}{}}\chi (A_i;B_l)\varphi (B_l;C_j).$$
(33)
In the context of quantum theory, the quantity $`\left|\varphi (B_i;C_j)\right|^2`$ is the probability that if the spin projection in the direction $`\widehat{𝐛}`$ is $`m_i\mathrm{}`$, a measurement of it in the direction $`\widehat{𝐜}`$ yields the result $`m_j\mathrm{}`$. Here $`\mathrm{}`$is Planck’s constant.
The parameters $`B`$ and $`C`$, as observed, are defined by the angles $`(\theta ^{},\phi ^{})`$ and $`(\theta ,\phi ),`$ and the indices of each parameter correspond to the eigenvalues. When $`(\theta ^{},\phi ^{})=(\theta ,\phi )`$ in Eqn. (33), $`A=C`$ and Eqn. (1) is satisfied. Eqn. (1) is a special form of Eqn. (33). Within the context of quantum measurement theory, all this has a ready interpretation. However, we stress that it is not necessary to understand this in order to use these functions to generate matrices customized as to eigenvalues and type.
With these probability amplitudes, the eigenvector for the eigenvalue $`\lambda _i`$ is
$$[\xi _i]=\left(\begin{array}{c}\varphi ^{}(B_1;C_i)\\ \varphi ^{}(B_2;C_i)\\ \varphi ^{}(B_3;C_i)\\ \varphi ^{}(B_4;C_i)\\ \varphi ^{}(B_5;C_i)\end{array}\right)=\left(\begin{array}{c}[\chi _{m=2}^{(\widehat{𝐛})}]^{}[\chi _{m_i}^{(\widehat{𝐜})}]\\ [\chi _{m=1}^{(\widehat{𝐛})}]^{}[\chi _{m_i}^{(\widehat{𝐜})}]\\ [\chi _{m=0}^{(\widehat{𝐛})}]^{}[\chi _{m_i}^{(\widehat{𝐜})}]\\ [\chi _{m=1}^{(\widehat{𝐛})}]^{}[\chi _{m_i}^{(\widehat{𝐜})}]\\ [\chi _{m=2}^{(\widehat{𝐛})}]^{}[\chi _{m_i}^{(\widehat{𝐜})}]\end{array}\right).$$
(34)
Of course, since the matrix Eqn. (19) is a particular case of the form of Eqn. (2), its eigenvectors are given by Eqn. (34); evidently, they correspond to the angles $`\theta ^{}=0`$ and $`\phi ^{}=0.`$ In other words, for this case, the unit vector $`\widehat{𝐛}`$ is in the $`z`$ direction.
The association between the eigenvalues of the matrix we are generating and their eigenvectors is through Eqn. (2), where the value of $`C`$ and the eigenvalue have the same index.
## 5 Results
A Fortran 77 program written to generate the matrices confirms that different families of matrices are obtained depending on the values of $`\theta ,\phi ,\theta ^{}`$ and $`\phi ^{},`$and on the character of the eigenvalues $`\{\lambda _i\}.`$ The following observations are made, and can be predicted from the structure of the probability amplitudes and of Eqn. (2).
If all the angles are zero or if $`\theta =\theta ^{}`$ and $`\phi =\phi ^{},`$ the resulting matrix is diagonal.
If the eigenvalues are real, the matrix is Hermitian, but its elements depend on what values of the angles are used.
Whatever the eigenvalues, the matrix is symmetric if $`\phi =\phi ^{}.`$
If $`\phi =\phi ^{},`$ the eigenvectors are real.
If all the eigenvalues are pure imaginary, and the arguments are arbitrary, the matrix is anti-Hermitian.
If all the eigenvalues are pure imaginary, and $`\phi =\phi ^{},`$ the matrix is imaginary but symmetric.
If the eigenvalues are arbitrary and the values of the arguments also arbitrary, the matrix is general.
The kinds of families that can be generated with different combinations of the eigenvectors and the arguments have not been fully investigated, but it seems probable that it does not require much ingenuity to generate such special forms as tridiagonal matrices, etc. Clearly, with this method, much of the inverse-eigenvalue problem is solved. Further investigation should show how to chose the values of the angles so as to obtain the desired matrices with specific values of certain elements.
As far as the diagonalization algorithm is concerned, it has only been partially validated at this time. The main complication here is that the non-linear equation (15) is in four variables. Therefore, solving it is somewhat complicated. The following was however achieved. A matrix was generated by means of Eqn. (2). In order to diagonalize it, three of the arguments were held at their actual values, and the remaining one was determined with the use of Eqn. (15). Which one of the variables to treat as unknown could be decided at pleasure. The value of this variable was determined from Eqn. (15) by means of the bisection method. That this approach was successful indicates that once a fast and reliable method of solving a non-linear equation in four variables is employed, the algorithm will prove to be of some utility.
## 6 Conclusion and Discussion
In this paper, we have presented a prescription for forming matrices in such a way that their eigenvalues and eigenvectors are known. The method is very general indeed, and simply by varying the arguments, different kinds of matrices can be obtained. Always, the normalised eigenvectors are simultaneously given. The eigenvectors can be chosen to be real if desired, and the any combination of eigenvalues can be used. By essentially reversing the matrix-generation procedure, we have proposed an algorithm for diagonalizing matrices formed this way.
A good amount of work is still needed in order to fully understand and utilize the methods presented. For example, it is necessary to classify more properly and completely according to the values of the arguments $`\theta ,\phi ,\theta ^{}`$ and $`\phi ^{}`$ the kinds of matrices that can be generated. Such information would be particularly helpful in the solution of the non-linear equation. If a matrix is such that one or more of the arguments must have certain values, this makes so much easier the job of solving the equation, since the number of variables is reduced.
The matrices dealt with here are of order $`5`$. In order to deal with matrices of higher order, it is necessary to have the functions $`\{\varphi \}`$ for those cases. A source of these functions will always be spin theory, but treatment of other $`N`$-dimensional quantum systems should produce other forms of the functions. Each such set of functions probably produces matrices of different characters. As such, it expands the range of uses to which we can put these matrices.
Acknowledgements
The author thanks the Staff and the Director at the Institute for Mathematical Sciences in Chennai, India, where this work was carried out.
## 7 References
1. G. H. Golub, H. A. van der Vorst, Eigenvalue computation in the 20th century, J. Comp. and Appl. Math. 123 (2000) 35-36
2. B. Boley, G. H. Golub, A survey of matrix inverse eigenvalue problems, Inverse Problems 3 (1987) 595-622
3. H. V. Mweene, Derivation of spin vectors and operators from first principles, quant-ph/9905012
4. See any standard text on quantum mechanics, such as B. H. Bransden and C. J. Joachain, Introduction to Quantum Mechanics, Longman Scientific & Technical, 1989.
5. H. V. Mweene, Generalized probability amplitudes for spin projection measurements on spin 2 systems, quant-ph/0502005
6. A. Landé, New Foundations of Quantum Mechanics, Cambridge University Press, 1965.
|
warning/0506/math0506018.html
|
ar5iv
|
text
|
# From triangulated categories to cluster algebras
## 1. Introduction
Cluster algebras were introduced by S. Fomin and A. Zelevinsky . They are subrings of the field $`(u_1,\mathrm{},u_m)`$ of rational fractions in $`m`$ indeterminates, and defined via a set of generators constructed recursively. These generators are called cluster variables and are grouped into subsets of fixed finite cardinality called clusters. The recursion process begins with a pair $`(𝐱,B)`$, called a seed, where $`𝐱`$ is an initial cluster and $`B`$ is a rectangular matrix with integer coefficients.
The first aim of the theory was to provide an algebraic framework for the study of total positivity and of Lusztig/Kashiwara’s canonical bases of quantum groups. The first result is the Laurent phenomenon which asserts that the cluster variables, and thus the cluster algebra they generate, are contained in the Laurent polynomial ring $`[u_1^{\pm 1},\mathrm{},u_m^{\pm 1}]`$.
Since its foundation, the theory of cluster algebras has witnessed intense activity, both, in its own foundations and in its connections with other research areas. One important aim has been to prove, as in , , … that many algebras encountered in the theory of reductive Lie groups have (at least conjecturally) a structure of cluster algebra with an explicit seed. On the other hand, a number of recent articles have been devoted to establishing links with subjects beyond Lie theory. These links mainly rely on the combinatorics by which cluster variables are grouped into clusters. Among the subjects concerned we find Poisson geometry , where clusters are interpreted in terms of integrable systems, Teichmüller theory , where clusters are viewed as systems of local coordinates, and tilting theory , where clusters are interpreted as sets of indecomposable factors of a tilting module.
A cluster algebra is said to be of finite type if the number of cluster variables is finite. In , S. Fomin and A. Zelevinsky classify the cluster algebras of finite type in terms of Dynkin diagrams. The cluster variables are then in bijection with the almost positive roots of the corresponding root system, i.e. the roots which are positive or opposite to simple roots. Note that this classification is analogous to P. Gabriel’s classification of representation-finite quivers; it is also analogous to the classification of finite dimensional semisimple Lie algebras in the theory of Kac-Moody algebras.
In finite type, the combinatorics of the clusters are governed by generalized associahedra. The purpose of the decorated categories of and, later, of the cluster categories of is to offer a better understanding of these combinatorics.
Let $`Q`$ be a quiver whose underlying graph is a simply laced Dynkin diagram and let $`\mathrm{𝗆𝗈𝖽}kQ`$ be the category of finite-dimensional representations of $`Q`$ over a field $`k`$. The cluster category $`𝒞`$ is the orbit category of the bounded derived category $`𝒟^b(\mathrm{𝗆𝗈𝖽}kQ)`$ under the action of a canonical automorphism. Thus, it only depends on the underlying Dynkin diagram and not on the orientation of the arrows in the quiver $`Q`$. With hindsight, the canonical automorphism is chosen so as to extend the bijection between indecomposable $`kQ`$-modules and positive roots to a bijection between indecomposable objects of $`𝒞`$ and almost positive roots and, hence, cluster variables. By the results of , this bijection also induces a bijection between clusters and ‘tilting objects’ of $`𝒞`$. These ‘coincidences’ lead us to the following
###### Question 1.
Can we realize a cluster algebra of finite type as a ‘Hall algebra’ of the corresponding category $`𝒞`$?
A first result in this direction is the cluster variable formula of . This formula gives an explicit expression for the cluster variable associated with a positive root $`\alpha `$ corresponding to an indecomposable module $`M_\alpha `$: The exponents of the Laurent monomials of the cluster variable $`X_\alpha `$ are provided by the homological form on $`\mathrm{𝗆𝗈𝖽}kQ`$ and the coefficients are Euler characteristics of Grassmannians of submodules of $`M_\alpha `$. In the present article, we use this formula to provide a more complete answer to question 1 and to obtain structural results on the cluster algebra, some of which constitute positive answers to conjectures by Fomin and Zelevinsky. We first describe our structural results:
* Canonical basis. We obtain a $``$-basis of the cluster algebra labelled by the set of so-called exceptional objects of the category $`𝒞`$. The results below point to an analogy of this basis with Lusztig/Kashiwara’s dual canonical bases of quantum groups.
* Positivity conjecture. We prove that the Laurent expansion of a cluster variable has positive coefficients, when the seed is associated to a quiver orientation.
* Good basis property and Toric degenerations. We prove that these bases are compatible with ‘good filtrations’ of the cluster algebra. This provides toric degenerations of the spectrum of cluster algebras of finite type, in the spirit of .
* Denominator conjecture. The formula enables us to prove the following: the denominator of the cluster variable associated to the positive root $`\alpha `$ in its irreducible fraction into polynomials in the $`u_i`$’s is $`_iu_i^{n_i}`$ where $`\alpha =_in_i\alpha _i`$ is the decomposition of $`\alpha `$ in the basis of simple roots.
Now, the main result of this article is the ‘cluster multiplication theorem’, Theorem 2. This result yields a more complete answer to Question 1. It provides a ‘Hall algebra type’ multiplication formula for the cluster algebra. The main part of the paper is devoted to the proof of this formula.
Recall that by a result of , the category $`𝒞`$ is triangulated. The cluster multiplication formula expresses the product of two cluster variables associated with objects $`L`$ and $`N`$ of the cluster category in terms of Euler characteristics of varieties of triangles with end terms $`L`$ and $`N`$. Repeated use of the formula leads to an expression of the structure constants of the cluster algebra in the basis provided by the exceptional objects in terms of Euler characteristics of varieties defined from the triangles of the category $`𝒞`$. Thus, the cluster algebra becomes isomorphic to what we call the ‘exceptional Hall algebra’ of $`𝒞`$.
This theorem can be compared with Peng and Xiao’s theorem , which realizes Kac-Moody Lie algebras as Hall algebras of a triangulated quotient of the derived category of a hereditary category. But a closer look reveals some differences. Indeed, we use the quotient of the set of triangles $`W_{N,M}^Y`$ of the form
$$MYNM[1],$$
by the automorphism group $`\mathrm{𝖠𝗎𝗍}Y`$ of the object $`Y`$, while Peng and Xiao use the quotient $`W_{N,M}^Y/\mathrm{𝖠𝗎𝗍}(M)\times \mathrm{𝖠𝗎𝗍}(N)`$. Hence, our approach is more a ‘dual Hall algebra’ approach as in Green’s quantum group realization . Another difference is that the associativity of the multiplication is not proved a priori but results from the isomorphism with the cluster algebra.
The paper is organized as follows: Generalities and auxiliary results on triangulated categories and, in particular, on the cluster category are given in section 2 and in the appendix, where we prove the constructibility of the sets $`W_{N,M}^Y/\mathrm{𝖠𝗎𝗍}(Y)`$ described above.
Section 3 deals with the cluster multiplication formula. We first reduce the proof to the case where the objects involved are indecomposable. Then the indecomposable case is solved. Here, the homology functor from $`𝒞`$ to the hereditary category of quiver representations plays an essential rôle. It allows us to bypass the ‘triangulated geometry’ of $`𝒞`$, which unfortunately is even out of reach of the methods of , because the graded morphism spaces of the category $`𝒞`$ are not of finite total dimension. The main ingredient of the proof is the Calabi-Yau property of the cluster category, which asserts a bifunctorial duality between $`\mathrm{𝖤𝗑𝗍}^1(M,N)`$ and $`\mathrm{𝖤𝗑𝗍}^1(N,M)`$ for any objects $`M`$ and $`N`$.
In section 4, we use Lusztig’s positivity results for canonical bases to prove the positivity theorem. Then we obtain the denominator theorem.
Section 5 deals with good bases for these cluster algebras. We provide a basis indexed by exceptional objects of the category $`𝒞`$, i.e. objects without self-extensions. The cluster variable formula yields that this basis has a ‘Groebner basis’ behaviour and provides toric degenerations.
The last part is concerned with conjectures for ‘non hereditary’ seeds of a cluster algebra of finite type. We formulate a generalization of the cluster variable formula, and other conjectures which would follow from it, such as results on positivity, simplicial fans … . We close the article with a positivity conjecture for the multiplication rule of the exceptional Hall algebra.
Acknowledgments. The authors thank F. Chapoton and C. Geiss for stimulating discussions. They are grateful to A. Zelevinsky for pointing out an error in an earlier version of this article.
## 2. The cluster category
### 2.1.
Let $`\mathrm{\Delta }`$ be a simply laced Dynkin diagram and $`Q`$ a quiver with underlying graph $`\mathrm{\Delta }`$. We denote the set of vertices of $`Q`$ by $`Q_0`$ and the set of arrows by $`Q_1`$. Let $`k`$ be a field. We denote by $`kQ`$ the path algebra of $`Q`$ and by $`\mathrm{𝗆𝗈𝖽}kQ`$ the category of finitely generated right $`kQ`$-modules. For $`iQ_0`$, we denote by $`P_i`$ the associated indecomposable projective $`kQ`$-module and by $`S_i`$ the associated simple module. The Grothendieck group $`𝖦_\mathrm{𝟢}(\mathrm{𝗆𝗈𝖽}kQ)`$ is free abelian on the classes $`[S_i]`$, $`iQ_0`$, and is thus isomorphic to $`^n`$, where $`n`$ is the number of vertices of $`Q`$. For any object $`M`$ in $`\mathrm{𝗆𝗈𝖽}kQ`$, the dimension vector of $`M`$, denoted by $`\underset{¯}{dim}(M)`$, is the class of $`M`$ in $`𝖦_\mathrm{𝟢}(\mathrm{𝗆𝗈𝖽}kQ)`$.
Recall that the category $`\mathrm{𝗆𝗈𝖽}kQ`$ is hereditary, i.e. we have $`\mathrm{𝖤𝗑𝗍}^2(M,N)=0`$ for any objects $`M`$, $`N`$ in $`\mathrm{𝗆𝗈𝖽}kQ`$. For all $`M`$, $`N`$ in $`\mathrm{𝗆𝗈𝖽}kQ`$, we put
$$[M,N]^0=dim\mathrm{𝖧𝗈𝗆}(M,N),[M,N]^1=dim\mathrm{𝖤𝗑𝗍}^1(M,N),M,N=[M,N]^0[M,N]^1.$$
In the sequel, for any additive category $``$, we denote by $`\mathrm{𝗂𝗇𝖽}()`$ the subcategory of $``$ formed by a system of representatives of the isomorphism classes of indecomposable objects in $``$. We know that there exists a partial ordering $`_Q`$, also denoted by $``$, on $`\mathrm{𝗂𝗇𝖽}(\mathrm{𝗆𝗈𝖽}kQ)`$ such that
$$[M,N]^00MN,M,N\mathrm{𝗂𝗇𝖽}(\mathrm{𝗆𝗈𝖽}kQ).$$
Denote by $`r`$ be the cardinality of $`\mathrm{𝗂𝗇𝖽}(\mathrm{𝗆𝗈𝖽}kQ)`$. We fix a numbering $`Z_k`$, $`1kr`$, of the objects in $`\mathrm{𝗂𝗇𝖽}(\mathrm{𝗆𝗈𝖽}kQ)`$ which is compatible with the ordering.
We recall from that there is a canonical $``$-linear category from which $`\mathrm{𝗆𝗈𝖽}kQ`$ is obtained by base change.
### 2.2.
Denote by $`𝒟=𝒟_Q=𝒟^b(\mathrm{𝗆𝗈𝖽}kQ)`$ the bounded derived category of the category of finitely generated $`kQ`$-modules and by $`S`$ its shift functor $`MM[1]`$. As shown in , the category $`𝒟`$ is a Krull-Schmidt category, and, up to canonical triangle equivalence, it only depends on the underlying graph $`\mathrm{\Delta }`$ of $`Q`$. We identify the category $`\mathrm{𝗆𝗈𝖽}kQ`$ with the full subcategory of $`𝒟`$ formed by the complexes whose homology is concentrated in degree $`0`$. We simply call ‘modules’ the objects in this subcategory. The indecomposable objects of $`𝒟`$ are the $`S^jZ_k`$, $`j`$, $`1kr`$.
We still denote by $`\underset{¯}{dim}(M)𝖦_\mathrm{𝟢}(𝒟)`$ the dimension vector of an object $`M`$ of $`𝒟`$ in the Grothendieck group of $`𝒟`$.
Let $`\tau `$ be the AR-translation of $`𝒟`$. It is the autoequivalence of $`𝒟`$ characterized by the Auslander-Reiten formula:
(2.1)
$$\mathrm{𝖤𝗑𝗍}_𝒟^1(N,M)D\mathrm{𝖧𝗈𝗆}_𝒟(M,\tau N),$$
where $`M`$, $`N`$ are any objects in $`𝒟`$ and where $`D`$ is the functor which takes a vector space to its dual. The AR-translation $`\tau `$ is a triangle equivalence and therefore induces an automorphism of the Grothendieck group of $`𝒟`$. If we identify this group with the root lattice of the corresponding root system, the Auslander-Reiten translation corresponds to the Coxeter transformation, cf. .
We now consider the orbit category $`𝒟/F`$, where $`F`$ is the autoequivalence $`S\tau ^1`$. The objects of the category $`𝒞=𝒞_Q=𝒟/F`$ are the objects of $`𝒟`$ and the morphisms are defined by
$$\mathrm{𝖧𝗈𝗆}_{𝒟/F}(M,N)=\underset{i}{}\mathrm{𝖧𝗈𝗆}_𝒟(F^iM,N).$$
The category $`𝒞_Q`$ was defined in , cf. also for the A<sub>n</sub>-case. It is the so-called cluster category. Like the derived category $`𝒟`$, up to canonical equivalence, the cluster category $`𝒞_Q`$ only depends on $`\mathrm{\Delta }`$ and not on the orientation of the quiver $`Q`$.
### 2.3.
We now review the basic properties of the category $`𝒞`$. The first two points of the following theorem were proved in and and the last two points in .
###### Theorem 1.
* The category $`𝒞`$ is triangulated and
* the natural functor $`\pi `$ : $`𝒟𝒞`$ is a triangle functor.
* The category $`𝒞`$ is a Krull-Schmidt category and
* we have $`\mathrm{𝖤𝗇𝖽}_𝒞(M)=k`$ for any indecomposable object $`M`$ of $`𝒞`$.
The shift functor of the triangulated category $`𝒞`$ will still be denoted by $`S`$. Often, we will omit the functor $`\pi `$ from the notations. With this convention, the objects in
$$\mathrm{𝗂𝗇𝖽}\mathrm{𝗆𝗈𝖽}kQ\{SP_i,\mathrm{\hspace{0.17em}1}in\}$$
form a set of representatives for the indecomposables of $`𝒞`$, as shown in Proposition 1.6 of . By formula 2.1, we have, for all objects $`M`$, $`N`$ of $`𝒟`$, that
$$\mathrm{𝖤𝗑𝗍}_𝒟^1(M,N)D\mathrm{𝖧𝗈𝗆}_𝒟(N,\tau M)D\mathrm{𝖧𝗈𝗆}_𝒟(\tau ^1N,M)D\mathrm{𝖤𝗑𝗍}^1(FN,M).$$
Hence, the category $`𝒞`$ is Calabi-Yau of CY-dimension $`1`$, which means that the functor $`\mathrm{𝖤𝗑𝗍}^1`$ is symmetric in the following sense:
$$\mathrm{𝖤𝗑𝗍}_𝒞^1(M,N)D\mathrm{𝖤𝗑𝗍}_𝒞^1(N,M).$$
In other words, there is an (almost) canonical non degenerate bifunctorial pairing
$$\varphi :\mathrm{𝖤𝗑𝗍}_𝒞^1(M,N)\times \mathrm{𝖤𝗑𝗍}_𝒞^1(N,M)k.$$
### 2.4.
In this section, we study the analogy between the (triangulated) category $`𝒞`$ and the (abelian) category $`\mathrm{𝗆𝗈𝖽}kQ`$. We will see that it can be useful to view $`𝒞`$ as glued together from copies of $`\mathrm{𝗆𝗈𝖽}kQ^{}`$, where $`Q^{}`$ runs through the set of orientations of $`\mathrm{\Delta }`$.
By the previous section, each object $`M`$ of $`𝒞`$ can be uniquely decomposed in the following way:
$$M=M_0SP_M,$$
where $`M_0`$ is the image under $`\pi `$ of a ‘module’ in $`𝒟`$, and where $`SP_M`$ is the image of the shift of a projective module. We will say that an object $`M`$ of $`𝒞`$ is a module if $`M=M_0`$, and that $`M`$ is the shift of a projective module if $`M=SP_M`$.
The module $`M_0`$ can be recovered using the functor
$$H^0=\mathrm{𝖧𝗈𝗆}_𝒞(kQ_{kQ},\mathrm{?}):𝒞\mathrm{𝗆𝗈𝖽}kQ.$$
Indeed, we have
$$H^0(M)=H^0(M_0)H^0(SP_M)=\mathrm{𝖧𝗈𝗆}_{\mathrm{𝗆𝗈𝖽}kQ}(kQ_{kQ},M_0)\mathrm{𝖧𝗈𝗆}_𝒞(_iP_i,SP_M)=M_0,$$
as the last factor is zero. The functor $`H^0`$ is a homological functor, i.e. it maps triangles in $`𝒞`$ to long exact sequences of $`kQ`$-modules.
We will deduce the following proposition from Proposition 1.7 of .
###### Proposition 1.
Let $`M`$, $`N`$ be indecomposable $`kQ`$-modules. Then
* $`\mathrm{𝖤𝗑𝗍}_𝒞^1(M,N)=\mathrm{𝖤𝗑𝗍}_{kQ}^1(M,N)\mathrm{𝖤𝗑𝗍}_{kQ}^1(N,M)`$ and at least one of the two direct factors vanishes.
* any short exact sequence of $`kQ`$-modules provides a (unique) triangle in $`𝒞`$,
* if $`MN`$ and if there exists a triangle in $`𝒞`$, then $`Y`$ is also a module and there exists a short exact sequence of $`kQ`$-modules . Moreover, if this sequence is non split, the modules $`M`$ and $`N`$ are non isomorphic and are not isomorphic to indecomposable factors of $`Y`$.
###### Proof.
Point (i) is proved in proposition 1.7 of . For points (ii) and (iii), we may assume that $`\mathrm{𝖤𝗑𝗍}_{kQ}^1(N,M)`$ does not vanish. This implies that there is a non zero morphism from $`M`$ to $`\tau N`$ and thus we have $`MN`$. If $`\mathrm{𝖤𝗑𝗍}_{kQ}^1(N,M)`$ was also non zero, we would also have $`MN`$, hence $`M=N`$ and $`\mathrm{𝖤𝗑𝗍}_{kQ}(M,N)=0`$, a contradiction. Now it follows from point (i) that the canonical map
$$\mathrm{𝖤𝗑𝗍}_{kQ}^1(M,N)\mathrm{𝖤𝗑𝗍}_𝒞^1(M,N)$$
is bijective. Points (ii) and the first assertion of (iii) therefore follow from the bijection between elements of $`\mathrm{𝖤𝗑𝗍}^1`$ and classes of short exact sequences in $`\mathrm{𝗆𝗈𝖽}kQ`$ respectively triangles in $`𝒞`$. For the last assertion of (iii), we first note that $`M`$ is not isomorphic to $`N`$ because no indecomposable module has selfextensions. Now since the sequence is non split, the map $`YN`$ factors through the middle term of the AR-sequence ending in $`N`$. Therefore each indecomposable factor of $`Y`$ strictly precedes $`N`$. Dually, $`M`$ must precede each indecomposable factor of $`Y`$. ∎
The following lemma is useful to understand extensions between indecomposable objects of $`𝒞`$ when the situation does not fit into the framework of the proposition above.
###### Lemma 1.
Let $`MYNSM`$ be a non split triangle, where $`M`$, $`N`$ are indecomposable objects of $`𝒞`$. Suppose that there exists no orientation $`Q^{}`$ of $`\mathrm{\Delta }`$ such that $`M`$, $`N`$ are simultaneously $`kQ^{}`$-modules with $`M_Q^{}N`$ via the embedding $`\mathrm{𝗆𝗈𝖽}kQ^{}𝒟𝒞`$. Then, $`N=SM`$ and $`Y=0`$.
###### Proof.
Using the AR-translation, it is always possible to choose an embedding of $`\mathrm{𝗆𝗈𝖽}kQ`$ in $`𝒟`$ such that $`M`$ is a projective module. By changing orientations, we can suppose that $`M`$ is simple projective, say $`P_i`$. By the hypothesis of the lemma, $`N`$ is not a module, and, as $`N`$ is indecomposable, $`N`$ is the shift of a projective $`SP_j`$. As the triangle above is non split, the morphism $`\epsilon `$ : $`NSM`$ is non zero. So there exists a non zero morphism from $`P_j`$ to $`P_i`$. The assumption on $`P_i`$ implies $`j=i`$. Hence, $`N=SM`$ and the morphism $`\epsilon `$ is an isomorphism. This forces $`Y`$ to be zero. ∎
### 2.5.
In this section, we give properties of group actions on triangles in $`𝒞`$. Actually, most of them are general facts valid in Krull-Schmidt triangulated categories.
Let $`Y`$ be an object of $`𝒞`$ and $`Y=_jY_j`$ be a decomposition into isotypical components. Suppose that, for each $`j`$, the object $`Y_j`$ is the sum of $`n_j`$ copies of an indecomposable. By Theorem 1(iv), the endomorphism algebra of each $`Y_j`$ is isomorphic to a matrix algebra over $`k`$. Therefore, the radical $`R`$ of $`\mathrm{𝖤𝗇𝖽}_𝒞(Y)`$ is formed by the endomorphisms $`f`$ all of whose components $`f_{jj}:Y_jY_j`$ vanish. We obtain the decomposition
$$\mathrm{𝖤𝗇𝖽}_𝒞(Y)=\underset{j}{}M_{n_j}(k)R.$$
Therefore, the group $`\mathrm{𝖠𝗎𝗍}(Y)`$ of invertible elements of $`\mathrm{𝖤𝗇𝖽}_𝒞(Y)`$ is isomorphic to $`L(Y)U`$, where $`L(Y)=_j\mathrm{𝖦𝖫}_{n_j}(k)`$ and where $`U`$ is the unipotent group $`1+R`$. We denote by $`W_{N,M}^Y`$ the set of triples $`(i,p,\eta )`$ of morphisms such that is a triangle. The group $`\mathrm{𝖠𝗎𝗍}(Y)`$ acts on $`W_{N,M}^Y`$ by $`g.(i,p,\eta )=(gi,pg^1,\eta )`$. There also exists an action of $`\mathrm{𝖠𝗎𝗍}(M)\times \mathrm{𝖠𝗎𝗍}(N)`$ on $`W_{N,M}^Y`$ given by $`(g,h).(i,p,\eta )=(ig^1,hp,Sg\eta h^1)`$. In particular, we can define an action of the group $`k^{}`$ on $`W_{N,M}^Y`$ given by $`\lambda .(i,p,\eta )=(\lambda ^1i,p,\lambda \eta )`$.
###### Lemma 2.
For any objects $`M`$, $`Y`$, $`N`$ of $`𝒞`$, the group $`\mathrm{𝖠𝗎𝗍}(Y)`$ acts on $`W_{N,M}^Y`$ with unipotent stabilizers.
###### Proof.
Fix a triple $`(i,p,\eta )`$ in $`W_{N,M}^Y`$ and let $`g`$ in the stabilizer of $`(i,p,\eta )`$ for the action of $`\mathrm{𝖠𝗎𝗍}(Y)`$. We can draw the following commuting diagram
As the identity $`I`$ lies in the stabilizer, we get the solid part of the commutative diagram
The long exact sequence obtained by applying $`\mathrm{𝖧𝗈𝗆}_𝒞(Y,\mathrm{?})`$ to the bottom triangle shows that there is a morphism $`f`$ with $`if=Ig`$. Similarly, there is morphism $`h`$ with $`hp=Ig`$. Therefore, we have $`(Ig)^2=(hp)(if)=0`$. This proves our assertion. ∎
Consider the third projection $`p_3`$ : $`W_{N,M}^Y\mathrm{𝖤𝗑𝗍}^1(N,M)`$. Define $`\mathrm{𝖤𝗑𝗍}^1(N,M)_Y`$ to be the image of $`W_{N,M}^Y`$ under $`p_3`$. For any $`\eta `$ in $`\mathrm{𝖤𝗑𝗍}^1(M,N)_Y`$, the group $`\mathrm{𝖠𝗎𝗍}(Y)`$ clearly acts on $`p_3^1(\eta )`$.
###### Lemma 3.
For any $`\eta `$ in $`\mathrm{𝖤𝗑𝗍}^1(N,M)_Y`$, the action of $`\mathrm{𝖠𝗎𝗍}(Y)`$ on $`p_3^1(\eta )`$ is transitive and has unipotent stabilizers.
###### Proof.
Let $`(i,p,\eta )`$ and $`(i^{},p^{},\eta )`$ be in $`p_3^1(\eta )`$. Then, by \[32, TR3\], there exists a morphism $`f`$ which makes the following diagram commutative.
By the 5-Lemma, $`f`$ is invertible. The last assertion is Lemma 2. ∎
### 2.6.
Let $`K_0^{split}(𝒞)`$ be the Grothendieck group of the underlying additive category of $`𝒞`$, i.e. the free abelian group generated by the isomorphism classes of indecomposable objects of $`𝒞`$. For each object $`Y`$ of $`𝒞`$, we still denote by $`Y`$ the corresponding element of $`K_0^{split}(𝒞)`$.
For a variety $`X`$, we define $`\chi _c(X)`$ to be the Euler-Poincaré characteristic of the étale cohomology with proper support of $`X`$, i.e. we have
$$\chi _c(X)=\underset{i=0}{\overset{\mathrm{}}{}}(1)^idimH_c^i(X,\overline{}_l).$$
Let $`M`$, $`N`$, $`Y`$ be objects of $`𝒞`$. By section 7, the subset $`\mathrm{𝖤𝗑𝗍}^1(M,N)_Y)`$ of the vector space $`\mathrm{𝖤𝗑𝗍}^1(M,N)`$ is constructible. Clearly it is conic. Thus its projectivization $`\mathrm{𝖤𝗑𝗍}^1(M,N)_Y)`$ is a variety and has a well-defined Euler characteristic $`\chi _c(\mathrm{𝖤𝗑𝗍}^1(M,N)_Y)`$.
For any pair $`(Z_j,Z_i)`$ of indecomposable object of $`𝒞`$, we call elementary vector associated to this pair any element $`Z_i+Z_jY_{ij}`$ in $`K_0^{split}(𝒞)`$ such that $`Y_{ij}`$ is the middle term of a non split triangle in $`𝒞`$. In this case, the number $`c_{ij}=\chi _c(\mathrm{𝖤𝗑𝗍}^1((Z_j,Z_i)_{Y_{ij}})`$ is called the multiplicity number of the elementary vector. The following proposition is an analogue of Theorem 3.3 in .
###### Proposition 2.
Let $`M`$, $`N`$, $`Y`$ be any objects of $`𝒞`$ such that $`YMN`$.
* If $`M+NY`$ is an elementary vector associated to the pair $`(Z_j,Z_i)`$, then we have
$$\chi _c(\mathrm{𝖤𝗑𝗍}^1(N,M)_Y)=z_iz_jc_{ij},$$
where $`z_j`$, resp. $`z_i`$, is the multiplicity of the indecomposable component $`Z_j`$, resp. $`Z_i`$, in $`N`$, resp. $`M`$, and where $`c_{ij}`$ is multiplicity number associated to the elementary vector.
* If $`M+NY`$ is not an elementary vector, then
$$\chi _c(\mathrm{𝖤𝗑𝗍}^1(N,M)_Y)=0.$$
###### Proof.
Let $`A`$, $`B`$ be two objects of $`𝒞`$. Since $`𝒞`$ is triangulated, for each morphism $`i`$ from $`A`$ to $`B`$, there is a triangle and the object $`C`$ is unique up to (non unique) isomorphism. Hence, we have a partition
$$\mathrm{𝖧𝗈𝗆}_𝒞(A,B)=\underset{C}{}\mathrm{𝖧𝗈𝗆}_𝒞(A,B)_C,$$
where $`C`$ runs through the isomorphism classes of $`𝒞`$ and
$$\mathrm{𝖧𝗈𝗆}_𝒞(A,B)_C=\{i\mathrm{𝖧𝗈𝗆}_𝒞(A,B)|\text{There is a triangle }\text{}\}.$$
Now, by \[32, TR2\] and Lemma 1.3 of , we have $`\mathrm{𝖤𝗑𝗍}^1(N,M)_Y=\mathrm{𝖧𝗈𝗆}(N,SM)_{SY}`$. This implies that
(2.2)
$$\mathrm{𝖤𝗑𝗍}^1(A,B)=\underset{C}{}\mathrm{𝖤𝗑𝗍}^1(A,B)_C.$$
Now we prove (i) and (ii). Denote by $`M=_im_iZ_i`$ and $`N=_jn_jZ_j`$ the decompositions of $`M`$ and $`N`$ into indecomposable objects. Consider the action of $`G:=\mathrm{𝖠𝗎𝗍}(M)\times \mathrm{𝖠𝗎𝗍}(N)`$ on $`\mathrm{𝖤𝗑𝗍}^1(N,M)`$. The group $`G`$ contains as a subgroup
$$H=\underset{i}{}\mathrm{𝖠𝗎𝗍}(m_iZ_i)\times \underset{j}{}\mathrm{𝖠𝗎𝗍}(n_jZ_j)=\underset{i}{}\mathrm{𝖦𝖫}_{m_i}\times \underset{j}{}\mathrm{𝖦𝖫}_{n_j}.$$
Under the isomorphism
$$\mathrm{𝖤𝗑𝗍}^1(N,M)=_{(i,j)}\mathrm{𝖤𝗑𝗍}^1(Z_j,Z_i)M_{n_j\times m_j},$$
the action of $`H`$ on $`\mathrm{𝖤𝗑𝗍}^1(N,M)`$ corresponds to the product of the canonical actions of the $`\mathrm{𝖦𝖫}_{m_i}\times \mathrm{𝖦𝖫}_{n_j}`$ on matrices. Let $`T`$ be the torus of $`G`$ corresponding to the diagonal matrices. We have $`THG`$ and $`T`$ acts on $`\mathrm{𝖤𝗑𝗍}^1(N,M)`$ with set of invariants
(2.3)
$$\mathrm{𝖤𝗑𝗍}^1(N,M)^T=\underset{(i,j)}{}\underset{r,s}{}\mathrm{𝖤𝗑𝗍}^1(Z_j,Z_i)kE_{rs},$$
where, for fixed $`(i,j)`$, the $`E_{rs}`$, $`1rn_j`$, $`1sm_i`$, are the elementary matrices in $`M_{n_j\times m_i}`$.
Set $`𝒳:=\mathrm{𝖤𝗑𝗍}^1(N,M)_Y`$. As $`G`$ acts on $`𝒳`$, we have an action of the torus $`T`$ and
$$𝒳^T=\mathrm{𝖤𝗑𝗍}^1(N,M)^T𝒳.$$
It follows from Białynicki-Birula’s results that we have $`\chi _c(𝒳)=\chi _c(𝒳^T)`$. We will compute $`\chi _c(𝒳^T)`$. If $`𝒳^T`$ is empty, we have $`\chi _c(𝒳)=\chi _c(𝒳^T)=0`$. Let us suppose from now on that $`𝒳^T`$ is non empty. Let the element $`[\eta ]E_{rs}`$ be in $`𝒳^T`$ with $`[\eta ]\mathrm{𝖤𝗑𝗍}^1(Z_j,Z_i)`$. As $`G`$, hence $`H`$, acts on $`𝒳`$, the element $`[\eta ]E_{r^{}s^{}}`$ is also in $`𝒳^T`$ for any $`r^{}`$, $`s^{}`$, $`1r^{}n_j`$, $`1s^{}m_i`$. Now, by 2.2, there exists a unique object $`Y_{ij}`$ such that $`[\eta ]`$ belongs to $`\mathrm{𝖤𝗑𝗍}^1(Z_j,Z_i)_{Y_{ij}}`$. And so, in $`K_0^{split}(𝒞)`$, we have
(2.4)
$$Y=M+NZ_iZ_j+Y_{ij}.$$
This implies that $`M+NY`$ is an elementary vector. Moreover, for any element $`[\eta ^{}]`$ of $`\mathrm{𝖤𝗑𝗍}^1(Z_j,Z_i)_{Y_{ij}}`$, the element $`[\eta ^{}]E_{kl}`$ belongs to $`\mathrm{𝖤𝗑𝗍}^1(N,M)_Y^{}`$, where
$$Y^{}=M+NZ_iZ_j+Y_{ij}=Y.$$
Thus, $`Y^{}`$ is isomorphic to $`Y`$. So we have proved the inclusion
$$\underset{k,l}{}\mathrm{𝖤𝗑𝗍}^1(Z_j,Z_i)_{Y_{ij}}E_{kl}𝒳^T.$$
The proposition will follow if we prove the reverse inclusion. Suppose that $`[\eta ^{}]`$ belongs to $`\mathrm{𝖤𝗑𝗍}^1(Z_j,Z_i)_{Y_{ij}^{}}`$ with $`[\eta ^{}]E_{kl}`$ belonging to $`𝒳^T`$. Then, we obtain $`Y=M+NZ_iZ_j+Y_{ij}^{}`$, which implies $`Y_{ij}^{}=Y_{ij}`$. Now, in order to finish the proof of the reverse inclusion, it is sufficient to prove that if we have two triangles
$$\text{},\text{},$$
with $`\eta \mathrm{𝖤𝗑𝗍}^1(Z_j,Z_i)`$ and $`\eta ^{}\mathrm{𝖤𝗑𝗍}^1(Z_j^{},Z_i^{})`$, then $`(i,j)`$ equals $`(i^{},j^{})`$. Let us prove this fact. The first triangle implies that $`M+NY=Z_j+Z_iY_{ij}`$ for some object $`Y_{ij}`$. The second one implies that $`M+NY=Z_j^{}+Z_i^{}Y_{i^{}j^{}}`$ for some object $`Y_{i^{}j^{}}`$. Comparing both equalities in $`K_0^{split}(𝒞)`$ we get $`Z_j+Z_iY_{ij}=Z_j^{}+Z_i^{}Y_{i^{}j^{}}`$. It follows from Lemma 1 and part (iii) of Proposition 1 that the modules $`Z_j`$ and $`Z_i`$ are non isomorphic and are not isomorphic to indecomposable factors of $`Y_{ij}`$, and similarly for $`Z_i^{}`$, $`Z_j^{}`$ and $`Y_{i^{}j^{}}`$. Therefore, the signs in the equality imply that $`(i,j)=(i^{},j^{})`$ or $`(i,j)=(j^{},i^{})`$. In the second case, we have $`Y_{ij}=Y_{i^{}j^{}}`$, but it follows from Lemma 1 and part (iii) of Proposition 1, that we cannot simultaneously have two triangles
$$\text{},\text{ and }\text{}.$$
Hence, we must have $`(i,j)=(i^{},j^{})`$ and we are done. ∎
## 3. The cluster multiplication theorem
### 3.1.
We now present the main theorem. The subsections 3.2, 3.3, 3.4 are devoted to the proof of the theorem.
For any $`kQ`$-module $`M`$, let $`\mathrm{𝖦𝗋}(M)`$ be the Grassmannian of $`kQ`$-submodules of $`M`$, and let $`\mathrm{𝖦𝗋}_e(M)`$ be the $`e`$-Grassmannian of $`M`$, i.e. the variety of submodules of $`M`$ with dimension vector $`e`$.
Following we define
$$X_\mathrm{?}:\mathrm{𝗈𝖻𝗃}(𝒞_Q)(x_1,\mathrm{},x_n),MX_M$$
to be the unique map with the following properties:
* $`X_M`$ only depends on the isomorphism class of $`M`$,
* we have
$$X_{MN}=X_MX_N$$
for all $`M,N`$ of $`𝒞_Q`$,
* if $`M`$ is isomorphic to $`SP_i`$ for the $`ith`$ indecomposable projective $`P_i`$, we have
$$X_M=x_i,$$
* if $`M`$ is the image in $`𝒞_Q`$ of an indecomposable $`kQ`$-module, we have
(3.1)
$$X_M=\underset{e}{}\chi _c(\mathrm{𝖦𝗋}_e(M))x^{\tau (e)\underset{¯}{dim}M+e},$$
where $`\tau `$ is the Auslander-Reiten translation on the Grothendieck group of $`𝒟_Q`$ and, for $`v^n`$, we put
$$x^v=\underset{i=1}{\overset{n}{}}x_i^{\underset{¯}{dim}S_i,v}.$$
From now on, we write $`\mathrm{𝖤𝗑𝗍}^1(N,M)`$ for $`\mathrm{𝖤𝗑𝗍}_𝒞^1(N,M)`$ for any objects $`N`$, $`M`$ of $`𝒞`$ .
###### Theorem 2.
For any objects $`M`$, $`N`$ of $`𝒞`$, we have
* If $`\mathrm{𝖤𝗑𝗍}^1(N,M)=0`$, then $`X_NX_M=X_{NM}`$,
* If $`\mathrm{𝖤𝗑𝗍}^1(N,M)0`$, then
$$\chi _c(\mathrm{𝖤𝗑𝗍}^1(N,M))X_NX_M=\underset{Y}{}(\chi _c(\mathrm{𝖤𝗑𝗍}^1(N,M)_Y)+\chi _c(\mathrm{𝖤𝗑𝗍}^1(M,N)_Y))X_Y,$$
where $`Y`$ runs through the isoclasses of $`𝒞`$.
Note that the first part is true by the definition of the map $`MX_M`$.
### 3.2.
We prove here that Theorem 2 is true if it is true for indecomposable $`kQ`$-modules $`M`$, $`N`$, and for any orientation $`Q`$ of $`\mathrm{\Delta }`$. So we suppose the following: For all indecomposable $`kQ`$-modules $`Z_i`$, $`Z_j`$ with $`\mathrm{𝖤𝗑𝗍}^1(Z_j,Z_i)0`$, we have
$`()`$
$$\chi _c(\mathrm{𝖤𝗑𝗍}^1(Z_j,Z_i))X_{Z_j}X_{Z_i}=\underset{Y_{ij}}{}(\chi _c(\mathrm{𝖤𝗑𝗍}^1(Z_j,Z_i)_{Y_{ij}})+\chi _c(\mathrm{𝖤𝗑𝗍}^1(Z_i,Z_j)_{Y_{ij}}))X_{Y_{ij}}.$$
By Proposition 2.6 of , the formula above is also true for all indecomposable objects $`Z_i`$, $`Z_j`$ such that $`Z_i=\tau Z_j=SZ_j`$. Hence, by Lemma 1, our hypothesis implies that the theorem is true for all indecomposable objects $`M`$ and $`N`$ of $`𝒞_Q`$.
Suppose now that $`M`$ and $`N`$ are arbitrary objects of $`𝒞`$, with decompositions into indecomposables $`M=m_iZ_i`$, $`N=n_jZ_j`$. Then we have
$$\chi _c((\mathrm{𝖤𝗑𝗍}^1(N,M)))=m_in_j\chi _c(\mathrm{𝖤𝗑𝗍}^1(Z_j,Z_i)).$$
Moreover, if an object $`Y`$ is such that $`\chi _c(\mathrm{𝖤𝗑𝗍}^1(N,M)_Y)`$ is non zero, then, by Proposition 2, there exist unique objects $`Z_i`$, $`Z_j`$, $`Y_{ij}`$ such that
$$\chi _c(\mathrm{𝖤𝗑𝗍}^1(N,M)_Y)=m_in_j\chi _c(\mathrm{𝖤𝗑𝗍}^1(Z_j,Z_i)_{Y_{ij}})$$
and $`Y_{ij}(n_j1)Z_j(m_i1)M_i=Y`$. Hence, by multiplying all equalities (\*) by $`m_in_jX_{(n_j1)Z_j(m_i1)M_i}`$ and adding all these equalities, we obtain the formula of Theorem 2.
### 3.3.
From now on, in the rest of the proof of theorem 2, we can suppose that $`M`$ and $`N`$ are indecomposable $`kQ`$-modules.
Suppose $`X_{𝔽_q}`$ is a family of sets indexed by all finite fields $`𝔽_q`$ such that the cardinality $`X_{𝔽_q}`$ is a polynomial $`P_X`$ in $`q`$ with integer coefficients. In particular, we will consider the case where $`X_{}`$ is a variety defined over $``$ and, for any field $`k`$, $`X_k`$ is the corresponding variety defined by base change. In this case, we will set $`\chi _1(X):=P_X(1)`$. Then it is a consequence of the Grothendieck trace formula, see , \[8, Lemma 3.5\], that $`\chi _1(X)`$ is exactly the Euler characteristic $`\chi _c(X_{})`$ of $`X_{}`$. We apply this to prove the following Lemma.
###### Lemma 4.
For any indecomposable modules $`M`$, $`N`$, any module $`Y`$, and any dimension vector $`e`$, we have $`\chi _c(\mathrm{𝖤𝗑𝗍}^1(M,N)_Y)=\chi _1(\mathrm{𝖤𝗑𝗍}^1(M,N)_Y)`$ and $`\chi _c(\mathrm{𝖦𝗋}_e(M))=\chi _1(\mathrm{𝖦𝗋}_e(M))`$.
###### Proof.
By part (i) of Proposition 1, we have
$$\mathrm{𝖤𝗑𝗍}^1(M,N)_Y=\mathrm{𝖤𝗑𝗍}_{kQ}^1(M,N)_Y\text{ or }\mathrm{𝖤𝗑𝗍}^1(M,N)_Y=\mathrm{𝖤𝗑𝗍}_{kQ}^1(N,M)_Y.$$
Therefore, we know from that this variety is obtained by base change from a variety defined over $``$ and that the cardinality of its set of $`𝔽_q`$-points is polynomial in $`q`$. The corresponding facts for $`\mathrm{𝖦𝗋}_e(M)`$ were shown in . ∎
This lemma allows us to work in the remainder of the section in the case where the base field $`k`$ is the finite field $`𝔽_q`$ with $`q`$ elements.
Let $`N,M`$ be indecomposable $`kQ`$-modules such that
$$0\mathrm{𝖤𝗑𝗍}_{kQ}^1(N,M).$$
This implies that $`\mathrm{𝖤𝗑𝗍}_{kQ}^1(M,N)`$ vanishes and hence that
$$\mathrm{𝖤𝗑𝗍}_{𝒞_Q}^1(N,M)=\mathrm{𝖤𝗑𝗍}_{kQ}^1(N,M)D\mathrm{𝖤𝗑𝗍}_{kQ}^1(M,N)=\mathrm{𝖤𝗑𝗍}_{kQ}^1(N,M).$$
We can suppose that we are in this case.
For a module $`Y`$, the group $`\mathrm{𝖠𝗎𝗍}(Y)`$ acts naturally on $`\mathrm{𝖦𝗋}(Y)`$ and this action stabilizes the subvarieties $`\mathrm{𝖦𝗋}(Y)`$. If $`Y`$ is any object, we define the action of $`L(Y):=\mathrm{𝖠𝗎𝗍}(H^0(Y))\times \mathrm{𝖠𝗎𝗍}(SP_Y)`$ on $`\mathrm{𝖦𝗋}(H^0(Y))`$ by letting the second factor act trivially.
If a group $`G`$ acts on the sets $`X`$ and $`Y`$, then $`G`$ acts on $`X\times Y`$ by $`g.(x,y)=(g.x,g.y)`$. We define as usual the set of orbits $`X\times _GY`$ for this action. By section 2.5, we obtain:
###### Lemma 5.
Let $`M`$, $`N`$, $`Y`$ be objects of $`𝒞`$. Suppose that $`\mathrm{𝖤𝗑𝗍}^1(N,M)_Y0`$. Then we have
$$\chi _1(W_{N,M}^Y\times _{(L(Y)\times k^{})}\mathrm{𝖦𝗋}(H^0(Y))=\chi _1(\mathrm{𝖤𝗑𝗍}^1(N,M)_Y\times \mathrm{𝖦𝗋}(H^0(Y)).$$
If $`Y`$ is a module, then we have
$$\chi _1(W_{N,M}^Y\times _{(\mathrm{𝖠𝗎𝗍}(Y)\times k^{})}\mathrm{𝖦𝗋}(Y))=\chi _1(\mathrm{𝖤𝗑𝗍}^1(N,M)_Y\times \mathrm{𝖦𝗋}(Y)).$$
Note that we have
$$\chi _c(\mathrm{𝖤𝗑𝗍}^1(N,M))=dim\mathrm{𝖤𝗑𝗍}^1(N,M).$$
so that we could ‘simplify’ the left hand side of the main formula in Theorem 2. Now, with the help of Lemma 5, we see that the specialization of the claimed formula at $`x_i=1`$, $`1in`$, is just the equality of the Euler characteristics $`\chi _1`$ of the following varieties
$$L=\mathrm{𝖤𝗑𝗍}^1(N,M)\times \mathrm{𝖦𝗋}(N)\times \mathrm{𝖦𝗋}(M)$$
and
$$R=\left(\underset{Y}{}W_{N,M}^Y\times _{\mathrm{𝖠𝗎𝗍}(Y)\times k^{}}\mathrm{𝖦𝗋}(Y)\right)\left(\underset{Y}{}W_{M,N}^Y\times _{L(Y)\times k^{}}\mathrm{𝖦𝗋}(H^0(Y))\right).$$
More precisely, the left hand side of the equality claimed in the theorem is a sum of terms
$$\chi _c((\mathrm{𝖤𝗑𝗍}^1(N,M)\times \mathrm{𝖦𝗋}_e(N))\times \mathrm{𝖦𝗋}_f(M))x^{\tau e\underset{¯}{dim}N+e+\tau f\underset{¯}{dim}M+f}.$$
Now the variety
$$L(e,f)=(\mathrm{𝖤𝗑𝗍}^1(N,M)\times \mathrm{𝖦𝗋}_e(N))\times \mathrm{𝖦𝗋}_f(M)$$
is the union of the subvariety $`L_1(e,f)`$ consisting of all those triples $`([\epsilon ],N^{},M^{})`$ such that there is a diagram of $`kQ`$-modules (since $`\mathrm{𝖤𝗑𝗍}_{kQ}^1(N,M)=\mathrm{𝖤𝗑𝗍}_{𝒞_Q}(N,M)`$)
and its complement $`L_2(e,f)`$. By Proposition 3 below, we see that the cardinality of $`L_2(e,f)`$ and hence of $`L_1(e,f)`$ over $`𝔽_q`$ is polynomial in $`q`$, so the numbers $`\chi _1(L_i(e,f))`$, $`i=1,2`$ make sense.
Now, the term
$$\chi _c(\mathrm{𝖤𝗑𝗍}^1(N,M)\times \mathrm{𝖦𝗋}_e(N))\times \mathrm{𝖦𝗋}_f(M))x^{\tau e\underset{¯}{dim}N+e+\tau f\underset{¯}{dim}M+f}$$
is the sum of
$$\chi _1(L_1(e,f))x^{\tau e\underset{¯}{dim}N+e+\tau f\underset{¯}{dim}M+f}$$
and
$$\chi _1(L_2(e,f))x^{\tau e\underset{¯}{dim}N+e+\tau f\underset{¯}{dim}M+f}.$$
Now we examine the right hand side of the equality of the theorem: First note that since we have $`\mathrm{𝖤𝗑𝗍}_{kQ}^1(N,M)=\mathrm{𝖤𝗑𝗍}_{𝒞_Q}^1(N,M)`$, the set $`\mathrm{𝖤𝗑𝗍}^1(N,M)_Y`$ is empty if $`Y`$ does not occur as the middle term of an extension
$$0MYN0$$
in the category of modules. Therefore, we have
$$\chi _c(\mathrm{𝖤𝗑𝗍}^1(N,M)_Y)X_Y=\underset{g}{}\chi _c(\mathrm{𝖤𝗑𝗍}^1(N,M)_Y\times \mathrm{𝖦𝗋}_g(Y))x^{\tau g\underset{¯}{dim}Y+g}.$$
For non empty $`\mathrm{𝖤𝗑𝗍}^1(N,M)_Y`$, by Proposition 1, one defines a map
$$W_{N,M}^Y\times \mathrm{𝖦𝗋}_g(Y)\underset{e+f=g}{}L_1(e,f),(i,p,\epsilon ,Y^{})([\epsilon ],Y^{}i(M),p(Y^{}))$$
This map descends to a map $`W_{N,M}^Y\times _{\mathrm{𝖠𝗎𝗍}(Y)\times k^{}}\mathrm{𝖦𝗋}_g(Y)_{e+f=g}L_1(e,f)`$ which is surjective with fibers which are affine spaces. One obtains the equality of the Euler characteristics and therefore
$$\chi _c(\mathrm{𝖤𝗑𝗍}^1(N,M)_Y)X_Y=\underset{e,f}{}\chi _1(L_1(e,f))x^{\tau e\underset{¯}{dim}N+e+\tau f\underset{¯}{dim}M+f}.$$
It remains to be proved that
$$\underset{e,f}{}\chi _1(L_2(e,f))x^{\tau e\underset{¯}{dim}N+e+\tau f\underset{¯}{dim}M+f}=\underset{Y}{}\chi _c(\mathrm{𝖤𝗑𝗍}^1(M,N)_Y)X_Y.$$
For this, we need a characterization of the points in $`L_2(e,f)`$. Let $`([\epsilon ],N^{},M^{})`$ be a point of
$$L(e,f)=\mathrm{𝖤𝗑𝗍}^1(N,M)\times \mathrm{𝖦𝗋}_e(N)\times \mathrm{𝖦𝗋}_f(M).$$
Let
$$\varphi :\mathrm{𝖤𝗑𝗍}_{𝒞_Q}^1(M,N)\times \mathrm{𝖤𝗑𝗍}_{𝒞_Q}^1(N,M)k$$
be the (almost) canonical duality pairing. Let
$$\beta :\mathrm{𝖤𝗑𝗍}^1(M,N^{})\mathrm{𝖤𝗑𝗍}^1(M,N)\mathrm{𝖤𝗑𝗍}^1(M^{},N^{})$$
be the map whose components are induced by the inclusions $`M^{}M`$ and $`N^{}N`$.
###### Proposition 3.
The following are equivalent
* $`([\epsilon ],N^{},M^{})`$ belongs to $`L_2(e,f)`$.
* $`\epsilon `$ is not orthogonal to $`\mathrm{𝖤𝗑𝗍}^1(M,N)\mathrm{𝖨𝗆}\beta `$.
* There is an $`\eta \mathrm{𝖤𝗑𝗍}^1(M,N)`$ such that $`\varphi (\eta \epsilon )0`$ and such that, if
is a triangle of $`𝒞_Q`$, then there is a diagram of $`kQ`$-modules
where $`Y^{}`$ is a submodule of $`H^0(Y)`$, $`N^{}`$ is the preimage of $`Y^{}`$ and $`M^{}`$ the image of $`Y^{}`$.
The proposition will be proved in section 3.4. We continue the proof of the theorem. Recall the equality we have to prove:
$$\underset{e,f}{}\chi _1(L_2(e,f))x^{\tau e\underset{¯}{dim}N+e+\tau f\underset{¯}{dim}M+f}=\underset{Y}{}\chi _c(\mathrm{𝖤𝗑𝗍}^1(M,N)_Y)X_Y.$$
Each object $`Y`$ of $`𝒞_Q`$ is isomorphic to the sum of $`H^0(Y)`$ with a module $`SP_Y`$ for some projective $`P_Y`$. With this notation, the right hand side equals
$$\underset{g,Y}{}\chi _c(\mathrm{𝖤𝗑𝗍}^1(M,N)_Y\times \mathrm{𝖦𝗋}_g(H^0(Y)))x^{\tau g\underset{¯}{dim}H^0(Y)+g}X_{SP_Y}.$$
To prove the equality, we will define a correspondence $`C`$ between
$$L_2=\underset{e,f}{}L_2(e,f)\mathrm{𝖤𝗑𝗍}^1(N,M)\times \mathrm{𝖦𝗋}(N)\times \mathrm{𝖦𝗋}(M)$$
and
$$R_2=\underset{Y,g}{}R_2(Y,g),\text{ where }R_2(Y,g)=W_{M,N}^Y\times _{L(Y)\times k^{}}\mathrm{𝖦𝗋}_g(H^0(Y)).$$
Namely, the correspondence $`CL_2\times R_2`$ is formed by all pairs consisting of a point $`([\epsilon ],M^{},N^{})`$ of $`L_2`$ and a point $`(i,p,\eta ,Y^{})`$ of $`R_2`$ such that $`\varphi (\eta \epsilon )0`$, $`N^{}=H^0(i)^1(Y^{})`$, $`M^{}=H^0(p)(Y^{})`$. Note that in this situation, we have a diagram
We say that a variety $`X`$ is an extension of affine spaces if there is a vector space $`V`$ and a surjective morphism $`XV`$ whose fibers are affine spaces of constant dimension.
###### Proposition 4.
The projection $`p_1:CL_2`$ is surjective and its fibers are extensions of affine spaces.
The projection $`p_2:CR_2`$ is surjective and its fibers are affine spaces of constant dimension.
If the pair formed by $`([\epsilon ],M^{},N^{})`$ and $`(i,p,\eta ,Y^{})`$ belongs to $`C`$, then
$$x^{\tau e\underset{¯}{dim}N+e+\tau f\underset{¯}{dim}M+f}=x^{\tau g\underset{¯}{dim}H^0(Y)+g}X_{SP_Y},$$
where $`Y=H^0(Y)SP_Y`$, $`P_Y`$ is projective, $`e=\underset{¯}{dim}M^{}`$, $`f=\underset{¯}{dim}N^{}`$, $`g=\underset{¯}{dim}Y^{}`$.
This proposition, which will be proved in section 3.5, allows us to conclude: Indeed, the variety $`C`$ is the disjoint union of the
$$C_{e,f,Y,g}=p_1^1(L_2(e,f))p_2^1(R_2(Y,g))$$
by parts a) and b) of the proposition, for each non empty $`C_{e,f,Y,g}`$, we have equality of cardinalities on $`𝔽_q`$
$$\mathrm{\#}(L_2(e,f))=\underset{Y,g}{}q^{m_{e,f,Y,g}}\mathrm{\#}(C_{e,f,Y,g})\text{ and }\underset{e,f}{}q^{n_{e,f,Y,g}}\mathrm{\#}(C_{e,f,Y,g})=\chi _c(R_2(Y,g)),$$
where $`m_{e,f,Y,g}`$ and $`n_{e,f,Y,g}`$ are (non positive) integers. Moreover, if $`C_{e,f,Y,g}`$ is non empty, then we have the equality of part c) of the proposition. Thus we have
$`{\displaystyle \underset{e,f}{}}\mathrm{\#}(L_2(e,f))x^{\tau e\underset{¯}{dim}N+e+\tau f\underset{¯}{dim}M+f}`$ $`={\displaystyle \underset{e,f,Y,g}{}}q^{m_{e,f,Y,g}}\mathrm{\#}(C_{e,f,Y,g})x^{\tau e\underset{¯}{dim}N+e+\tau f\underset{¯}{dim}M+f}`$
$`={\displaystyle \underset{e,f,Y,g}{}}q^{m_{e,f,Y,g}}\mathrm{\#}(C_{e,f,Y,g})x^{\tau g\underset{¯}{dim}H^0(Y)+g}X_{SP_Y}`$
$`={\displaystyle \underset{e,f,Y,g}{}}q^{n_{e,f,Y,g}}\mathrm{\#}(C_{e,f,Y,g})x^{\tau g\underset{¯}{dim}H^0(Y)+g}X_{SP_Y}+Z`$
$`={\displaystyle \underset{Y,g}{}}\mathrm{\#}(R_2(Y,g))x^{\tau g\underset{¯}{dim}H^0(Y)+g}X_{SP_Y}+Z,`$
where $`Z`$ is a term which vanishes at $`q=1`$. Hence,
$`{\displaystyle \underset{e,f}{}}\chi _1(L_2(e,f))x^{\tau e\underset{¯}{dim}N+e+\tau f\underset{¯}{dim}M+f}`$ $`={\displaystyle \underset{Y,g}{}}\chi _1(R_2(Y,g))x^{\tau g\underset{¯}{dim}H^0(Y)+g}X_{SP_Y}`$
$`={\displaystyle \underset{Y}{}}\chi _c(\mathrm{𝖤𝗑𝗍}^1(M,N)_Y)X_Y.`$
### 3.4.
This section is devoted to the proof of Proposition 3.
Before proving the equivalence of (i) and (ii), we need some preparation: consider the diagram
Note that (i) holds if and only if there is no $`\epsilon ^{}`$ which makes the left hand square commutative. We formalize this as follows: The diagram yields two complexes
where we write $`(,)`$ for $`\mathrm{𝖧𝗈𝗆}_{𝒞_Q}(,)`$ and where
$$\alpha ^{}=\left[\begin{array}{c}(i_M^{})_{}\\ (S^1i_N^{})^{}\end{array}\right],\beta ^{}=[(S^1i_N^{})^{},(i_M^{})_{}],\alpha =[(i_M^{})_{},(Si_N^{})^{}],\beta =\left[\begin{array}{c}(Si_N^{})^{}\\ (i_M^{})_{}\end{array}\right].$$
The two complexes are in duality via the pairings
$$(\eta ,\epsilon )\varphi (\eta \epsilon )\text{ and }(\eta ^{},\epsilon ^{})\varphi ^{}(\eta ^{}\epsilon ^{})$$
given by the (almost) canonical forms
$$\varphi :\mathrm{𝖧𝗈𝗆}_{𝒞_Q}(S^1N,SN)k\text{ and }\varphi ^{}:\mathrm{𝖧𝗈𝗆}_{𝒞_Q}(S^1N^{},SN^{})k.$$
Now let us prove the equivalence of (i) and (ii). Let $`p`$ denote the projection
$$(S^1N,M)(S^1N^{},M^{})(S^1N,M).$$
Then (i) says that $`\epsilon `$ does not belong to $`p(\mathrm{𝗄𝖾𝗋}\beta ^{})`$. This holds if and only if $`\epsilon `$ is not orthogonal to the orthogonal of $`p(\mathrm{𝗄𝖾𝗋}\beta ^{})`$ in $`(M,SN)(M^{},SN^{})`$. Now the orthogonal of the image of the map
is the kernel of its transpose
and this is precisely $`(M,SN)\mathrm{𝖨𝗆}\beta `$. So (i) holds if and only if $`\epsilon `$ is not orthogonal to $`(M,SN)\mathrm{𝖨𝗆}\beta `$, which means that (i) and (ii) are equivalent.
Let us prove that (ii) implies (iii). We choose a morphism $`f:MSN^{}`$ such that $`\beta (f)`$ belongs to $`\mathrm{𝖤𝗑𝗍}^1(M,N)`$ and is not orthogonal to $`\epsilon `$. This means that we have
$$(Si_N^{})f=\eta ,fi_M^{}=0,\varphi (\eta \epsilon )0.$$
Now we form triangles on $`\eta :MSN`$ and $`0:M^{}SN^{}`$. Thanks to the fact that $`\eta i_M^{}=0`$, we have a morphism of triangles
By applying $`H^0`$ we obtain a morphism of long exact sequences
Now we let $`Y^{}`$ be the image of $`N^{}M^{}`$ in $`H^0Y`$. An easy diagram chase then shows that $`N^{}`$ is the preimage of $`Y^{}`$ under $`H^0i`$ and $`M^{}`$ is the image of $`Y^{}`$ under $`H^0p`$.
Let us prove that (iii) implies (ii). We are given $`\eta `$ such that $`\varphi (\eta \epsilon )0`$, a triangle of $`𝒞_Q`$
and a diagram of $`kQ`$-modules
where $`Y^{}`$ is a submodule of $`H^0(Y)`$, $`N^{}`$ is the preimage of $`Y^{}`$ and $`M^{}`$ the image of $`Y^{}`$. We will show that $`\eta `$ belongs to $`\mathrm{𝖤𝗑𝗍}^1(M,N)\mathrm{𝖨𝗆}\beta `$. For this we consider the larger diagram
Here $`H^0(S^1\eta )`$ factors through $`N^{}`$ since its image is contained in the kernel of $`H^0i`$, which is contained in $`N^{}`$. Moreover, $`H^0\eta `$ vanishes on $`M^{}`$ since $`M^{}`$ is contained in the image of $`H^0p`$. Now recall that $`M`$ and $`N`$ are indecomposable $`kQ`$-modules and $`\mathrm{𝖤𝗑𝗍}_{kQ}^1(N,M)0`$. Therefore $`M`$ is non injective and so $`S^1M=\tau ^1M`$ is still a module (and not just an object of $`𝒞_Q`$) and similarly, $`N`$ is non projective and so $`SN=\tau N`$ is still a module. Moreover, $`M^{}`$ cannot have an injective direct factor (since that would also be an injective direct factor of $`M`$) and so $`S^1M^{}=\tau ^1M^{}`$ is still a module and similarly $`SN^{}=\tau N^{}`$ is still a module.
We would like to show that $`\eta \mathrm{𝖧𝗈𝗆}_{𝒞_Q}(M,\tau N)`$ comes from a morphism of modules. For this, recall that we have
$$\mathrm{𝖧𝗈𝗆}_{𝒞_Q}(U,V)=\underset{n}{}\mathrm{𝖧𝗈𝗆}_{𝒟^b(kQ)}(F^nU,V)$$
for arbitrary modules $`U,V`$, where $`F=\tau ^1S`$. Moreover, if $`U`$ and $`V`$ are indecomposable and either $`U`$ or $`V`$ does not lie on a cycle of $`𝒟^b(kQ)`$, then by part b) of proposition 1.5 of , there is at most one $`n`$ such that
$$\mathrm{𝖧𝗈𝗆}_{𝒟^b(kQ)}(F^nU,V)0.$$
We know that indecomposable postprojective $`kQ`$-modules do not lie on cycles of $`𝒟^b(kQ)`$. Thus if $`U`$ and $`V`$ are indecomposable and one of them is postprojective, we have
$$\mathrm{𝖧𝗈𝗆}_{kQ}(U,V)0\mathrm{𝖧𝗈𝗆}_{kQ}(U,V)\stackrel{_{}}{}\mathrm{𝖧𝗈𝗆}_{𝒞_Q}(U,V).$$
In the case where $`Q`$ is a Dynkin quiver (which we assume), all modules are postprojective.
Now $`M`$ and $`\tau N`$ are indecomposable and we have
$$\mathrm{𝖧𝗈𝗆}_{kQ}(M,\tau N)=D\mathrm{𝖤𝗑𝗍}_{kQ}^1(N,M)0$$
and hence
$$\mathrm{𝖧𝗈𝗆}_{kQ}(M,\tau N)=\mathrm{𝖧𝗈𝗆}_{𝒞_Q}(M,\tau N).$$
In particular, $`\eta `$ comes from a morphism of modules so that we have $`\eta =H^0\eta `$. Since $`H^0\eta `$ vanishes on $`M^{}M`$, the composition of $`\eta `$ with $`i_M^{}`$ vanishes. It remains to be shown that $`\eta `$ factors through $`SN^{}`$. Now $`\tau ^1M`$ and $`N`$ are also indecomposable and we have
$$\mathrm{𝖧𝗈𝗆}_{kQ}(\tau ^1M,N)=D\mathrm{𝖤𝗑𝗍}_{kQ}^1(N,M)0$$
and hence
$$\mathrm{𝖧𝗈𝗆}_{kQ}(\tau ^1M,N)=\mathrm{𝖧𝗈𝗆}_{𝒞_Q}(\tau ^1M,N).$$
Thus $`\tau ^1\eta `$ comes from a morphism of modules and $`\tau ^1\eta =H^0(\tau ^1\eta )`$. Since $`H^0(\tau ^1\eta )`$ factors as $`i_N^{}f`$ for a morphism $`f:\tau ^1MN^{}`$, we have $`\tau ^1(\eta )=i_N^{}f`$ and therefore
$$\eta =\tau (i_N^{})\tau f=(Si_N^{})(Sf).$$
### 3.5.
We now give a proof of Proposition 4.
We prove part a). The projection $`p_1:CL_2`$ is surjective by the equivalence between (i) and (iii) in Proposition 3. Let $`([\epsilon ],M^{},N^{})`$ be in $`L_2`$ and pick an element of $`p_2(p_1^1([\epsilon ],M^{},N^{}))`$. Recall that it is an $`(L(Y)\times k^{})`$-orbit and by construction, for each point $`(i,p,\eta ,Y^{})`$ of the orbit, we have a diagram
where $`N^{}`$ is the preimage of $`Y^{}`$ under $`H^0(i)`$ and $`M^{}`$ the image of $`Y^{}`$ under $`H^0(p)`$. We have a morphism of short exact sequences
Thus, if we fix the triangle
$$\text{},$$
then the possible submodules $`Y^{}`$ form an affine space endowed with a simply transitive action by
$$\mathrm{𝖧𝗈𝗆}_{kQ}(H^0(p)Y^{},(H^0i)(N)/(H^0i)(N^{}))=\mathrm{𝖧𝗈𝗆}_{kQ}(M^{},N/N^{}),$$
where we have the last equality because $`H^0(p)(Y^{})=M^{}`$ and $`\mathrm{𝗄𝖾𝗋}(H^0i)N^{}N`$. Thus the space does not depend on the choice of a point $`(i,p,\eta ,Y^{})`$ in the orbit. Moreover, the set of the possible straight lines
$$[\eta ](\mathrm{𝖤𝗑𝗍}_{𝒞_Q}^1(M,N)\mathrm{𝖨𝗆}\beta )\mathrm{𝖤𝗑𝗍}^1(M,N)$$
is the complement of the hyperplane defined by $`\varphi (\mathrm{?}\eta )=0`$ inside
$$(\mathrm{𝖤𝗑𝗍}_{𝒞_Q}^1(M,N)\mathrm{𝖨𝗆}\beta ).$$
Thus these $`\eta `$ also form an affine space. Therefore, the variety $`p_1^1([\epsilon ],M^{},N^{})`$ is an extension of affine spaces by Lemma 3.
We prove part b). Let $`(i,p,\eta ,Y^{})`$ be a point of an orbit in $`R_2`$ and $`([\eta ],M^{},N^{})`$ in $`p_1(p_2^1([\eta ],Y^{}))`$. Then, $`N^{}:=H^0(i)^1(H^0(Y^{}))`$ and $`M^{}:=H^0(p)(H^0(Y^{}))`$ only depend on the choice of the orbit. Thus the set $`p_1p_2^1([\eta ],Y^{})`$ is parametrized by the $`[\epsilon ]`$ with $`\varphi (\eta \epsilon )0`$. These form an affine space inside $`\mathrm{𝖤𝗑𝗍}_{𝒞_Q}^1(M,N)`$
We prove part c). Consider the diagram of $`kQ`$-modules
Since $`N^{}`$ is the preimage of $`Y^{}`$ under $`H^0(i)`$, the kernels of $`N^{}Y^{}`$ and $`H^0(i)`$ are isomorphic. We denote both by $`K`$. Dually, since $`M^{}`$ is the image of $`Y^{}`$ under $`H^0(p)`$, the cokernels of $`(H^0Y)/Y^{}M/M^{}`$ and of $`H^0(p)`$ are isomorphic. We denote both by $`C`$. Then in the Grothendieck group of $`kQ`$-modules, we have the following equalities
$$\tau (N^{})+\tau (M^{})=\tau (Y^{})+\tau (K)\text{ and }N/N^{}+M/M^{}=(H^0Y)/Y^{}+C.$$
Therefore we have
$$\tau N^{}N+N^{}+\tau M^{}M+M^{}=\tau (Y^{})+\tau (K)(H^0Y)/Y^{}C.$$
In the notation of the proposition, it remains to be shown that
$$x^{\underset{¯}{dim}\tau (K)\underset{¯}{dim}C}=X_{SP_Y}.$$
Now in fact, it is easy to see that $`X_{SP_i}=x^{I_i}`$, where $`I_i`$ is the injective module associated to $`i`$. Hence, for each projective $`kQ`$-module $`P`$,
$$X_{SP}=x^{\nu P},$$
where $`\nu `$ is the Nakayama functor. So, what we have to prove is the equality
$$\tau (K)C=\nu P_Y$$
in the Grothendieck group of $`kQ`$-modules. For this, we first note that by the triangle
$$NYMSN$$
of $`𝒞_Q`$, the module $`K`$ is a quotient of $`H^0(S^1M)=H^0(\tau ^1M)`$. Since $`M`$ is indecomposable non injective, $`\tau ^1M`$ is still a module so that $`K`$ is a quotient of $`\tau ^1M`$ and $`\tau K`$ a quotient of $`M`$. In particular, $`\tau K`$ is still a module and $`\tau K=H^0(\tau K)`$. Thus it suffices to prove that
$$H^0(\tau K)C=\nu P_Y$$
in the Grothendieck group of $`kQ`$-modules. For this, we first note that we have a triangle
in $`𝒟^b(kQ)`$ and thus in $`𝒞_Q`$. Secondly, we have a split triangle
in $`𝒞_Q`$; and thirdly, we have the triangle
$$NYMSN$$
in $`𝒞_Q`$. Note that $`H^0i`$ is the composition of the morphism $`NY`$ with the projection $`YH^0Y`$. If we form the octahedron associated with this composition, the three triangles we have just mentioned appear among its faces, as well as a new triangle, namely
$$\text{}.$$
Note that $`S^2P_Y=\nu P_Y`$ by the Calabi-Yau property. If we apply $`H^{}`$ to this triangle, we obtain the exact sequence of $`kQ`$-modules
$$\text{}.$$
Since $`M`$ is an indecomposable module, $`\tau M`$ is either an indecomposable non injective module or zero. The image of $`\nu P_Y\tau M=H^0\tau M`$ is injective (as a quotient of an injective module). Hence it is zero and we get an exact sequence
$$\text{}.$$
In the Grothendieck group, this yields
$$0=M\mathrm{𝖼𝗈𝗄}(H^0i)H^0(\tau K)+\nu P_Y=CH^0(\tau K)+\nu P_Y$$
and this is what we had to prove.
### 3.6.
We give here some examples which illustrate Theorem 2.
Example 1.
Suppose that $`M`$ and $`N`$ are indecomposable objects such that $`dim\mathrm{𝖤𝗑𝗍}_𝒞^1(N,M)=1`$. As in , let $`B`$ and $`B^{}`$ be the unique objects such that there exist non split triangles $`MBNSM`$ and $`NB^{}MSN`$. In this case, we have
$$\mathrm{𝖤𝗑𝗍}^1(N,M)_B=k^{},\mathrm{𝖤𝗑𝗍}^1(M,N)_B^{}=k^{}.$$
The cluster multiplication theorem then asserts that $`X_NX_M=X_B+X_B^{}`$. Note that in this particular case, this formula was a conjecture of and is since a theorem of , .
Example 2.
If $`Q`$ is the following quiver of type A<sub>2</sub>:
(3.2)
$$\text{}.$$
Set $`M=S_1S_1`$, $`N=S_2S_2`$. If $`Y`$ is an object such that $`\mathrm{𝖤𝗑𝗍}^1(N,M)_Y`$ is not empty then $`Y`$ is either $`S_1P_2S_2`$ or $`P_2P_2`$ and it is an easy exercise to prove that the cardinality of $`\mathrm{𝖤𝗑𝗍}^1(N,M)_Y`$ on $`𝔽_q`$ is respectively $`q^2+2q+1`$ and $`q(q^21)`$. In a dual way, if $`Y`$ is an object such that $`\mathrm{𝖤𝗑𝗍}^1(M,N)_Y`$ is not empty then $`Y`$ is either $`S_1S_2`$ or $`0`$ and the cardinality of $`\mathrm{𝖤𝗑𝗍}^1(N,M)_Y`$ on $`𝔽_q`$ is respectively $`q^2+2q+1`$ and $`q(q^21)`$.
The cluster multiplication theorem gives:
$$X_NX_M=X_{S_1P_2S_2}+X_{S_1S_2}.$$
Then, applying again the formula – note that we now in the case of the previous example– yields:
$$X_NX_M=X_{2P_2}+2X_{P_2}+1.$$
Note that this can be easily verified by taking squares in $`X_{S_2}X_{S_1}=X_{P_2}+1`$.
Example 3.
We give an example where the two indecomposable objects $`M`$ and $`N`$ are such that their first extension space has dimension 2.
We consider the following quiver $`Q`$ of type D<sub>4</sub>:
(3.3)
Set $`\alpha _i=\underset{¯}{dim}(S_i)`$. Let $`R`$, $`S`$, $`T`$, $`U`$ be the indecomposable $`kQ`$-modules with respective dimension vectors $`\alpha _2+\alpha _3+\alpha _4`$, $`\alpha _1+\alpha _2+\alpha _4`$, $`\alpha _1+\alpha _2+\alpha _3`$, $`\alpha _1+2\alpha _2+\alpha _3+\alpha _4`$. Consider the injective module $`M=I_2`$ and the simple module $`N=S_2`$. If $`Y`$ is an object such that $`\mathrm{𝖤𝗑𝗍}^1(M,N)_Y`$ is not empty then $`Y`$ is either $`U`$, $`RP_1`$, $`SP_3`$ or $`TP_4`$ and the cardinality of $`\mathrm{𝖤𝗑𝗍}^1(M,N)_Y`$ on $`𝔽_q`$ is respectively $`q2`$, 1, 1 and 1. Symmetrically, if $`Y`$ is an object such that $`\mathrm{𝖤𝗑𝗍}^1(N,M)_Y`$ is not empty then $`Y`$ is either $`SP_2`$, $`SP_1+S_1`$, $`SP_3+S_3`$ or $`SP_4+S_4`$ and the cardinality of $`\mathrm{𝖤𝗑𝗍}^1(N,M)_Y`$ on $`𝔽_q`$ is respectively $`q2`$, 1, 1 and 1.
Hence, the cluster multiplication theorem gives
$$2X_NX_M=X_U+X_{P_1}X_R+X_{P_3}X_S+X_{P_4}X_TX_{SP_2}+X_{SP_1}X_{S_1}+X_{SP_3}X_{S_3}+X_{SP_4}X_{S_4}.$$
Applying the formula again gives
$$2X_NX_M=X_U+3(X_U+1)X_{SP_2}+3(X_{SP_2}+1),$$
and finally
$$X_NX_M=X_U+3+X_{SP_2}.$$
## 4. Finite cluster algebras and positivity
### 4.1.
In order to go further, we have to recall some terminology on cluster algebras. More precise and complete information can be found in .
Let $`nm`$ be two positive integers. We fix the coefficient field $`_0=(u_{n+1},\mathrm{}u_m)`$ and the ambient field $`=(u_1,\mathrm{},u_m)`$, where the $`u_i`$’s are indeterminates. Let $`𝐱`$ be a free generating set of $``$ over $`_0`$ and let $`\stackrel{~}{B}=(b_{ij})`$ be an $`m\times n`$ integer matrix such that the submatrix $`B=(b_{ij})_{1i,jn}`$ is antisymmetric. Such a pair $`(𝐱,\stackrel{~}{B})`$ is called a seed.
Let $`(𝐱,\stackrel{~}{B})`$ be a seed and let $`x_j`$, $`1jn`$, be in $`𝐱`$. We define a new seed as follows. Let $`x_j^{}`$ be the element of $``$ defined by the exchange relation:
$$x_jx_j^{}=\underset{b_{ij}>0}{}x^{b_{ij}}+\underset{b_{ij}<0}{}x^{b_{ij}},$$
where, by convention, we have $`x_i=u_i`$ for $`i>n`$. Set $`𝐱^{}=𝐱\{x_j^{}\}\backslash \{x_j\}`$. Let $`\stackrel{~}{B}^{}`$ be the $`m\times n`$ matrix given by
$$b_{ik}^{}=\{\begin{array}{cc}b_{ik}\hfill & \text{if }i=j\text{ or }k=j\hfill \\ b_{ik}+\frac{1}{2}(|b_{ij}|b_{jk}+b_{ij}|b_{jk}|)\hfill & \text{ otherwise.}\hfill \end{array}$$
Then a result of Fomin and Zelevinsky asserts that $`(𝐱^{},\stackrel{~}{B}^{})`$ is a seed. It is called the mutation of the seed $`(𝐱,\stackrel{~}{B})`$ in the direction $`x_j`$. We consider all the seeds obtained by iterated mutations. The free generating sets occurring in the seeds are called clusters, and the variables they contain are called cluster variables. By definition, the cluster algebra $`𝒜(𝐱,\stackrel{~}{B})`$ associated to the seed $`(𝐱,\stackrel{~}{B})`$ is the $`[u_{n+1},\mathrm{},u_m]`$-subalgebra of $``$ generated by the set of cluster variables. The Laurent phenomenon, see , asserts that the cluster variables are Laurent polynomials with integer coefficients in the $`x_i`$, $`1im`$. So, we have $`𝒜(𝐱,B)[x_1^{\pm 1},\mathrm{},x_m^{\pm 1}]`$.
Except in section 5.4, we will be concerned with cluster algebras such that $`n=m`$, i.e. $`\stackrel{~}{B}=B`$. Note that an antisymmetric matrix $`B`$ defines a quiver $`Q=Q_B`$ with vertices corresponding to its rows (or columns) and which has $`b_{ij}`$ arrows from the vertex $`i`$ to the vertex $`j`$ whenever $`b_{ij}>0`$. The cluster algebra associated to the seed $`(𝐱,B)`$ will be also denoted by $`𝒜(Q)`$.
An important result of asserts that a cluster algebra is finite, i.e. has a finite number of cluster variables, if and only if there exists a seed associated to a quiver of simply laced Dynkin type. In this case, the Dynkin type is unique.
Now fix a quiver $`Q`$ of simply laced Dynkin type. Then, by , the $``$-module generated by the variables $`X_M`$, where $`M`$ runs over the set of objects of $`𝒞_Q`$, is an algebra; it is the cluster algebra $`𝒜(Q)`$ and the cluster variables are the $`X_M`$’s, where $`M`$ runs through the indecomposable objects of $`𝒞_Q`$.
An object $`M`$ of $`𝒞`$ is called exceptional if it has no selfextensions, i.e. $`\mathrm{𝖤𝗑𝗍}^1(M,M)=0`$. An object $`T`$ of $`𝒞`$ is a tilting object if it is exceptional , multiplicity free, and has the following maximality property: if $`M`$ is an indecomposable object such that $`\mathrm{𝖤𝗑𝗍}^1(M,T)=0`$, then $`M`$ is a direct factor of a direct sum of copies of $`T`$. Note that a tilting object can be identified with a maximal set of indecomposable objects $`T_1,\mathrm{},T_n`$ such that $`\mathrm{𝖤𝗑𝗍}^1(T_i,T_j)=0`$ for all $`i,j`$.
In view of , the main result of can be stated as follows: the map $`X_\mathrm{?}`$ : $`MX_M`$ induces a bijection from the set of tilting objects to the set of clusters of $`𝒜(Q)`$.
### 4.2.
Here we prove a positivity theorem that was conjectured in , see also .
###### Theorem 3.
For any object $`M`$ of $`𝒞_Q`$, the variable $`X_M`$ is in $`_0[u_i^{\pm 1}]`$.
###### Proof.
It is sufficient to prove that for any $`M`$ in $`\mathrm{𝗆𝗈𝖽}kQ`$, and for any $`e`$ in $`^n`$, we have $`\chi _c(\mathrm{𝖦𝗋}_e(M))0`$. For this, we recall the construction of the (classical) Hall algebra $`(Q)`$ of $`\mathrm{𝗆𝗈𝖽}(Q)`$: The algebra $`(Q)`$ is the vector space with basis $`\{e_M\}_M`$, where $`M`$ runs through the set of isoclasses of finite-dimensional $`kQ`$-modules. The multiplication rule on $`(Q)`$ is given by
$$e_Me_N=\underset{X}{}P_{M,N}(1)e_X,$$
where $`P_{M,N}^X`$ is the Hall polynomial defined by
$$P_{M,N}^X(q)=\mathrm{\#}\{Y,Y\mathrm{𝖦𝗋}(X),YN,X/YM\}_{𝔽_q}.$$
It is known from that $`(Q)`$ is an associative algebra, isomorphic to the enveloping algebra $`U(𝔫)`$ of a maximal nilpotent subalgebra $`𝔫`$ of the semisimple Lie algebra $`𝔤`$ associated to the Dynkin diagram underlying $`Q`$. Via the isomorphism $`(Q)U(𝔫)`$, the basis $`\{e_M\}_M`$ is identified with a Poincaré-Birkoff-Witt basis of $`U(𝔫)`$ (in the sense of ).
For any dimension vector $`e`$, set
$$b_e=\underset{\underset{¯}{dim}(N)=e}{}e_N(Q).$$
Then we have
$$b_e^{}b_e=\underset{\underset{¯}{dim}(N^{})=e^{}}{}e_N^{}\underset{\underset{¯}{dim}(N)=e}{}e_N=\underset{\underset{¯}{dim}M=e+e^{}}{}\left(\underset{\underset{¯}{dim}N^{}=e^{},\underset{¯}{dim}N=e}{}P_{N^{},N}^M(1)\right)e_M.$$
Hence,
$$b_e^{}b_e=\underset{\underset{¯}{dim}m=e+e^{}}{}\chi _c(\mathrm{𝖦𝗋}_e(M))e_M,$$
by Lemma 4. Now by \[21, 7.3\], for any dimension vector $`e`$, the element $`b_e`$ is in Lusztig’s canonical basis of $`U(𝔫)`$, when the quantification parameter $`q`$ is equal to 1. Moreover, by \[22, par. 14\], the product of two elements of the canonical basis has positive coefficients in its expansion in the canonical basis. Finally, by \[21, 7.11\], an element of the canonical basis has positive coefficients in its expansion in the PBW-base $`\{e_M\}`$. Hence we have $`\chi _c(\mathrm{𝖦𝗋}_e(M))0`$. ∎
We can have more by noting that the element $`b_e`$ of the proof is the element of the canonical basis associated to the dense orbit of the moduli space of dimension vector $`e`$. This easily implies that $`\chi _c(\mathrm{𝖦𝗋}_e(M))>0`$ if $`\mathrm{𝖦𝗋}_e(M)\mathrm{}`$. It would be interesting to prove that the variety $`\mathrm{𝖦𝗋}_e(M)`$ has a cellular decomposition and to find a combinatorial way to calculate its Euler characteristic.
As a particular case of the theorem, we obtain the
###### Corollary 1.
For any quiver $`Q`$ of simply laced Dynkin type, the cluster variables of $`𝒜(Q)`$ are Laurent polynomials in the variables $`x_i`$ with positive integer coefficients.
We can also generalize Fomin and Zelevinky’s denominator theorem , see also \[9, Theorem 3.6\], to any quiver $`Q`$ of simply laced Dynkin type:
###### Corollary 2.
Let $`M`$ be an indecomposable $`kQ`$-module and set $`\underset{¯}{dim}M=_im_i\underset{¯}{dim}S_i`$. Then the denominator of $`X_M`$ as an irreducible fraction of integral polynomials in the variables $`u_i`$ is $`u_i^{m_i}`$.
###### Proof.
By 3.1 and the positivity theorem, $`X_M`$ is a linear combination with positive integer coefficients of terms $`u_i^{n_i}`$, $`n_i`$. These terms are indexed by the set of dimension vectors of submodules $`N`$ of $`M`$, and for each submodule $`N`$, we have, by the Serre duality formula, that
$$n_i=N,S_iS_i,M/N.$$
So, it is sufficient to prove that
* for all $`i`$, we have $`N,S_i+S_i,M/N(\underset{¯}{dim}M)_i`$ and
* for all $`i`$, there exists a submodule $`N`$ such that the equality holds.
First recall that for each module $`X`$, we have $`X,I_i=P_i,X=(\underset{¯}{dim}X)_i`$. Now, as $`\mathrm{𝗆𝗈𝖽}kQ`$ is hereditary, the injective resolution of $`S_i`$ yields
$$N,S_iN,I_i.$$
Dually, we have
$$S_i,M/NP_i,M/N.$$
Adding both inequalities and using the formula above gives the first point. Now fix a vertex $`i`$ of the quiver and let $`J`$ be the set of vertices $`j`$ such that there exists a path from $`i`$ to $`j`$. Define the subspace $`N`$ of $`M`$ to be the sum of the subspaces $`e_jM`$, $`jJ`$, where $`e_j`$ is the idempotent associated with $`j`$. Then, by construction, $`N`$ is a submodule of $`M`$ with the following properties: a) $`(\underset{¯}{dim}N)_j=0`$ if there is a path $`ji`$, b) $`(\underset{¯}{dim}M/N)_j=0`$ if there is a path $`ij`$. Considering the injective resolution
$$0S_iI_iI0,$$
we obtain the equality $`N,S_i=N,I_iN,I=N,I_i`$, by a). Dually, property b) implies that $`S_i,M/M_i=P_i,M/M_i`$. So we obtain the equality $`M_i,S_i+S_i,M/M_i=(\underset{¯}{dim}M)_i`$ as required.
## 5. Filtrations and bases
As in the previous section, we assume that $`Q`$ is a quiver of simply laced Dynkin type. Recall that the elements of the generating set $`X_M`$, $`M\mathrm{𝗈𝖻𝗃}(𝒞_Q)`$, of $`𝒜(Q)`$ can be written
$$X_M=\underset{e}{}\chi _c(\mathrm{𝖦𝗋}_e(M))x_i^{e,\underset{¯}{dim}S_i\underset{¯}{dim}S_i,\underset{¯}{dim}Me}.$$
We will show that this formula provides ‘good’ filtrations for finite cluster algebras.
### 5.1.
Fix a quiver $`Q`$ of simply laced Dynkin type and let $`B`$ be the antisymmetric matrix such that $`Q=Q_B`$. We can view $`B`$ as an endomorphism of $`𝖦_\mathrm{𝟢}(\mathrm{𝗆𝗈𝖽}kQ)`$ endowed with the basis $`\underset{¯}{dim}S_i`$, $`1in`$.
###### Lemma 6.
We have $`Be=_i(e,\underset{¯}{dim}S_i\underset{¯}{dim}S_i,e)\underset{¯}{dim}S_i`$.
###### Proof.
Recall that $`B=(b_{ij})`$ with
$$b_{ij}=\{\begin{array}{cc}1\hfill & \text{ if }ij\hfill \\ 1\hfill & \text{ if }ji\hfill \\ 0\hfill & \text{ otherwise }\hfill \end{array}.$$
Hence,
$$B(\underset{¯}{dim}S_j)=\underset{ij}{}\underset{¯}{dim}S_i\underset{ji}{}\underset{¯}{dim}S_i.$$
Note now that
$$\underset{¯}{dim}S_j,\underset{¯}{dim}S_i\underset{¯}{dim}S_i,\underset{¯}{dim}S_j=\{\begin{array}{cc}1\hfill & \text{ if }ij\hfill \\ 1\hfill & \text{ if }ji\hfill \\ 0\hfill & \text{ otherwise. }\hfill \end{array}$$
This proves the lemma. ∎
We need the
###### Lemma 7.
Let $`M`$ be an indecomposable $`kQ`$-module and let $`N`$ be a proper submodule of $`M`$. Then, $`B(\underset{¯}{dim}N)0`$.
###### Proof.
Suppose that $`NM`$ and $`\underset{¯}{dim}N,\underset{¯}{dim}S_i\underset{¯}{dim}S_i,\underset{¯}{dim}N=0`$ for all $`i`$. This implies that for any $`kQ`$-module $`X`$, we have $`N,XX,N=0`$.
Suppose first that $`N`$ has no injective component. Then, $`N`$ has an injective hull $`I`$, with the following property: $`[N,I]^1=0=[I,N]^0`$. Hence,
$$N,IN,I[N,I]^0>0,$$
which contradicts the above formula.
Suppose that $`N`$ has a non zero injective component $`J`$. Then since $`N`$ is a proper submodule of $`M`$, the module $`J`$ is a proper direct factor of $`M`$, in contradiction with the assumption that $`M`$ is indecomposable. ∎
For any Laurent polynomial $`X`$ in the set of variables $`\{x_i\}`$, the support $`\mathrm{𝗌𝗎𝗉𝗉}(X)`$ of $`X`$ is by definition the set of points $`\lambda =(\lambda _1,\mathrm{},\lambda _n)`$ of $`^n`$ such that the $`\lambda `$-component, i.e. the coefficient in $`_ix^{\lambda _i}`$, of $`X`$ is non zero. For any point $`\lambda `$ in $`^n`$, identified with $`𝖦_\mathrm{𝟢}(\mathrm{𝗆𝗈𝖽}kQ)`$, let $`C_\lambda `$ be the convex cone with vertex $`\lambda `$, and whose edge vectors are generated by $`B(\underset{¯}{dim}S_i)`$. The previous lemma easily implies the following proposition.
###### Proposition 5.
Fix an indecomposable object $`M`$ of $`𝒞_Q`$, and let $`M=M_0SP_M`$ be its decomposition as in 2.4. Then, $`\mathrm{𝗌𝗎𝗉𝗉}(X_M)`$ is in $`C_{\lambda _M}`$ with $`\lambda _M:=(\underset{¯}{dim}S_i,\underset{¯}{dim}M_0+\underset{¯}{dim}P_M,\underset{¯}{dim}S_i)`$. Moreover, the $`\lambda _M`$-component of $`X_M`$ is 1.
### 5.2.
The following proposition rephrases a result of .
###### Proposition 6.
The map $`\lambda _\mathrm{?}`$ : $`𝒞^n`$, $`M\lambda _M`$ is surjective. Any fiber of a point in $`^n`$ contains a unique exceptional object. The cones generated by the images of tilting objects provide a complete simplicial fan.
###### Proof.
We first describe the exceptional objects of $`𝒞`$. For any $`M`$ in $`\mathrm{𝗈𝖻𝗃}(𝒞)`$, we denote by $`I_M`$ the set of $`i`$ such that $`P_i`$ is a component of $`P_M`$. The following fact is clear :
The object $`M=M_0SP_M`$ is exceptional if and only if $`M_0`$ is exceptional and $`(\underset{¯}{dim}M_0)_i=0`$ for any $`i`$ in $`I_M`$.
Recall now that for any dimension vector $`d`$ in $`𝖦_\mathrm{𝟢}(\mathrm{𝗆𝗈𝖽}kQ)`$, there exists a unique exceptional module $`M_d`$ such that $`\underset{¯}{dim}(M_d)=d`$.
Let $`E`$ be the set of exceptional modules. It decomposes into the disjoint union $`E=E_I`$, where $`I`$ runs over the set of partitions of $`\{1,\mathrm{},n\}`$ and where $`E_I:=\{ME,I_M=I\}`$.
For any object $`M=M_0(_im_iSP_i)`$, we set $`\underset{¯}{dim}(M)=\underset{¯}{dim}(M_0)(m_1,\mathrm{},m_n)`$.
On the one hand, it is known by that the cones generated by the images under $`\underset{¯}{dim}`$ of tilting objects of $`𝒞`$ provide a complete simplicial fan in $`^n`$. On the other hand, by the assertion above, $`\underset{¯}{dim}`$ provides a bijection from $`E`$ to $`^n`$, and via this bijection, the map $`\lambda _\mathrm{?}`$ is piecewise linear – the domains of linearity are the $`E_I`$’s. Moreover, on $`E_I`$, the matrix of $`\lambda `$ is triangular and the diagonal components are
$$d_i=\{\begin{array}{cc}1\hfill & \text{ if }iI\hfill \\ 1\hfill & \text{ if }iI\hfill \end{array}.$$
This proves the proposition. ∎
### 5.3.
Under the following hypothesis on $`Q`$, we will now define a filtration of the cluster algebra $`𝒜(Q)[x_1^{\pm 1},\mathrm{},x_n^{\pm 1}]`$.
HYPOTHESIS: There exists a form $`ϵ`$ on $`^n`$ such that
$$ϵ(B_Q\underset{¯}{dim}S_i)_{<0},\text{for all }i.$$
Note that for any Dynkin diagram except A<sub>1</sub>, there exists an orientation $`Q`$ satifying our hypothesis. Indeed, let $`Q_{alt}`$ be an alternating quiver. Then the matrix $`B_{alt}=(b_{ij})`$ associated to this quiver satisfies $`b_{ij}0`$ if $`i`$ is a source, and $`b_{ij}0`$ if $`i`$ is a sink. Moreover, each row of $`B_{alt}`$ is non zero. So we can take any form $`ϵ`$ whose coordinates in the dual basis of $`^n`$ satisfy $`ϵ_i<0`$ if $`i`$ is a source and $`ϵ_i>0`$ if $`i`$ is a sink. Note also that we have
$$Be=\underset{i}{}e+\tau e,\underset{¯}{dim}S_i\underset{¯}{dim}S_i$$
for all $`e^n`$ so that the above hypothesis holds iff the image of the positive cone of $`𝖦_\mathrm{𝟢}(\mathrm{𝗆𝗈𝖽}kQ)`$ under the map $`\tau +\mathrm{𝟏}`$ is strictly contained in a halfspace.
For any $`n`$ in $``$, set
$$F_n=(_{ϵ(\mu )n}x_i^{\mu _i})𝒜(Q).$$
Using Proposition 5 and Proposition 6, we obtain:
###### Proposition 7.
The set $`(F_n)_n`$ defines a filtration of $`𝒜(Q)`$. The graded algebra associated to this filtration is isomorphic to $`[u_1^{\pm 1},\mathrm{},u_n^{\pm 1}]`$.
###### Proof.
As we have $`F_nF_mF_{n+m}`$, the sequence $`(F_n)`$ is a filtration of $`𝒜(Q)`$. Moreover, Proposition 5 implies that for any indecomposable module $`M`$, we have
(5.1)
$$\text{gr}X_M=\text{gr}\underset{i}{}u_i^{(\lambda _M)_i}.$$
The result then follows from Proposition 6.
This implies
###### Corollary 3.
For any Dynkin quiver, the set $`:=\{X_M,M\mathrm{𝗈𝖻𝗃}(𝒞),\mathrm{𝖤𝗑𝗍}^1(M,M)=0\}`$ of variables corresponding to exceptional objets of the category $`𝒞`$ is a $``$-basis of the cluster algebra $`𝒜(Q)`$.
###### Proof.
This is obviously true for a quiver of type A<sub>1</sub>. Now, for any quiver, we have $`𝒜(Q)=𝒜(Q^{})`$ for some quiver $`Q^{}`$ which satisfies the hypothesis above. In this case, the proposition together with formula 5.1 imply that $``$ is $``$-free.
Let us prove now that $``$ generates $`𝒜(Q)`$ as a $``$-module. We first define a degeneration ordering $`_e`$ in $`\mathrm{𝗈𝖻𝗃}(𝒞)`$. Let $`M`$, $`M^{}`$ be objects of $`𝒞`$. We say that there is an elementary degeneration $`M^{}_eM`$ if $`MM^{}`$ is an elementary vector in $`K_0^{split}(𝒞)`$. We have
###### Lemma 8.
Let $`M`$, $`M^{}`$ be objects of $`𝒞`$.
* We have $`M^{}_eM`$ iff there are decompositions $`M=LUV`$ and $`M^{}=LE`$ where $`U`$ and $`V`$ are indecomposable and $`E`$ is the middle term of a non split triangle
* If we have $`M^{}_eM`$, then
(5.2)
$$0dim\mathrm{𝖤𝗑𝗍}^1(M^{},M^{})<dim\mathrm{𝖤𝗑𝗍}^1(M,M).$$
###### Proof.
a) If the condition holds, then $`M^{}M`$ equals the elementary vector $`U+VE`$. Conversely, if $`M^{}M`$ is an elementary vector, we have $`M^{}M=U+VE`$ where $`U`$, $`V`$ are indecomposable and $`E`$ is the middle term of a non split triangle as in the assertion. Then we have $`M^{}+E=M+U+V`$. It follows from Proposition 1 and Lemma 1 that $`U`$ and $`V`$ are not direct factors of $`E`$. Moreover, they are non isomorphic since no indecomposable has selfextensions. Thus, the object $`M^{}`$ decomposes as the sum of some $`L`$ and $`UV`$ so that we obtain $`L+U+V+E=M+U+V`$ and $`L+E=M`$.
b) Let $`U`$, $`V`$ and $`L`$ be as in a). For any objects $`N,N^{}`$ of $`𝒞`$, set $`[N,N^{}]^1=dim\mathrm{𝖤𝗑𝗍}_𝒞^1(N,N^{})`$. We view $`[\mathrm{?},\mathrm{?}]^1`$ as a symmetric bilinear form on $`K_0^{split}(𝒞)`$. For any object $`N`$ of $`𝒞`$, the long exact sequence obtained by applying $`\mathrm{𝖧𝗈𝗆}(\mathrm{?},SN)`$ to the triangle
shows that we have
$`()`$
$$[E,N]^1[U,N]^1+[V,N]^1.$$
Moreover, for $`N=U`$, we have the strict inequality
$$[E,U]<[U,U]^1+[V,U]^1$$
since in the sequence
$$\mathrm{𝖧𝗈𝗆}(V,SU)\mathrm{𝖧𝗈𝗆}(E,SU)\mathrm{𝖧𝗈𝗆}(U,SU),$$
the first map is not injective: its kernel contains $`e`$. By the inequality $`()`$, we have
$$[L,L]^1+2[L,E]^1[L,L]^1+2[L,U+V]$$
and it only remains to be shown that
$$[E,E]^1<[U+V,U+V]^1.$$
For this, we note that by the above inequalities, we have
$$[E,E]^1[E,U]^1+[E,V]^1,[E,V]^1[U,V]^1+[V,V]^1\text{ and }[E,U]^1<[U,U]^1+[V,U]^1.$$
Let us finish the proof of the corollary. It remains to be proved that each $`X_M`$ is in $``$. If $`M`$ is indecomposable, then $`M`$ is exceptional and hence is in $``$. If $`M`$ is not indecomposable, say $`M=M^{}M^{\prime \prime }`$, then by the cluster multiplication theorem and the lemma above, $`X_M`$ expands into a $``$-linear combination of terms $`X_Y`$ for objects $`Y`$ such that $`0\mathrm{𝖤𝗑𝗍}^1(Y,Y)<\mathrm{𝖤𝗑𝗍}^1(M,M)`$. By induction, we obtain that $`X_M`$ is in $``$. As the coefficients $`\chi _c(\mathrm{𝖦𝗋}_e(M))`$ in the cluster variable formula are integers, we obtain, using induction on the filtration, that $`X_M`$ is in $``$.
### 5.4.
In this section, we consider the more general case of finite cluster algebras associated to a rectangular $`m\times n`$ matrix $`\stackrel{~}{B}`$. We want to prove that our construction provides a toric degeneration of the spectrum of finite cluster algebras.
Let $`B`$ be the antisymmetric submatrix associated to $`\stackrel{~}{B}`$. We have a projection $`\pi `$: $`𝒜(\stackrel{~}{B})𝒜(B)`$ such that $`u_iu_i`$, if $`1in`$, and $`u_i1`$, if $`n+1im`$. This projection gives a one-to-one correspondence between the cluster variables of the two cluster algebras. For any indecomposable object $`M`$ of $`𝒞`$, we denote by $`\stackrel{~}{X}_M`$ the cluster variable such that $`\pi (\stackrel{~}{X}_M)=X_M`$. We fix a quiver $`Q`$ as in the previous section and we suppose without loss of generality that $`B=B_Q`$. Let $`F_n`$ be the filtration of $`𝒜(B)`$ constructed from $`Q`$. We now consider the filtration $`\stackrel{~}{F}_n=\pi ^1(F_n)`$, $`n`$, induced by $`\pi `$ from the filtration $`F_n`$.
###### Theorem 4.
The graded algebra gr$`𝒜(\stackrel{~}{B})`$ associated to the filtration $`\stackrel{~}{F}_n`$ is isomorphic to a subalgebra of $`[u_i^{\pm 1},\mathrm{\hspace{0.17em}1}im]`$ generated by a finite set of unitary monomials.
###### Proof.
For any $`z`$ in $`𝒜(\stackrel{~}{B})`$, we denote by gr$`z`$ the corresponding element in the graded algebra gr$`𝒜(\stackrel{~}{B})`$. It is sufficient to prove that for any indecomposable object $`M`$ of $`𝒞`$, gr$`\stackrel{~}{X}_M`$ is a unitary monomial in the gr$`u_i`$’s. This is true for $`M=SP_i`$ as in this case, gr$`\stackrel{~}{X}_M`$ is the monomial $`u_i`$. Now, we make an induction with the help of the Hom-ordering $``$ in $`\mathrm{𝗂𝗇𝖽}\mathrm{𝗆𝗈𝖽}kQ`$. By the exchange relation as in \[8, 3.4\], we have
$$\stackrel{~}{X}_{\tau (M)}\stackrel{~}{X}_M=p\underset{i}{}\stackrel{~}{X}_{B_i}+q,$$
where $`p`$, $`q`$ are unitary monomials in the $`u_i^{}s`$, $`n+1im`$. In this relation, $`\tau (M)`$ and $`B_i`$ are indecomposable objects which verify the induction hypothesis. Suppose that gr$`\stackrel{~}{X}_{B_i}`$ and 1 have the same degree in gr$`𝒜(\stackrel{~}{B})`$. Then, the monomial gr$`X_{B_i}`$ and $`1`$ have the same degree in gr$`𝒜(B)`$. But this would imply that the coefficient of the monomial gr$`X_M`$ is not $`1`$, in contradiction with Proposition 5. Hence, gr$`\stackrel{~}{X}_{B_i}`$ and 1 do not have the same degree in gr$`𝒜(\stackrel{~}{B})`$. This implies that either gr$`\stackrel{~}{X}_{\tau (M)}`$gr$`\stackrel{~}{X}_M=`$gr$`p`$ gr$`\stackrel{~}{X}_{B_i}`$ or gr$`\stackrel{~}{X}_{\tau (M)}`$gr$`\stackrel{~}{X}_M=`$gr$`q`$. In both cases, the induction process is proved.
This is a classical corollary of the theorem, see .
###### Corollary 4.
The spectrum of a finite cluster algebra has a toric degeneration.
### 5.5.
With the help of the basis of Corollary 3, we can reformulate Theorem 2. Actually, we can give a complete realization of the cluster algebra $`𝒜(Q)`$ from the cluster category $`𝒞_Q`$.
We recall the degeneration ordering $`_e`$ in $`\mathrm{𝗈𝖻𝗃}(𝒞)`$ defined for the proof of Lemma 8. Suppose that there is an elementary degeneration $`M^{}_eM`$ with elementary vector $`Z_i+Z_jY_{ij}`$. Then, by the lemma 8, we can define the ratio
$$r(M,M^{}):=cz_iz_j/dim\mathrm{𝖤𝗑𝗍}^1(M,M),$$
where $`c`$ is the multiplicity number associated to the elementary vector, and where $`z_i`$, resp. $`z_j`$, are the multiplicity of $`Z_i`$, resp. $`Z_j`$, in $`M`$. Let $``$ be the ordering generated by $`_e`$.
Note the surprising fact that in the category $`𝒞`$, the composition of elementary degenerations can again be an elementary degeneration.
Note that, by 5.2, any chain of elementary degenerations descending from an object $`M`$ is finite.
For any pair of objects $`M`$, $`N`$ of $`𝒞`$ and for any chain from $`N`$ to $`M`$
$$\mathrm{\Gamma }:N=M_0_eM_1_e\mathrm{}_eM_k=M,$$
we set
$$r(M,N,\mathrm{\Gamma })=\underset{i=1}{\overset{k}{}}r(M_i,M_{i1}),r(M,N)=\underset{\mathrm{\Gamma }}{}r(M,N,\mathrm{\Gamma }),$$
where $`\mathrm{\Gamma }`$ runs through the chains from $`N`$ to $`M`$.
We define the exceptional Hall algebra of $`𝒞`$ to be the rational vector space $`_{exc}(𝒞)`$ of $``$-valued functions on the isomorphism classes of exceptional objects of $`𝒞`$ endowed with the multiplication given by
$$\chi _M\chi _N=\underset{K}{}r(MN,K)\chi _K,$$
where the sum runs over the isomorphism classes of exceptional objects $`K`$ and $`\chi _K`$ denotes the characteristic function. From Theorem 2 and section 3.2, we easily deduce the
###### Theorem 5.
The exceptional Hall algebra $`_{exc}(𝒞_Q)`$ is an associative $``$-algebra with unit element $`\chi _0`$. The map $`\chi _MX_M`$ induces an isomorphism between the exceptional Hall algebra $`_{exc}(𝒞_Q)`$ and the cluster algebra $`𝒜_Q`$.
## 6. Conjectures
### 6.1.
The results in and in this article are concerned with finite cluster algebras, with a fixed seed corresponding to a Dynkin quiver. We conjecture some generalizations for any seed.
Fix a quiver $`Q`$ of Dynkin type and a tilting object $`T`$ of $`𝒞_Q`$. Consider the so-called tilted algebra $`A_T:=\mathrm{𝖤𝗇𝖽}_𝒞(T)^{opp}`$ and the category $`\mathrm{𝗆𝗈𝖽}A_T`$ of finite dimensional $`A_T`$-modules.
We consider the form
$$N,M=dim\mathrm{𝖧𝗈𝗆}(N,M)dim\mathrm{𝖤𝗑𝗍}^1(N,M),N,M\mathrm{𝗆𝗈𝖽}A_T.$$
Remark that in general this form does not descend to the Grothendieck group of the category $`\mathrm{𝗆𝗈𝖽}A_T`$. One defines the antisymmetrized form:
$$N,M_a=N,MM,N,N,M\mathrm{𝗆𝗈𝖽}A_T.$$
We know that there exists a seed $`(𝐱_T,B_T)`$ of the cluster algebra $`𝒜(Q)`$ associated to the tilting object $`T`$, cf. . Moreover, by , the set $`\mathrm{𝗂𝗇𝖽}(\mathrm{𝗆𝗈𝖽}A_T)`$ is in bijection with the set of cluster variables which do not belong to $`𝐱_T`$.
###### Conjecture 1.
The form $`,_a`$ descends to the Grothendieck group $`𝖦_\mathrm{𝟢}(\mathrm{𝗆𝗈𝖽}A_T)`$. Its matrix for the basis $`(\underset{¯}{dim}S_i)`$ is $`B_T`$.
Set $`𝐱_T=\{x_1,\mathrm{},x_n\}`$. The following conjecture describes the bijection explained above. It can be seen as a generalization of Theorem 3.4 of .
###### Conjecture 2.
To any indecomposable module $`M`$ in $`\mathrm{𝗆𝗈𝖽}A_T`$, we assign
$$X_M:=\underset{e}{}\chi _c(\mathrm{𝖦𝗋}_e(M))\underset{i}{}x_i^{(B_Te)_iS_i,M}.$$
Then the set $`\{X_M,M\mathrm{𝗂𝗇𝖽}\mathrm{𝗆𝗈𝖽}A_T\}`$ is exactly the set of cluster variables which do not belong to $`𝐱_T`$.
Via a conjectural extension of theorem 3, this conjecture would yield positivity properties.
### 6.2.
As before, fix a tilting object $`T=_{i=1}^nT_i`$ of $`𝒞_Q`$, with $`T_i`$ indecomposable. By the discussion above, the set $`\mathrm{𝗂𝗇𝖽}(𝒞)`$ can be seen as a disjoint union
$$\mathrm{𝗂𝗇𝖽}(𝒞)=\mathrm{𝗂𝗇𝖽}\mathrm{𝗆𝗈𝖽}A_T\{T_i,\mathrm{\hspace{0.17em}1}in\}.$$
Hence, as in 2.4, each object $`M`$ of $`𝒞`$ has a unique decomposition $`M=M_0T_M`$, where $`M_0`$ is in $`\mathrm{𝗆𝗈𝖽}A_T`$ and where $`T_M`$ is a direct factor of a sum of copies of $`T`$.
Suppose that Conjecture 2 is true. Then, for any object $`M=M_0(_im_iT_i)`$, we can define the variable
$$X_M:=\underset{e}{}\chi _c(\mathrm{𝖦𝗋}_e(M_0))\underset{i}{}x_i^{(B_Te)_iS_i,M+m_i}.$$
Define the map $`\lambda _\mathrm{?}`$ : $`𝒞^n`$ to be given by $`M(S_i,Mm_i)`$.
###### Conjecture 3.
The cones generated by the images of tilting objects under $`\lambda _\mathrm{?}`$ provide a complete simplicial fan.
Note that if we replace the map $`\lambda _\mathrm{?}`$ by the dimension vector map, we obtain a complete fan which is in general not simplicial.
### 6.3.
We finish with a positivity conjecture which can be seen as an analogue of Lusztig’s positivity theorem, cf. , for the dual canonical basis.
###### Conjecture 4.
For any object $`M`$ and any exceptional object $`K`$ of $`𝒞`$, the integer $`r(M,K)`$ is non negative.
In particular, the conjecture implies that the coefficients in Theorem 5 are positive. Note that the rational numbers $`r(M,N,\mathrm{\Gamma })`$ defined in section 5 can be negative.
## 7. Appendix on constructibility
We present a general proof for the constructibility of the sets $`\mathrm{𝖤𝗑𝗍}^1(M,N)_Y`$ in a triangulated category with finitely many isoclasses of indecomposables.
Let $`k`$ be a field and $`𝒯`$ a $`k`$-linear triangulated category with suspension functor $`S`$ such that
* all $`\mathrm{𝖧𝗈𝗆}`$-spaces in $`𝒯`$ are finite-dimensional,
* each indecomposable of $`𝒯`$ has endomorphism ring $`k`$,
* each object is a finite direct sum of indecomposables,
* $`𝒯`$ has Serre duality, i.e. there is an equivalence $`\nu :𝒯𝒯`$ such that we have
$$D\mathrm{𝖧𝗈𝗆}(X,\mathrm{?})\stackrel{_{}}{}\mathrm{𝖧𝗈𝗆}(\mathrm{?},\nu X)$$
for each $`X𝒯`$, where $`D`$ denotes the functor $`\mathrm{𝖧𝗈𝗆}_k(\mathrm{?},k)`$.
It is not hard to show that the last condition is a consequence of the first three. The conditions imply that $`𝒯`$ has Auslander-Reiten triangles and that the Auslander-Reiten translation $`\tau `$ is given by $`S^1\nu `$.
For objects $`X,Y,Z`$ of $`𝒯`$, let $`\mathrm{𝖧𝗈𝗆}(X,Y)_Z`$ be the set of morphisms $`f:XY`$ such that there is a triangle
###### Proposition 8.
The set $`\mathrm{𝖧𝗈𝗆}(X,Y)_Z`$ is constructible.
###### Proof.
Recall that a split triangle is a triangle which is a direct sum of triangles one of whose three morphisms is an isomorphism. Let us call a triangle minimal if it does not have a non zero split triangle as a direct factor. A triangle
is minimal iff, in the category of morphisms, $`f`$ does not admit non zero factors of the following forms
$$\text{},\text{},\text{}.$$
Let us call such morphisms $`f`$ minimal. We now proceed by induction on the sum $`s`$ of the numbers of indecomposable modules occurring in the decompositions of $`X`$ and $`Y`$ into indecomposables. Clearly the assertion holds if $`s=0`$ i.e. $`X=Y=0`$. Let us suppose $`s>0`$. Then $`\mathrm{𝖧𝗈𝗆}(X,Y)_Z`$ is the disjoint union of the set $`M`$ of morphisms $`f`$ such that the triangle
is minimal and of the set $`M^{}`$ of morphisms admitting a non zero direct factor of one of the above forms. The set $`M^{}`$ is the union of orbits under $`\mathrm{𝖠𝗎𝗍}(X)\times \mathrm{𝖠𝗎𝗍}(Y)`$ of morphisms of the forms
$$f^{}\mathrm{𝟏}_U:X^{}UY^{}U,[f^{},0]:X^{}UY^{},[f^{},0]^t:X^{}Y^{}U,$$
where $`U`$ is non zero and $`f^{}`$ runs through the sets $`\mathrm{𝖧𝗈𝗆}(X^{},Y^{})_Z^{}`$ for suitable $`X^{}`$, $`Y^{}`$, $`Z^{}`$, of which there are only a finite number. It therefore follows from the induction hypothesis that $`M^{}`$ is constructible. It remains to be shown that the set $`M`$ is constructible. We work in the category $`\mathrm{𝗆𝗈𝖽}𝒯`$ of finitely presented functors on $`𝒯`$ with values in the category of $`k`$-vector spaces. It is an abelian category. Its projective objects are the representable functors $`\widehat{U}=\mathrm{𝖧𝗈𝗆}(\mathrm{?},U)`$ and these are also the injective objects. If $`U`$ is indecomposable in $`𝒯`$ and $`S_U`$ is the simple top of the indecomposable projective $`\widehat{U}`$, then $`S_U`$ admits the minimal projective presentation
where
is an Auslander-Reiten triangle. Moreover, $`S_U`$ is also the simple socle of the indecomposable injective
$$\widehat{S\tau U}=\widehat{\nu U}.$$
If $`f:XY`$ is a minimal morphism, then the morphisms
$$\widehat{X}\mathrm{𝗂𝗆}(\widehat{f}),\mathrm{𝗂𝗆}(\widehat{f})\widehat{Y}$$
induced by $`f`$ are a projective cover and an injective hull, respectively. Moreover, if $`f`$ is minimal and
is a triangle, then $`g`$ is also minimal, so that $`\widehat{Z}`$ is an injective hull of $`\mathrm{𝗂𝗆}(\widehat{g})=\mathrm{𝖼𝗈𝗄}(\widehat{f})`$. Therefore, the multiplicity $`m_U`$ of an indecomposable object $`\nu U`$ in the decomposition of $`Z`$ into indecomposables equals the multiplicity of the simple $`S_U`$ in the socle of $`\mathrm{𝖼𝗈𝗄}(\widehat{f})`$. Since $`U`$ has endomorphism algebra $`k`$, this also holds for $`S_U`$ and the multiplicity $`m_U`$ equals
$$dim\mathrm{𝖧𝗈𝗆}(S_U,\mathrm{𝖼𝗈𝗄}(\widehat{f})).$$
Now we have projective presentations
and
Thus, we can compute the space $`\mathrm{𝖧𝗈𝗆}(S_U,\mathrm{𝖼𝗈𝗄}(\widehat{f}))`$ as the quotient of the space of morphisms
modulo the subspace formed by the morphisms of the form $`(a,b)=(cp_U,fc)`$ for some morphism $`c:UX`$. The condition
$$dim\mathrm{𝖧𝗈𝗆}(S_U,\mathrm{𝖼𝗈𝗄}(\widehat{f}))=m_U$$
then translates into conditions on the ranks of the linear maps
$$(a,b)bp_Ufa\text{ and }c(cp_U,fc).$$
Clearly, the $`f\mathrm{𝖧𝗈𝗆}(X,Y)`$ satisfying these rank conditions for all indecomposables $`U`$ form a constructible subset. The intersection of this subset with the complement of $`M^{}`$ is still constructible (since $`M^{}`$ is) and clearly equals $`M`$. So $`M`$ is constructible. ∎
|
warning/0506/q-bio0506005.html
|
ar5iv
|
text
|
# Asymptotic construction of pulses in the discrete Hodgkin-Huxley model for myelinated nerves
## I Introduction
Understanding wave propagation in discrete excitable media is challenging because of poorly understood phenomena associated with spatial discreteness sleeman ; cbfhn ; keener1 ; fath ; cbn . The study of the transmission of nerve impulses along myelinated axons is a paradigmatic example. Myelinated nerve fibers, such as the motor axons of vertebrates, are covered almost everywhere by a thick insulating coat of myelin. Only a fraction of the active membrane is exposed, at small active nodes called Ranvier nodes. The myelinated axons of motor nerves can be very long, and contain hundreds or thousands of nodes biophys . The wave activity jumps from one node to the next one giving rise to “saltatory” propagation of impulses rushton . Saltatory conduction on myelinated nerve models has two important features. One is the possibility of increasing the speed of the nerve impulse while decreasing the diameter of the nerve fiber scott1 . The other is propagation failure when the myelin coat is damaged nature , which causes diseases such as multiple sclerosis.
The propagation of nerve impulses along a myelinated fiber can be described by the spatially discrete Hodgkin-Huxley system (HH) scott1 ; keener2 . For typical experimental data, this system couples equations for two fast variables and two slow variables. Most analytical studies of action potentials have focused on discrete FitzHugh-Nagumo (FHN) models for one fast and one slow variables kunov ; erneux2 ; sleeman ; hastings ; cbfhn ; tonnelier and discrete bistable equations for the leading edge richer ; zinner ; keener1 ; erneux1 ; fath ; cbn . Careful computational studies of models including more biological detail were carried out in fh ; goldman ; moore78 .
Numerical simulations show that reduced models involving only one fast and one slow variables are quantitatively inaccurate. Fig. 1 compares pulse solutions of the full discrete Hodgkin-Huxley model (circles) and a FitzHugh-Nagumo type reduction (asterisks) generated by exciting the left end of a fiber. The pulse solutions of the HH model are slower and narrower than the pulse solutions of the FHN-like reduction. Discarding one of the two fast variables produces a pulse with an increased speed, as illustrated by Figure 1. To describe the propagation of the pulse leading edge we need to keep the two fast variables. Their evolution is described by a discrete bistable equation coupled to an ordinary differential equation. This system yields a Nagumo type equation (a single discrete bistable equation) only in a very particular limit. Similarly, discarding one of the slow variables produces a slightly wider pulse than the one described by the two original slow variables. Moreover, FHN pulses and HH pulses have different structures. Pulse solutions of FHN-like models typically consist of two rigidly moving wave fronts cbfhn . HH pulses are “triangular waves”: they are formed by a leading wave front followed by a smooth region, as shown in Figure 1.
In this paper we introduce an asymptotic strategy to construct solitary pulses and wave trains in the discrete HH model. Our asymptotic study exploits time scale separation to split the variables in two blocks. The leading edge of the pulses is a wave front solution of the reduced system involving the two fast variables. This selects the speed of the wave. The two slow variables become relevant to determine the width of the peaks. Our asymptotic constructions agree reasonably well with numerical solutions for typical experimental data. For continuous HH models, a quantitative approximation scheme exploiting time scale separation was proposed in muratov . In the discrete case, traveling impulses do not appear as solutions of ordinary differential equations. Instead, more complicated differential-difference equations have to be analyzed.
The paper is organized as follows. In Section II we describe the discrete Hodgkin-Huxley model for myelinated nerves. We present a few numerical solutions and discuss the reasons for the poor performance of FitzHugh-Nagumo type reductions. Solitary pulses are constructed in Section III. Section IV studies wave train generation at the boundary by periodic firing. In Section V, we describe the two main mechanisms for propagation failure, related to damage in the myelin sheath and the action of chemicals. Section VI contains our conclusions. In Appendix A we recall the derivation of the model. Appendix B explains the nondimensionalization procedure. Appendices C, D and E contain additional material on pulses.
## II The Hodgkin-Huxley model for myelinated nerves
### II.1 Dimensionless equations
The dimensionless Hodgkin-Huxley (HH) model for a myelinated nerve axon is:
$`{\displaystyle \frac{dv_k}{dt}}+I(v_k,m_k,n_k,h_k)=D(v_{k+1}2v_k+v_{k1}),`$ (1)
$`{\displaystyle \frac{dm_k}{dt}}=\mathrm{\Lambda }_m(v_k)\left[m_{\mathrm{}}(v_k)m_k\right],`$ (2)
$`{\displaystyle \frac{dn_k}{dt}}=ϵ\mathrm{\Lambda }_n(v_k)\left[n_{\mathrm{}}(v_k)n_k\right],`$ (3)
$`{\displaystyle \frac{dh_k}{dt}}=ϵ\lambda \mathrm{\Lambda }_h(v_k)\left[h_{\mathrm{}}(v_k)h_k\right],`$ (4)
with
$`\begin{array}{ccc}I(v,m,n,h)\hfill & =& g_Kn^4(vV_K)+g_{Na}m^3h(v1)\hfill \\ & & +g_L(vV_L).\hfill \end{array}`$ (7)
Here, $`v_k`$ is the ratio of the deviation from rest of the membrane potential to a reference potential, $`n_k`$ is the potassium activation, $`m_k`$ is the sodium activation and $`h_k`$ the sodium inactivation. Appendix A recalls the derivation of the model. Appendix B details the procedure we have followed to nondimensionalize the system. The time scale is chosen by looking at the relaxation times $`\tau _n`$, $`\tau _m`$ and $`\tau _h`$ for $`n_k`$, $`m_k`$ and $`h_k`$. Typically, $`\tau _m\tau _n`$ and $`\tau _n\tau _h`$. Thus, $`ϵ=\frac{\tau _m}{\tau _n}`$ is small.
The rate functions and the stationary states in (2)-(4) can be fitted to experimental data. Figure 2 plots their shape for the motor axon of a frog. Analytic expressions for these functions and typical values of the parameters for the frog nerve are collected in Appendix B and will be used in our numerical tests. The values for $`g_{Na},g_K,g_L`$ have orders of magnitude $`1,10^1,10^2`$, respectively. The coupling parameter $`D10^1`$ and $`ϵ10^2`$ with $`\lambda 1`$. This means that we have two separate time scales and that discreteness effects are relevant.
### II.2 Numerical solutions
System (1)-(4) displays excitable behavior when it has a unique constant stationary state $`(v^{},m^{},n^{},h^{})`$, which is stable. Figures 3 and 4 show solitary pulses and wave train solutions generated by solving (1)-(4) numerically for the parameter values in Appendix B. After a short transient, the system relaxes to a traveling wave: $`v_k(t)=v(kct)`$, $`m_k(t)=m(kct)`$, $`n_k(t)=n(kct)`$ and $`h_k(t)=h(kct)`$. All nodes undergo the same evolution with a time delay $`\frac{1}{c}`$.
As shown in Figures 3 and 4, pulses and wave trains are composed of “sharp” interfaces and “smooth” regions. Let us describe the temporal profiles of a pulse as we move from left to right starting from the equilibrium region (1) in Figure 3. The leading edge of each pulse is a sharp interface, marked as (2) in Fig. 3. At this leading interface, $`n_k`$ and $`h_k`$ remain almost constant whereas $`v_k`$ and $`m_k`$ undergo abrupt changes. The evolution of $`v_k`$ and $`m_k`$ is described by a reduced bistable system and the leading edge is a wave front solution of this system. A smooth region follows, where $`n_k,h_k`$ vary slowly and $`v_k`$ and $`m_k`$ are quasi-static: $`m_k=m_{\mathrm{}}(v_k)`$ and $`I(v_k,m_{\mathrm{}}(v_k);n_k,m_k)=0`$. It is marked as (3) in Fig. 3. At the end of this region the approximation $`m_k=m_{\mathrm{}}(v_k)`$ breaks down, see the thin solid line in Figure 3. A trailing interface develops, where $`m_k`$ changes abruptly whereas $`n_k,h_k`$ remain almost constant. This new region is marked as (4) in Fig. 3. Notice that the fast variable $`v_k`$ happens to be near its equilibrium value $`v^{}`$ and decreases smoothly towards it. Diffusion is negligible and the evolution of $`v_k`$ and $`m_k`$ at the trailing edge is governed by a system of ordinary differential equations. The bistable structure is lost and no trailing wave front is formed.
The variables seem to be split in two blocks: slow (evolving in the dimensionless time scale $`T=ϵt`$) and fast (evolving in the dimensionless time scale $`t`$). This suggest the possibility of finding an asymptotic description of pulses by exploiting time scale separation. The difficulty of dealing with a discrete space variable is overcome by noticing that the traveling wave profiles are smooth functions of a continuous variable and solve a set of differential-difference equations (see Appendix C). Let us first check whether the number of variables can be reduced.
### II.3 Failure of simple FitzHugh-Nagumo type reductions
It is quite tempting to look for a simple asymptotic description of pulses in terms of one fast and one slow variable, as in the FitzHugh-Nagumo model fhn ; cbfhn . These reduced models usually assume that $`m_k`$ relaxes instantaneously to its equilibrium state $`m_k=m_{\mathrm{}}(v_k)`$ and that the sum of the two slow variables is constant during an action potential $`n_k+h_k=r`$ keener2 :
$`{\displaystyle \frac{dv_k}{dt}}+g_Kn_k^4(v_kV_K)+g_{Na}m_{\mathrm{}}(v_k)^3(rn_k)(v_k1)`$
$`+g_L(v_kV_L)=D(v_{k+1}2v_k+v_{k1}),`$ (8)
$`{\displaystyle \frac{dn_k}{dt}}=ϵ\mathrm{\Lambda }_n(v_k)\left[n_{\mathrm{}}(v_k)n_k\right].`$ (9)
In view of the numerical results described in Section II.2, these assumptions are inaccurate. The solid line in Fig. 3 shows that the difference $`m_km_{\mathrm{}}(v_k)`$ is not negligible at the leading wave front. Setting $`m_k=m_{\mathrm{}}(v_k)`$ distorts the speed, as shown in Figure 1. On the other hand, $`n_k+h_k`$ is not constant during at action potential. We may select $`r=n^{}+h^{}`$ to fit the leading edge but a slightly different value of $`r`$ is required for the trailing part of the pulse. Keeping $`n_k+h_k=r`$ during the action potential alters the width of the pulse, as shown in Figure 1.
In conclusion, the FHN type reduction (8)-(9) may be useful to gain insight on pulse motion and propagation failure in excitable media, but it is quantitatively inaccurate for the HH system with realistic parameter values.
## III Asymptotic construction of pulses
Accurate descriptions of pulse waves in the HH model have to deal with the full set of equations. In this Section, we find an approximation of their temporal profiles by matched asymptotic expansions as $`ϵ0`$. This construction yields predictions for the speed and width of the pulses, as well as a characterization of the parameter ranges in which propagation fails. We explain the procedure for the parameter values indicated in (74). The structure of the pulse may change slightly for other parameter values. Other possible structures are discussed in Appendix E.
In a traveling pulse we distinguish five regions (illustrated in Figure 3). In each of them, a reduced description holds. The whole profile is reconstructed by matching the approximated solutions found in each region at zeroth order (see lager for a description of this technique). Our reference time scale is the slow time scale $`T=ϵt`$. The technical details of the matching are given in Appendix D.
The resulting temporal pulse profiles with $`k`$ fixed have the following structure:
1. Front of the pulse. This is region (1) in Figure 3. Here $`v_kv^{}`$, $`m_km^{}`$, $`n_kn^{}`$ and $`h_kh^{}`$, $`(v^{},m^{},n^{},h^{})`$ being the stable equilibrium state of the system.
2. Leading edge of the pulse, located at $`T=T_0`$ and marked as (2) in Figure 3. The slow variables, $`n_k`$ and $`h_k`$, remain essentially constant but the fast variables, $`v_k`$ and $`m_k`$, undergo rapid changes in the time scale $`\overline{t}=\frac{TT_0}{ϵ}(\mathrm{},\mathrm{})`$. To leading order,
$`\begin{array}{ccc}\frac{dv_k}{d\overline{t}}+I(v_k,m_k,n_k,h_k)=D(v_{k+1}2v_k+v_{k1}),\hfill & & \\ \frac{dm_k}{d\overline{t}}=\mathrm{\Lambda }_m(v_k)\left[m_{\mathrm{}}(v_k)m_k\right],\hfill & & \\ \frac{dn_k}{d\overline{t}}=0,\frac{dh_k}{d\overline{t}}=0,\hfill & & \end{array}`$ (13)
Thus, $`n_k`$ and $`n_k`$ remain almost constant. Matching with the previous region, $`n_k=n^{}`$ and $`h_k=h^{}`$ (see Appendix D). The evolution of $`v_k`$ and $`m_k`$ is described by the ‘fast reduced system’:
$`{\displaystyle \frac{dv_k}{d\overline{t}}}+I(v_k,m_k,n_k,h_k)=D(v_{k+1}2v_k+v_{k1}),`$ (14)
$`{\displaystyle \frac{dm_k}{d\overline{t}}}=\mathrm{\Lambda }_m(v_k)\left[m_{\mathrm{}}(v_k)m_k\right],`$ (15)
with $`n_k=n^{}`$ and $`h_k=h^{}`$. This system displays bistable behavior. Let us denote by $`v=\nu ^{(i)}(n,h),i=1,2,3`$ the three solutions of
$`f(v;n,h)=I(v,m_{\mathrm{}}(v),n,h)=0`$ (16)
in a neighborhood of $`(n^{},h^{})`$. The system (14)-(15) has a unique stable traveling wave front solution $`v_k(t)=v(kc\overline{t})`$, $`m_k=m(kc\overline{t})`$ joining the two stable constant states $`\nu ^{(3)}(n^{},h^{})`$, $`m_{\mathrm{}}(\nu ^{(3)})`$ and $`\nu ^{(1)}(n^{},h^{})`$, $`m_{\mathrm{}}(\nu ^{(1)})`$, which propagates with a definite speed $`c=c_+(n^{},h^{})`$. This wave front solution is the leading edge of the pulse.
3. Top of the pulse. This is region (3) in Figure 3. Here, the slow variables evolve in the time scale $`T`$ and the fast variables relax instantaneously to their equilibrium values: $`m_k=m_{\mathrm{}}(v_k)`$, in which $`v_k`$ solves $`f(v_k;n_k,h_k)=0`$. The evolution of the slow variables is governed by the ’slow reduced system’:
$`\begin{array}{c}\frac{dn_k}{dT}=\mathrm{\Lambda }_n(v_k)\left[n_{\mathrm{}}(v_k)n_k\right],\hfill \\ \frac{dh_k}{dT}=\lambda \mathrm{\Lambda }_h(v_k)\left[h_{\mathrm{}}(v_k)h_k\right],\hfill \end{array}`$ (19)
for $`T_0<T<T_1`$. Matching with the previous stage we get $`v_k=\nu ^{(3)}(n_k,h_k)`$, $`n_k=n^{}`$ and $`h_k=h^{}`$ at $`T=T_0`$ (see Appendix D). For the parameter values indicated in (74), the third branch of roots $`v_k=\nu ^{(3)}(n_k,h_k)`$ of $`f(v_k;n_k,h_k)=0`$ disappears at $`(n_k,h_k)=(n^{[1]},h^{[1]})`$, colliding with the second branch: $`\nu ^{(3)}(n^{[1]},h^{[1]})=\nu ^{(2)}(n^{[1]},h^{[1]})=v^{[1]}`$. At the corresponding time $`T=T_1`$, region (3) ends. The time $`T_1`$ is characterized by $`n_k(T_1)=n^{[1]}`$ and $`h_k(T_1)=h^{[1]}`$ nota .
4. Trailing edge of the pulse, located at $`T_1`$ and marked as (4) in Figure 3. In this region, $`v_k`$ is no longer at equilibrium since the two largest roots of $`f(v;n_k,h_k)=0`$ are lost. The fast variables evolve in the time scale $`\overline{t}=\frac{TT_1}{ϵ}(\mathrm{},\mathrm{})`$, whereas the slow variables remain essentially constant. Matching with the previous stage, $`n_k=n^{[1]}`$, $`h_k=h^{[1]}`$. Due to the particular shape of $`f`$ (see the dashed line in Figure 5 (a)), $`\nu ^{(2)}(n^{[1]},h^{[1]})`$ is close enough to $`\nu ^{(1)}(n^{[1]},n^{[1]})`$ for $`v_kv_{k1}`$ to be small. Thus, we may neglect the discrete differences $`D(v_{k+1}2v_k+v_{k1})`$ and the fast variables are governed by a system of ordinary differential equations:
$`\begin{array}{c}\frac{dv_k}{d\overline{t}}=I(v_k,m_k,n^{[1]},h^{[1]}),\hfill \\ \frac{dm_k}{d\overline{t}}=\mathrm{\Lambda }_m(v_k)\left[m_{\mathrm{}}(v_k)m_k\right].\hfill \end{array}`$ (22)
This system has one stable equilibrium point: $`v^{[2]}=\nu ^{(1)}(n^{[1]},h^{[1]}),`$ $`m^{[2]}=m_{\mathrm{}}(v^{[2]})`$. The fast variables evolve from their initial values $`v_k=v^{[1]}`$ and $`m_k=m^{[1]}=m_{\mathrm{}}(v^{[1]})`$ to the equilibrium point.
5. Pulse tail. This is region (5) in Figure 3. In the pulse tail, the fast variables relax instantaneously to their equilibrium values: $`m_k=m_{\mathrm{}}(v_k)`$ and $`v_k=\nu ^{(1)}(n_k,h_k).`$ The slow variables solve (19) with $`v_k=\nu ^{(1)}(n_k,h_k)`$ for $`T_1<T<\mathrm{}`$ and evolve smoothly from $`(n^{[1]},h^{[1]})`$ to their equilibrium values $`(h^{},n^{})`$ as $`T\mathrm{}`$.
Uniform approximations to the temporal profiles obtained by gluing together the approximated solutions in each region are given in Appendix D. Figure 6 compares the asymptotic reconstruction to the actual profiles. The agreement improves as $`ϵ`$ decreases.
Our construction characterizes the speed of the traveling pulse: it is the speed of the wave front solution of (14)-(15) with $`n_kn^{}`$ and $`h_kh^{}`$. The speed of these fronts can be predicted by a depinning analysis for small speeds (as in cbn ) or in terms of continuous waves for large speeds (as in keener1 ).
Let us now compute the width of the pulse. In the peak, $`(n_k,h_k)`$ lies in the integral curve of:
$`\begin{array}{c}\frac{dh}{dn}=\frac{\lambda \mathrm{\Lambda }_h\left(\nu ^{(3)}(n,h)\right)\left[h_{\mathrm{}}\left(\nu ^{(3)}(n,h)\right)h\right]}{\mathrm{\Lambda }_n\left(\nu ^{(3)}(n,h)\right)\left[n_{\mathrm{}}\left(\nu ^{(3)}(n,h)\right)n\right]},\hfill \end{array}`$ (24)
selected by $`h(n^{})=h^{}`$ and $`h(n^{[1]})=n^{[1]}`$. Going back to the time scale $`t=\frac{T}{ϵ}`$ in (19), the peak width is:
$`\begin{array}{c}𝒯_1=\frac{T_1T_0}{ϵ}=_n^{}^{n^{[1]}}\frac{ds}{ϵ\mathrm{\Lambda }_n\left(\nu ^{(3)}(s,h(s))\right)\left[n_{\mathrm{}}\left(\nu ^{(3)}(s,h(s))\right)s\right]}\hfill \end{array}`$ (26)
where the integral is calculated along the solution $`h(n)`$ of (24). Similarly, the duration of tail is:
$`𝒯_2={\displaystyle _{n^{[1]}}^n^{}}{\displaystyle \frac{ds}{ϵ\mathrm{\Lambda }_n\left(\nu ^{(1)}(s,h(s))\right)\left[n_{\mathrm{}}\left(\nu ^{(1)}(s,h(s))\right)s\right]}},`$ (27)
with
$`\begin{array}{c}\frac{dh}{dn}=\frac{\lambda _h\mathrm{\Lambda }_h\left(\nu ^{(1)}(n,h)\right)\left[h_{\mathrm{}}\left(\nu ^{(1)}(n,h)\right)h\right]}{\mathrm{\Lambda }_n\left(\nu ^{(1)}(n,h)\right)\left[n_{\mathrm{}}\left(\nu ^{(1)}(n,h)\right)n\right]},h(n^{[1]})=h^{[1]}.\hfill \end{array}`$ (29)
The integral $`𝒯_2`$ diverges due to a singularity at $`n^{}`$. However, we can use it to predict how long does it take for the tail to get close enough to $`n^{},h^{}`$ replacing $`n^{}`$ by $`n^{}\eta `$.
The spatial length of the peak is found using the traveling wave structure: $`v_k(t)=v(kct).`$ Thus, $`𝒯_1`$ is the time elapsed from the instant at which the leading front reaches a point $`k`$ to the instant when the end of the peak crosses the same point $`k`$. The number of nodes $`L_1`$ in the pulse peak is approximately the integer part of:
$`\begin{array}{c}L_1c_+(n^{},h^{})𝒯_1.\hfill \end{array}`$ (31)
Our asymptotic construction is consistent when $`L_11`$. This yields a restriction on the size of $`ϵ`$ for the existence of pulses:
$`ϵc_+{\displaystyle _n^{}^{n^{[1]}}}{\displaystyle \frac{ds}{\mathrm{\Lambda }_n\left(\nu ^{(3)}(s,h(s))\right)\left[n_{\mathrm{}}\left(\nu ^{(3)}(s,h(s))\right)s\right]}}.`$ (32)
A similar argument can be applied in the infinite tail to quantify the number of nodes at which $`n_k,h_k`$ depart noticeably from equilibrium.
The values predicted by our asymptotic theory for the parameters in Appendix B are $`c=c_+=0.069`$, $`n^{[1]}=0.777`$, $`h^{[1]}=0.099`$, $`L_1=12`$, in good agreement with the numerical measurements. The leading and trailing edges contain about $`4`$ more nodes, that we have neglected.
Figure 7 compares the predicted and numerically measured widths and speeds as the parameters $`D`$ and $`g_{Na}`$ change. Notice that the values $`(n^{},h^{})`$ and $`(n^{[1]},h^{[1]})`$ are independent of $`D`$. Thus, the dependence of the width on $`D`$ comes through the speed. This explains the similar curves observed in Figure 7 (a) and (b). The triangles in Figure 7 (a) represent the number of nodes in the leading edge, that must be added to compute the total length of a peak. Changes in $`g_{Na}`$ affect the curves $`\nu ^{(i)}(n,h)`$, $`i=1,2,3`$, and the speed of the fronts in the reduced fast system. As Figures 7 (c) and (d) show, the quantitative agreement between predicted and numerically measured widths and speeds is reasonable, as long as we are not close to the critical thresholds $`D_c`$ and $`ϵ_c`$ for propagation failure. We will discuss this point further in Section V.
For the parameters indicated in (74), FHN type reductions perform poorly. The widths predicted as $`D`$ changes are particularly bad. We find 160 or 250 points in the peak, when 18 or 34 are expected. The predictions for the speed are better, see the triangles in Figure 7 (b). Similar comments apply when $`g_{Na}`$ is modified.
## IV Asymptotic construction of wave trains
When a nerve fiber is periodically excited, we expect propagation of signals in form of wave trains. Wave trains resemble a sequence of identical pulses periodically spaced keener3 . The asymptotic description of each of these pulses is similar to the construction of solitary pulses given in Section III. However, there are several differences. First, the front of the pulses is another pulse and not a region at equilibrium. Second, the leading edge is a wave front solution of the fast reduced system (14)-(15) with $`n_k=N`$, $`h_k=H`$, $`(N,H)(n^{},h^{})`$. This wave front solution selects the speed $`c=c_+(N,H)`$ of the wave train. It has to match the tail of the previous pulse, described by the slow reduced system (19) with $`v_k=\nu ^{(1)}(n_k,h_k)`$. Thus, we find a family of wave trains for couples $`(N,H)`$ lying on an integral curve of:
$`{\displaystyle \frac{dh}{dn}}={\displaystyle \frac{\lambda \mathrm{\Lambda }_h(\nu ^{(1)}(n,h))\left[h_{\mathrm{}}(\nu ^{(1)}(n,h))h\right]}{\mathrm{\Lambda }_n(\nu ^{(1)}(n,h))\left[n_{\mathrm{}}(\nu ^{(1)}(n,h))n\right]}}.`$ (33)
The curve is selected by observing that each pulse in the wave train has one driving leading edge, as in Section III. This means that $`h(n^{[1]})=h^{[1]}`$, where $`(n^{[1]},h^{[1]})`$ are the values for which $`\nu ^{(2)}(n^{[1]},h^{[1]})=\nu ^{(3)}(n^{[1]},h^{[1]})`$.
The spatial period is approximated by adding the lengths of the smooth regions: $`L(N,H)=c_+(N,H)𝒯(𝒩,)`$ where $`𝒯(N,H)=𝒯_1(N,H)+𝒯_2(N,H)`$ is the time spent in the excited and recovery branches. The temporal lengths $`𝒯_1,𝒯_2`$ are defined as in Section III, with $`n^{},h^{}`$ replaced by $`N,H`$. Now, $`𝒯_2(N,H)`$ is finite.
The values predicted by our asymptotic theory for Figure 4 are $`c=c_+=0.052`$, $`n^{[1]}=0.755`$, $`h^{[1]}=0.09`$, $`L3738`$, in good agreement with the numerical measurements. Numerically, we obtain $`L43`$ for the spatial period. The difference is approximately the number of points in the edges, that we have neglected.
## V Propagation failure
Pulses may fail to propagate at least by two reasons: weak coupling and inadequate time scale separation between the different variables. Weak couplings cause propagation failure by pinning the leading edge of the pulse. A small time scale separation between the two blocks of slow and fast variables produces pulses of vanishing width. Other scenarios for failure might arise when more than two time scales are present. We focus here on mechanisms for failure associated with changes in the parameters due to illness or drugs.
### V.1 Propagation failure due to damage in the myelin sheath
The loss of myelin alters the value of the coupling coefficient $`D`$. Then, the leading wave front can only propagate if the $`D`$ is large enough to avoid the pinning in the fast reduced system (14)-(15) with $`n_k=n^{}`$ and $`h_k=h^{}`$. The critical coupling $`D_c`$ depends on the shape of the nonlinear sources, which is controlled by the parameters $`g_{Na}`$,$`g_K`$,$`g_L`$, $`V_{Na}`$,$`V_K`$ and $`V_L`$. The general rule is that the area $`A_{23}`$ enclosed by $`f(v)`$ between its second and the third zeroes has to be large enough (depending on $`D`$) for the leading edge to propagate. The proximity of failure is detected by the fact that the wave profiles develop ‘steps’. Figure 8(b) illustrates generation of ‘steps’ in the time profile of the pulse for $`v_k`$ near propagation failure $`D=0.0072D_c`$. As for bistable equations, the depinning transition is associated with a bifurcation in the system cbn .
### V.2 Propagation failure due to the action of chemicals
Many drugs block the propagation of nerve impulses by reducing sodium and potassium conductivities scott1 ; keener2 . Figure 5(b) illustrates the impact of decreasing $`g_{Na}`$ on the shape of $`f`$. The enclosed area decreases and so does the propagation speed, up to a critical value at which the width of the pulses (given by formula (31)) vanishes. Only decremental pulses are observed, see Figure 9.
## VI Conclusion
We have introduced an asymptotic strategy to construct solitary pulses and wave trains in the discrete Hodgkin-Huxley model. Unlike FHN reductions, our asymptotic descriptions agree reasonably well with numerical solutions of the discrete HH system. We have discussed two mechanisms for propagation failure. The first one is related to damage in the myelin sheath and is reminiscent of the depinning transitions for wave fronts in bistable systems cbn as the coupling decreases. The second mechanism reflects the impact of chemicals on the time scale separation in the model. Our analysis applies to isolated nerve fibers. However, real motor nerves of vertebrates comprise several hundred of interacting fibers scott2 . It would be interesting to extend our asymptotic predictions to bundles of fibers.
Our asymptotic construction may be useful to understand systems with a mathematical structure similar to HH: models for propagation of impulses through cardiac tissue beeler , models of charge transport in semiconductor superlattices bonilla or the more complex Frankenhauser-Huxley model for myelinated nerves goldman .
###### Acknowledgements.
This work has been supported by the Spanish MCyT through grant BFM2002-04127-C02-02, and by the European Union under grant HPRN-CT-2002-00282. The author thanks one of the unknown referees for bringing reference muratov to her attention.
## Appendix A The discrete Hodgkin-Huxley model for myelinated nerves
Myelinated nerve fibers are covered almost everywhere by an insulating coat of myelin. Only at small active sites (nodes of Ranvier) can the membrane function in the normal way. Figure 10(a) illustrates the structure of a myelinated fiber. The length of the myelin sheath is typically $`1`$ to $`2`$ $`mm`$ (close to $`100d`$ where $`d`$ is the fiber diameter) and the width of the nodes is about $`1\mu m`$. The nodes have conduction properties similar to the unmyelinated nerve membrane, while the myelin has a much higher resistance and lower capacitance than the axonal membrane keener2 . Myelinated nerve fibers can be described by a linear diffusion equation which is periodically loaded by the active nodes pickard66 ; markin1967 ; fh . This picture can be simplified by lumping the internode capacitance of the myelin together with the nodal capacitance scott1 . The myelin is considered to be a perfect insulator. This leads to the equivalent circuit in Figure 10(b). $`C`$ and $`R`$ represent lumped resistance and capacitance. $`V_k`$, $`I_k`$ and $`I_{ion}(k)`$ represent the membrane potential, internodal current and ionic current at the $`k`$-th node. Applying Kirchoff’s laws to the circuit yields:
$`V_{k1}V_k=RI_k,I_kI_{k+1}=C{\displaystyle \frac{dV_k}{dt}}+I_{ion}(k)`$ (34)
Adopting at each node the Hodgkin-Huxley expression for the ion current hh , we obtain the discrete Hodgkin-Huxley model for a myelinated axon:
$`\begin{array}{c}C\frac{dV_k}{dT}+I_{ion}(V_k,M_k,N_k,H_k)=\hfill \\ \overline{D}(V_{k+1}2V_k+V_{k1}),\hfill \end{array}`$ (37)
$`\begin{array}{c}\frac{dM_k}{dT}=\overline{\lambda }_M\overline{\mathrm{\Lambda }}_M(V_k)(M_{\mathrm{}}(V_k)M_k),\hfill \\ \frac{dN_k}{dT}=\overline{\lambda }_N\overline{\mathrm{\Lambda }}_N(V_k)(N_{\mathrm{}}(V_k)N_k),\hfill \\ \frac{dH_k}{dT}=\overline{\lambda }_H\overline{\mathrm{\Lambda }}_H(V_k)(H_{\mathrm{}}(V_k)H_k),\hfill \end{array}`$ (41)
where the index $`k`$ designs the $`k`$-th node of the fiber. Here, $`V_k`$ is the deviation from rest of the membrane potential, $`N_k`$ is the potassium activation, $`M_k`$ is the sodium activation and $`H_k`$ the sodium inactivation. The ion current is given by:
$`\begin{array}{c}I_{ion}(V,M,N,H)=\overline{g}_{Na}M^3H(V\overline{V}_{Na,R})\hfill \\ +\overline{g}_L(V\overline{V}_{L,R})+\overline{g}_KN^4(V\overline{V}_{K,R}).\hfill \end{array}`$ (44)
The fraction of open $`K^+`$ channels is computed as $`N_k^4`$. The fraction of open $`Na^+`$ channels is approximated by $`M_k^3H_k`$. The parameters have the following interpretation. $`\overline{g}_{Na}`$ and $`\overline{g}_K`$ are the maximum conductance values for $`Na^+`$ and $`K^+`$ pathways, respectively. $`\overline{g}_L`$ is a constant leakage conductance. The corresponding equilibrium potentials are $`\overline{V}_{Na}`$, $`\overline{V}_K`$ and $`\overline{V}_L`$, respectively. Then, $`\overline{V}_{Na,R}=\overline{V}_{Na}\overline{V}_R`$, $`\overline{V}_{K,R}=\overline{V}_K\overline{V}_R`$ and $`\overline{V}_{L,R}=\overline{V}_L\overline{V}_R,`$ where $`\overline{V}_R`$ is the resting potential. $`C`$ is the membrane capacitance. The coefficient $`\overline{D}=\frac{1}{L(r_i+r_e)}=\frac{1}{R}`$, where $`L`$ is the length of the myelin sheath between nodes and $`r_i,r_e`$ the resistances per unit length of intracellular and extra-cellular media.
This model is adequate for the long axons of peripheral myelinated nerves. More biological detail can be included by adding an equation for the membrane potential $`V(x,t)`$ across the myelin sheath in the internodes fh :
$`c{\displaystyle \frac{V}{T}}={\displaystyle \frac{1}{r_i+r_e}}{\displaystyle \frac{^2V}{^2x}}{\displaystyle \frac{V}{r}},x(x_k,x_{k+1}),t>0`$ (45)
$`V(x_k,t)=V_k(t),V(x_{k+1},t)=V_{k+1}(t)`$ (46)
coupled with (41) and:
$`C{\displaystyle \frac{dV_k}{dT}}+I_{ion}(V_k,M_k,N_k,H_k)=I_k(t),`$ (47)
$`I_k(t)={\displaystyle \frac{1}{r_i+r_e}}[{\displaystyle \frac{V}{x}}(x_k^+,t){\displaystyle \frac{V}{x}}(x_k^{},t)].`$ (48)
This model produces a good quantitative approximation of the conduction velocity for toad axons fh . Numerical simulations of the sensitivity to different parameters (diameter, nodal area…) produce results in agreement with experiments moore78 ; goldman . The discrete model (37)-(41) is recovered by assuming that the axial currents along the myelin sheath $`\frac{V}{x}(x,t)`$ are constant in each internode. Then, $`\frac{V}{x}(x,t)=\frac{V_{k+1}(t)V_k(t)}{L}`$ in $`[x_k,x_{k+1}]`$ with $`L=x_{k+1}x_k`$. As a result, $`I_k(t)=\frac{1}{L(r_i+r_e)}(V_{k+1}V_k+V_{k1})`$. This approximation is reasonable in view of the numerical results in fh (see Figure 2 therein).
## Appendix B Dimensionless equations
For our numerical tests we have selected the parameters and coefficient functions of a frog. For the motor nerve of a frog, the data in cole can be fitted by the following rate and stationary state functions:
$`\begin{array}{c}\overline{\mathrm{\Lambda }}_M(V)=0.03\left[\frac{2.50.1V}{\mathrm{exp}(2.50.1V)1}+4\mathrm{exp}(\frac{V}{18})\right],\hfill \\ M_{\mathrm{}}(V)=\left[1+4\mathrm{exp}(\frac{V}{18})\frac{\mathrm{exp}(2.50.1V)1}{(2.50.1V)}\right]^1,\hfill \\ \overline{\mathrm{\Lambda }}_H(V)=\left[0.07\mathrm{exp}(\frac{V}{20})+\frac{1}{\mathrm{exp}(30.1V)+1}\right],\hfill \\ H_{\mathrm{}}(V)=\left[1+\frac{\mathrm{exp}(\frac{V}{20})}{0.07\left(\mathrm{exp}(30.1V)+1\right)}\right]^1,\hfill \\ \overline{\mathrm{\Lambda }}_N(V)=0.79\left[\frac{0.10.01V}{\mathrm{exp}(10.1V)1}+0.125\mathrm{exp}(\frac{V}{80})\right],\hfill \\ N_{\mathrm{}}(V)=\left[1+0.125\mathrm{exp}(\frac{V}{80})\frac{\mathrm{exp}(30.1V)1}{(0.10.01V)}\right]^1.\hfill \end{array}`$ (55)
Typical values of the parameters cole ; scott1 are given below:
$`\begin{array}{cccc}& & & \\ \overline{D}& \overline{g}_{Na}& \overline{g}_K& \overline{g}_L\\ & & & \\ 1/28(M\mathrm{\Omega })^1& 0.57\mu mho& 0.104\mu mho& 0.025\mu mho\\ & & & \\ \overline{V}_R& \overline{V}_{Na}& \overline{V}_K& \overline{V}_L\\ & & & \\ 75mV& 47mV& 75mV& 75mV\\ & & & \\ C& \overline{\lambda }_M& \overline{\lambda }_H& \overline{\lambda }_N\\ & & & \\ 2.64.7pF& 127(ms)^1& 1.76(ms)^1& 2(ms)^1\end{array}`$ (62)
We nondimensionalize the model by choosing as new variables $`v_k=\frac{V_k}{\overline{V}_{Na,R}}`$, $`t=T\overline{\lambda }_M`$, $`m_k=M_k`$, $`n_k=N_k`$, $`h_k=H_k`$. The dimensionless equations are:
$`\begin{array}{c}\frac{dv_k}{dt}+g_Kn_k^4(v_kV_K)+g_{Na}m_k^3h_k(v_k1)+\hfill \\ g_L(v_kV_L)=D(v_{k+1}2v_k+v_{k1}),\hfill \\ \frac{dm_k}{dt}=\mathrm{\Lambda }_m(v_k)\left[m_{\mathrm{}}(v_k)m_k\right],\hfill \\ \frac{dn_k}{dt}=\lambda _n\mathrm{\Lambda }_n(v_k)\left[n_{\mathrm{}}(v_k)n_k\right],\hfill \\ \frac{dh_k}{dt}=\lambda _h\mathrm{\Lambda }_h(v_k)\left[h_{\mathrm{}}(v_k)h_k\right].\hfill \end{array}`$ (68)
Set $`G=C\overline{\lambda }_M`$. Then, the dimensionless parameters are given by:
$`\begin{array}{ccccccccc}& & & & & & & & \\ g_{Na}& g_K& g_L& D& V_K& V_L& \lambda _n& \lambda _m& \\ & & & & & & & & \\ \frac{\overline{g}_{Na}}{G}& \frac{\overline{g}_K}{G}& \frac{\overline{g}_L}{G}& \frac{\overline{D}}{G}& \frac{\overline{V}_{K,R}}{\overline{V}_{Na,R}}& \frac{\overline{V}_{L,R}}{\overline{V}_{Na,R}}& \frac{\overline{\lambda }_N}{\overline{\lambda }_M}& \frac{\overline{\lambda }_H}{\overline{\lambda }_M}& \end{array}`$ (71)
The new rate functions and stationary states are obtained from (55) replacing $`V`$ by $`v\overline{V}_{Na,R}`$. In dimensionless units, the parameters for a frog nerve become:
$`\begin{array}{cccccccc}& & & & & & & \\ D& g_{Na}& g_K& g_L& V_K& V_L& \lambda _h& \lambda _n\\ & & & & & & & \\ 0.093& 1.49& 0.27& 0.065& 0& 0& 0.014& 0.015\end{array}`$ (74)
In our asymptotic analysis, we choose $`\lambda _n=ϵ`$ as small parameter and write $`\lambda _h=ϵ\lambda `$, $`\lambda =\frac{\lambda _h}{\lambda _n}1`$.
## Appendix C Equations for the wave profiles
The wave profiles and speeds solve an eigenvalue problem for a system of differential-difference equations:
$`\begin{array}{ccc}cv_z(z)\hfill & =\hfill & D(v(z+1)2v(z)+v(z1))\hfill \\ & & I(v(z),m(z),n(z),h(z)),\hfill \\ cm_z(z)\hfill & =\hfill & \mathrm{\Lambda }_m(v(z))\left[m_{\mathrm{}}(v(z))m(z)\right],\hfill \\ cn_z(z)\hfill & =\hfill & ϵ\mathrm{\Lambda }_n(v(z))\left[n_{\mathrm{}}(v(z))n(z)\right],\hfill \\ ch_z(z)\hfill & =\hfill & ϵ\lambda \mathrm{\Lambda }_h(v(z))\left[h_{\mathrm{}}(v(z))h(z)\right].\hfill \end{array}`$ (80)
In a solitary pulse, the profiles tend to the equilibrium states as $`z\pm \mathrm{}`$. In a wave train, the profiles are periodic: $`v(z)=v(z+L)`$, $`m(z)=m(z+L)`$, $`n(z)=n(z+L)`$ and $`h(z)=h(z+L)`$, $`L`$ being the spatial period.
## Appendix D Reconstruction of the temporal profile
We describe below the matching conditions and the uniform reconstruction of the pulse profiles in the different regions. The superscripts (I),…,(V) refer to the reduced descriptions corresponding to regions (1),…,(5). We have:
* The matching conditions for the reduced descriptions of the front of the pulse and the leading edge at $`T=T_0`$ are:
$`\begin{array}{c}v^{}v_k^{(II)}(\frac{TT_1}{ϵ})1,m^{}m_k^{(II)}(\frac{TT_1}{ϵ})1,\hfill \end{array}`$ (82)
if $`ϵT_0T1`$. The uniform approximations of the profiles for $`TT_0`$ are:
$`\begin{array}{cc}v_k^{unif}=\hfill & v_k^{(II)}(\frac{TT_0}{ϵ}),n_k^{unif}=n^{},\hfill \\ m_k^{unif}=\hfill & m_k^{(II)}(\frac{TT_0}{ϵ}),h_k^{unif}=h^{}.\hfill \end{array}`$ (85)
* The matching conditions for the reduced descriptions of the peak of the pulse and the leading edge at $`T=T_0`$ are:
$`\begin{array}{c}\nu _1^{(3)}(n_k^{(III)}(T),h_k^{(III)}(T))v_k^{(II)}(\frac{TT_0}{ϵ})1,\hfill \\ m_{\mathrm{}}\left(\nu _1^{(3)}(n_k^{(III)}(T),h_k^{(III)}(T))\right)m_k^{(II)}(\frac{TT_0}{ϵ})1,\hfill \\ n_k^{(III)}(T)n^{}1,h_k^{(III)}(T)h^{}1,\hfill \end{array}`$ (89)
if $`ϵTT_01`$. The uniform approximations of the profiles in regions (II)-(III) are:
$`\begin{array}{cc}v_k^{unif}=\hfill & v_k^{(II)}(\frac{TT_0}{ϵ})+\nu _1^{(3)}(n_k^{(III)}(T),h_k^{(III)}(T))\hfill \\ & \nu _1^{(3)}(n^{},h^{}),\hfill \\ m_k^{unif}=\hfill & m_k^{(II)}(\frac{TT_0}{ϵ})+m_{\mathrm{}}\left(\nu _1^{(3)}(n_k^{(III)}(T),h_k^{(III)}(T))\right)\hfill \\ & m_{\mathrm{}}\left(\nu _1^{(3)}(n^{},h^{})\right),\hfill \\ n_k^{unif}=\hfill & n_k^{(III)}(T),h_k^{unif}=h_k^{(III)}(T).\hfill \end{array}`$ (95)
* The matching conditions for the reduced descriptions of the peak of the pulse and the trailing edge at $`T=T_1`$ are:
$`\begin{array}{c}\nu _1^{(3)}(n_k^{(III)}(T),h_k^{(III)}(T))v_k^{(IV)}(\frac{TT_1}{ϵ})1,\hfill \\ m_{\mathrm{}}\left(\nu _1^{(3)}(n_k^{(III)}(T),h_k^{(III)}(T))\right)m_k^{(IV)}(\frac{TT_1}{ϵ})1,\hfill \\ n_k^{(II)}(T)n^{}1,h_k^{(II)}(T)h^{}1,\hfill \end{array}`$ (99)
if $`ϵT_1T1`$. The uniform approximations of the profiles in regions (II)-(III)-(IV) are:
$`\begin{array}{cc}v_k^{unif}=\hfill & v_k^{(II)}(\frac{TT_0}{ϵ})+v_k^{(IV)}(\frac{TT_1}{ϵ})\hfill \\ & +\nu _1^{(3)}(n_k^{(III)}(T),h_k^{(III)}(T))\hfill \\ & \nu _1^{(3)}(n^{},h^{})\nu _1^{(3)}(n^{[1]},h^{[1]}),\hfill \\ m_k^{unif}=\hfill & m_k^{(II)}(\frac{TT_0}{ϵ})+m_k^{(IV)}(\frac{TT_1}{ϵ})\hfill \\ & +m_{\mathrm{}}\left(\nu _1^{(3)}(n_k^{(III)}(T),h_k^{(III)}(T))\right)\hfill \\ & m_{\mathrm{}}\left(\nu _1^{(3)}(n^{},h^{})\right)m_{\mathrm{}}\left(\nu _1^{(3)}(n^{[1]},h^{[1]})\right),\hfill \\ n_k^{unif}=\hfill & n_k^{(III)}(T),h_k^{unif}=h_k^{(III)}(T).\hfill \end{array}`$ (107)
* The matching conditions for the reduced descriptions of the tail of the pulse and the trailing edge at $`T=T_1`$ are:
$`\begin{array}{c}\nu _1^{(1)}(n_k^{(V)}(T),h_k^{(V)}(T))v_k^{(IV)}(\frac{TT_1}{ϵ})1,\hfill \\ m_{\mathrm{}}\left(\nu _1^{(1)}(n_k^{(V)}(T),h_k^{(V)}(T))\right)m_k^{(IV)}(\frac{TT_1}{ϵ})1,\hfill \\ n_k^{(V)}(T)n^{}1,h_k^{(V)}(T)h^{}1,\hfill \end{array}`$ (111)
if $`ϵTT_11`$. The uniform approximations of the profiles for $`TT_1`$ are:
$`\begin{array}{cc}v_k^{unif}=\hfill & v_k^{(IV)}(\frac{TT_1}{ϵ})+\nu _1^{(1)}(n_k^{(V)}(T),h_k^{(V)}(T))\hfill \\ & \nu _1^{(1)}(n^{[1]},h^{[1]}),\hfill \\ m_k^{unif}=\hfill & m_k^{(IV)}(\frac{TT_1}{ϵ})+m_{\mathrm{}}\left(\nu _1^{(1)}(n_k^{(V)}(T),h_k^{(V)}(T))\right)\hfill \\ & m_{\mathrm{}}\left(\nu _1^{(1)}(n^{[1]},h^{[1]})\right),\hfill \\ n_k^{unif}=\hfill & n_k^{(V)}(T),h_k^{unif}=h_k^{(V)}(T).\hfill \end{array}`$ (117)
## Appendix E Pulses formed by two wavefronts
For the choice of parameters indicated in (74), the third region of the pulse ends when two branches of roots of the cubic source collapse. For other parameters, the situation might be different. It might happen that along the integral curve (24) we find a couple $`(n^{[1]},h^{[1]})`$ such that there are wave front solutions of the reduced fast system traveling with speed $`c_{}(n^{[1]},h^{[1]})=c_+(n^{},h^{})`$. Then, the top of the pulse ends at this point and the fourth region is now a traveling wave front. The pulse is formed by two rigidly moving traveling wave fronts.
Whether a traveling wave front is formed in the back of the pulse or not, can be guessed from the shape of the nonlinear source $`f`$. Figure 5(a) depicts $`f(v;n,h)`$ with the parameters values (74) when $`(n,h)=(n^{},h^{})`$. It is strongly asymmetric. The magnitude of the speed of the leading wave front is intuitively related to size of the area enclosed by $`f(v;n^{},h^{})`$ between its second and third zeroes, $`A^{}`$. If we vary $`(n,h)`$ along the curve (24), the speed of a back front is related to the area $`A`$ enclosed by $`f(v;n,h)`$ between its first and the second zeroes. We find that such areas are always smaller than $`A^{}`$ and the cubic structure is finally lost. A trailing wave front moving at the same speed as the leading wave front cannot be formed. Back wave fronts can only be observed for more symmetrical sources, when varying $`(n,h)`$ along the curve (24) we can make $`A`$ equal to $`A^{}`$.
We describe here how to modify the asymptotic construction in Sections III and IV to account for pulses formed by two rigidly moving wave fronts. In the asymptotic description of these pulses we distinguish again five regions. The first three are similar:
* The front of the pulse is described by $`v_kv^{}`$, $`m_km^{}`$, $`n_kn^{}`$ and $`h_kh^{}`$.
* The leading edge of the pulse is a wave front solution of the fast reduced system (14)-(15) with $`n_k=n^{}`$ and $`h_k=h^{}`$ joining $`(\nu ^{(1)}(n^{},h^{}),m_{\mathrm{}}(v^{}))=(v^{},m^{})`$ and $`(\nu ^{(3)}(n^{},h^{}),m_{\mathrm{}}(\nu ^{(3)}(n^{},h^{}))`$. This front propagates with a definite speed $`c=c_+(n^{},h^{})`$.
* In the transition between interfaces, $`v_k=\nu ^{(3)}(n_k,h_k)`$, $`m_k=m_{\mathrm{}}(v_k)`$ and $`n_k,h_k`$ solve the slow reduced system (19), evolving from $`(n^{},h^{})`$ to $`(n^{[1]},h^{[1]})`$.
The fourth region is different:
* The trailing edge is the wave front solution for the fast reduced system (14)-(15) with $`n_k=n^{[1]}`$ and $`h_k=h^{[1]}`$, joining $`(\nu ^{(3)}(n^{[1]},h^{[1]}),m_{\mathrm{}}(\nu ^{(3)}(n^{[1]},h^{[1]}))`$ and $`(\nu ^{(1)}(n^{[1]},h^{[1]}),m_{\mathrm{}}(\nu ^{(1)}(n^{[1]},h^{[1]}))`$, respectively. $`n^{[1]}`$ and $`h^{[1]}`$ are now selected in such a way that this front travels with speed $`c=c_{}(n^{[1]},h^{[1]})=c_+(n^{},h^{})`$.
The fifth is similar:
* In the pulse tail, $`v_k=\nu ^{(1)}(n_k,h_k)`$, $`m_k=m_{\mathrm{}}(v_k)`$ and $`n_k,h_k`$ solve the slow reduced system (19), evolving from $`(n^{[1]},h^{[1]})`$ to $`(n^{},h^{})`$ as $`t\mathrm{}`$.
The temporal and spatial width of the peak can be computed as in Section III. Now, the number of points in the leading and trailing wave fronts are neglected.
For wave trains, the construction in Section IV has to be modified as follows: $`(N,H)`$ and $`(n^{[1]},h^{[1]})`$ must lie in the same integral curve of (33) and satisfy $`c=c_{}(n^{[1]},h^{[1]})=c_+(N,H)`$.
|
warning/0506/cond-mat0506431.html
|
ar5iv
|
text
|
# On the nature of the magnetic ground-state wave function of V2O3.
## I Introduction
In the last 35 years V<sub>2</sub>O<sub>3</sub> has been the subject of wide experimental and theoretical investigations, after its identification as the prototype of Mott-Hubbard systems rice . At the beginning of the seventies an intensive set of measurements (resistivity mcwhan1 , specific heat mcwhan2 , magnetic susceptibility mcwhan3 , x-rays dernier , and polarized neutron diffraction moon ) was performed in order to clarify its crystal and magnetic structure as well as its transport properties. This marked a break-point with all previous theories adler1 ; rice1 that described the ”metal-semiconductor” phase transition of V<sub>2</sub>O<sub>3</sub>. Actually, two metal-insulator transitions were found: one from a paramagnetic metallic (PM) to an antiferromagnetic insulating (AFI) phase, below $`150`$ K, associated with a corundum-to-monoclinic crystal distortion; the other from the PM to a paramagnetic insulating (PI) phase at higher temperatures (above $`500`$ K): among transition metal oxides it is the only known example of PM-PI transition weibao , with no structural deformations and no magnetic degrees of freedom involved. The PI phase can be also obtained by means of a small ($`1÷2\%`$) Cr-doping: in this case Cr ions act as a negative pressure, thus enlarging the average cation-cation distance and inducing a Mott-insulator transition. As described in spalek , all transitions are due to the interplay between band formation and electron Coulomb correlation.
In 1978 Castellani, Natoli and Ranninger (CNR) attempted, in a series of papers cnr , a realistic description of the complex magnetic properties and phase diagram of V<sub>2</sub>O<sub>3</sub>. CNR focused on the peculiar magnetic structure observed in the AFI phase, that breaks the high-temperature trigonal symmetry and can not be explained in terms of a single-band Hubbard model. They realized that the introduction of the extra degree of freedom represented by t<sub>2g</sub>-vanadium orbitals was necessary to describe correctly the spin structure. The driving mechanism was the ordered pattern of these t<sub>2g</sub>-orbitals throughout the whole crystal, the so-called ”orbital ordering”. Twenty years later, the advent of modern third generation synchrotron radiation sources stimulated an interesting prediction fabrizio to look for the expected orbital ordering by means of resonant x-ray scattering (RXS). Soon after, its experimental detection was claimed to be found paolasini . The subsequent theoretical and experimental debate led to the abandonment of the CNR model, that was based on a spin $`S_V=1/2`$ on each vanadium ion and thus in contradiction with the experimental result of $`S_V=1`$ paolasini ; park , in favour of a picture where the two nearest V-ions are linked together in a stable molecule with spin $`S_M=2`$ mila12 ; dimatteo ; tanaka . This picture has been recently criticized by Elfimov et al. elfimov . Yet, the LDA and LDA+U schemes at the basis of this work are unable to describe the physics of V<sub>2</sub>O<sub>3</sub>, due to a fundamental weakness of local density approximations to mimic the behaviour of the correlation potential, as we shall see in Section 4. Thus, I believe that, especially in the version of Tanaka tanaka that includes the spin-orbit interaction, the ”molecular” model catches the main energy scales and correctly describes many relevant ground state properties: spin magnitude and direction, softening of the C<sub>44</sub> phonon mode, spin to orbital moment ratio and relative orbital anisotropies ($`a_{1g}`$ to $`e_g`$ ratios) in the three PM, PI, and AFI phases.
The paper is divided as follows: in Section 2 the implications of the RXS experiments paolasini ; paolasini1 about the nature of V<sub>2</sub>O<sub>3</sub> magnetic space group are introduced. Section 3 is devoted to a different interpretation of the ”non-reciprocal” linear dichroism experiment in Ref. goulon , in order to reconcile it with the magnetic space group as determined by RXS. Finally, in section 4, the breakdown of both local density approximation and LDA+U is illustrated, due to the non-local correlation potential generated by the molecular ground-state wave function. The molecular ground state is also at the basis of a simple explanation, based on symmetry arguments only, of the reason why the magnetic moment should lie in the glide-plane, in spite of the absence of a local symmetry element at vanadium sites.
## II Interpretation of resonant x-ray scattering experiments.
The RXS experiments paolasini ; paolasini1 were predicted to occur on the basis of CNR model, and had of course to be reinterpreted once the failure of such a model had been recognized. Such a task was undertaken by Tanaka tanaka and Lovesey and collaborators lovesey , who, independently, identified the RXS signal as a signature of the magnetic distribution and not of the orbital ordering of vanadium electrons. Their interpretations are based on the commonly accepted magnetic space group (MSG), $`P2/a+\widehat{T}\{E|t_0\}P2/a`$, derived from x-ray dernier and polarized neutron moon experiments, that contains explicitly the time-reversal symmetry $`\widehat{T}`$. Here $`t_0`$ is the body-centered translation and $`P2/a`$ the monoclinic group containing the identity $`\widehat{E}`$, the inversion $`\widehat{I}`$, the two-fold rotation about the monoclinic b<sub>m</sub>-axis $`\widehat{C}_{2b}`$ and the reflection $`\widehat{m}_b`$ in the plane perpendicular to this axis. Of course, the appropriate translation is associated to each of these operators, as detailed below:
$`\begin{array}{ccc}1)\widehat{E},\widehat{I}\hfill & \hfill & \mathrm{No}\mathrm{translation}\hfill \\ 2)\widehat{C}_{2b},\widehat{m}_b\hfill & \hfill & \frac{1}{2}(\stackrel{}{b}_m+\stackrel{}{c}_m)\hfill \\ 3)\widehat{T},\widehat{T}\widehat{I}\hfill & \hfill & \frac{1}{2}(\stackrel{}{a}_m+\stackrel{}{b}_m+\stackrel{}{c}_m)t_0\hfill \\ 4)\widehat{T}\widehat{C}_{2b},\widehat{T}\widehat{m}_b\hfill & \hfill & \frac{1}{2}\stackrel{}{a}_m.\hfill \end{array}`$ (5)
The origin of the axes has been chosen in the inversion center shown in Fig. 1. Referring to the figure labels, V<sub>1</sub> and V<sub>2</sub> are connected by the $`\widehat{C}_{2b}`$-rotation through their midpoint, V<sub>2</sub> and V<sub>3</sub> are related by inversion and V<sub>1</sub> and $`V_3`$ by the glide-plane operation $`\widehat{m}_b`$, within the plane of the figure. The unit cell contains eight V-ions: the other four are related to these by the time-reversal operation and the body-centered translation. Introducing the atomic scattering factor (ASF) as:
$$A\underset{n}{}\frac{\psi _g|\widehat{O}^{}|\psi _n\psi _n|\widehat{O}|\psi _g}{\mathrm{}\omega (E_nE_g)i\mathrm{\Gamma }_n}$$
(6)
the RXS amplitude can be written as: $`A_{j=1}^8e^{i\stackrel{}{Q}\stackrel{}{R}_j}f_j`$ where the sum is extended over the 8 inequivalent vanadium ions of the magnetic unit cell, at positions $`\stackrel{}{R}_j`$, whose ASF is $`f_j`$. Here $`\psi _g`$ and $`\psi _n`$ are the ground state and the photoexcited state, whose energies are $`E_g`$ and $`E_n`$, respectively; the operator $`\widehat{O}`$ represents the matter-radiation interaction, $`\mathrm{}\omega `$ is the photon energy, $`\mathrm{\Gamma }_n`$ the inverse lifetime of the excited states and $`\stackrel{}{Q}(h,k,l)`$ is the exchanged Bragg-vector. Reflections with $`h+k+l=`$odd and $`h=`$odd paolasini ; paolasini1 are those of interest. Taking into account of the tensorial character of $`f_j`$ dimatteo ; lovesey ; carra and the properties of the MSG $`P2/a+\widehat{T}\{E|t_0\}P2/a`$, it is possible to express all the ASF in terms of just one of them, through the appropriate symmetry operations nota5 . We get:
$$A=(e^{i\alpha }+\widehat{I}e^{i\alpha })(1\widehat{T}\widehat{m}_b)(1\widehat{T})f_1$$
(7)
where $`\alpha 2\pi (hu+kv+lw)`$, and $`(u,v,w)`$ are the fractional coordinates of vanadium ions in the unit of monoclinic axes. Equation (7) has two important consequences: first, it shows that all reflections of kind $`h+k+l=`$odd and $`h=`$odd have a magnetic origin, in such a way that the factor $`(1\widehat{T})`$ be non-zero. In second place, it indicates that they are sensitive to magnetic multipoles of higher order than the magnetic moment, because of the factor $`(1\widehat{T}\widehat{m}_b)`$. This latter observation stems from the fact that the magnetic moment lies on the glide plane $`\widehat{m}_b`$, thus being an eigenvector for $`\widehat{T}\widehat{m}_b`$, with eigenvalue +1. Such a picture clearly shows that the presence of the time-reversal symmetry plays a crucial role in the determination of the RXS origin.
Yet, little before the publication of tanaka and lovesey , Goulon et al. goulon had interpreted the results of their linear dichroism experiment in the monoclinic phase of V<sub>2</sub>O<sub>3</sub> as an evidence of the non-reciprocal gyrotropic effect, whose implications are in striking contradictions with the previous conclusions. In fact, a necessary condition to detect the non-reciprocal gyrotropy tensor is that the system be magnetoelectric, implying, in turn, that neither time-reversal nor inversion symmetry can separately belong to the MSG. If the linear dichroism experiment had correctly been interpreted, this would have implied a reduction of $`P2/a+\widehat{T}\{E|t_0\}P2/a`$ to a subgroup not containing $`\widehat{T}`$ and $`\widehat{I}`$, like $`(\widehat{E},\widehat{T}\widehat{I},\widehat{T}\widehat{C}_{2b},\widehat{m}_b)`$ or $`(\widehat{E},\widehat{T}\widehat{I},\widehat{C}_{2b},\widehat{T}\widehat{m}_b)`$.
Thus, one might wonder whether the RXS experiment could be explained with similar arguments as above even in the framework of the above magnetoelectric space groups, in order to be consistent with non-reciprocal effects. My answer is negative, and is motivated by the results of Refs. dimatteo ; dimatteo2 ; dimatteo3 ; yvesprb , that I summarize here.
For practice, we shall distinguish two physical mechanisms that can lead to a MSG reduction. The first is of ”charge” origin, i.e., due to any kind of orbital ordering (not necessarily that of CNR); the second concerns the presence of an out-of-plane component of the magnetic moment directed along the $`b_m`$ axis. The first mechanism has been definitively ruled out by a numerical simulation of the RXS experiment yvesprb , based on the finite difference method yves1 . In fact, in this case, due to the breakdown of the time-reversal symmetry in Eq. (7), for $`h+k+l=`$odd, $`h=`$odd reflections a signal would appear at $`4p`$-energies much bigger than the one at the pre K edge, whereas no such a signal is detected experimentally. The second possibility has been long debated in the literature moon ; word ; yethirai , and yet is still very controversial. The importance of an out-of-plane component of the magnetic moment is related to the fact that it could break the glide-plane symmetry. Three independent polarised neutron scattering experiments moon ; word ; yethirai were not able either to discard or accept this hypotesis. It is interesting to note here that in a crystal field approach, due to the absence of symmetry at vanadium sites, the magnetic moment would be allowed to have any direction. Nonetheless, this picture turns out to be too simple: the symmetry of the molecular wave-function is such as to force the moment in the glide plane (see Section 4). In any case, an out-of-plane component would again imply the presence of a signal at the (111)<sub>m</sub> reflection in the energy region of the conduction $`p`$-band, contrary to what experimentally detected. In fact, as described in Refs. dimatteo ; tanaka ; lovesey , the extinction rule that forbids this signal comes from the time-reversed glide-plane symmetry, due to the factor $`(1\widehat{T}\widehat{m}_b)`$. This latter is active only when the magnetic moment lies within the mirror plane. For this reason, we can conclude that the absence of any features at $`4p`$-energies of $`h+k+l=`$odd, $`h=`$odd reflections is an indication that the correct MSG of V<sub>2</sub>O<sub>3</sub> is $`P2/a+\widehat{T}\{E|t_0\}P2/a`$.
Finally, I would like to underline a relatively minor, but still important difference between tanaka and lovesey . In Tanaka’s paper the RXS signal is a consequence of the dipole-quadrupole (E1-E2) interference, thus sensitive to magnetic, parity-odd physical quantities (like a toroidal moment or a magnetic quadrupole - which of them is left unspecified). On the contrary Lovesey and collaborators were able to fit the experimental results of Refs. paolasini ; paolasini1 in the different case of purely quadrupolar reflections (E2-E2), where the extracted information concerns the magnetic octupole of the system. These two views seem conflicting each other, but probably ”in medio stat veritas”. In fact, a third more recent result on this subject yvesprb , based on ab-initio calculations, shows that both channels can contribute to the global signal: by means of a relativistic extension of the multiple scattering theory, implemented in the FDMNES package yves1 , the authors of Ref. yvesprb have been able to find a fair agreement in energy and azimuthal scans with the experimental results paolasini ; paolasini1 . Both E1-E2 and E2-E2 channels are present in the global intensity and their relative contribution depends basically on the energy and on the azimuthal angle at which the reflection is measured. Yet, the single-particle character of these calculations still leaves open the door to future improvements in the description of the pre K edge RXS.
## III Interpretation of the linear dichroism experiment.
As sketched above and already discussed in the literature dimatteo3 , the interpretations of RXS tanaka ; lovesey and non-reciprocal dichroism goulon in their original forms require incompatible MSG. Nonetheless, I believe that there is a way to explain both experiments within a unified theoretical framework, which consists in a different interpretation of the linear dichroism experiment, as suggested in dimatteo3 .
This alternative view is based on the possibility that the measured linear dichroism goulon is not a gyrotropic signal, but is related to the electric quadrupole, and thus of pure dipole-dipole (E1-E1) origin. The difference between the physical operators associated with these two kinds of dichroism was first derived in Ref. dimatoli , where their formal expression was given. In a reference frame where $`z`$ is the direction of propagation of the x-ray beams, and $`x`$ and $`y`$, respectively, those of the associated polarizations, the pure E1-E1 dichroism is described by the electric quadrupole operator $`L_x^2L_y^2`$, while the non-reciprocal (E1-E2) signal is determined by the expectation value of $`(L_x^2L_y^2)\mathrm{\Omega }_zc.c.`$. Here $`\stackrel{}{L}`$ is the angular momentum operator and $`\stackrel{}{\mathrm{\Omega }}`$ is the toroidal moment radescu ; carraprb . The ab-initio simulations dimatteo3 , based on either the finite difference method or the multiple scattering theory, clearly demonstrated the presence of a dichroic signal of E1-E1 origin, the size of which is some percent of the absorption signal. This calculated intensity has the same shape and order of magnitude as the one measured by Goulon et al. goulon . Indeed, apart from the ab-initio calculation results, a dichroic signal of E1-E1 origin and of this order of magnitude had to be expected, as the low-temperature distortion of the corundum hexagonal planes along orthogonal directions, where the dichroism is measured, can be up to 4$`\%`$. These simple considerations explain the findings of Ref. dimatteo3 also at a qualitative level.
The reason why Goulon et al. goulon have interpreted the dichroic signal as ”non-reciprocal” is mainly due to the fact that they found an inverted signal when their experiment was performed with a reversed applied magnetic field. This seems to imply that the measured dichroic signal has a time-reversal odd character (like that of a non-reciprocal tensor), contrary to the case of the electric quadrupole $`L_x^2L_y^2`$. In the alternative explanation given in Ref. dimatteo3 , the change of sign measured in goulon , when the magnetic field is reversed, is interpreted as due to the same electric quadrupole, but in a different twin chrystallographic domain. In fact, V<sub>2</sub>O<sub>3</sub> has the tendency to form big monodomains in its monoclinic phase, in an unpredictable way yethirai , and there are some specific angles at which the measured intensity in one of the other two monoclinic domains appears exactly reversed. I shall not go into details here, referring to the original paper dimatteo3 for technical aspects. But I want to underline the following two points. First of all, this interpretation is coherent with all past failures to detect a magnetoelectric effect in V<sub>2</sub>O<sub>3</sub> astrov ; jansen , and does not require an ad-hoc mechanism to justify a magnetoelectric space group in the ground state of V<sub>2</sub>O<sub>3</sub>. In second place, such an interpretation can be checked by some very simple experiments, that I shall describe below. I indeed believe that this is the only way to really settle the question.
There are basically three ways to get some insight into the problem. The first, and probably best, is to repeat the experiment performed in Ref. goulon , but paying particular care to check the monoclinic twin after each transition to the AFI phase, for example by means of an x-ray-scattering equipment: this would give a full geometrical control of the system. The role of the magnetoelectric annealing goulon ; dimatteo3 , in this sense, should also be analyzed. Another possible experimental test to confirm, or reject, the interpretation given in Ref. dimatteo3 , is to measure the linear dichroism in the AFI phase in the same geometrical conditions as in goulon , ie, with the x-ray direction parallel to the corundum axis, but without any magnetoelectric annealing, that would be irrelevant to the purpose: in this way the E1-E1 origin of the signal can be proved or disproved. Moreover, performing an azimuthal scan of the dichroism around the corundum axis, it is possible to compare directly the signal to the simulations performed in dimatteo3 . Finally, another way to unravel the contradictions is to study the dependence of twin formations on the applied electric and magnetic fields by means of an x-ray-diffraction equipment, in order to determine if their presence can influence the phase transition from PI to AFI and establish whether the ground-state or some excited state properties are investigated by this procedure.
In the light of all what stated above, and given the implications of the non-reciprocal effect, if confirmed, for the physics of V<sub>2</sub>O<sub>3</sub>, its unambigous experimental determination is in my opinion of the utmost importance.
## IV The ground-state in the AFI phase: correlated magnetic moment and non-local exchange.
In this section I shall focus on two apparently unrelated facts that are nonetheless founded on a common background: i), the reasons why LDA (and related LDA+U) are completely unreliable in the description of the ground-state wave function of V<sub>2</sub>O<sub>3</sub>; and ii), the correct symmetry analysis to explain why the magnetic moments lie in the glide-plane. The first task is achieved through a critical review of all Refs. dimatteo ; tanaka ; elfimov ; ezhov ; allen ; the second is attained by simply analyzing the local symmetries as determined by the molecular nature of the ground-state wave function, whose point group does not coincide with that at the vanadium site. Both effects are a consequence of the non-local correlations in the wave-function (whose specific form will be precised below).
Before starting the theoretical analysis of the ground-state wave function, I prefer to introduce the experimental results at the basis of the molecular model, that are probably much more indicative allen ; good . In 1976 J.W. Allen allen wrote that ”The magnetic and optical properties of all the phases of V<sub>2</sub>O<sub>3</sub> show a loss of V<sup>3+</sup>-ion identity”. This statement was based on the optical experimental study of (V<sub>2</sub>O<sub>3</sub>)<sub>x</sub>(Al<sub>2</sub>O<sub>3</sub>)<sub>1-x</sub> mixtures ($`0.01x0.08`$). The paper continues with: ”Optical studies \[…\] show that V<sup>3+</sup>-V<sup>3+</sup> interactions between second and further near neighbors are too weak to destroy the V<sup>3+</sup>-ion identity. This result suggests the possibility that the basic localized-electrons unit in the antiferromagnetic insulating phase may be the nearest neighbor pairs, with the electrons delocalized within the pair”. I think that it can be instructive to compare this statement to the one, of opposite tenor, based on LDA(+U) calculations, from Elfimov et al. elfimov : ”\[…\] we show that there are other hopping integrals which are equally important for the band shape as the integral for hopping between the partners of the pair”. Why do LDA (+U) results differ so much from optical data and from all theoretical descriptions mila12 ; dimatteo ; tanaka ? The answer to this question is already contained, implicitly, in the discussion of section V in Ref. dimatteo , to which we refer for a quantitative treatment: here we just try to delineate qualitatively the physics behind it, trying to add new elements for a better comprehension. As a result, we shall be able to define clearly the reasons behind the failure of local density approximations to describe the physics of V<sub>2</sub>O<sub>3</sub>.
As illustrated in Ref. dimatteo , a good starting point to evaluate the energetics of the vertical molecule is through an orbitally-degenerate Hubbard Hamiltonian. We shall discuss a posteriori why it is possible to consider simply the vertical pair, that is the contested point in Ref. elfimov , and why our results do not depend on the approximations that are impicit in the Hubbard model. The parameters that play the most relevant role are: the Coulomb on-site repulsions $`U_1`$ and $`U_2`$, between electrons on the same orbital and on different orbitals, respectively; the Hund’s coupling constant $`J_H`$ and the trigonal field splitting $`\mathrm{\Delta }_t`$. The relation $`U_1=U_2+2J_H`$ holds with very good approximation. In V<sub>2</sub>O<sub>3</sub> every V<sup>3+</sup>-ion is octahedrally surrounded by a cage of oxygens and, thus, the two valence electrons are forced to a three-fold $`t_{2g}`$ configuration. The effect of the trigonal field splitting is to lift the three-fold $`t_{2g}`$-degeneracy and to push higher in energy the $`a_{1g}`$ state, leaving as a ground state the degenerate doublet $`e_{u,v}`$. The explicit expression of $`a_{1g}`$, $`e_{u,v}`$ in terms of 3$`d`$-orbital wave-functions can be found, for example, in tanaka . The main contribution to the kinetic energy in the molecule comes from the terms describing $`a_{1g}`$-$`a_{1g}`$ hopping, $`\rho 0.82`$ eV, and $`e_{u,v}`$-$`e_{u,v}`$ hopping, $`\mu 0.2`$ eV mattheiss (notice that they have opposite signs). In the high trigonal-field-splitting limit the ground state is given, thus, by $`|\mathrm{\Psi }_1|e_ue_v;e_ue_v`$ (the two groups of $`e_{u,v}`$ labels refer to the electron occupancy of one or the other ion in the molecule, e.g., in Fig. 1, $`V_1`$ or $`V_2`$). Even for the trigonal splitting $`\mathrm{\Delta }_t0.4`$ eV, deduced from Ref. ezhov , the ground-state wave function is characterized by a high percentage of $`e_g`$-doublet composition. Notice that similar results were already obtained in 1978 by CNR, in their second paper cnr , through an unrestricted Hartree-Fock + Hubbard U calculation, which is very akin to the modern LDA+U. A complete description of the molecular behaviour within a big range of variation of parameters can be found in Ref. dimatteo . Here we focus on the values given in Ref. ezhov , even if $`J_H`$ is clearly overestimated dimatteo ; jcond in the light of the atomic calculations of Tanabe and Sugano ($`J_H=0.79`$ eV) tanabe : $`\mathrm{\Delta }_t0.4`$ eV; $`J_H0.93`$ eV; $`U_22.8`$ eV. Consider now the FM correlated molecular wave function nota2 , first derived by Mila et al. mila12 :
$$|\mathrm{\Psi }_0=\frac{1}{\sqrt{2}}(|a_{1g}e_u;e_ve_u+|e_ve_u;a_{1g}e_u)|S_M=2$$
(8)
Its energy, in our description, is given by: $`E_0=\mathrm{\Delta }_t(\rho \mu )^2/(U_2J_H)0.16`$ eV. Notice that the reference energy $`E_1=0`$ has been assigned to the ground state of the LDA+U procedure $`\mathrm{\Psi }_1=|e_ue_v;e_ue_v`$: here the absence of $`a_{1g}`$ electrons gives no trigonal-field contribution and Pauli principle prevents any hopping in the ferromagnetic vertical molecule. One might wonder why is $`E_0`$ so lower, having used the same parameters as obtained by the local density calculations. To this aim, consider the wave function $`|\mathrm{\Psi }_2|e_ue_v;a_{1g}e_u`$. The associated energy is given by: $`E_2=\mathrm{\Delta }_t(\rho ^2+\mu ^2)/(U_2J_H)+0.02`$ eV, higher than both $`E_0`$ and $`E_1`$. There are $`180`$ meV of difference nota1 between the ”mean-field” state $`|\mathrm{\Psi }_2`$ and the ”correlated” state $`|\mathrm{\Psi }_0`$, coming from the hopping interference term $`2\mu \rho /(U_2J_H)`$. The origin of this energy is due to the correlated structure of $`|\mathrm{\Psi }_0`$ that makes possible for an $`e_v`$-electron on V<sub>1</sub> to hop on V<sub>2</sub> and simultaneously for an $`a_{1g}`$ electron on V<sub>2</sub> to hop on V<sub>1</sub> (and vice-versa), without changing state. This form of superexchange interaction is basically a non-local electronic correlation and, as such, it cannot be taken into account in the framework of local density calculations. These simple considerations show why LDA+U results fail to catch the ground-state wave function: as non-local correlations are neglected, the interference energy $`2\mu \rho /(U_2J_H)`$ is completely missed ($``$180/2=90 meV per ion !) and the purely $`e_{u,v}`$ state $`|\mathrm{\Psi }_1`$ is found as ground state.
In this picture, if we even consider that there are three in-plane second nearest neighbors and introduce further nearest-neighbor hoppings, as done in Ref. elfimov , still the non-local correlation energy of the vertical molecule represents a relevant (if not the main) contribution. Notice that the corundum geometry is such that further-nearest-neighbors hopping integrals cannot give rise to any other correlated hopping of relevant strenght: for this reason all inter-molecular interactions can be treated in mean field (or local density) approximation. The energy gain due to the correlated molecule is so big that it can be hardly conceived as an artefact of the Hubbard Hamiltonian, or that the corresponding state dissolves when further nearest neighbor hopping-integrals are taken into account, as Ref. elfimov seems to suggest, provided, of course, all local and non-local forms of energy are correctly considered. Moreover, it is stable against rather large variations of $`\mathrm{\Delta }_t`$, $`U_2`$ and $`J_H`$ around the chosen values dimatteo . When this correlation energy is simply neglected, as in local density approximations, the ground-state wave function is mainly determined by the trigonal splitting $`\mathrm{\Delta }_t`$, that pushes higher the one-particle energy level of the $`a_{1g}`$-electrons, thus advantaging $`e_{u,v}`$. In my opinion, this is a signature of the failure of LDA+U to describe the ground state of V<sub>2</sub>O<sub>3</sub>. Indeed, the results of LDA+U calculations ezhov are lacking for many aspects: the spin exchanges along the different bonds are strongly underestimated when compared to the experimental results word , as recognized by the authors themselves. The orbital occupancy of mainly $`e_ue_v`$ kind does not reproduce the findings of Park et al. park interpreted with a substantial $`a_{1g}`$-occupancy about 20-25$`\%`$. Notice that this is in keeping with the experimental evidence from Rubinstein rubin that found no detectable anisotropy in the Knight shift of the NMR spectrum, no observable quadrupole splitting in the NMR spectrum, and zero anysotropy in single-crystal susceptibility measurements for various cristalline orientations with respect to the magnetic field. All these findings apparently point to a negligeable trigonal field splitting, ie, an almost cubic crystal field. Only the molecular picture, through the interference effect, is able to reconcile these latter experiments with the sizeable value of the trigonal distortion, because of the average orbital occupancy $`a_{1g}=25\%`$, $`e_{u,v}=37.5\%`$, close to the statistical ratio ($`33\%`$) of the cubic field. In the light of the previous discussion, the main statement of Elfimov et al. elfimov looses its strenght: these authors performed a tight-binding fitting of their LDA and LDA+U band structure, and concluded that a proper description requires second, third and forth neighbor hoppings. Yet, the evident drawback of LDA+U correlated description makes such a conclusion very weak. Moreover, they focused on the $`a_{1g}`$ band only, whereas the three bands should strongly interfere.
My conclusion in this respect is that V<sub>2</sub>O<sub>3</sub> is a very peculiar strongly correlated electron system, where ordinary ab-initio codes may loose their original strenght elfimov ; anisimov ; held . The case of LDA+U is emblematical: this approach was initially conceived to describe the insulating magnetic materials whose energy gap was not determined by the Stoner parameter, but by the Coulomb repulsion $`U`$ anisimov . Even though it works correctly in these cases anisimov , it is nonetheless based on a local exchange and correlation potential, and, as such, it cannot treat properly a system like V<sub>2</sub>O<sub>3</sub>, where non-local correlations amount to 90 meV per ion. The same arguments apply to the recent LDA+DMFT held , as clear from the following statement, from Keller et al. held1 : ”\[…\] this small trigonal splitting strongly determines the orbital ground state of the V-ion obtained from LDA+DMFT calculations \[…\]”. This method, in spite of the progress coming from the DMFT, can handle just the local dynamics of the system, while non-local correlations are still neglected. This is the reason why the orbital nature of the ground state seems to be determined by the trigonal field splitting.
It may be true that the physical idea of the ”weakly” interacting gas of vertical molecules dimatteo could be oversimplified and that a different picture for the band structure may be required, as in the original idea of Elfimov et al. elfimov : yet, to this aim, an extension of the correlation potential V<sub>c</sub> in a non-local direction is, in my opinion, essential to describe the physics of V<sub>2</sub>O<sub>3</sub>. All past experimental results, like Allen’s, must be explained by any theory, and not simply dismissed, as seems to be the case of LDA-based approaches. In order to get reliable results, it is necessary to go back to the original form of Kohn-Sham equations kohn and find out a good approximation for the correlation potential $`V_c`$: some hints in this sense could be taken from the form of the non-local molecular wave function (8). In this way new informations about V<sub>2</sub>O<sub>3</sub> band structure can be extracted, that do not neglect the ”correlated hopping” energy, and, probably, can also explain, among the others, Allen’s results allen .
All the previous analysis was based on the non-local nature of superexchange correlations, only, and did not take into account the spin-orbit interaction, that, nonetheless, was fundamental to explain the results of RXS tanaka ; yvesprb . Notice that, even if very important to describe the RXS experiments and to remove the orbital degeneracy, this interaction is about one order of magnitude lower than the correlated hopping ($`25`$ meV) and, thus, it can be introduced adiabatically in the ground-state manifold of $`|\mathrm{\Psi }_0`$. In this way, its inclusion, though it removes the orbital degeneracy of $`|\mathrm{\Psi }_0`$ tanaka , leaves unaltered all the previous conclusions. The ground state is modified in:
$$|\mathrm{\Psi }_0^{}=\frac{1}{\sqrt{2}}(|a_{1g}e^+;e^+e^{}+|e^+e^{};a_{1g}e^+)|S_{Mz}=\pm 2$$
(9)
where $`e^+\frac{1}{\sqrt{2}}(e_u+ie_v)`$ and $`e^{}\frac{1}{\sqrt{2}}(e_uie_v)`$. Here the spin part of the ground state has been written explicitly. Such a state is characterized by a global spin $`S_M=2`$, with $`S_{Mz}=\pm 2`$. As discussed previously, a long debate arose in the literature about a possible component of the spin of V<sub>2</sub>O<sub>3</sub> out of the glide plane, especially for its consequences on the magnetic space group. From a theoretical point of view, on the basis of usual crystal field models, such an out-of-plane component could even be expected: in fact, vanadium ions, in their AFI phase, are characterized by the absence of any point symmetry, thus allowing the local magnetic moment to follow, in principle, any direction. Yet, the molecular nature of the ground state plays a fundamental role also in the description of the correct symmetry. In fact, as any physical variable is determined through its expectation value on the ground-state wave function, and as the electronic wave function is ”molecular”, then the magnetic moment must be invariant with respect to all the symmetry operations of the molecular point group and not of the ion point group. In other words, we identify the molecular spin $`S_M=S_{V_1}+S_{V2}=2`$ as the physical variable detected by neutron measurements, even if it is customary to speak about the magnetic moment per ion, $`S_V=1`$, in an average sense. This interpretation parallels and extends to the four-electron case of the molecule the S<sub>V</sub>=1 atomic picture: here, of the whole space spanned by the two $`S=1/2`$-electrons on the vanadium ion, the S<sub>V</sub>=0 subspace is projected out, and only the spin-1 is considered for the ground-state properties. Similarly, in the molecule, only the $`S_M=2`$ subspace has a physical reality, and all the other spin-states ($`S_M=0,1`$) are projected out.
Based on these premises, we are now ready to repeat the previous symmetry considerations from a different viewpoint. In fact, the electronic ground-state of the molecule now has a local symmetry element, contrary to the case of the single ion: this symmetry element is the $`\widehat{C}_{2b}`$-axis shown in Fig. 1, orthogonal to the molecular axis, and running through the midpoint of the molecule. It follows that the molecular magnetic moment must lie either along the $`\widehat{C}_2`$-axis, or perpendicular to it (this latter case arises when the time-reversal operation is associated to the $`\widehat{C}_2`$-axis). No ”mixed” configurations are possible. Neutron experiments allow us to choose: the magnetic moment lies orthogonal to the $`\widehat{C}_{2b}`$-axis and, thus, the correct local symmetry for the molecule is $`\widehat{T}\widehat{C}_2`$. The same conclusion is reached by considering Tanaka’s wave function (9). In fact, under $`\widehat{C}_2`$ symmetry, $`a_{1g}a_{1g}`$, $`e^+e^+`$ and $`e^{}e^{}`$ (see tanaka , pag. 3), and time-reversal symmetry changes sign to the imaginary unit. Thus, $`\mathrm{\Psi }_0^{}|L_M|\mathrm{\Psi }_0^{}=\mathrm{\Psi }_0^{}|(\widehat{T}\widehat{C}_2)^{}L_M\widehat{T}\widehat{C}_2|\mathrm{\Psi }_0^{}`$ and the molecular angular moment is invariant under $`\widehat{T}\widehat{C}_2`$ symmetry, so that it must lie on the glide plane and force the spin moment in the same direction. Notice, in this respect, that when Tanaka speaks of a possible out-of-plane magnetic component he refers to the single-ion moment: when combined throughout the unit cell, all the ions belonging to the same vertical molecule have opposite out-of-plane moments and thus the global molecular spin is always in the glide-plane, as the symmetry constraint requires.
In conclusion, I have shown that the non-local, molecular, wave function can explain different physical phenomena, apparently unrelated and inexplicable in the framework of an average, ionic picture.
I would like to thank Stephen W. Lovesey who strongly supported and stimulated the present work, and Calogero R. Natoli who shared with me his knowledge on V<sub>2</sub>O<sub>3</sub>.
|
warning/0506/cond-mat0506367.html
|
ar5iv
|
text
|
# Liquid Polymorphism and Double Criticality in a Lattice Gas Model
## I Introduction
The recently acknowledged possibility of the existence of single component systems which display coexistence between two different liquid phases Po92 Gl99 has opened many interesting questions regarding the link between the interaction potential and the presence of double criticality .
Network-forming fluids are primary candidates for liquid-liquid transitions. They exhibit directional, intermolecular attractions that lead to the formation of bonds between molecules. As a result, locally structured regions have lower density than unbonded regions. These structures at the fluid phase are transient and local but become permanent at lower temperatures.
The case of water is probably the network-forming liquid most intensively studied, due to its ubiquity in nature. An unstable liquid-liquid transition ending in a critical point was initially proposed to explain the anomalous behavior of network-forming liquids such as water Po92 De98 -An72 . Closely associated with the possibility of a liquid-liquid phase transition in supercooled water is the phenomenon of polyamorphism occurring below the glass transition temperature, $`T_g`$. Polyamorphism refers to the occurrence of distinct amorphous solid forms. There is by now a general consensus that water displays a transition between two different amorphous states in the supercooled region of its phase diagram Mi98 . Experiments carried out in water at $`T130K`$ show an abrupt change of volume as a function of pressure which indicates the existence of a first-order transition Mi84 . The two amorphous phases of water might be related to two different liquid phases at higher temperatures. At coexistence, these two liquid phases determine a first-order transition line ending at a critical point. Simulations and experiments predict that the liquid-liquid transition is in an experimentally inaccessible region of the phase-diagram Po92 Ta96 Mi00 Fr03 .
Notwithstanding of its confirmation for metastable water, the coexistence of two liquid phases was uncovered as a possibility for a few both network-forming and non-bonding liquids. Computer simulations for realistic models for carbon Gl99 , phosphorus Mo01 , germanium Du02 , silica ($`SiO_2`$) An97 Sa01 La00 Hu04 and silicon An96 Alex04 suggest the existence of a first-order transition between two liquid phases. Recent experiments for phosphorus Ka00 Mo03 or phosphate compounds Ku04 , and for carbon To97 confirm these predictions.
Substances that are structurally similar to water, such as Si, Ge, $`GeO_2`$ and silica, can exhibit not only the liquid-liquid transition but also the other thermodynamic and dynamic anomalies present in water. Particular attention has been given to silica because its technological applications. Many intriguing properties of liquid silica and water occur at low temperatures. Examples include negative thermal expansion coefficients and polymorphism for both liquids. Compression experiments on silica show a non-trivial change in the microscopic structure Gr84 He86 . This behavior, reproduced by simulations Sa93 -Ji93 , was interpreted as an indication of a first-order transition. Inspired by these results, new simulations for different models for silica were performed An97 Sa01 La00 , giving support to the hypothesis of a first-order transition in silica. However, experiments must sill confirm a discontinuous transition between these amorphous phases Gr84 Wi88 .
In resume, a full understanding of the effects of the number, spatial orientation, and strength of bonds on the global fluid-phase behavior of network fluids is still lacking.
In order to gain some understanding, a number of simple models have been proposed. Since the work of Bernal Be33 , the water anomalies have been described in terms of the the presence of an extensive hydrogen bond network which persists in the fluid phase Er02 . In the case of lattice models, the main strategy has been to associate the hydrogen bond disorder with bond Fr03 Sa96 or site Sa93b Ro96b Potts states. In the former case coexistence between two liquid phases may follow from the presence of an order-disorder transition and a density anomaly is introduced *ad hoc* by the addition to the free energy of a volume term proportional to a Potts order parameter. In the second case, it may arise from the competition between occupational and Potts variables introduced through a dependency of bond strength on local density states.
A different approach is to represent hydrogen bonds through ice variablesHu83 attard Na91 Gu00 , so successful in the description of ice Li67 entropy, for dense systems. In this case, an order-disorder transition is absent. Recently a description based also on ice variables but which allows for a low density ordered structure He05 was proposed. Competition between the filling up of the lattice and the formation of an open four-bonded orientational structure is naturally introduced in terms of the ice bonding variables and no *ad hoc* introduction of density or bond strength variations is needed. Our approach bares some resemblance to that of some continuous modelsSi98 Tr99 Tr02 , which, however, lack entropy related to hydrogen distribution on bonds. Also, the reduction of phase-space imposed by the lattice allows construction of the full phase diagram from simulations, not always possible for continuous models Si98 . In our previous publication, we have shown that this model is able to exhibit for a convenient set of parameters both density anomalies and the two liquid phases. Here we explore the phase space for various values of the orientational energy term. We show that by varying the relative strength of the orientational part, we can go at a fixed temperature from two coexisting liquid phases as observed in amorphous water to a smooth transition between two amorphous structures as might be the case of silica.
The remainder of the paper goes as follows. In sec. II the model is revised, for clarity. In sec. III mean-field results are given. Sec. IV has the simulations results. Conclusions end this session.
## II The Model
Consider a lattice gas on a triangular lattice with sites which may be full or empty. Besides the occupational variables, $`\sigma _i`$, associated to each particle $`i`$, there are six other variables, $`\tau _i^{ij}`$, pointing to neighboring sites $`j`$: four are the usual ice bonding arms, two donor, with $`\tau _i^{ij}=1`$, and two acceptor, with $`\tau _i^{ij}=1`$, while two additional opposite arms are taken as inert (non-bonding), $`\tau _i^{ij}=0`$, as illustrated in Fig. 1. Therefore each occupied site is allowed to be in one of eighteen possible states.
Two kinds of interactions are considered: isotropic “van der Waals” and orientational hydrogen bonding. An energy $`v`$ is attributed to each pair of occupied neighboring sites that form a hydrogen bond, while non-bonding pairs have an energy, $`v+2u`$ (for $`u>0`$), which makes $`2u`$ the energy of a hydrogen bond. The overall model energy is given by
$$E=\underset{(i,j)}{}\{(v+2u)\sigma _i\sigma _j+u\sigma _i\sigma _j\tau _i^{ij}\tau _j^{ji}(1\tau _i^{ij}\tau _j^{ji})]\}$$
(1)
where $`\sigma _i=0,1`$ are occupation variables and $`\tau _i^{ij}=0,\pm 1`$ represent the arm states described above. Note that each particle may have six neighbors, but the number of bonds per molecule is limited to four. For $`u/v>1/2`$, the ¨van der Waals¨ forces become repulsive. As a result, each molecule attracts four neighbors, if properly oriented, and repeals the other two. An interpretation for this ¨repulsion¨ would be that the presence of the two extra neighbors distorts the electronic orbitals, thus weakening the hydrogen bonds.
Inspection of the model properties allows the prediction of two ordered states, as shown in Fig. 1. For low chemical potential, the soft core repulsion becomes dominant, $`\rho =0.75`$, and energy ¨volume¨ density is given by $`e=E/V=3v/2`$, where V is the number of lattice sites. If the chemical potential is high, $`\rho =1`$, and energy density $`e=3v+2u`$. At zero temperature, the low density liquid (LDL) coexists with the high density liquid (HDL) at chemical potential $`\mu /v=6+8u/v`$, obtained by equating the grand potential density (or pressure) associated with each one of these phases. Similarly the coexistence pressure at zero temperature is given by $`p/v=3+6u/v`$. Besides these two liquid states, a gas phase is also found and it coexists with the low density liquid at chemical potential $`\mu /v=2`$ and pressure $`p=0`$. The condition for the presence of the two liquid phases is therefore $`u/v>0.5`$, i.e., that the ”Van der Waals” interactions are repulsive.
The two ordered structures could be qualitatively associated to different ice phases. Under pressure, hydrogen bonds reorganize in more dense phases eisenberg .
Our model may be interpreted in terms of some sort of average soft-core potential for large hydrogen bond energies. The low density phase implies average interparticle distance $`\overline{d_{LD}}=\rho _{LD}^{1/2}=2/\sqrt{3}`$, whereas for the high density phase we have $`\overline{d_{HD}}=\rho _{HD}^{1/2}=1`$. The corresponding energies per pair of particles is $`v`$ and $`v+2u/3`$. The hard core is offered by the lattice. For $`u/v>3/2`$, the shoulder becomes repulsive, making the potential soft-core. Figure 2 illustrates, in a schematic way, the average pair energy $`e_p/v`$ dependence on distance between particles, for $`u/v=0.5,1,2`$. Notice that for $`u/v=2`$ the potential at high density phases becomes repulsive. For $`u/v=1,0.5`$, the shoulder is attractive.
## III The Mean-Field Analysis
Designing a correct mean-field version of the model is not an obvious task, due to the special orientational ice-like interactions. We have therefore looked at model properties on the Bethe lattice, for which exact relations may be given. Each site of the usual Cayley tree (of coordination six) is replaced by a hexagon and ($`\eta ,\tau `$) variables are attributed to its vertices (see Fig. 3), with $`\eta `$ and $`\tau `$ defined as in the previous section. This representation is inspired on the Bethe solution for a generalization of the square water model Gu00 ; square2 presented by Izmailian et al. izmailian . For an occupied site-hexagon, we will have two vertices with $`(\eta ,\tau )=(1,1)`$, two with $`(\eta ,\tau )=(1,1)`$ and another two with $`(\eta ,\tau )=(1,0)`$. For an empty site-hexagon the six vertices have $`(\eta ,\tau )=(0,0)`$.
The partition function in the grand canonical ensemble is given by:
$`\mathrm{\Xi }(T,\mu )=18e^{\beta \mu }g_N^2(1,+)g_N^2(1,)g_N^2(1,0)+g_N^6(0,0),`$ (2)
where $`\mu `$ is the chemical potential and $`g_N(1,+)`$ is the partial partition function (generation $`N`$) for a central site with $`(\eta ,\mu )=(1,1)`$. Similarly for $`g_N(1,)`$, $`g_N(1,0)`$ and $`g_N(0,0)`$.
On each lattice line we have two $`(\eta ,\tau )`$ pairs of variables connecting two consecutive generations of the lattice. The Boltzmann weights for each lattice line are $`\omega =1`$, if at least one of the connecting sites is empty, and, if both sites are occupied, either $`\omega =e^{\beta v}`$, if arm variables are of opposite sign, and $`\omega =e^{\beta (v2u)}`$, otherwise.
The density of water molecules at the central site of the tree is given by
$$\rho (T,\mu )=\frac{18z}{18z+x_N^2y_N^2r_N^2},$$
(3)
where $`z=e^{\beta \mu }`$ is the activity and
$$x_N\frac{g_N(0,0)}{g_N(1,+)},y_N\frac{g_N(0,0)}{g_N(1,)},r_N=\frac{g_N(0,0)}{g_N(1,0)}.$$
Eqn.3 gives us a relation between the water density $`\rho `$ and the activity $`z`$: $`6z=\gamma (x_Ny_Nr_N)^2`$, with $`\gamma =\rho /(3(1\rho ))`$. Using this result the recursion relations at the fixed point may be written as
$`x={\displaystyle \frac{\gamma (x+y+r)+1}{\gamma e^{\beta (v2u)}(x+e^{2\beta u}y+r)+1}},`$
$$y=\frac{\gamma (x+y+r)+1}{\gamma e^{\beta (v2u)}(e^{2\beta u}x+y+r)+1},$$
(4)
$`r={\displaystyle \frac{\gamma (x+y+r)+1}{\gamma e^{\beta (v2u)}(x+y+r)+1}}.`$
Under the the physical conditions $`x0`$, $`y0`$ and $`r0`$, since $`x`$, $`y`$ and $`r`$ are ratios between partition functions, we must have $`x=y`$. Eqs.4 reduce to two equations which are solved numerically for specific values of the model parameters $`u/v`$. Chemical potential vs density isotherms are then obtained from Eq.3, as shown in Figures 4 and 5.
The first striking result is that temperature destabilizes the liquid-liquid transition. Van der Waals loops are present only for the gas-liquid, but not for the liquid-liquid transition. However, for larger bond energies $`u/v>0.5`$, the liquid phase is of low density $`(\rho =0.75)`$, whereas for smaller bond energies, the liquid phase is of high density $`(\rho =1)`$. A typical isotherm for the first case, for which the gas-liquid transition is to a phase of low density, is shown in Fig 4, for $`u/v=1`$ . It is to be noted that the chemical potential presents a very abrupt rise, just above the gas-liquid transition, as if signaling a quasi liquid-liquid transition. Fig 5 shows a typical low temperature isotherm for the second case, for $`u/v=0.25`$, for which no low density phase is found.
Inspite of the absence of a liquid-liquid transition, present in the results of simulations (see He05 and next section), both the LD and the HD phases appear, for different strengths of the hydrogen bonds. Also, the abrupt rise of the chemical potential in Fig.4 indicates that this mean-field treatment comes near to but misses the LD-HD transition. Similar problems were noted for bonding ice-type models izmailian square2 . Cyclic ordering of edges around a vertex, in the lattice ordered structure, cannot be represented on the tree, which may lead to analytical free-energies for the same parameters for which exact izmailian or simulation He05 studies predict phase transitions.
## IV The Monte Carlo Results
The model properties for finite temperatures were obtained through Monte Carlo simulations in the grand-canonical ensemble using the Metropolis algorithm. Particle insertion and exclusion were tested with transition probabilities given by $`w(insertion)=exp(\mathrm{\Delta }\varphi )`$ and $`w(exclusion)=1`$ if $`\mathrm{\Delta }\varphi >0`$ or $`w(insertion)=1`$ and $`w(exclusion)=exp(+\mathrm{\Delta }\varphi )`$ if $`\mathrm{\Delta }\varphi <0`$ with $`\mathrm{\Delta }\varphi \mathrm{exp}\{\beta (e_{particle}\mu )\mathrm{ln}(18)\}`$ where $`e_{particle}`$ is the particle energy. Since the empty and full sites are visited randomly, the factor 18 is required in order to guarantee detailed balance.
A detailed study of the model properties was presented in a previous publication for a particular value of the energy parameter, $`u/v=1`$ He05 . Here, we present the evolution of the phase diagram under variation of this energy parameter which represents relative energy of bond strength. Note that the relevant parameter range is $`u/v>1/2`$. For $`u<v/2`$, the LDL disappears. Our study is for L=10 and runs were of the order of $`10^6`$ Monte Carlo steps.
Chemical potential vs. density isotherms for $`u/v=1`$, and a large set of temperatures are illustrated in Fig.6. The figure shows the first-order phase transitions between the gas and the low density liquid phase and between the low density liquid and high density liquid phases. The coexistence lines are estimated from the mid-gaps in the chemical potentials. Fig. 7 illustrates the chemical potential vs. temperature phase-diagram for $`u/v=0.55,0.6,0.75,1`$. The gas-LDL coexistence lines for the different $`u/v`$ seem to collapse into a single line that at zero temperature gives $`\mu /v=2`$. The different LDL-HDL lines are also illustrated. At zero temperature, exact calculations locate the end-points of these lines at, respectively, $`\mu /v=1.6,1.2,0,2`$. The error bars here are a measure of the size of the hysteresis loops for each temperature. The statistical errors are of similar size as the symbols. As $`u/v0.5`$ the liquid-liquid coexistence lines approach the gas-liquid transition, merging with it at $`u/v=0.5`$.
Pressure was computed from numerical integration of the Gibbs Duhem equation, at fixed temperature, from zero pressure at zero density. Fig. 8 shows the corresponding pressure vs temperature phase diagrams.
At zero temperature, we have the exact values $`p/v=3,1.5,0.6,0.3`$. An inversion of the LDL-HDL phase boundary, close to the critical point, is to be noted. The LDL-HDL line has positive slope for $`u/v=1,0.75`$, and negative slope for $`u/v=0.55`$. This indicates that for $`u/v=1,0.75`$ the low density liquid phase is more entropic than the high density liquid phase as it would be the case for most liquids. However, for $`u/v=0.55`$, the LDL phase is less entropic than the HDL phase, a behavior that is expected for water. The line of temperature of maximum densities, $`TMD`$, is present for all $`u/v>0.5`$ values. The figure illustrates the $`TMD`$ only for the $`u/v=1`$ case.
In summary, we have shown, both from mean-field and from simulations that the model we proposed in a previous publication may present two liquid phases and the associated double criticality, depending on the ratio of bond strength to energy penalty for distortion, represented by our parameter $`u/v`$. Double criticality disappears in case the price paid for increasing the coordination is too high. A second consequence of varying the relative energies is that the slope of the liquid-liquid line may become negative, around the critical point.
Analysis of the model behavior shows that the two liquids are present if the two competing distances are sufficiently separated, but the ¨shoulder¨ need not be negative (see Fig. 2), in contradiction to a recent proposal Fr01 . In other words, two characteristic attractive distances might also generate the liquid-liquid transition.
Experimental and numerical evidence suggests that water possesses a first-order transition between a low density liquid and a high density liquid. The present model shows two liquid phases and consequently two critical points and a line of density anomalies.
Experimental evidence for a first-order transition in silica is still missing. According to our model, as the bond energy varies, the critical point temperature decreases. For temperatures above the critical point, under compression the liquid goes from the low density phase to high density phase in a continuous way as it is observed experimentally in silica. Since the liquid-liquid critical point moves to lower temperatures as the bond energy becomes weaker, one could suggest that the first-order transition predicted by simulations is not observed in experiments because the coexistence line and the second critical point in the case of silica might be located at very low temperatures.
Water and silica are network bonding systems, and present low-coordination solids with transition to higher coordinated solids at high pressures. Ice I is tetrahedrally bonded, with bonds suffering distortion, for higher pressures. At still higher pressures, the tetrahedral structures interpenetrate, to satisfy both energy and pressure requirements. Silica transitions between tetrahedrally and octahedrally coordinated structures. It would be interesting to establish energy criteria for bond distortions in both cases, in order to test the ideas we propose in this study.
Acknowledgments
This work was supported by the Brazilian science agencies CNPq, FINEP, Fapesp and Fapergs.
|
warning/0506/quant-ph0506054.html
|
ar5iv
|
text
|
# Improvement of stabilizer based entanglement distillation protocols by encoding operators
## 1 Introduction
In various methods in quantum communication, we have to share a maximally entangled state. Bennett et al. proposed the entanglement distillation protocol (EDP), which is a scheme for sharing a maximally entangled state by spatially separated two parties with local operations and classical communication. Classical communication in EDPs can be either one-way or two-way, and two-way EDPs can distill more entanglement than one-way EDPs.
In , the stabilizer based EDP is proposed, which is constructed from the quantum stabilizer code, and is generalization of the CSS code based EDP . By using an $`[[n,k]]`$ stabilizer code, we can construct EDPs that distill $`k`$ Bell states from $`n`$ Bell states. The recurrence protocol and the QPA protocol are special cases of stabilizer based EDPs, which are constructed from $`[[2,1]]`$ stabilizer codes \[20, Section 4\].
By now, we arbitrarily choose one of many encoding operators with a stabilizer based EDP. However, in construction of EDPs from quantum stabilizers, choice of encoding operators for stabilizer codes make large differences in performances of constructed EDPs. Even though there exist infinitely many encoding operators for a given quantum stabilizer, we cannot implement all encoding operators efficiently. The reason is as follows. Any unitary operator can be approximated by using only elementary operators, the Hadamard operator, the phase operator, the controlled not operator, and the $`\frac{\pi }{8}`$ operator. However in general, most unitary operators require exponentially many elementary operators to be approximated in high accuracy \[21, Section 4.5\].
The Clifford group is the set of unitary operators generated by the Hadamard operator, the phase operator, and the controlled not operator. In particular, each element in the Clifford group that acts on $`n`$ qubits is products of at most $`O(n^2)`$ previously described three operators . Thus, for a given stabilizer, encoding operators in the Clifford group are efficiently implementable. It is also known that operators in the Clifford group can be efficiently simulated on a classical computer (Gottesman-Knill Theorem) .
There is another method to construct two-way EDPs, which is the permutation based EDP . Permutation based EDPs utilize local operations chosen from the Clifford group, and it is known that choices of local operations make difference in performances of permutation based EDPs. When encoding operators of stabilizer based EDPs are restricted to operators in the Clifford group, the classes of stabilizer based EDPs and permutation based EDPs are equivalent . Elements of the Clifford group are described in terms of symplectic geometry, which enable us to enumerate all local operations for permutation based EDPs .
In this paper, we construct a method for enumerating all encoding operators in the Clifford group for a given stabilizer. Furthermore, we classify encoding operators into the equivalence classes such that EDPs constructed from encoding operators in the same equivalence class have the same performance. Such a classification has not been considered for either the stabilizer based EDP nor the permutation based EDP until now. By this classification, for given parameters, the number of candidates for good EDPs is significantly reduced. For example, in the case of EDPs constructed from the $`[[4,2]]`$ stabilizer code, the number of candidates is reduced by $`1/12288`$. It took one week to find the best EDP among EDPs constructed from the $`[[4,2]]`$ stabilizer code with computer search, so we need about $`200`$ years to find the best EDP without our result.
As a result, we find the best EDP over wide range of fidelity among EDPs constructed from $`[[4,2]]`$ stabilizer code. This EDP has a better performance than previously known EDPs over wide range of fidelity.
This paper is organized as follows. In Section 2, we review the stabilizer code and the stabilizer based EDP. In Section 3, we show our main theorems. In Section 4, we show the best EDP found by using our main theorems.
## 2 Preliminary
In this section, we review the stabilizer code, the encoding operator of the stabilizer code, the construction of entanglement distillation protocols (EDP) from stabilizer codes, and previously known results about two-way EDPs and the Clifford group. To make our argument general, we use the $`p`$–dimensional Hilbert space (qudit) instead of the two–dimensional space (qubit).
### 2.1 Stabilizer code
In this section, we review the non-binary generalization of the stabilizer code .
Let $``$ be the $`p`$-dimensional complex linear spaces with an orthonormal basis $`\{|0,\mathrm{},|p1\}`$, where $`p`$ is a prime number. We define two matrices $`X`$ and $`Z`$ by
$$X|i=|i+1modp,Z|i=\omega ^i|i$$
with a complex primitive $`p`$-th root $`\omega `$ of $`1`$. The matrices $`X`$ and $`Z`$ have the following relation
$`ZX=\omega XZ.`$ (1)
Let $`𝐙_p=\{0`$, …, $`p1\}`$ with addition and multiplication taken modulo $`p`$, and $`𝐙_p^n`$ be the $`n`$-dimensional vector space over $`𝐙_p`$. For a vector $`\stackrel{}{a}=(a_1,\mathrm{},a_n|b_1,\mathrm{},b_n)𝐙_p^{2n}`$, let
$$\mathrm{𝖷𝖹}^n(\stackrel{}{a})=X^{a_1}Z^{b_1}\mathrm{}X^{a_n}Z^{b_n}.$$
Note that eigenvalues of $`X^{a_i}Z^{b_i}`$ are powers of $`\omega `$ for $`p3`$, and $`\{\pm 1,\pm 𝐢\}`$ for $`p=2`$, where $`𝐢`$ is the imaginary unit. For a vector $`\stackrel{}{c}=(c_1,\mathrm{},c_n)𝐙_p^n`$, we denote
$`|\stackrel{}{c}=|c_1\mathrm{}|c_n.`$
###### Definition 1
Let
$`𝒫_n=\{\omega ^i\mathrm{𝖷𝖹}^n(\stackrel{}{a})i𝐙_p,\stackrel{}{a}𝐙_p^{2n}\}`$ (2)
for $`p3`$,
$`𝒫_n=\{\mu \mathrm{𝖷𝖹}^n(\stackrel{}{a})\mu \{\pm 1,\pm 𝐢\},\stackrel{}{a}𝐙_p^{2n}\}`$
for $`p=2`$, and $`S`$ a commutative subgroup of $`𝒫_n`$. The group $`𝒫_n`$ is called the Pauli group and the subgroup $`S`$ is called a stabilizer.
Suppose that $`\{\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_1)`$, …, $`\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_{nk})`$ (and possibly some power of $`\omega I_{p^n}`$ for $`p3`$ and some power of $`𝐢I_{p^n}`$ for $`p=2`$) $`\}`$ is a generating set of the group $`S`$, where $`\stackrel{}{\xi }_1`$, $`\mathrm{}`$, $`\stackrel{}{\xi }_{nk}`$ are linearly independent over $`𝐙_p`$. From now, we fix a generating set of $`S`$ as $`\stackrel{}{\xi }_1,\mathrm{},\stackrel{}{\xi }_{nk}`$.
A stabilizer code $`Q`$ is a joint eigenspace of $`S`$ in $`^n`$. There are many joint eigenspaces of $`S`$ and we can distinguish an eigenspace by its eigenvalue of $`\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_i)`$ for $`i=1`$, …, $`nk`$. Hereafter we fix a joint eigenspace $`Q(\stackrel{}{0})`$ of $`S`$ and suppose that $`Q(\stackrel{}{0})`$ belongs to the eigenvalue $`\lambda _i`$ of $`\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_i)`$ for $`i=1`$, …, $`nk`$. Note that $`\lambda _i\{\omega ^aa𝐙_p\}`$ for $`p3`$, and $`\lambda _i\{\pm 1,\pm 𝐢\}`$ for $`p=2`$. For a vector $`\stackrel{}{x}=(x_1,\mathrm{},x_{nk})𝐙_p^{nk}`$, we denote $`Q(\stackrel{}{x})`$ as a joint eigenspace that belongs to the eigenvalue $`\lambda _i\omega ^{x_i}`$ of $`\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_i)`$ for $`i=1`$, …, $`nk`$.
###### Definition 2
For two vectors $`\stackrel{}{x}=(a_1`$,$`\mathrm{}`$,$`a_n|b_1`$,…,$`b_n)`$ and $`\stackrel{}{y}=(c_1`$,…,$`c_n|d_1`$,…,$`d_n)`$, the symplectic inner product is defined by
$`\stackrel{}{x},\stackrel{}{y}={\displaystyle \underset{i=1}{\overset{n}{}}}b_ic_ia_id_i.`$
Suppose that we sent $`|\phi Q(\stackrel{}{0})`$, and received $`\mathrm{𝖷𝖹}^n(\stackrel{}{e})|\phi `$. We can tell which eigenspace of $`S`$ contains the state $`\mathrm{𝖷𝖹}^n(\stackrel{}{e})|\phi `$ by measuring an observable whose eigenspaces are the same as those of $`\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_i)`$. Then the measurement outcome always indicates that the measured state $`\mathrm{𝖷𝖹}^n(\stackrel{}{e})|\phi `$ belonging to the eigenspace that belongs to eigenvalue $`\lambda _i\omega ^{\stackrel{}{\xi }_i,\stackrel{}{e}}`$.
### 2.2 Encoding operator
In this section, we review encoding operators of stabilizer codes. An encoding operator of a stabilizer code is a unitary matrix that maps the canonical basis of $`^n`$ to joint eigenvectors of a stabilizer $`S`$.
###### Definition 3
Let $`^n(\stackrel{}{e})`$ be the subspace of $`^n`$ such that $`^n(\stackrel{}{e})`$ is spanned by
$`\left\{|\stackrel{}{e}|\stackrel{}{x}\stackrel{}{x}𝐙_p^k\right\},`$
where $`\stackrel{}{e}=(e_1,\mathrm{},e_{nk})𝐙_p^{nk}`$.
Let $`\{|\vartheta (\stackrel{}{e},\stackrel{}{x})\stackrel{}{x}𝐙_p^k\}`$ be an orthonormal basis of $`Q(\stackrel{}{e})`$.
###### Definition 4
An encoding operator $`U`$ of a stabilizer code is a unitary operator on $`^n`$ that maps an orthonormal basis of $`(\stackrel{}{e})`$ to an orthonormal basis of $`Q(\stackrel{}{e})`$ for all $`\stackrel{}{e}𝐙_p^{nk}`$, i.e.,
$`U:(\stackrel{}{e})|\stackrel{}{e}|\stackrel{}{x}|\vartheta (\stackrel{}{e},\stackrel{}{x})Q(\stackrel{}{e})`$
for $`\stackrel{}{e}𝐙_p^{nk}`$ and $`\stackrel{}{x}𝐙_p^k`$.
Note that a state $`_{\stackrel{}{x}𝐙_p^k}\alpha _\stackrel{}{x}|\stackrel{}{x}`$ of $`^k`$ is encoded into
$`{\displaystyle \underset{\stackrel{}{x}𝐙_p^k}{}}\alpha _\stackrel{}{x}|\vartheta (\stackrel{}{e},\stackrel{}{x})`$
by $`U`$ with ancilla qudits $`|e`$.
### 2.3 Stabilizer based EDP
In this section, we review the stabilizer based EDP. We define the maximally entangled states in $`_A^n_B^n`$ by
$`|\beta ^n(\stackrel{}{v})=I_{p^n}\mathrm{𝖷𝖹}^n(\stackrel{}{v}){\displaystyle \frac{1}{\sqrt{p^n}}}{\displaystyle \underset{i=0}{\overset{p^n1}{}}}|i_A|i_B,`$
where $`\stackrel{}{v}𝐙_p^{2n}`$.
Suppose that Alice and Bob share a mixed state $`\rho 𝒮(_A^n_B^n)`$, where $`𝒮(_A^n_B^n)`$ is the set of density operators on $`_A^n_B^n`$. The goal of an entanglement distillation protocol is to extract as many pairs of particles with state close to $`|\beta ^1(\stackrel{}{0})`$ as possible from $`n`$ pairs of particles in the state $`\rho `$, where
$`|\beta ^1(\stackrel{}{0})={\displaystyle \frac{1}{\sqrt{p}}}{\displaystyle \underset{i=0}{\overset{p1}{}}}|i_A|i_B.`$
For $`\stackrel{}{\xi }_i=(a_1,\mathrm{},a_n|b_1,\mathrm{},b_n)`$, we define $`\stackrel{}{\xi }_i^{}=(a_1`$, $`b_1`$, …, $`a_n`$, $`b_n)`$. Since the complex conjugate of $`\omega `$ is $`\omega ^1`$, we can see that $`\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_i^{})`$ is the component-wise complex conjugated matrix of $`\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_i)`$. Let $`S^{}`$ be the subgroup of $`𝒫_n`$ generated by $`\{\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_1^{})`$, …, $`\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_{nk}^{})\}`$. Easy computation shows that $`S^{}`$ is again commutative. So we can consider joint eigenspaces of $`S^{}`$. There exists a joint eigenspace $`Q^{}(\stackrel{}{x})`$ of $`S^{}`$ whose eigenvalue of $`\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_i^{})`$ is $`\overline{\lambda }_i\omega ^{x_i}`$, where $`\overline{\lambda }_i`$ is the complex conjugate of $`\lambda _i`$. For a state
$$|\phi =\alpha _0|0+\mathrm{}+\alpha _{p^n1}|p^n1^n,$$
we define
$$\overline{|\phi }=\overline{\alpha }_0|0+\mathrm{}+\overline{\alpha }_{p^n1}|p^n1,$$
where $`\overline{\alpha }_i`$ is the complex conjugate of $`\alpha _i`$.
With those notation, our protocol is executed as follows.
1. Alice measures an observable corresponding to $`\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_i^{})`$ for each $`i`$, and let $`\overline{\lambda }_i\omega ^{a_i}`$ be the eigenvalue of an eigenspace of $`S^{}`$ containing the state after measurement. In what follows we refer to $`(a_1`$, …, $`a_{nk})𝐙_p^{nk}`$ as a *measurement outcome*.
2. Bob measures an observable corresponding to $`\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_i)`$ for each $`i`$, and let $`\lambda _i\omega ^{b_i}`$ be the eigenvalue of an eigenspace of $`S`$ containing the state after measurement. In what follows we also refer to $`(b_1`$, …, $`b_{nk})𝐙_p^{nk}`$ as a *measurement outcome*.
3. Alice sends $`(a_1`$, …, $`a_{nk})`$ to Bob.
4. If the difference of measurement outcomes $`(b_1a_1,\mathrm{},b_{nk}a_{nk})T`$ for a previously specified set $`T𝐙_p^{nk}`$, then they abort the protocol.
5. Bob performs the error correction process according to $`a_1`$, …, $`a_{nk}`$ as follows: Bob finds a matrix $`M𝒫_n`$ such that $`MQ(\stackrel{}{b})=Q(\stackrel{}{a})`$. There are many matrices $`M`$ with $`MQ(\stackrel{}{b})=Q(\stackrel{}{a})`$, and Bob choose $`M`$ providing the highest fidelity among those matrices. See for details. He applies $`M`$ to his particles.
6. Alice and Bob apply the inverse of encoding operators $`\overline{U}^{}`$ and $`U^{}`$ of the quantum stabilizer codes respectively, where $`U^{}`$ is the adjoint operator of the encoding operator $`U`$ and $`\overline{U}^{}`$ is the component-wise complex conjugate operator of $`U^{}`$. We stress that Alice applies $`\overline{U}^{}`$ instead of $`U^{}`$ .
7. Alice and Bob discards $`nk`$ ancilla qudits.
Note that, when we start with the state $`|\beta ^n(\stackrel{}{u})`$, the state becomes
$$(I_{p^n}\mathrm{𝖷𝖹}^n(\stackrel{}{u}))\underset{\stackrel{}{x}𝐙_p^k}{}\overline{|\vartheta (\stackrel{}{a},\stackrel{}{x})}|\vartheta (\stackrel{}{a},\stackrel{}{x})$$
(3)
after Step 1 \[20, proof of Lemma 1\].
### 2.4 Clifford group
###### Definition 5
Let $`𝒰_n`$ be the set of all unitary operators on $`^n`$, and $`N(𝒫_n)`$ be the normalizer of $`𝒫_n`$ in $`𝒰_n`$, i.e.,
$`N(𝒫_n)=\{UU𝒰_n,UMU^{}𝒫_nM𝒫_n\},`$
which is called the Clifford group, where $`U^{}`$ is the adjoint operator of $`U`$.
The unitary operators in the Clifford group $`N(𝒫_n)`$ are decomposed into products of the elementary operators, where elementary operators for $`p=2`$ are the Hadamard operator, the phase operator, and the controlled not operator , and the elementary operators for $`p>2`$ are the $`p`$–dimensional discrete Fourier transform operator, the sum operator, the $`p`$–dimensional phase operator, and the $`S`$ operator . The required number of the elementary operators to represent an operator in the Clifford group is at most $`O(n^2)`$.
## 3 Construction of encoding operators
In this section, we present a method to enumerate all encoding operators in the Clifford group for a given stabilizer (Definitions 12 and 15). Then, we show relations between Bell states and encoded Bell states (Lemma 20, Corollary 21, and Corollary 22). Then, we classify encoding operators into equivalence classes such that EDPs constructed from encoding operators in the same equivalence class have the same performances (Definition 28, Theorems 29 and 30). Finally, we show the method to enumerate all equivalence classes (Theorem 35).
### 3.1 Construction method
For a given stabilizer $`S`$, we define $`(S)`$ as the set of all encoding operators, which maps the subspace $`(\stackrel{}{e})`$ to the subspace $`Q(\stackrel{}{e})`$ for all $`\stackrel{}{e}𝐙_p^{nk}`$ (See Definition 4).
###### Definition 6
Let $`_{\mathrm{cl}}(S)`$ be the subset of $`(S)`$ defined by
$`_{\mathrm{cl}}(S)=(S)N(𝒫_n).`$
$`_{\mathrm{cl}}(S)`$ is the set of all encoding operators that are contained in the Clifford group. The goal of this section is to present a method for enumerating all elements of $`_{\mathrm{cl}}(S)`$. Although the method for enumerating all elements of the Clifford group is known , the method for enumerating all elements of $`_{\mathrm{cl}}(S)`$ for a given stabilizer $`S`$ is not known.
###### Definition 7
The linear space $`𝐙_p^{2n}`$ with symplectic inner product defined in Definition 2 is called the symplectic space.
###### Definition 8
Let $`\{\stackrel{}{x}_1,\mathrm{},\stackrel{}{x}_n,`$ $`\stackrel{}{y}_1,\mathrm{},\stackrel{}{y}_n\}`$ be a basis of a symplectic space $`𝐙_p^{2n}`$. If $`\stackrel{}{x}_i`$ and $`\stackrel{}{y}_i`$ satisfy
$`\stackrel{}{x}_i,\stackrel{}{y}_j`$ $`=`$ $`\delta _{ij},`$
$`\stackrel{}{x}_i,\stackrel{}{x}_j`$ $`=`$ $`0,`$
$`\stackrel{}{y}_i,\stackrel{}{y}_j`$ $`=`$ $`0`$
for all $`i`$ and $`j`$, then the basis $`\{\stackrel{}{x}_1,\mathrm{},\stackrel{}{x}_n,`$ $`\stackrel{}{y}_1,\mathrm{},\stackrel{}{y}_n\}`$ is called a hyperbolic basis.
###### Lemma 9
There exists vectors $`\stackrel{}{\xi }_{nk+1},\mathrm{},\stackrel{}{\xi }_n`$ and $`\stackrel{}{\eta }_1,\mathrm{},\stackrel{}{\eta }_n`$ such that
$`\stackrel{}{\xi }_i,\stackrel{}{\eta }_j`$ $`=`$ $`\delta _{ij},`$
$`\stackrel{}{\xi }_i,\stackrel{}{\xi }_j`$ $`=`$ $`0,`$ (4)
$`\stackrel{}{\eta }_i,\stackrel{}{\eta }_j`$ $`=`$ $`0,`$
i.e., $`\{\stackrel{}{\xi }_1,\mathrm{},\stackrel{}{\xi }_n,`$ $`\stackrel{}{\eta }_1,\mathrm{},\stackrel{}{\eta }_n\}`$ form a hyperbolic basis of $`𝐙_p^{2n}`$.
*Proof.* The assertion of this lemma follows from the standard fact in symplectic geometry . ∎
###### Lemma 10
Let $`C`$ be a linear subspace of $`𝐙_p^{2n}`$ spanned by $`\stackrel{}{\xi }_1`$, …, $`\stackrel{}{\xi }_{nk}`$, and $`C^{}`$ be the orthogonal space of $`C`$ with respect to the symplectic inner product. Let $`C_{\mathrm{max}}`$ be a linear subspace of $`𝐙_p^{2n}`$ spanned by $`\stackrel{}{\xi }_1,\mathrm{},\stackrel{}{\xi }_n`$. Then,
$`C_{\mathrm{max}}`$ $`=`$ $`C_{\mathrm{max}}^{},`$
$`C`$ $`C_{\mathrm{max}}`$ $`C^{},`$
and $`C^{}`$ is spanned by $`\stackrel{}{\xi }_1,\mathrm{},\stackrel{}{\xi }_n,\stackrel{}{\eta }_{nk+1},\mathrm{},\stackrel{}{\eta }_n`$.
*Proof.* The assertion of this lemma follows from the property of a hyperbolic basis. ∎
###### Definition 11
For $`p=2`$, we define $`\mu (\stackrel{}{\xi }_i),\mu (\stackrel{}{\eta }_i)\{\pm 1,\pm 𝐢\}`$ for each $`\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_i),\mathrm{𝖷𝖹}^n(\stackrel{}{\eta }_i)`$ as follows, where $`𝐢`$ is the imaginary unit. For a vector $`\stackrel{}{\xi }_i=(a_1,\mathrm{},a_n|b_1,\mathrm{},b_n)`$, we define $`m(\stackrel{}{\xi }_i)=|\{ia_i=b_i=1\}|`$, i.e., the number of $`XZ`$s in $`\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_i)`$. We define $`\mu (\stackrel{}{\xi }_i)`$ as
$`\mu (\stackrel{}{\xi }_j)=𝐢^{m(\stackrel{}{\xi }_i)}.`$
$`\mu (\stackrel{}{\eta }_i)`$ is defined in the same way.
For example, in case of $`n=4`$ and $`\mathrm{𝖷𝖹}^4(\stackrel{}{\xi }_j)=XXZXZXZ`$, $`\mu (\stackrel{}{\xi }_j)=𝐢`$. In case of $`n=3`$ and $`\mathrm{𝖷𝖹}^3(\stackrel{}{\xi }_j)=XZI_2XZ`$, $`\mu (\stackrel{}{\xi }_j)=1`$. We need $`\mu (\stackrel{}{\xi }_j)`$ so that $`\left(\mu (\stackrel{}{\xi }_j)\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_j)\right)^2=I_{2^n}`$. For $`p3`$, we do not need $`\mu (\stackrel{}{\xi }_j)`$ and $`\mu (\stackrel{}{\eta }_j)`$.
###### Definition 12
Let $`S_{\mathrm{max}}`$ be a subgroup of $`𝒫_n`$ generated by $`\{\mathrm{𝖷𝖹}^n(\stackrel{}{x})\stackrel{}{x}C_{\mathrm{max}}\}`$. Let $`Q_{\mathrm{min}}(\stackrel{}{0})`$ be a stabilizer code defined by $`S_{\mathrm{max}}`$ contained in $`Q(\stackrel{}{0})`$. We have $`dimQ_{\mathrm{min}}(\stackrel{}{0})=1`$. Let $`|\psi (\stackrel{}{0})Q_{\mathrm{min}}(\stackrel{}{0})`$ be a state vector of unit norm.
Let
$`\stackrel{~}{𝖷}^n(\stackrel{}{f}_i)`$ $`=`$ $`\theta _x(\stackrel{}{f}_i)\mathrm{𝖷𝖹}^n(\stackrel{}{\eta }_i),`$ (5)
$`\stackrel{~}{𝖹}^n(\stackrel{}{f}_i)`$ $`=`$ $`\theta _z(\stackrel{}{f}_i)\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_i)`$ (6)
for $`p3`$, and
$`\stackrel{~}{𝖷}^n(\stackrel{}{f}_i)`$ $`=`$ $`\theta _x(\stackrel{}{f}_i)\mu (\stackrel{}{\eta }_i)\mathrm{𝖷𝖹}^n(\stackrel{}{\eta }_i),`$ (7)
$`\stackrel{~}{𝖹}^n(\stackrel{}{f}_i)`$ $`=`$ $`\theta _z(\stackrel{}{f}_i)\mu (\stackrel{}{\xi }_i)\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_i)`$ (8)
for $`p=2`$, where $`\stackrel{}{f}_i`$ is a vector such that the $`i`$-th element is $`1`$ and the other elements are $`0`$, $`\theta _x()`$ is an arbitrary power of $`\omega `$, and we choose $`\theta _z()`$ so that $`\stackrel{~}{𝖹}^n(\stackrel{}{f}_i)|\psi (\stackrel{}{0})=|\psi (\stackrel{}{0})\text{for }i=1,\mathrm{},n`$.
Let
$`\stackrel{~}{𝖷}^n(\stackrel{}{u})`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{n}{}}}(\stackrel{~}{𝖷}(\stackrel{}{f}_i))^{u_i}`$ (9)
$`\stackrel{~}{𝖹}^n(\stackrel{}{v})`$ $`=`$ $`{\displaystyle \underset{i=1}{\overset{n}{}}}(\stackrel{~}{𝖹}(\stackrel{}{f}_i))^{v_i}`$ (10)
for $`\stackrel{}{u}=(u_1,\mathrm{},u_n)𝐙_p^n`$ and $`\stackrel{}{v}=(v_1,\mathrm{},v_n)𝐙_p^n`$.
We define our encoding operator $`U_e`$ by
$`U_e:𝖷^n(\stackrel{}{u})|\stackrel{}{0}\stackrel{~}{𝖷}^n(\stackrel{}{u})|\psi (\stackrel{}{0}),`$ (11)
where $`𝖷^n(\stackrel{}{u})=X^{u_1}\mathrm{}X^{u_n}`$. We define $`𝖹^n(\stackrel{}{u})`$ in a similar manner.
###### Remark 13
From Lemma 16, we find that Eqs. (9) and (10) are a generalization of encoded $`𝖷^n(\stackrel{}{u})`$ operator and encoded $`𝖹^n(\stackrel{}{v})`$ operator defined in .
###### Remark 14
The construction of the encoding operator depends on the choice of $`\stackrel{}{\xi }_{nk+1},\mathrm{},\stackrel{}{\xi }_n`$ and $`\stackrel{}{\eta }_1,\mathrm{},\stackrel{}{\eta }_n`$ that satisfy Eq. (4), $`Q_{\mathrm{min}}(\stackrel{}{0})Q(\stackrel{}{0})`$, and phase factors $`\theta _x()`$. An example will be given in Section 4.
###### Definition 15
For a given stabilizer $`S`$, we define $`_g(S)`$ as the set of encoding operators $`U_e`$ for all choices of $`\stackrel{}{\xi }_{nk+1},\mathrm{},\stackrel{}{\xi }_n`$, $`\stackrel{}{\eta }_1,\mathrm{},\stackrel{}{\eta }_n`$, $`Q_{\mathrm{min}}(\stackrel{}{0})Q(\stackrel{}{0})`$, and $`\theta _x()`$.
$`_g(S)`$ is the set of all encoding operators that are constructed by the method in Definition 12. Next, we show $`_g(S)`$ is equal to $`_{\mathrm{cl}}(S)`$.
###### Lemma 16
For $`\stackrel{~}{𝖷}^n(\stackrel{}{s}),\stackrel{~}{𝖹}^n(\stackrel{}{t}),U_e`$ defined by Eqs. (9), (10), and (11), we have
$`U_e𝖷^n(\stackrel{}{s})U_e^{}`$ $`=`$ $`\stackrel{~}{𝖷}^n(\stackrel{}{s})𝒫_n\stackrel{}{s}𝐙_p^n,`$ (12)
$`U_e𝖹^n(\stackrel{}{t})U_e^{}`$ $`=`$ $`\stackrel{~}{𝖹}^n(\stackrel{}{t})𝒫_n\stackrel{}{t}𝐙_p^n.`$ (13)
*Proof.* For $`\stackrel{}{u}𝐙_p^n`$, let $`|\phi (\stackrel{}{u})=U_e|\stackrel{}{u}=\stackrel{~}{𝖷}^n(\stackrel{}{u})|\psi (\stackrel{}{0})`$. For $`\stackrel{}{s}𝐙_p^n`$, we have
$`U_e𝖷(\stackrel{}{s})U_e^{}|\phi (\stackrel{}{u})`$ $`=`$ $`U_e𝖷^n(\stackrel{}{s})|\stackrel{}{u}`$
$`=`$ $`U_e|\stackrel{}{u}+\stackrel{}{s}`$
$`=`$ $`\stackrel{~}{𝖷}^n(\stackrel{}{u}+\stackrel{}{s})|\psi (\stackrel{}{0})`$
$`=`$ $`\stackrel{~}{𝖷}^n(\stackrel{}{s})\stackrel{~}{𝖷}^n(\stackrel{}{u})|\psi (\stackrel{}{0})`$
$`=`$ $`\stackrel{~}{𝖷}^n(\stackrel{}{s})|\phi (\stackrel{}{u}).`$
Since $`\{|\phi (\stackrel{}{u})\stackrel{}{u}𝐙_p^n\}`$ form an orthonormal basis of $`^n`$, we have
$`U_e𝖷^n(\stackrel{}{s})U_e^{}=\stackrel{~}{𝖷}^n(\stackrel{}{s})𝒫_n\stackrel{}{s}𝐙_p^n.`$
Next, for $`\stackrel{}{t}𝖹_p^n`$, we have
$`U_e𝖹^n(\stackrel{}{t})U_e^{}|\phi (\stackrel{}{u})`$ $`=`$ $`U_e𝖹^n(\stackrel{}{t})|\stackrel{}{u}`$ (14)
$`=`$ $`\omega ^{(\stackrel{}{t},\stackrel{}{u})}U_e|\stackrel{}{u}`$
$`=`$ $`\omega ^{(\stackrel{}{t},\stackrel{}{u})}\stackrel{~}{𝖷}^n(\stackrel{}{u})|\psi (\stackrel{}{0}),`$
where $`(,)`$ is the standard inner product. From the definition of $`\stackrel{~}{𝖹}^n(\stackrel{}{t})`$, we have $`\stackrel{~}{𝖹}^n(\stackrel{}{t})|\psi (\stackrel{}{0})=|\psi (\stackrel{}{0})`$. Since $`\stackrel{}{\xi }_i`$ and $`\stackrel{}{\eta }_j`$ satisfy Eq. (4), we have
$`\mathrm{𝖷𝖹}^n(\stackrel{}{\eta }_i)\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_j)`$ $`=`$ $`\omega ^1\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_j)\mathrm{𝖷𝖹}^n(\stackrel{}{\eta }_i),`$
$`\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_i)\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_j)`$ $`=`$ $`\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_j)\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_i),`$
$`\mathrm{𝖷𝖹}^n(\stackrel{}{\eta }_i)\mathrm{𝖷𝖹}^n(\stackrel{}{\eta }_j)`$ $`=`$ $`\mathrm{𝖷𝖹}^n(\stackrel{}{\eta }_j)\mathrm{𝖷𝖹}^n(\stackrel{}{\eta }_i),`$
and
$`\stackrel{~}{𝖷}^n(\stackrel{}{u})\stackrel{~}{𝖹}^n(\stackrel{}{t})=\omega ^{(\stackrel{}{t},\stackrel{}{u})}\stackrel{~}{𝖹}^n(\stackrel{}{t})\stackrel{~}{𝖷}^n(\stackrel{}{u}).`$
Since $`\stackrel{~}{𝖹}(\stackrel{}{t})|\psi (\stackrel{}{0})=|\psi (\stackrel{}{0})`$, Eq. (14) is equal to
$`\omega ^{(\stackrel{}{t},\stackrel{}{u})}\stackrel{~}{𝖷}^n(\stackrel{}{u})\stackrel{~}{𝖹}^n(\stackrel{}{t})|\psi (\stackrel{}{0})`$
$`=`$ $`\omega ^{(\stackrel{}{t},\stackrel{}{u})}\omega ^{(\stackrel{}{t},\stackrel{}{u})}\stackrel{~}{𝖹}^n(\stackrel{}{t})\stackrel{~}{𝖷}^n(\stackrel{}{u})|\psi (\stackrel{}{0})`$
$`=`$ $`\stackrel{~}{𝖹}^n(\stackrel{}{t})|\phi (\stackrel{}{u}).`$
Thus, we have
$`U_e𝖹^n(\stackrel{}{t})U_e^{}=\stackrel{~}{𝖹}^n(\stackrel{}{t})𝒫_n\stackrel{}{t}𝐙_p^n.`$
###### Corollary 17
For any $`U_e_g(S)`$, we have $`U_eN(𝒫_n)`$.
*Proof.* Since $`\{𝖷^n(\stackrel{}{s})\}`$ and $`\{𝖹^n(\stackrel{}{t})\}`$ are the generator set of $`𝒫_n`$ for $`p3`$, and $`\{𝖷^n(\stackrel{}{s})\}`$ and $`\{𝖹^n(\stackrel{}{t})\}`$ and $`𝐢I_{p^n}`$ are the generator set of $`𝒫_n`$ for $`p=2`$, from Lemma 16, we have $`U_eN(𝒫_n)`$
###### Lemma 18
For $`U_c_{\mathrm{cl}}(S)`$, there exists $`U_e_g(S)`$ such that $`U_c=U_e`$.
*Proof.* We will construct $`U_e_g(S)`$ such that $`U_e=U_c`$. We set $`|\psi `$ by
$`|\psi `$ $`=`$ $`U_c|\stackrel{}{0},`$
and set $`\stackrel{}{\xi }_{nk+1},\mathrm{},\stackrel{}{\xi }_n`$ and $`\theta _z()`$ by
$`\stackrel{~}{𝖹}^n(\stackrel{}{f}_i)`$ $`=`$ $`\theta _z(\stackrel{}{f}_i)\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_i)\text{for }i=1,\mathrm{},nk`$ (15)
$`\stackrel{~}{𝖹}^n(\stackrel{}{f}_i)`$ $`=`$ $`\theta _z(\stackrel{}{f}_i)\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_i)=U_c𝖹^n(\stackrel{}{f}_i)U_c^{}`$ (16)
$`\text{for }i=nk+1,\mathrm{},n,`$
for $`p3`$, and
$`\stackrel{~}{𝖹}^n(\stackrel{}{f}_j)`$ $`=`$ $`\theta _z(\stackrel{}{f}_j)\mu (\stackrel{}{\xi }_j)\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_j)\text{for }j=1,\mathrm{},nk,`$ (17)
$`\stackrel{~}{𝖹}^n(\stackrel{}{f}_j)`$ $`=`$ $`\theta _z(\stackrel{}{f}_j)\mu (\stackrel{}{\xi }_j)\mathrm{𝖷𝖹}(\stackrel{}{\xi }_j)=U_c𝖹^n(\stackrel{}{f}_j)U_c^{}`$ (18)
$`\text{for }j=nk+1,\mathrm{},n,`$
for $`p=2`$, where $`\stackrel{}{f}_i`$ is a vector such that the $`i`$-th element is $`1`$ and the other elements are $`0`$. Note that $`\stackrel{}{\xi }_{nk+1},\mathrm{},\stackrel{}{\xi }_n`$ are determined by $`U_c`$, while $`\stackrel{}{\xi }_1,\mathrm{},\stackrel{}{\xi }_{nk}`$ are fixed basis of $`C`$ as we stated in Section 2. From Definition 4, we have $`|\psi Q(\stackrel{}{0})`$. For $`i=1,\mathrm{},nk`$, we set $`\theta _z(\stackrel{}{f}_i)`$ so that $`\stackrel{~}{𝖹}^n(\stackrel{}{f}_i)|\psi =|\psi `$. Specifically, since $`Q(\stackrel{}{0})`$ is an eigenspace that belongs to an eigenvalue $`\lambda _i`$ of $`\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_i)`$ for $`i=1,\mathrm{},nk`$, we set $`\theta _z(\stackrel{}{f}_i)=\overline{\lambda }_i`$ for $`p3`$ and $`\theta _z(\stackrel{}{f}_i)=\overline{\lambda }_i\overline{\mu (\stackrel{}{\xi }_i)}`$ for $`p=2`$.
Set $`\stackrel{}{\eta }_1,\mathrm{},\stackrel{}{\eta }_n`$, and $`\theta _x()`$ by
$`\stackrel{~}{𝖷}^n(\stackrel{}{f}_i)=\theta _x(\stackrel{}{f}_i)\mathrm{𝖷𝖹}^n(\stackrel{}{\eta }_i)=U_c𝖷^n(\stackrel{}{f}_i)U_c^{}\text{for }i=1,\mathrm{},n,`$ (19)
for $`p3`$, and
$`\stackrel{~}{𝖷}^n(\stackrel{}{f}_j)`$ $`=`$ $`\theta _x(\stackrel{}{f}_j)\mu (\stackrel{}{\eta }_j)\mathrm{𝖷𝖹}(\stackrel{}{\eta }_j)=U_c𝖷^n(\stackrel{}{f}_j)U_c^{}\text{for }j=1,\mathrm{},n`$ (20)
for $`p=2`$. Then we have
$`\stackrel{~}{𝖷}^n(\stackrel{}{u})={\displaystyle \underset{i=1}{\overset{n}{}}}\left(\stackrel{~}{𝖷}^n(\stackrel{}{f}_i)\right)^{u_i}=U_c𝖷^n(\stackrel{}{u})U_c^{}\text{for }\stackrel{}{u}𝐙_p^n.`$
We also have
$`U_c|\stackrel{}{u}`$ $`=`$ $`U_c𝖷^n(\stackrel{}{u})|\stackrel{}{0}`$
$`=`$ $`U_c𝖷^n(\stackrel{}{u})U_c^{}U_c|\stackrel{}{0}`$
$`=`$ $`\stackrel{~}{𝖷}^n(\stackrel{}{u})|\psi .`$
Next, we show
$`\stackrel{~}{𝖹}^n(\stackrel{}{f}_i)=U_c𝖹^n(\stackrel{}{f}_i)U_c^{}\text{for }i=1,\mathrm{},nk.`$ (21)
Let $`\stackrel{}{u}=(e_1,\mathrm{},e_{nk},x_1,\mathrm{},x_k)𝐙_p^n`$, $`\stackrel{}{e}=(e_1,\mathrm{},e_{nk})`$, and $`|\phi (\stackrel{}{u})=U_c|\stackrel{}{u}`$. Then, since $`|\phi (\stackrel{}{u})Q(\stackrel{}{e})`$ and $`\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_i)|\phi (\stackrel{}{u})=\lambda _i\omega ^{e_i}|\phi (\stackrel{}{u})`$, we have
$`\stackrel{~}{𝖹}^n(\stackrel{}{f}_i)|\phi (\stackrel{}{u})`$ $`=`$ $`\theta _z(\stackrel{}{f}_i)\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_i)|\phi (\stackrel{}{u})`$
$`=`$ $`\omega ^{e_i}|\phi (\stackrel{}{u})`$
$`=`$ $`\omega ^{e_i}U_c𝖷^n(\stackrel{}{u})U_c^{}|\psi `$
$`=`$ $`\omega ^{e_i}U_c𝖷^n(\stackrel{}{u})U_c^{}U_c𝖹^n(\stackrel{}{f}_i)|\stackrel{}{0}`$
$`=`$ $`\omega ^{e_i}U_c𝖷^n(\stackrel{}{u})U_c^{}U_c𝖹^n(\stackrel{}{f}_i)U_c^{}|\psi `$
$`=`$ $`\omega ^{e_i}\omega ^{e_i}U_c𝖹^n(\stackrel{}{f}_i)U_c^{}U_c𝖷^n(\stackrel{}{u})U_c^{}|\psi `$
$`=`$ $`U_c𝖹^n(\stackrel{}{f}_i)U_c^{}|\phi (\stackrel{}{u}),`$
for $`i=1,\mathrm{},nk`$. Since $`\{|\phi (\stackrel{}{u})\stackrel{}{u}𝐙_p^n\}`$ form an orthonormal basis of $`^n`$, Eq. (21) is satisfied.
From Eqs. (1), (15), (16), (19) and (21), we have
$`\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_i)\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_j)`$ $`=`$ $`\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_j)\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_i)`$
$`\mathrm{𝖷𝖹}^n(\stackrel{}{\eta }_i)\mathrm{𝖷𝖹}^n(\stackrel{}{\eta }_j)`$ $`=`$ $`\mathrm{𝖷𝖹}^n(\stackrel{}{\eta }_j)\mathrm{𝖷𝖹}^n(\stackrel{}{\eta }_i)`$
$`\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_i)\mathrm{𝖷𝖹}^n(\stackrel{}{\eta }_i)`$ $`=`$ $`\omega \mathrm{𝖷𝖹}^n(\stackrel{}{\eta }_i)\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_i)`$
$`\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_i)\mathrm{𝖷𝖹}^n(\stackrel{}{\eta }_j)`$ $`=`$ $`\mathrm{𝖷𝖹}^n(\stackrel{}{\eta }_j)\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_i),`$
which mean that $`\stackrel{}{\xi }_1,\mathrm{},\stackrel{}{\xi }_n`$ and $`\stackrel{}{\eta }_1,\mathrm{},\stackrel{}{\eta }_n`$ satisfy Eq. (4). It is easy to check that $`|\psi `$ is an eigenvector of $`\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_1),\mathrm{},\mathrm{𝖷𝖹}^n(\stackrel{}{\xi }_n)`$, thus we can write $`|\psi =|\psi (\stackrel{}{0})Q_{\mathrm{min}}(\stackrel{}{0})`$ for some $`Q_{\mathrm{min}}(\stackrel{}{0})`$.
Consequently, we can construct an encoding operator $`U_e_g(S)`$ such that $`U_e=U_c`$. ∎
From Corollary 17 and Lemma 18, we have the following theorem.
###### Theorem 19
For a given stabilizer $`S`$, $`_g(S)=_{\mathrm{cl}}(S)`$.
### 3.2 Classification of encoding operators
In this section, we show the correspondence between Bell states and Bell states encoded by encoding operators (Lemma 20, Corollary 21, and Corollary 22). Then we show the output state of our EDPs is always a probabilistic mixture of Bell states if the input state of protocols is a probabilistic mixture of Bell states (Theorem 27). Then, we classify encoding operators into equivalence classes such that EDPs constructed from encoding operators in the same equivalence class have the same performance when the input of EDPs are the probabilistic mixture of Bell states (Definition 28, and Theorems 29, 30).
###### Lemma 20
The Bell state $`|\beta ^k(\stackrel{}{0})`$ with ancilla qubits $`|\stackrel{}{e}_A|\stackrel{}{e}_B`$, i.e.,
$`|\beta ^k(\stackrel{}{0}),\stackrel{}{e}={\displaystyle \frac{1}{\sqrt{p^k}}}{\displaystyle \underset{\stackrel{}{v}𝐙_p^k}{}}|\stackrel{}{e}_A|\stackrel{}{v}_A|\stackrel{}{e}_B|\stackrel{}{v}_B,`$
is mapped by $`\overline{U_e}U_e`$ to
$`|\varphi (\stackrel{}{e})={\displaystyle \frac{1}{\sqrt{p^k}}}{\displaystyle \underset{\stackrel{}{u}\stackrel{}{e}\times 𝐙_p^k}{}}\overline{\stackrel{~}{𝖷}}^n(\stackrel{}{u})\overline{|\psi (\stackrel{}{0})}\stackrel{~}{𝖷}(\stackrel{}{u})|\psi (\stackrel{}{0}),`$ (22)
where $`\overline{\stackrel{~}{𝖷}}^n(\stackrel{}{u})`$ is the complex conjugated matrix of $`\stackrel{~}{𝖷}(\stackrel{}{u})`$, and $`\stackrel{}{e}\times 𝐙_p^k`$ is the subset $`\{(e_1,\mathrm{},e_{nk},x_1,\mathrm{},x_k)x_i𝐙_p\}`$ of $`𝐙_p^n`$.
*Proof.* It is obvious from Eq. (11) in the definition of the encoding operator $`U_e`$. ∎
###### Corollary 21
A Bell state
$`I_{p^k}𝖷^k(\stackrel{}{\mathrm{}})𝖹^k(\stackrel{}{m})|\beta ^k(\stackrel{}{0})`$ (23)
with ancilla qubits $`|\stackrel{}{e}_A|\stackrel{}{e}_B`$, i.e.,
$`|\beta ^k(\stackrel{}{\mathrm{}},\stackrel{}{m}),\stackrel{}{e}={\displaystyle \frac{1}{\sqrt{p^k}}}{\displaystyle \underset{\stackrel{}{v}𝐙_p^k}{}}|\stackrel{}{e}_A|\stackrel{}{v}_A|\stackrel{}{e}_B𝖷^k(\stackrel{}{\mathrm{}})𝖹^k(\stackrel{}{m})|\stackrel{}{v}_B,`$ (24)
is mapped by $`\overline{U_e}U_e`$ to
$`I_{p^n}\mathrm{𝖷𝖹}^n(\stackrel{}{\mathrm{}}G+\stackrel{}{m}H)|\varphi (\stackrel{}{e}),`$ (25)
multiplied by a scalar of unit absolute value, where the matrices $`G`$ and $`H`$ are
$`G=\left(\begin{array}{c}\stackrel{}{\eta }_{nk+1}\\ \mathrm{}\\ \stackrel{}{\eta }_n\end{array}\right),H=\left(\begin{array}{c}\stackrel{}{\xi }_{nk+1}\\ \mathrm{}\\ \stackrel{}{\xi }_n\end{array}\right).`$
*Proof.* Let
$`\stackrel{}{\mathrm{}}^{}`$ $`=`$ $`(0,\mathrm{},0,\mathrm{}_1,\mathrm{},\mathrm{}_k)𝐙_p^n`$ (27)
$`\stackrel{}{m}^{}`$ $`=`$ $`(0,\mathrm{},0,m_1,\mathrm{},m_k)𝐙_p^n.`$ (28)
From Eqs. (12) and (13),
$`U_e𝖷^n(\stackrel{}{\mathrm{}}^{})𝖹^n(\stackrel{}{m}^{})U_e^{}=\stackrel{~}{𝖷}^n(\stackrel{}{\mathrm{}}^{})\stackrel{~}{𝖹}^n(\stackrel{}{m}^{}).`$
Thus, a state in Eq. (24) is mapped by $`\overline{U_e}U_e`$ to
$`(\overline{U_e}U_e)|\beta ^k(\stackrel{}{\mathrm{}},\stackrel{}{m}),\stackrel{}{e}`$
$`=`$ $`(\overline{U_e}U_e)(I_{p^n}𝖷^n(\stackrel{}{\mathrm{}}^{})𝖹^n(\stackrel{}{m}^{}))|\beta ^k(\stackrel{}{0}),\stackrel{}{e}`$
$`=`$ $`(\overline{U_e}U_e)(I_{p^n}𝖷^n(\stackrel{}{\mathrm{}}^{})𝖹^n(\stackrel{}{m}^{}))(\overline{U_e}^{}U_e^{})(\overline{U_e}U_e)|\beta ^k(\stackrel{}{0}),\stackrel{}{e}`$
$`=`$ $`I_{p^n}\stackrel{~}{𝖷}^n(\stackrel{}{\mathrm{}}^{})\stackrel{~}{𝖹}^n(\stackrel{}{m}^{})|\varphi (\stackrel{}{e})`$
$`\stackrel{\left(a\right)}{}`$ $`I_{p^n}\mathrm{𝖷𝖹}^n(\stackrel{}{\mathrm{}}G)\mathrm{𝖷𝖹}^n(\stackrel{}{m}H)|\varphi (\stackrel{}{e})`$
$``$ $`I_{p^n}\mathrm{𝖷𝖹}^n(\stackrel{}{\mathrm{}}G+\stackrel{}{m}H)|\varphi (\stackrel{}{e}),`$
where $``$ denotes that one vector is equal to another vector multiplied by a scalar of unit absolute value. Note that (a) follows from Eqs. (1), (9), and (10). ∎
###### Corollary 22
The state
$`I_{p^n}\mathrm{𝖷𝖹}^n(\stackrel{}{\mathrm{}}G+\stackrel{}{m}H)|\varphi (\stackrel{}{e})`$
is mapped by $`\overline{U_e}^{}U_e^{}`$ to
$`|\beta ^k(\stackrel{}{\mathrm{}},\stackrel{}{m}),\stackrel{}{e}={\displaystyle \frac{1}{\sqrt{p^k}}}{\displaystyle \underset{\stackrel{}{v}𝐙_p^k}{}}|\stackrel{}{e}_A|\stackrel{}{v}_A|\stackrel{}{e}_B𝖷^k(\stackrel{}{\mathrm{}})𝖹^k(\stackrel{}{m})|\stackrel{}{v}_B`$
multiplied by a scalar of unit absolute value, i.e., $`|\beta ^k(\stackrel{}{w})`$ with ancilla qudits $`|\stackrel{}{e}_A|\stackrel{}{e}_B`$, where $`\stackrel{}{w}=(\mathrm{}_1,\mathrm{},\mathrm{}_k|m_1,\mathrm{},m_k)`$. ∎
###### Definition 23
For a vector $`\stackrel{}{s}=(s_1,\mathrm{},s_{nk})`$, we define the set $`D(\stackrel{}{s})`$ by
$`D(\stackrel{}{s})=\{\stackrel{}{t}𝐙_p^{2n}\stackrel{}{\xi }_i,\stackrel{}{t}=s_i\}.`$
###### Lemma 24
When we apply Steps 15 of our distillation protocol to the state $`|\beta ^n(\stackrel{}{t})`$ and Alice and Bob do not abort the protocol in Step 4, the resulting quantum state is
$`I_{p^k}\mathrm{𝖷𝖹}^n(f(\stackrel{}{t}))|\varphi (\stackrel{}{a}),`$
where $`f()`$ is the mapping from $`𝐙_p^{2n}`$ to $`C^{}`$ and depends on the error correction process in Step 5. Specifically, $`f()`$ is defined as follows. Let $`\stackrel{}{t}^{}`$ be the most likely error in $`D(\stackrel{}{b}\stackrel{}{a})`$. The mapping $`f()`$ is defined as
$`f:D(\stackrel{}{b}\stackrel{}{a})\stackrel{}{x}\stackrel{}{x}\stackrel{}{t}^{}C^{}`$ (29)
for each $`D(\stackrel{}{b}\stackrel{}{a})`$. Note that $`_{\stackrel{}{s}𝐙_p^{nk}}D(\stackrel{}{s})=𝐙_p^{2n}`$.
*Proof.* After Steps 1 and 2, the state becomes
$`𝖯^{}(\stackrel{}{a})𝖯(\stackrel{}{b})|\beta ^n(\stackrel{}{t})=I_{p^k}\mathrm{𝖷𝖹}^n(\stackrel{}{t})|\varphi (\stackrel{}{a})Q^{}(\stackrel{}{a})Q(\stackrel{}{b}),`$
where $`𝖯^{}(\stackrel{}{a})`$ and $`𝖯(\stackrel{}{b})`$ represent the projection on to $`Q^{}(\stackrel{}{a})`$ and $`Q(\stackrel{}{b})`$ respectively. In Step 5, Bob decides the most likely error $`\stackrel{}{t}^{}D(\stackrel{}{b}\stackrel{}{a})`$ and applies $`M=\mathrm{𝖷𝖹}^n(\stackrel{}{t}^{})`$. Then the state becomes
$`I_{p^k}\mathrm{𝖷𝖹}^n(\stackrel{}{t}\stackrel{}{t}^{})|\varphi (\stackrel{}{a})=I_{p^k}\mathrm{𝖷𝖹}^n(f(\stackrel{}{t}))|\varphi (\stackrel{}{a})Q^{}(\stackrel{}{a})Q(\stackrel{}{a}).`$
The condition $`MQ(\stackrel{}{b})=Q(\stackrel{}{a})`$ implies $`\stackrel{}{t}\stackrel{}{t}^{}C^{}`$. ∎
###### Remark 25
The mapping $`f()`$ does not depends on the choice of a basis $`\{\stackrel{}{\xi }_1,\mathrm{},\stackrel{}{\xi }_{nk}\}`$ of $`C`$ or the joint eigenspace $`Q(\stackrel{}{0})`$. Since there exists one to one correspondence between $`D(\stackrel{}{s})`$ and a coset of $`𝐙_p^{2n}/C^{}`$, the mapping $`f()`$ is defined only by a representative $`\stackrel{}{t}^{}`$ of each coset of $`𝐙_p^{2n}/C^{}`$ in Eq. (29).
###### Lemma 26
When we apply Step 17 of our distillation protocol to the state $`|\beta ^n(\stackrel{}{t})`$ and Alice and Bob do not abort the protocol in Step 4, the resulting quantum state is
$`|\beta ^k(\stackrel{}{w})=|\beta ^k(gf(\stackrel{}{t})),`$
where the mapping $`g`$ is the mapping from $`C^{}`$ to $`𝐙_p^{2k}`$, more precisely
$`g:C^{}\stackrel{}{\mathrm{}}G+\stackrel{}{m}H+\stackrel{}{v}\stackrel{}{w}=(\mathrm{}_1,\mathrm{},\mathrm{}_k|m_1,\mathrm{},m_k)𝐙_p^{2k}\stackrel{}{v}C.`$ (30)
*Proof.* From Lemma 24, after Steps 15 the resulting quantum state is
$`I_{p^k}\mathrm{𝖷𝖹}^n(f(\stackrel{}{t}))|\varphi (\stackrel{}{a}),`$
with $`f(\stackrel{}{t})C^{}`$. Since $`\stackrel{}{\xi }_1,\mathrm{},\stackrel{}{\xi }_n,\stackrel{}{\eta }_{nk+1},\mathrm{},\stackrel{}{\eta }_n`$ form a basis of $`C^{}`$ and $`\stackrel{}{\xi }_1,\mathrm{},\stackrel{}{\xi }_{nk}`$ form a basis of $`C`$, $`f(\stackrel{}{t})`$ can be written as a linear combination
$`f(\stackrel{}{t})={\displaystyle \underset{i=1}{\overset{k}{}}}\mathrm{}_i\stackrel{}{\eta }_{nk+i}+m_i\stackrel{}{\xi }_{nk+i}+\stackrel{}{v},`$ (31)
where $`\stackrel{}{v}C`$. Since $`|\varphi (\stackrel{}{a})`$ is a joint engeinvector of $`S`$,
$`I_{p^k}\mathrm{𝖷𝖹}^n(f(\stackrel{}{t}))|\varphi (\stackrel{}{a})`$ $`=`$ $`I_{p^n}\mathrm{𝖷𝖹}^n(\stackrel{}{\mathrm{}}G+\stackrel{}{m}H+\stackrel{}{v})|\varphi (\stackrel{}{a})`$
$``$ $`I_{p^n}\mathrm{𝖷𝖹}^n(\stackrel{}{\mathrm{}}G+\stackrel{}{m}H)|\varphi (\stackrel{}{a})`$
By Corollary 22, after Step 6 and 7 the quantum state becomes $`|\beta ^k(\stackrel{}{w})=|\beta ^k(gf(\stackrel{}{t}))`$, where $`\stackrel{}{w}=(\mathrm{}_1,\mathrm{},\mathrm{}_k|m_1,\mathrm{},m_k)`$. ∎
###### Theorem 27
When the input to our distillation protocol is a probabilistic mixture of Bell states $`|\beta ^n(\stackrel{}{t})`$ for $`\stackrel{}{t}𝐙_p^{2n}`$, i.e.,
$`\rho _{in}={\displaystyle \underset{\stackrel{}{t}𝐙_p^{2n}}{}}P_{in}(\stackrel{}{t})|\beta ^n(\stackrel{}{t})\beta ^n(\stackrel{}{t})|`$ (32)
and the difference of Alice and Bobs’ measurement result is $`\stackrel{}{b}\stackrel{}{a}T`$, then the output from our distillation protocol is also probabilistic mixture of Bell states $`|\beta ^k(\stackrel{}{w})`$ for $`\stackrel{}{w}𝐙_p^{2k}`$, i.e.,
$`\rho _{out}={\displaystyle \underset{\stackrel{}{w}𝐙_p^{2k}}{}}P_{out}(\stackrel{}{w})|\beta ^k(\stackrel{}{w})\beta ^k(\stackrel{}{w})|,`$
where $`P_{out}(\stackrel{}{w})`$ is given by
$`P_{out}(\stackrel{}{w})={\displaystyle \underset{\stackrel{}{t}D(\stackrel{}{b}\stackrel{}{a}):gf(\stackrel{}{t})=\stackrel{}{w}}{}}P_{in}^{}(\stackrel{}{t}),`$ (33)
and $`P_{in}^{}(\stackrel{}{t})`$ is normalized as
$`P_{in}^{}(\stackrel{}{t})={\displaystyle \frac{P_{in}(\stackrel{}{t})}{_{\stackrel{}{t}D(\stackrel{}{b}\stackrel{}{a})}P_{in}(\stackrel{}{t})}}.`$
*Proof.* After Steps 1–4 of our distillation protocol, from the linearity of the measurement and the error correction, the input state $`\rho _{in}`$ becomes
$`\rho ^{}={\displaystyle \underset{\stackrel{}{t}D(\stackrel{}{b}\stackrel{}{a})}{}}P_{in}^{}(\stackrel{}{t})\left(I_{p^k}\mathrm{𝖷𝖹}^n(f(\stackrel{}{t}))|\varphi (\stackrel{}{a})\varphi (\stackrel{}{e})|I_{p^k}\mathrm{𝖷𝖹}^n(f(\stackrel{}{t}))^{}\right).`$
After applying the inverse of the encoding operator, the state $`\rho ^{}`$ becomes
$`\rho _{out}`$ $`=`$ $`{\displaystyle \underset{\stackrel{}{t}D(\stackrel{}{b}\stackrel{}{a})}{}}P_{in}^{}(\stackrel{}{t})|\beta ^k(gf(\stackrel{}{t}))\beta ^k(gf(\stackrel{}{t}))|`$ (34)
$`=`$ $`{\displaystyle \underset{\stackrel{}{w}𝐙_p^{2k}}{}}P_{out}(\stackrel{}{w})|\beta ^k(\stackrel{}{w})\beta ^k(\stackrel{}{w})|,`$ (35)
where $`P_{out}(\stackrel{}{w})`$ is given by
$`P_{out}(\stackrel{}{w})={\displaystyle \underset{\stackrel{}{t}D(\stackrel{}{b}\stackrel{}{a}):gf(\stackrel{}{t})=\stackrel{}{w}}{}}P_{in}^{}(\stackrel{}{t}).`$
When the input of EDPs are the probabilistic mixture of Bell states, the performance of the distillation protocol only depends on the coefficients $`P_{out}(\stackrel{}{w})`$ of the output of the protocol. Hereafter, we fix the stabilizer $`S`$ and the error correction process $`f()`$.
###### Definition 28
For two stabilizer based EDPs constructed from encoding operators $`U_e`$ and $`V_e`$ respectively, let the mapping $`g_U`$ be determined by $`U_e`$ in Eq. (30) and $`g_V`$ be determined by $`V_e`$ in Eq. (30). If $`g_U()=g_V()`$, then we define two encoding operators $`U_e`$ and $`V_e`$ are similar and denote it by $`U_eV_e`$.
###### Theorem 29
For two stabilizer based EDPs constructed from encoding operators $`U_e`$ and $`V_e`$ respectively, let
$`\rho _{out,U_e}={\displaystyle \underset{\stackrel{}{w}𝐙_p^{2k}}{}}P_{out,U_e}(\stackrel{}{w})|\beta ^k(\stackrel{}{w})\beta ^k(\stackrel{}{w})|`$
and
$`\rho _{out,V_e}={\displaystyle \underset{\stackrel{}{w}𝐙_p^{2k}}{}}P_{out,V_e}(\stackrel{}{w})|\beta ^k(\stackrel{}{w})\beta ^k(\stackrel{}{w})|`$
be output states of each protocols when inputs of each protocol are Eq. (32). If $`U_eV_e`$, then we have
$`P_{out,U_e}(\stackrel{}{w})=P_{out,V_e}(\stackrel{}{w})\stackrel{}{w}𝐙_p^{2k},`$ (36)
i.e., performances of two protocols are the same.
*Proof.* From Eq. (33) and the fact that $`g_U()=g_V()`$, for any $`\stackrel{}{w}𝐙_p^{2k}`$
$`P_{out,U_e}(\stackrel{}{w})`$ $`=`$ $`{\displaystyle \underset{\stackrel{}{t}D(\stackrel{}{b}\stackrel{}{a}):g_Uf(\stackrel{}{t})=\stackrel{}{w}}{}}P_{in}^{}(\stackrel{}{t})`$
$`=`$ $`{\displaystyle \underset{\stackrel{}{t}D(\stackrel{}{b}\stackrel{}{a}):g_Vf(\stackrel{}{t})=\stackrel{}{w}}{}}P_{in}^{}(\stackrel{}{t})=P_{out,V_e}(\stackrel{}{w}).`$
###### Theorem 30
If Eq. (36) holds for any input state of the form in Eq. (32), then $`U_eV_e`$.
*Proof.* We prove the contraposition of this statement, i.e., if $`U_e\sim ̸V_e`$, then Eq. (36) does not hold for some input states. Since $`g_U()g_V()`$, there exists $`\stackrel{}{u}C^{}`$ such that $`g_U(\stackrel{}{u})g_V(\stackrel{}{u})`$. Consider the following input state. Let
$`P_{in}(\stackrel{}{t})=\{\begin{array}{cc}\frac{1}{|\{\stackrel{}{s}𝐙_p^{2n}f(\stackrel{}{s})=\stackrel{}{u}\}|}\hfill & \text{if }f(\stackrel{}{t})=\stackrel{}{u}\hfill \\ 0\hfill & \text{if }f(\stackrel{}{t})\stackrel{}{u}\hfill \end{array}.`$
Then we have
$`P_{out,U_e}(\stackrel{}{w})`$ $`=`$ $`\{\begin{array}{cc}1\hfill & \text{if }\stackrel{}{w}=g_U(\stackrel{}{u})\hfill \\ 0\hfill & \text{if }\stackrel{}{w}g_U(\stackrel{}{u})\hfill \end{array},`$
$`P_{out,V_e}(\stackrel{}{w})`$ $`=`$ $`\{\begin{array}{cc}1\hfill & \text{if }\stackrel{}{w}=g_V(\stackrel{}{u})\hfill \\ 0\hfill & \text{if }\stackrel{}{w}g_V(\stackrel{}{u})\hfill \end{array},`$
and Eq. (36) does not holds. ∎
### 3.3 Enumeration of equivalence classes of encoding operators
Classify $`_g(S)`$ into equivalence classes by $``$, and denote the representative set of the equivalence classes by $`\widehat{}_g(S)`$. In this section, we show how to enumerate all elements of $`\widehat{}_g(S)`$ in Theorem 35.
###### Lemma 31
Let two encoding operators $`U_e`$ and $`V_e`$ be constructed from $`\{\stackrel{}{\xi }_{nk+1},\mathrm{},\stackrel{}{\xi }_n`$, $`\stackrel{}{\eta }_{nk+1},\mathrm{},\stackrel{}{\eta }_n\}`$ and $`\{\stackrel{}{\xi }_{nk+1}^{},\mathrm{},\stackrel{}{\xi }_n^{}`$, $`\stackrel{}{\eta }_{nk+1}^{},\mathrm{},\stackrel{}{\eta }_n^{}\}`$ respectively and the other parameters (a) $`\theta _x()`$, (b) $`\stackrel{}{\eta }_1,\mathrm{},\stackrel{}{\eta }_{nk}`$, and (c) $`Q_{\mathrm{min}}(\stackrel{}{0})`$ be the same. Further assume that $`\stackrel{}{\xi }_i\stackrel{}{\xi }_i^{}(modC)`$ for all $`nk+1in`$ and $`\stackrel{}{\eta }_i\stackrel{}{\eta }_i^{}(modC)`$ for all $`nk+1in`$. Then $`U_eV_e`$.
*Proof.* Let $`g_U`$, $`G_U`$, and $`H_U`$ be determined by $`U_e`$ in Eq. (30), and $`g_V`$, $`G_V`$, and $`H_V`$ be determined by $`V_e`$ in Eq. (30). For any vector $`\stackrel{}{u}C^{}`$, we have
$`\stackrel{}{u}=\stackrel{}{\mathrm{}}G_U+\stackrel{}{m}H_U+\stackrel{}{v}=\stackrel{}{\mathrm{}}G_V+\stackrel{}{m}H_V+\stackrel{}{v}^{}\stackrel{}{v},\stackrel{}{v}^{}C.`$
Thus, we have
$`g_U(\stackrel{}{u})=g_V(\stackrel{}{u})\stackrel{}{u}C^{}.`$
and $`U_eV_e`$. ∎
###### Lemma 32
Let two encoding operators $`U_e`$ and $`V_e`$ be constructed from $`\{\stackrel{}{\xi }_{nk+1},\mathrm{},\stackrel{}{\xi }_n`$, $`\stackrel{}{\eta }_{nk+1},\mathrm{},\stackrel{}{\eta }_n\}`$ and $`\{\stackrel{}{\xi }_{nk+1}^{},\mathrm{},\stackrel{}{\xi }_n^{}`$, $`\stackrel{}{\eta }_{nk+1}^{},\mathrm{},\stackrel{}{\eta }_n^{}\}`$ respectively and the other parameters (a). $`\theta _x()`$, (b). $`\stackrel{}{\eta }_1,\mathrm{},\stackrel{}{\eta }_{nk}`$, and (c). $`Q_{\mathrm{min}}(\stackrel{}{0})`$ are the same. If $`g_U()=g_V()`$, i.e., $`U_eV_e`$, then $`\stackrel{}{\xi }_i\stackrel{}{\xi }_i^{}(modC)`$ for all $`nk+1in`$ and $`\stackrel{}{\eta }_i\stackrel{}{\eta }_i^{}(modC)`$ for all $`nk+1in`$.
*Proof.* For $`\stackrel{}{u}C^{}`$ such that
$`g_U(\stackrel{}{u})=g_V(\stackrel{}{u})=(\stackrel{}{f}_i|\stackrel{}{0})𝐙_p^{2k},`$
from Eq. (30), we have
$`\stackrel{}{u}=\stackrel{}{\xi }_i+\stackrel{}{v}=\stackrel{}{\xi }_i^{}+\stackrel{}{v}^{}\stackrel{}{v},\stackrel{}{v}^{}C.`$
Thus, we have
$`\stackrel{}{\xi }_i\stackrel{}{\xi }_i^{}=\stackrel{}{v}\stackrel{}{v}^{}C,`$
which means $`\stackrel{}{\xi }_i\stackrel{}{\xi }_i^{}(modC)`$ for $`nk+1in`$. Similarly, for $`\stackrel{}{u}C^{}`$ such that
$`g_U(\stackrel{}{u})=g_V(\stackrel{}{u})=(\stackrel{}{0}|\stackrel{}{f}_i)𝐙_p^{2k},`$
from Eq. (30), we have
$`\stackrel{}{u}=\stackrel{}{\eta }_i+\stackrel{}{v}=\stackrel{}{\eta }_i^{}+\stackrel{}{v}^{}\stackrel{}{v},\stackrel{}{v}^{}C.`$
Thus, we have
$`\stackrel{}{\eta }_i\stackrel{}{\eta }_i^{}=\stackrel{}{v}\stackrel{}{v}^{}C,`$
which means $`\stackrel{}{\eta }_i\stackrel{}{\eta }_i^{}(modC)`$ for $`nk+1in`$. ∎
###### Definition 33
Let $`\stackrel{}{x}+C`$ and $`\stackrel{}{y}+C`$ be elements of the coset $`C^{}/C`$. Define a symplectic inner product of $`\stackrel{}{x}+C`$ and $`\stackrel{}{y}+C`$ as
$`\stackrel{}{x}+C,\stackrel{}{y}+C=\stackrel{}{x},\stackrel{}{y}.`$ (40)
Note that this inner product does not depend on choices of a representatives $`\stackrel{}{x}`$ of $`\stackrel{}{x}+C`$ or $`\stackrel{}{y}`$ of $`\stackrel{}{y}+C`$.
###### Lemma 34
The linear space $`C^{}/C`$ is a $`2k`$–dimensional symplectic space with respect to the symplectic inner product in Eq. (40), and $`\{\stackrel{}{\xi }_{nk+1}+C,\mathrm{},\stackrel{}{\xi }_n+C`$, $`\stackrel{}{\eta }_{nk+1}+C,\mathrm{},\stackrel{}{\eta }_n+C\}`$ form a hyperbolic basis of $`C^{}/C`$.
*Proof.* It is easy to check that $`\{\stackrel{}{\xi }_{nk+1}+C,\mathrm{},\stackrel{}{\xi }_n+C`$, $`\stackrel{}{\eta }_{nk+1}+C,\mathrm{},\stackrel{}{\eta }_n+C\}`$ form a basis of $`C^{}/C`$. From Eqs. (4), we have
$`\stackrel{}{\xi }_i+C,\stackrel{}{\eta }_j+C`$ $`=`$ $`\delta _{ij},`$
$`\stackrel{}{\xi }_i+C,\stackrel{}{\xi }_j+C`$ $`=`$ $`0`$
$`\stackrel{}{\eta }_i+C,\stackrel{}{\eta }_j+C`$ $`=`$ $`0`$
for $`i,j\{nk+1,\mathrm{},n\}`$. ∎
As a consequence of Lemmas 31, and 32, we have the following theorem.
###### Theorem 35
There is one-to-one correspondence between Elements of $`\widehat{}_g(S)`$ and choices of hyperbolic bases of $`C^{}/C`$ with respect to the inner product in Eq. (40). Specifically, if two encoding operators $`U_e`$ and $`V_e`$ are different only by (a) $`\theta _x()`$, (b) $`\stackrel{}{\eta }_1,\mathrm{},\stackrel{}{\eta }_{nk}`$, or (c) $`Q_{\mathrm{min}}(\stackrel{}{0})`$, then $`U_eV_e`$. Two encoding operators $`U_e`$ and $`V_e`$ are $`U_eV_e`$ if and only if $`\stackrel{}{\xi }_i\stackrel{}{\xi }_i^{}(modC)`$ for all $`nk+1in`$ and $`\stackrel{}{\eta }_i\stackrel{}{\eta }_i^{}(modC)`$ for all $`nk+1in`$, i.e., two hyperbolic bases $`\{\stackrel{}{\xi }_{nk+1}+C,\mathrm{},\stackrel{}{\xi }_n+C`$ $`\stackrel{}{\eta }_{nk+1}+C,\mathrm{},\stackrel{}{\eta }_n+C\}`$ and $`\{\stackrel{}{\xi }_{nk+1}^{}+C,\mathrm{},\stackrel{}{\xi }_n^{}+C`$ $`\stackrel{}{\eta }_{nk+1}^{}+C,\mathrm{},\stackrel{}{\eta }_n^{}+C\}`$ of $`C^{}/C`$ are equal (Lemmas 31 and 32).
###### Remark 36
When the input of the protocol is a probabilistic mixture of Bell states, we can find the best stabilizer based EDP as follows. For a given parameter $`n`$ and $`k`$, find appropriate values for the following parameters.
1. a stabilizer $`S`$: a self-orthogonal subspace $`C𝐙_p^{2n}`$.
2. decision rule whether or not to abort the protocol in Step 4: a set $`T𝐙_p^{nk}`$.
3. error correction process: mapping $`f()`$ from $`𝐙_p^{2n}`$ to $`C^{}`$.
4. an equivalence class of encoding operator: a hyperbolic basis of $`C^{}/C`$.
Remark 25 and Theorem 35 significantly reduce the number of candidates of good EDPs. Indeed, for a given parameter $`n`$ and $`k`$, we enumerate $`nk`$ dimensional self-orthogonal subspaces $`C`$ (enumerating stabilizers $`S`$) and all hyperbolic bases of $`C^{}/C`$ for each $`C`$ (enumerating the equivalence classes of encoding operators), instead of all hyperbolic bases of $`𝐙_p^{2n}`$ (enumerating stabilizers $`S`$ and all encoding operators). The number of all hyperbolic bases of $`𝐙_p^{2n}`$ is equal to the cardinality of the set of symplectic mappings on $`𝐙_p^{2n}`$, i.e., $`|\mathrm{Sp}_{2n}(𝐙_p)|=p^{n^2}_{i=1}^n(p^{2i}1)`$ \[22, Theorem 3.1.2\]. While, the number of $`nk`$ dimensional self-orthogonal subspace of $`𝐙_p^{2n}`$ is $`_{i=0}^{nk1}(p^{2ni}p^i)/(p^{nk}p^i)`$ (see remark 37), and the number of all hyperbolic bases of $`C^{}/C`$ is equal to $`|\mathrm{Sp}_{2k}(𝐙_p)|=p^{k^2}_{i=1}^k(p^{2i}1)`$. Thus the number of candidates of EDPs is reduced by $`1/\{p^{n^2k^2}_{i=1}^{nk}(p^i1)\}`$. For example, the number of candidates of EDPs is reduced by $`1/12288`$ when $`n=4`$, $`k=2`$, and $`p=2`$. Note that the number of permutation based EDPs for a given parameter $`n`$, $`k`$, and $`p`$ is also same as the number of all hyperbolic bases of $`𝐙_p^{2n}`$.
###### Remark 37
The number of $`nk`$ dimensional self-orthogonal subspace of $`𝐙_p^{2n}`$ is the number of $`nk`$ mutually orthonormal vectors $`_{i=0}^{nk1}(p^{2ni}p^i)`$ divided by the number of bases of $`nk`$ dimensional self-orthogonal subspace $`_{i=0}^{nk1}(p^{nk}p^i)`$.
## 4 EDP with good performance
We can improve the performance of the protocol proposed in by choosing an optimal encoding operator. The improved protocol has the best performance over the range of fidelity greater than $`0.6`$ for a parameter $`n=4,k=2`$, $`p=2`$, and $`T=\{\stackrel{}{0}\}`$. Note that there is no choice of error correction process when $`T=\{\stackrel{}{0}\}`$. We calculated the performance by using the protocol appropriate times iteratively followed by the hashing protocol. The performance is plotted in Fig. 1 and is compared to the performance of the protocol in . The proposed protocol is also compared to the performance of the QPA protocol in Fig. 2, and has a better performance than the QPA protocol over the wide range of fidelity. We remark that the QPA protocol has the best performance among EDPs constructed from $`[[2,1]]`$ stabilizer codes.
The proposed protocol is constructed from a stabilizer code with a stabilizer
$`S=\{XXXX,ZZZZ\}.`$
The encoding operator is constructed as follows. The vector representation of the stabilizer is
$`\begin{array}{cc}\stackrel{}{\xi }_1=(1111|0000),\hfill & \stackrel{}{\xi }_2=(0000|1111).\hfill \end{array}`$
Then we choose $`\stackrel{}{\xi }_3,\stackrel{}{\xi }_4`$ and $`\stackrel{}{\eta }_1,\mathrm{},\stackrel{}{\eta }_4`$ to be
$`\begin{array}{cc}\stackrel{}{\xi }_3=(1100|0000),\hfill & \stackrel{}{\xi }_4=(1010|0000),\hfill \\ \stackrel{}{\eta }_1=(0000|1110),\hfill & \stackrel{}{\eta }_2=(1110|0000),\hfill \\ \stackrel{}{\eta }_3=(0000|1010),\hfill & \stackrel{}{\eta }_4=(1010|1100).\hfill \end{array}`$
We choose
$`\begin{array}{cc}\stackrel{~}{𝖷}^4(\stackrel{}{f}_1)=ZZZI_2\hfill & \stackrel{~}{𝖷}^4(\stackrel{}{f}_2)=XXXI_2\hfill \\ \stackrel{~}{𝖷}^4(\stackrel{}{f}_3)=ZI_2ZI_2\hfill & \stackrel{~}{𝖷}^4(\stackrel{}{f}_4)=𝐢XZZXI_2\hfill \end{array}`$
and
$`\begin{array}{cc}\stackrel{~}{𝖹}^4(\stackrel{}{f}_1)=XXXX\hfill & \stackrel{~}{𝖹}^4(\stackrel{}{f}_2)=ZZZZ\hfill \\ \stackrel{~}{𝖹}^4(\stackrel{}{f}_3)=XXI_2I_2\hfill & \stackrel{~}{𝖹}^4(\stackrel{}{f}_4)=XI_2XI_2.\hfill \end{array}`$
We choose one of joint eigenspaces $`Q(\stackrel{}{0})`$ spanned by
$`\{|0000+|1111,|0011+|1100,|1001+|0110,|0101+|1010\},`$
and choose $`Q_{\mathrm{min}}(\stackrel{}{0})`$ as
$`Q_{\mathrm{min}}(\stackrel{}{0})`$ $`=`$ $`\{|0000+|1111+|0011+|1100`$
$`+|1001+|0110+|0101+|1010\}.`$
## 5 Conclusion
In this paper, we showed a method for enumerating all encoding operators in the Clifford group for a given stabilizer code systematically. We further classified those encoding operators into equivalence classes such that EDPs constructed from encoding operators in the same equivalence class have the same performance when the input of EDPs is a probabilistic mixture of Bell states. By this classification, we can search EDPs with good performances efficiently. As a result, we found the best EDP among EDPs constructed from $`[[4,2]]`$ stabilizer codes. Although in this paper we employed $`T=\{\stackrel{}{0}\}`$, i.e., we abort the protocol if Alice and Bobs’ measurement outcomes disagree, performances of stabilizer EDPs may be improved by employing $`T\{\stackrel{}{0}\}`$, i.e., we decide whether to abort or perform the error correction according to the difference of Alice and Bobs’ measurement outcome. Exploring the potential of $`T\{\stackrel{}{0}\}`$ is a future research agenda.
## 6 Acknowledgment
This research is in part supported by International Communication Foundation, Japan. The authors deeply acknowledge the financial support.
|
warning/0506/astro-ph0506742.html
|
ar5iv
|
text
|
# Galaxy formation and cosmic-ray acceleration in a magnetized universe
## Abstract
We study the linear magneto-hydrodynamical behaviour of a Newtonian cosmology with a viscous magnetized fluid of finite conductivity and generalise the Jeans instability criterion. The presence of the field favors the anisotropic collapse of the fluid, which in turn leads to further magnetic amplification and to an enhanced current-sheet formation in the plane normal to the ambient magnetic field. When the currents exceed a certain threshold, the resulting electrostatic turbulence can dramatically amplify the resistivity of the medium (anomalous resistivity). This could trigger strong electric fields and subsequently the acceleration of ultra-high energy cosmic rays (UHECRs) during the formation of protogalactic structures.
large scale structure of the universe — galaxies: formation — acceleration of particles — magnetic fields — turbulence
A wide variety of astrophysical and cosmological problems are currently interpreted on the basis of gravitational instability. The current large-scale structure and galactic evolution theories are some of the best known examples. Relatively few of the available studies, however, consider the role of magnetic fields, despite the widespread presence of the latter. Magnetic fields observed in galaxies and galaxy clusters are in energy equipartition with the gas and the cosmic rays. The origin of these fields, which can be astrophysical, cosmological or both, remains an unresolved issue (Kronberg, 1994; Han & Wielebinski, 2002). If magnetism has a cosmological origin, as observations of $`\mu `$G fields in galaxy clusters and high redshift protogalaxies seem to suggest, it could have affected the evolution of universe (Grasso & Rubinstein, 2001; Widrow, 2002; Giovannini, 2004). Studies of large-scale magnetic fields and their potential implications for the formation of the observed structure have been given by several authors (see Thorn (1967); Jacobs (1968); Ruzmaikina & Ruzmakin (1971); Wasserman (1978); Papadopoulos & Esposito (1982); Zeldovich et al. (1983); Adams et al (1996); Barrow et al. (1997); Tsagas & Barrow (1997); Jedamzik et al. (2000) for a representative, though incomplete, list). Most of the early treatments were Newtonian, with the relativistic studies making a relatively recent appearance in the literature. A common factor between almost all the approaches is the use of the MHD approximation, namely the assumption that the magnetic field is frozen into an effectively infinitely conductive cosmic medium. With few exceptions (Fennelly, 1980; Jedamzik et al., 1998), the role of kinetic viscosity and the possibility of finite conductivity have been largely marginalized. Nevertheless, these aspects are essential for putting together a comprehensive picture of the magnetic behavior, particularly during the nonlinear regime. In this article we consider a Newtonian expanding magnetized fluid and assume that both viscosity and resistivity are finite. At first, we look into the linear evolution of small inhomogeneities in the cosmic medium and examine how the field and the fluid viscosity affect the characteristic scales of the gravitational instability. We then discuss the electrodynamic properties of the collapsing fluid, the resulting magnetic amplification and the formation of unstable current sheets. Central to our discussion is the concept of “anomalous resistivity”, which is triggered by electrostatic instabilities in the plasma and can substantially reduce the electrical conductivity of the latter. We argue that such changes in the resistivity of the protogalactic medium will lead to the formation of strong electric fields during the galactic collapse. These fields can then accelerate the abundant free electrons and ions to ultra high energies.
Let us consider an expanding, incompressible, magnetized fluid with $`p=p(\rho )`$, where $`p`$ and $`\rho `$ are respectively the pressure and the density of the matter. This medium obeys the standard Newtonian MHD equations, which in comoving coordinates read
$`{\displaystyle \frac{\rho }{t}}`$ $`=`$ $`3{\displaystyle \frac{\dot{a}}{a}}\rho {\displaystyle \frac{1}{a}}(\rho \stackrel{}{u})`$ (1)
$`{\displaystyle \frac{\stackrel{}{u}}{t}}`$ $`=`$ $`{\displaystyle \frac{\dot{a}}{a}}\stackrel{}{u}{\displaystyle \frac{1}{a}}(\stackrel{}{u})\stackrel{}{u}{\displaystyle \frac{c_\mathrm{s}^2}{a\rho }}\rho +{\displaystyle \frac{1}{a}}\varphi +{\displaystyle \frac{1}{4\pi a\rho }}(\times \stackrel{}{B})\times \stackrel{}{B}+{\displaystyle \frac{\nu }{a^2\rho }}^2\stackrel{}{u}`$ (2)
$`^2\varphi `$ $`=`$ $`4\pi Ga^2\rho ,`$ (3)
$`{\displaystyle \frac{\stackrel{}{B}}{t}}`$ $`=`$ $`2{\displaystyle \frac{\dot{a}}{a}}\stackrel{}{B}+{\displaystyle \frac{1}{a}}\times (\stackrel{}{u}\times \stackrel{}{B})+{\displaystyle \frac{\eta }{a^2}}^2\stackrel{}{B},`$ (4)
$`\stackrel{}{B}`$ $`=`$ $`0.`$ (5)
In the above $`a`$ is the cosmological scale factor, $`\stackrel{}{u}`$ is the fluid peculiar velocity (with $`\stackrel{}{u}=0`$), $`c_\mathrm{s}^2=\mathrm{d}p/\mathrm{d}\rho `$ is the square of the sound speed, $`\varphi `$ is the gravitational potential, $`\stackrel{}{B}`$ is the magnetic field vector, $`\nu `$ is the viscosity coefficient of the medium and $`\eta `$ is its electric resistivity. The system (1)-(5) accepts a homogeneous solution with $`\rho =\rho _0(t)a^3`$, $`\stackrel{}{B}=\stackrel{}{B}_0(t)a^2`$ and $`\stackrel{}{u}=\stackrel{}{u}_0=0`$. This solution describes a weakly magnetized (i.e. $`B_0^2/\rho _01`$) Newtonian FRW universe, which also defines our unperturbed background.
Following Eq. (2), the magnetic effects are confined orthogonal to $`\stackrel{}{B}`$ (recall that $`[(\times \stackrel{}{B})\times \stackrel{}{B}]\stackrel{}{B}=0`$), which ensures that there is no magnetic effect along the field’s force lines. Given this, we align the background magnetic field along the z-axis of an orthonormal frame and consider its effects in the x-y plane. We do so by perturbing Eqs. (1)-(5) around the zero-order FRW solution so that $`\rho =\rho _0+\rho _1`$, $`\stackrel{}{B}=\stackrel{}{B}_0+\stackrel{}{B}_1`$, $`\varphi =\varphi _0+\varphi _1`$ and $`\stackrel{}{u}0`$. Assuming wave-like perturbations (i.e. $`\rho _1(\stackrel{}{r},t)=\stackrel{~}{\rho }_1(t)e^{i\stackrel{}{k}\stackrel{}{r}}`$, $`\stackrel{}{B}_1=\stackrel{~}{\stackrel{}{B}}_1(t)e^{i\stackrel{}{k}\stackrel{}{r}}`$, etc) and using expressions (1)-(5), the time derivative of Eq. (2) gives
$`\ddot{\stackrel{}{u}}`$ $`=`$ $`\left(H+{\displaystyle \frac{\nu k^2}{a^2\rho _0}}\right)\dot{\stackrel{}{u}}+\left[8\pi G\rho _0{\displaystyle \frac{k^2}{a^2}}\left(c_\mathrm{s}^2+c_\mathrm{a}^2+{\displaystyle \frac{\nu H}{\rho _0}}\right)\right]\stackrel{}{u}`$ (6)
$`+i\left[{\displaystyle \frac{c_\mathrm{s}^2H}{a\rho _0}}\rho _1{\displaystyle \frac{8\pi GaH}{k^2}}\rho _1+{\displaystyle \frac{H}{2\pi a\rho _0}}\left(\stackrel{}{B}_0\stackrel{}{B}_1\right)+{\displaystyle \frac{k^2\eta }{4\pi a^3\rho _0}}\left(\stackrel{}{B}_0\stackrel{}{B}_1\right)\right]\stackrel{}{k},`$
where $`H=\dot{a}/a`$ is the Hubble parameter, $`c_\mathrm{a}^2=B_0^2/4\pi \rho _0`$ is the Alfvén speed squared and we have dropped the tildas for simplicity. Also, to reduce the algebra we have only considered perturbations orthogonal to the background magnetic field. The real component of the above provides a wave equation for the peculiar velocity vector, which describes a damped oscillation. In particular,
$`\ddot{\stackrel{}{u}}`$ $`=`$ $`\left(H+{\displaystyle \frac{\nu k^2}{a^2\rho _0}}\right)\dot{\stackrel{}{u}}+\left[8\pi G\rho _0{\displaystyle \frac{k^2}{a^2}}\left(c_\mathrm{s}^2+c_\mathrm{a}^2+{\displaystyle \frac{\nu H}{\rho _0}}\right)\right]\stackrel{}{u},`$ (7)
where the first term in the right shows the damping due to the expansion and of the fluid viscosity. The latter effect is scale-dependent and vanishes on large enough scales (i.e. as $`k0`$). The last term in Eq. (7) demonstrates the conflict between gravity on the one hand and fluid pressure and viscosity on the other. On large scales gravity always wins and the perturbations collapse. Small wavelength fluctuations, however, oscillate.
Accordingly, the magnetic presence adds to the supporting effects of pressure and viscosity only orthogonal to $`\stackrel{}{B}_0`$. This means that the first scales to collapse along the magnetic field lines are smaller than those normal to them. The two critical wavelengths are the associated Jeans scales
$`\lambda _{}\sqrt{{\displaystyle \frac{c_\mathrm{s}^2+c_\mathrm{a}^2+\nu H/\rho _0}{8\pi G\rho _0}}}\mathrm{and}\lambda _{}\sqrt{{\displaystyle \frac{c_\mathrm{s}^2+\nu H/\rho _0}{8\pi G\rho _0}}}`$ (8)
orthogonal and parallel to $`\stackrel{}{B}_0`$ respectively. Overall, the magnetic presence induces a degree of anisotropy in the collapse. Note that for a pressureless, dust-like medium the Jeans length along $`\stackrel{}{B}_0`$ depends entirely on the viscosity and the Hubble rate.
As the collapse proceeds, one expects the gradual formation of turbulent motions within the magnetized medium. The associated eddy viscosity is proportional to $`\nu _{turb}\rho _0u_1l_{mix}`$ where $`u_1`$ is the velocity perturbation and $`l_{mix}`$ the turbulent mixing length (e.g. see Biskamp (2003)). Assuming that $`u_1`$ reaches values close to $`c_\mathrm{a}`$, that the mixing length is a fraction of the magnetically induced Jeans length (i.e. $`l_{mix}\lambda _{}`$) and given the low thermal temperature of the post-recombination universe, the characteristic scaling on the velocities is $`c_\mathrm{s}^2<\nu H/\rho _0<c_\mathrm{a}^2`$. Note that we have also adopted the typical values of $`B10^7`$ G, $`\rho _010^{29}`$ $`\mathrm{gr}\mathrm{cm}^3`$ and $`H=100h`$ $`\mathrm{km}\mathrm{sec}^1\mathrm{Mpc}^1`$, where $`0.4h1`$. Then, $`\lambda _{}/\lambda _{}\sqrt{(c_\mathrm{a}^2\rho _0)/(\nu _{turb}H)}1`$, with $`\lambda _{}`$ of the order of a (comoving) Mpc. Following the standard structure formation scenarios, the initial collapse of this very large structure will be followed by successive fragmentation into smaller scale formations with characteristic lengths $`\lambda \lambda _{}`$. Moreover, as we will outline next, the anisotropy of the collapse will further increase the magnetic field trapped into the gravitating medium.
In the case of an almost spherically symmetric collapse, linear inhomogeneities in the magnetic energy density amplify in tune with those in the density of the matter so that $`\delta B^2\delta \rho `$, where $`\delta B^2=B_1^2/B_0^2`$ and $`\delta \rho =\rho _1/\rho _0`$ (Tsagas & Barrow, 1998; Tsagas & Maartens, 2000). Therefore, even within spherical symmetry, the formation of matter condensations in the post-recombination universe also signals the amplification of any magnetic field that happens to be present at the time. We have seen, however, that the generically anisotropic nature of the field will inevitably induce some degree of anisotropy to the collapse. Moreover, the magnetically induced anisotropy in the collapse will backreact and affect the evolution of the field itself. The magnetic evolution during the nonlinear regime of a generic, non-spherical protogalactic collapse, has been considered by a number of authors (Zeldovich, 1970; Zeldovich et al., 1983; Bruni et al., 2003; Siemieniec-Ozieblo & Golda, 2004; Dolag et al., 1999, 2002; Roettiger et al., 1999). The approaches are both analytical and numerical and agree that shearing effects increase the strength of the final field, while confining it to the protogalactic plane. Compared to the magnetic strengths of the spherical collapse scenario, the anisotropic increase of $`B`$ is stronger by at least one order of magnitude. Thus, protogalactic structures can be endowed with magnetic fields stronger than those previously anticipated.
So far we have seen how the magnetic presence modifies the way gravitational collapse proceeds, by changing the overall stability of the magnetized fluid. This in turn affects the evolution of the field itself and can trigger a chain of nonlinear effects on certain scales. Next we will argue that this selective amplification of certain perturbative modes can play a important role during the nonlinear stages of protogalactic collapse, helping the instability to reach its saturation point. The current induced by the total field $`B`$ is
$$\stackrel{}{J}=\frac{c}{4\pi }\times \stackrel{}{B}=\frac{c}{4\pi }\alpha \stackrel{}{B},$$
(9)
where $`\times \stackrel{}{B}=\alpha \stackrel{}{B}`$ and $`\alpha `$ measures the magnetic torsion (see Parker (1993) for details). Initially $`\alpha `$ is small. However, the subsequent fragmentation of the protogalactic cloud will increase $`\times \stackrel{}{B}`$ and strengthen the induced current. For example, a large-scale magnetic field with magnitude $`B10^7`$ G at the time of the collapse will lead to $`J10^2\alpha `$. The latter can reach appreciable strengths for reasonable values of $`\alpha `$.
When the current exceeds the critical value $`J_c\rho (e/m_p)c_\mathrm{s}`$, where $`c_\mathrm{s}10^4\sqrt{T(^{}K)}`$ cm/sec, $`m_p`$ is the ion mass and $`e`$ is the electron charge, the excitation of low frequency electrostatic turbulence will increase the resistivity of the medium by several orders of magnitude (Galeev & Sagdeev, 1984; Kulsrud, 1998). Note that the typical critical current is very small in the early post-recombination universe (i.e. $`J_cnec_s10^{10}\text{statamper/cm}^2`$). Consequently, very small values of $`\alpha `$ will lead to $`J>J_c`$, thus making the plasma electrostatically unstable. The effect, which is known as ‘anomalous resistivity’, can be explained through the development of current driven electrostatic instabilities in the plasma. The latter lead to the excitation of waves and oscillations of different kinds. The absorption of these waves by the ions is an additional way of transferring momentum from the electrons to the ions, along with the usual momentum loss from the former species to the latter. The average momentum loss by the electron per unit time can be written as an effective collision term in the form $`nm_e\nu _{eff}\stackrel{}{u}_e=\stackrel{}{F}_{fr}`$, where $`F_{fr}`$ is the average friction force and $`n=\rho /m_p`$ the ambient number density of the plasma particles. The friction force is proportional to the linear growth rate of the electrostatic waves ($`\gamma _k`$) and the energy of the excited waves ($`W_k`$). The effective collision frequency is estimated to be $`\nu _{eff}=\omega _e(W_{sat}/k_BT)`$. Then, the anomalous resistivity will be
$$\eta _{an}\frac{\nu _{eff}}{\omega _e^2}\left(\frac{W_{sat}}{k_BT}\right)\frac{1}{\omega _e},$$
(10)
where $`\omega _e=5.6\times 10^4\sqrt{n}\mathrm{sec}^1`$ is the plasma frequency, $`k_B`$ the Boltzmann constant and $`(W_{sat}/k_BT)1`$ is the saturated level of the electrostatic waves (Galeev & Sagdeev, 1984). For certain types of current driven waves, the anomalous resistivity is several orders of magnitude above the classical one, as confirmed in numerous laboratory experiments (Hamberger & Friedman, 1968; Yamada et al., 1975).
This sudden switch to high electrical resistivity will inevitably lead to the formation of strong electric currents and therefore to a fast magnetic dissipation and intense plasma heating. The electric fields induced by the gravitational collapse will be $`Ec_\mathrm{a}B/c+\eta _{an}J_c\eta _{an}J_c`$, given that $`c_\mathrm{a}/c1`$. Thus, in this scenario, the gravitational collapse of the magnetized, post-recombination cloud amplifies the magnetic field and indirectly generates strong electric currents localized on the protogalactic plane. The anisotropy of the collapse enhances the local currents further and eventually drives the resistivity towards anomalously high values. The inevitable result is strong electric fields accelerating the abundant free electrons. The energy gain by an electron travelling a length $`\lambda \lambda _{}`$ is $`W_{kin}eE\lambda _{}e\eta _{an}J_c\lambda _{}`$ and the relativistic factor $`\gamma =[1(v/c)^2]^1`$ is given by
$`\gamma 1`$ $`=`$ $`{\displaystyle \frac{W_{kin}}{m_ec^2}}{\displaystyle \frac{e\eta _{an}J_c\lambda _{}}{m_ec^2}}{\displaystyle \frac{e^2\eta _{an}nc_\mathrm{s}\lambda _{}}{m_ec^2}}{\displaystyle \frac{e^2B\sqrt{T}}{m_em_pc^2\sqrt{G}\sqrt{n}}}`$ (11)
$``$ $`10^{11}\left({\displaystyle \frac{n}{10^4\mathrm{cm}^3}}\right)^{{\scriptscriptstyle \frac{1}{2}}}\left({\displaystyle \frac{T}{1^{}K}}\right)^{{\scriptscriptstyle \frac{1}{2}}}\left({\displaystyle \frac{B}{10^7\mathrm{G}}}\right),`$
in CGS units. Recall that $`J_cenc_\mathrm{s}`$, $`c_\mathrm{s}10^4\sqrt{T}`$ and that $`\lambda _{}B/nm_p\sqrt{G}`$ when $`c_\mathrm{s}^2<\nu H/\rho <c_\mathrm{a}^2`$ (see (8b)). Also, we have set $`W_{sat}/k_BT1`$ in Eq. (10), which means that $`\eta _{an}1/10^4\sqrt{n}`$. Accordingly, the typical energy gain by a free electron can reach extremely high values within short timescales ($`t_{acc}\lambda _{}/c10^6`$ yrs), even for relatively weak magnetic fields. Clearly, one can extend this process to proton acceleration and show that protogalactic collapse can also source UHECRs.
We also anticipate a few particles drifting in and out these “primordial” current sheets (and the associated strong E-fields). If fragmentation has already taken place these particles will diffuse along the different current sheets and possibly form the observed power law distribution. The detailed acceleration processes, however, is beyond the scope of this article (see (Azner & Vlahos, 2004; Vlahos et al., 2004) for the diffusion of particles in many acceleration sites).
The role of unstable currents sheets along the giant radio galaxies on the acceleration of cosmic ray acceleration has already been pointed out in the literature (Kronberg et al., 2004; Colgate et al., 2001; Nodes et al., 2003). Particles gain and lose (through synchrotron and inverse Compton emission) energy continuously, by travelling at speeds close to the speed of light. The suggestion made here is that ultra high energy cosmic-ray acceleration and propagation may have started almost simultaneously with the formation of galaxies through the electrodynamic characteristics of the gravitational instability and continue, through the same processes till today, since the previously described instability is active on all cosmic scales. It is also worth pointing out that the anomalous resistivity mechanism can easily dissipate strong galactic magnetic fields, in the form of bursty heating and particle acceleration, whenever $`\eta _{an}^2\stackrel{}{B}>\times (\stackrel{}{u}\times \stackrel{}{B})`$.
The role of cosmic magnetism during the early evolution of the first structures in our universe has been a subject of research and debate for many decades. Most of the available studies, however, operate within the limits of the MHD approximation, that is they assume a highly conducting cosmic medium. As a result, the potential large-scale implications of a magnetic presence within a resistive environment are still relatively uncertain. In this letter, have considered a simple scenario which starts from the gravitational instability of a Newtonian, expanding, magnetized, viscous and resistive fluid and discusses the implications of the field’s presence during the early phases of what one might call the mild nonlinear protogalactic collapse. Focusing on the role of viscosity and especially on that of electrical resistivity, we have looked into the electrodynamic properties of the aforementioned gravitating medium and discussed issues such as current sheet formation, anomalous resistivity and particle acceleration on large scales.
We begun by outlining the ways in which a magnetic presence and a finite fluid viscosity can alter the standard picture of gravitational instability. We then discussed how the preferential, anisotropic magnetic amplification will also increase the currents on the plane perpendicular to the main axis of the collapse. These gravitationally induced current sheets will in turn trigger electrostatic instabilities, which can then lead to anomalous resistivity values and subsequently to strong electric fields. We argue that the latter can be strong enough to accelerate the free electrons to ultra high energies.
The influence of magnetic fields on cosmic-ray propagation has been the subject of research in the past (see Sigl et al. (2004) and references therein). To the best of our knowledge, however, this is the first time that a direct connection between gravitational instability and cosmic-ray acceleration has been suggested and discussed. We have outlined the basic features of this connection in a simple scenario that involves only standard Newtonian magnetohydrodynamics. Given that we are in the post-recombination era and that the scales of interest are well within the horizon, we do not anticipate any general relativistic corrections. Special relativistic effects may need to be accounted for, but not before the particle velocities are an appreciable fraction of the light speed. Clearly, a detailed study of the acceleration mechanism proposed here should also consider nonlinear effects and the possible implications of a varying electrical resistivity. The latter has been treated as a slowly changing variable, relative to the acceleration timescale. In any case, the key requirement for this simple scenario to work is the presence of a magnetic field coherent on the scale of the collapsing protogalaxy. Our calculations argue that the required strength of this field is comparable to those observed in high redshift protogalaxies. If such magnetic fields are widespread, as current observations indicate, their amplification during the nonlinear regime of galaxy formation can trigger a range of nontrivial effects. In this letter we suggest that these effects can include the formation of strong large-scale current sheets and electric fields. The latter could act as driving sources for the cosmic rays observed in our universe today.
The authors would like to thank Prof. Jan Kuijpers and Dr Heinz Isliker for helpful discussions. GT would also like to thank the Astronomical Observatory at the University of Thessaloniki, where most of this work was done, for their hospitality. This project was supported by the Greek Ministry of Education through the PYTHAGORAS program.
|
warning/0506/astro-ph0506280.html
|
ar5iv
|
text
|
# Discovery of the Binary Pulsar PSR B1259-63 in Very-High-Energy Gamma Rays around Periastron with H.E.S.S.
## 1 Introduction
PSR B1259$``$63 / SS 2883 is a binary system consisting of a $``$48 ms pulsar in orbit around a massive B2e companion star (Johnston et al., 1992a, b). The highly eccentric orbit of the pulsar places it just $`10^{13}\mathrm{cm}`$ from the companion during periastron every $``$3.4 years. Be stars are known to have non-isotropic stellar winds forming an equatorial disk with enhanced mass outflow (e.g. Waters et al., 1988). In the case of PSR B1259$``$63, timing measurements suggest that the disk is inclined with respect to the orbital plane (Wex et al., 1998), probably because the neutron star received a substantial birth kick, causing the pulsar to cross the disk two times near periastron. These unique properties make the binary system PSR B1259$``$63 an excellent laboratory for the study of pulsar winds interacting with a changing environment in the presence of an extremely intense photon field. The synchrotron origin of optically thin unpulsed radio emission detected from this source, in particular during the periastron passage (e.g. Johnston et al., 1999; Connors et al., 2002), indicates acceleration of electrons to relativistic energies. The acceleration process has been argued to be most efficient when the pulsar passes through the equatorial disk (Ball et al., 1999). The basic features of such a system (“binary plerion”) in the context of higher energy X- and $`\gamma `$-radiation components have been comprehensively discussed by Tavani & Arons (1997).
The intense photon field provided by the companion star not only plays an important role in the cooling of relativistic electrons but also serves as perfect target for the production of high energy $`\gamma `$-rays through inverse Compton (IC) scattering (Tavani et al., 1996; Kirk et al., 1999; Ball & Kirk, 2000; Ball & Dodd, 2001; Murata et al., 2004). Some of these emission models predict wind powered shock acceleration of electrons to multi-TeV energies, radiating predominantly through the synchrotron and IC channels, with the main energy release in the X- and high energy $`\gamma `$-ray bands, respectively.
The unpulsed non-thermal X-ray emission detected from PSR B1259$``$63 throughout its orbital phase in 1992 to 1996 by the ROSAT and ASCA satellites (Cominsky et al., 1994; Kaspi et al., 1995; Hirayama et al., 1996) generally supports the synchrotron origin of X-rays. The spectrum of the synchrotron radiation seems to extend to hard X-rays/low energy $`\gamma `$-rays as shown by OSSE (Grove et al., 1995) and recently confirmed by observations with the INTEGRAL satellite (Shaw et al., 2004).
Since the companion star provides the dominant source of photons for IC scattering, the target photon density is well known throughout the entire orbit. Therefore, the ratio of X-ray flux to high energy $`\gamma `$-ray flux depends only on the strength of the ambient magnetic field. Although the latter can be estimated within a general magneto-hydrodynamic treatment of the problem, it contains large uncertainties which affect the estimate of the IC $`\gamma `$-ray flux as $`F_\gamma B^2`$.
Kirk et al. (1999) studied the light curves of very-high-energy (VHE) $`\gamma `$-rays under the assumption of a $`1/r`$ dependence of the magnetic field which implies that the ratio of the energy density of the photon field to that in the magnetic field $`B`$ is independent of orbital phase. They also assumed that the position of the termination shock as well as the strength of the magnetic field is not affected by the disk of the B2e star. Under such assumptions, they predicted an asymmetric $`\gamma `$-ray light curve with respect to periastron (because of the inclination of the orbit with respect to the line of sight and the dependence of the inverse Compton $`\gamma `$-ray emissivity on the scattering angle), with an increase towards periastron and monotonic decrease after the passage of periastron. However, one might possibly expect significant deviation from such a simplified picture given the apparent strong impact of the disk on the pulsar wind termination as seen in the X-ray light curve (Tavani & Arons, 1997). Moreover, during the time periods of interaction of the pulsar wind with the equatorial disk one may expect, in addition to the IC $`\gamma `$-rays, a new component of $`\gamma `$-radiation associated with interactions of accelerated electrons and possibly also protons with the dense ambient gas (Kawachi et al., 2004). Up to now, the theoretical understanding of the properties of this complex system, involving pulsar and stellar winds interacting with each other, is quite limited because of the lack of constraining observations.
Nevertheless, the fortunate combination of: (1) the high spin-down luminosity of the pulsar, $`L=8.3\times 10^{35}\mathrm{erg}/\mathrm{s}`$, which is partially converted into populations of ultra-relativistic particles, (2) the presence of the intense target photon field provided by the companion star with energy density $`\mathrm{\hspace{0.17em}0.9}(R(t)/10^{13}\mathrm{cm})^2\mathrm{erg}/\mathrm{cm}^3`$ (where $`R(t)`$ is the spatial separation between the pulsar and the companion star), and (3) the relatively small distance to the source ($`d1.5\mathrm{kpc}`$) makes this object a very attractive candidate for VHE $`\gamma `$-ray emission.
Previous observations of PSR B1259$``$63 in VHE $`\gamma `$-rays, apart from its periastron passage were performed using the CANGAROO I and CANGAROO II detectors, but did not result in significant signals and provided upper limits at $`13\%`$ of the flux from the Crab Nebula (see Kawachi et al., 2004, and references therein). The first significant detection in VHE $`\gamma `$-rays based on preliminary analysis results was reported in Beilicke et al. (2004) a few days prior to the periastron passage in February 2004 by the High Energy Stereoscopic System (H.E.S.S.) to allow for target of opportunity observations of other instruments. The analysis of the data from this initial detection and the data obtained in the subsequent H.E.S.S. observation campaign on PSR B1259$``$63 is presented in this paper.
## 2 VHE $`\gamma `$-ray Observations and Results
### 2.1 Observations and Analysis
The observations from February to June 2004 were performed with the High Energy Stereoscopic System (H.E.S.S.), consisting of four imaging atmospheric Cherenkov telescopes (Hinton, 2004) located in Namibia, at $`23\mathrm{°}16^{}`$ S $`16\mathrm{°}30^{}`$ E in 1800 m above sea level. Each telescope has a tesselated spherical mirror with 13 m diameter and $`107\mathrm{m}^2`$ area (Bernlöhr et al., 2003; Cornils et al., 2003) and is equipped with a camera of 960 $`0.16\mathrm{°}`$-photomultiplier tubes providing a total field of view of $`5\mathrm{°}`$ in diameter (Vincent et al., 2003). During the stereoscopic observations, an array trigger requires the simultaneous detection of air-showers by several telescopes at the hardware level, allowing a suppression of background events (Funk et al., 2004).
All observations were carried out in moonless nights tracking sky positions with an alternating offset of typically $`\pm 0.5^{}`$ in declination relative to the source (the wobble mode) in time intervals of 28 minutes duration. This allows to determine the background from the same field of view and one can omit off-source observations, effectively doubling the observation time. Due to the serendipitous discovery of another source in the field of view around PSR B1259$``$63 (see Aharonian et al., 2005a), for all observations subsequent to the 14th of May 2004 (MJD 53139) the array pointing was changed to a position $`0.6`$° north of PSR B1259$``$63 and an alternating wobble offset of $`0.5`$° in right ascension (instead of declination) was used.
The data set, selected on the basis of standard quality criteria, has a dead time corrected exposure (live time) of 48.6 h and a mean zenith angle of 42.7$`\mathrm{°}`$. The corresponding mean threshold energy defined by the peak $`\gamma `$-ray detection rate for a source with a Crab-like spectrum (Aharonian et al., 2000) after selection cuts was estimated to be 380 GeV. In the phase prior to the periastron passage (7.8 h live time), due to technical problems, data from only three telescopes were considered, and for the post-periastron phase (41.9 h live time) data from the full telescope array were used.
After the calibration of the recorded air shower data (Aharonian et al., 2004), each telescope image was parametrised by its centre of gravity and second moments (Hillas, 1985) followed by the stereoscopic reconstruction of the shower geometry providing an angular resolution of $`0.1\mathrm{°}`$ for individual $`\gamma `$-rays. The $`\gamma `$-ray energy was estimated from the image intensity and the shower geometry with a typical resolution of $`15\%`$. In order to reject the vast background of cosmic-ray showers, $`\gamma `$-ray candidates are selected using cuts on image shape scaled with their expectation values obtained from Monte Carlo simulations. The cuts used for this analysis were optimised on simulations of a $`\gamma `$-ray point source with a flux level of 10% of the Crab Nebula VHE $`\gamma `$-ray flux, allowing 41.2% of the $`\gamma `$-rays to be retained while rejecting more than 99.9% of the cosmic-ray air showers. A more detailed description of the analysis techniques can be found in Aharonian et al. (2005b).
### 2.2 Detection
Figure 1 shows the distribution of the squared angular distance $`\theta ^2`$ of excess events relative to the position of PSR B1259$``$63 for the whole data set. The circle with a radius corresponding to the angular cut $`\theta _{\mathrm{cut}}^2=0.02\mathrm{deg}^2`$ in the field of view around the source position was considered as the on-region. The background was estimated from several non-overlapping circles of the same radius (off-regions) with the same angular distance from the camera centre, allowing corrections due to the varying camera acceptance to be omitted (Aharonian et al., 2001). The clear excess in the direction of the pulsar has a significance of 13.8 $`\sigma `$ and is consistent with a distribution obtained from a simulated $`\gamma `$-ray point source. The $`\gamma `$-ray signal from the direction of PSR B1259$``$63 was detected in most of the darkness periods from February to June 2004, for which the results are summarised in Table 1.
A 2D-analysis of the H.E.S.S. field of view around PSR B1259$``$63 was performed using the ring background technique as alternative background estimation method. In this method, for each bin, the number of on-events $`N_{\mathrm{on}}`$ was derived by integrating the bin content within a circle of radius $`\theta _{\mathrm{cut}}`$ (on-region). The number of background events $`N_{\mathrm{off}}`$ was estimated from a ring around the bin position with mean radius $`0.6\mathrm{°}`$ and an area 7 times larger than the area of the on-region. The normalisation $`\alpha `$ was corrected for the decrease of the radial acceptance of the cameras towards the edge of the field of view. Figure 2 shows the significance sky-map of a $`2.3\mathrm{°}`$ field of view around PSR B1259$``$63 for the February data. The excess at the pulsar position ($`N_\gamma =N_{\mathrm{on}}\alpha N_{\mathrm{off}}=227\pm 27`$, with a significance $`S=9.1\sigma `$) is consistent with the excess given in Table 1. The additional broad excess $`0.6\mathrm{°}`$ north of the pulsar is the unidentified TeV source HESS J1303$``$631 discovered in the same field of view (Aharonian et al., 2005a). The resulting bias in the background estimation due to both sources was corrected in the analysis by excluding events from a circle with radius $`0.4\mathrm{°}`$ around each source from the background estimation.
In order to derive the position of the pulsar excess and to check for possible source extension, the data was reanalysed using hard cuts requiring a minimal camera image intensity of 200 photo-electrons which significantly improves the angular resolution and drastically reduces the cosmic ray background at the expense of a higher energy threshold of 750 GeV. The uncorrelated two dimensional excess distribution for the whole data set was fitted assuming a radially symmetric, Gaussian source intensity profile
$$I(\theta )e^{\theta ^2/2\sigma _\mathrm{s}^2}$$
convolved with the point spread function of the detector. A possible bias from the second source within the field of view was corrected for by including an additional non-symmetric two dimensional Gaussian excess distribution in the fit, centred at the position of HESS J1303$``$631 and allowing the orientation and width to vary freely. The resulting pulsar excess position of $`\mathrm{RA}\mathrm{\hspace{0.17em}13}^\mathrm{h}2^\mathrm{m}49\stackrel{s}{.}3\pm 2\stackrel{s}{.}3_{\mathrm{stat}}`$, $`\mathrm{Dec}63\mathrm{°}49\mathrm{}53\mathrm{}\pm 17\mathrm{}_{\mathrm{stat}}`$ is consistent with the position of PSR B1259$``$63 ($`\mathrm{\Delta }\mathrm{RA}=9\mathrm{}\pm 15\mathrm{}_{\mathrm{stat}}`$, $`\mathrm{\Delta }\mathrm{Dec}=16\mathrm{}\pm 17\mathrm{}_{\mathrm{stat}}`$) within statistical errors. The systematic uncertainty on the absolute pointing of the telescopes has been estimated in Gillessen (2003) to be $`20\mathrm{}`$. A limit for the source extension $`\sigma _\mathrm{s}`$ was found to be $`<33\mathrm{}`$ at $`95\%`$ confidence level which corresponds to 0.24 pc at an assumed distance of 1.5 kpc. Note that a possible source confusion with the variable hard X-ray source 1RXP J130159.6$``$635806 (Kaspi et al., 1995), located $`9\mathrm{}`$ southeast from the pulsar, can be firmly excluded.
### 2.3 Energy Spectra
The measured differential energy spectrum derived from all darkness periods with a significant detection of the source (February to May) is shown in Fig. 3. A power-law fit
$$F(E)=\mathrm{d}N/\mathrm{d}E=F_0\left(\frac{E}{1\mathrm{TeV}}\right)^\mathrm{\Gamma }$$
(1)
of this spectrum yields a photon index $`\mathrm{\Gamma }=2.7\pm 0.2_{\mathrm{stat}}`$ and $`F_0=(1.3\pm 0.1_{\mathrm{stat}})\times 10^{12}\mathrm{cm}^2\mathrm{s}^1\mathrm{TeV}^1`$. The integral flux above the mean threshold energy is $`F(E>380\mathrm{GeV})=(4.0\pm 0.4_{\mathrm{stat}})\times 10^{12}\mathrm{cm}^2\mathrm{s}^1`$, equivalent to $`4.9\%`$ of the Crab Nebula flux (Aharonian et al., 2000) above this threshold.
For each darkness period for which the significance of the observed $`\gamma `$-ray excess exceeded $`6\sigma `$ (February, March, and April) a differential spectrum was derived (see Fig. 4) and the results of the corresponding power law fit according to Eq. (1) are listed in Table 2. Within statistical errors, there is no indication of time variability of the photon index. However, the changing flux normalisation $`F_0`$ indicates source flux variability (see next section). Systematic errors were estimated to be $`\delta \mathrm{\Gamma }0.2`$ and $`\delta F/F30\%`$, dominated by the precision of the energy calibration of the instrument and variations in the atmospheric extinction of the Cherenkov light.
### 2.4 Flux Variability
The daily integral flux of $`\gamma `$-rays above the mean threshold of 380 GeV is shown in Fig. 5 (lower panel). Since the spectrum cannot be derived on a daily basis due to limited statistics, the integral flux $`F(E>E_{\mathrm{th}}=380\mathrm{GeV})`$ was obtained by integrating Eq. (1) assuming the spectral index of the time-averaged spectrum $`\mathrm{\Gamma }=2.7`$. The flux normalisation $`F_0`$ was calculated using the measured excess of $`\gamma `$-rays $`N_\gamma `$ given by
$$N_\gamma =F_0d\mathrm{\Theta }t_{\mathrm{live}}(\mathrm{\Theta })dEA_{\mathrm{eff}}(E,\mathrm{\Theta })\left(\frac{E}{1\mathrm{TeV}}\right)^\mathrm{\Gamma },$$
with the daily live time $`t_{\mathrm{live}}(\mathrm{\Theta })`$ as a function of the zenith angle $`\mathrm{\Theta }`$, and the effective area $`A_{\mathrm{eff}}(E,\mathrm{\Theta })`$ after selection cuts obtained from simulations. Note that this method does not depend on the energy reconstruction and so the complete sample of excess events can be used. An alternative method, using the energy reconstruction to calculate the flux based on the effective area for each event above the energy threshold $`E_{\mathrm{th}}`$ according to
$$F(>E_{\mathrm{th}})=\frac{1}{t_{\mathrm{live}}}\left\{\underset{i,\mathrm{on}}{\overset{E_i^{}>E_{\mathrm{th}}}{}}\frac{1}{A_{\mathrm{eff}}(E_i^{},\mathrm{\Theta }_i)}\alpha \underset{j,\mathrm{off}}{\overset{E_j^{}>E_{\mathrm{th}}}{}}\frac{1}{A_{\mathrm{eff}}(E_j^{},\mathrm{\Theta }_j)}\right\},$$
with $`E^{}`$ as the reconstructed energy and $`\alpha `$ as the background normalisation, yields consistent results but introduces a bigger statistical error since only events above the threshold are considered.
The daily light curve in Fig. 5 clearly indicates a variable flux. This can be quantified by a fit of a constant flux to the data yielding a $`\chi ^2`$ of 90.9 per 35 degrees of freedom corresponding to a $`\chi ^2`$ probability of $`1.2\times 10^7`$.
In order to investigate the flux trend in each darkness period, the individual light curves were fitted separately by a straight line and the significance for an increasing or decreasing flux is listed in Table 3. The low flux state after periastron is followed by a distinct rise beginning at $`\tau +15`$ days and a slow decrease until $`\tau +75`$ days where the observed excess is no longer significant.
The upper panel of Fig. 5 shows the light curve of the transient unpulsed radio emission obtained from 2004 observations of PSR B1259$``$63 (Johnston et al., 2005) for the same time range as for the TeV band. There is some correlation visible between the radio and the TeV bands, especially a low flux state around periastron and a high flux state at $`\tau +20`$ days. The pulsed radio emission from the pulsar was eclipsed during the periastron passage (Johnston et al., 1992a), interpreted as due to scattering in the dense disk material and allowing the dates of the disk crossings to be estimated, in particular $`\tau 16`$ days and $`\tau +13`$ days for the 2004 periastron passage (Johnston et al., 2005). The observed dates of enhanced VHE $`\gamma `$-ray flux roughly correspond to orbital phases a few days after the assumed disk crossings. However, the coverage of the considered time interval by H.E.S.S. observations is limited and does not allow firm conclusions about VHE $`\gamma `$-ray flux maxima, especially pre-periastron.
## 3 Discussion and Conclusions
The detection of VHE $`\gamma `$-rays from PSR B1259$``$63 gives the first unambiguous and model-independent evidence of particle acceleration to multi-TeV energies in this unique binary system. The (most likely) synchrotron origin of X-rays (e.g. Tavani & Arons, 1997) also indicates the presence of ultra-relativistic particles, in the form of electrons, but other possible explanations of the X-radiation, e.g. due to the bulk motion Comptonisation of the stellar photons by the pulsar wind with a moderate Lorentz factor of 10–100 (Chernyakova & Illarionov, 1999), cannot be firmly excluded.
In contrast to the X-ray emission, the extension of the $`\gamma `$-ray spectrum to several TeV necessarily implies that the parent particles, electrons and/or protons, are accelerated to at least 10 TeV. The point-like feature of the detected signal constrains the extension of the $`\gamma `$-ray production region to $`R0.24\mathrm{pc}=7.4\times 10^{17}\mathrm{cm}`$. The information contained in the time variability of the signal on timescales of days provides a hundred times more stringent upper limit on the size of the $`\gamma `$-ray source of about $`10^{16}\mathrm{cm}`$.
The most likely scenario of particle acceleration and radiation in this system is a variety of the standard model of Pulsar Driven Nebulae (see e.g. Rees & Gunn, 1974; Kennel & Coroniti, 1984) which postulates that the deceleration of the ultra-relativistic pulsar wind (with a bulk motion Lorentz factor of $`10^6`$$`10^7`$) by the pressure forces of the external medium enforces a termination shock. In this context electrons are accelerated to multi-TeV energies. These electrons radiate in the magnetic and photon fields in which they propagate and thus produce synchrotron and Inverse-Compton nebulae with a typical size exceeding 0.1 pc, commonly called plerions.
The mean energy flux contained in the VHE $`\gamma `$-ray emission derived from the time-averaged energy spectrum is $`EF_E3\times 10^{12}\mathrm{erg}\mathrm{cm}^2\mathrm{s}^1`$ which represents a $`\gamma `$-ray luminosity of $`L_\gamma 8\times 10^{32}\mathrm{erg}\mathrm{s}^1`$ for an assumed distance of the binary system of $`1.5`$kpc. This corresponds to $`0.1\%`$ of the pulsar spin-down luminosity. Interestingly, this value matches the typical ratio between the pulsed and unpulsed component of X-ray radiation from isolated pulsars with associated pulsar wind nebulae and their spin-down luminosity (see e.g. Cheng et al., 2004), supporting the suggestion that the energy for the observed $`\gamma `$-rays is provided by the pulsar.
However, in the case of PSR B1259$``$63 both the spatial and temporal scales are expected to be rather different from those in plerions around isolated pulsars. Also, dynamical flow effects can play a different role since the pulsar moves through the stellar wind and the photon field of the massive companion star. Due to the high pressure of the stellar wind, the pulsar wind terminates very close to the pulsar and thus the electrons are accelerated well within the binary system. Moreover, because of severe adiabatic and radiative losses, the TeV electrons have a very short lifetime. As a result, the radiation is emitted in a rather compact region not far from the acceleration site, and for any given time the emission originates from a quite short sector of the pulsar trajectory.
In the following we summarise the physical processes expected to be at work, and put our observational results into this perspective.
### 3.1 Characteristic Cooling Times
The synchrotron loss time
$$t_{\mathrm{sy}}=6\pi \frac{m_\mathrm{e}^2c^4}{\sigma _\mathrm{T}cϵB^2}400ϵ_{\mathrm{TeV}}^1B_\mathrm{G}^2\mathrm{s},$$
(2)
of an electron with energy $`ϵ`$ depends on the local magnetic field strength $`B`$ (where $`B_G=B/1G`$) which is not known a priori. We shall estimate $`B`$ below by combining the INTEGRAL X-ray data and our H.E.S.S. TeV $`\gamma `$-ray data. In contrast, we can estimate the inverse Compton cooling rate of electrons with very good accuracy at any given time, i.e. at any location of the pulsar in its orbit (Tavani & Arons, 1997; Kirk et al., 1999).
Indeed, the inverse Compton losses are dominated by the scattering of electrons on the light of the companion star with luminosity $`L_{\mathrm{star}}=3.3\times 10^{37}\mathrm{erg}/\mathrm{s}`$ and effective blackbody temperature $`T_{\mathrm{eff}}=2.3\times 10^4\mathrm{K}`$. At any distance $`R`$, the black-body radiation decreases by a factor of $`(R/R_{})^2`$, where $`R_{}=4\times 10^{11}\mathrm{cm}`$ is the radius of the companion star. Thus at the distance of the pulsar to the star $`R_{13}=R/10^{13}\mathrm{cm}`$, the energy density of the radiation is
$$w_\mathrm{r}=\frac{L_{\mathrm{star}}}{4\pi R^2c}0.87R_{13}^2\mathrm{erg}/\mathrm{cm}^3,$$
(3)
and correspondingly the electron cooling time is determined in the deep relativistic Klein-Nishina regime which applies here for TeV energies, given the high energies of the seed photons. For a Planckian seed photon spectrum, and using the formalism of Blumenthal & Gould (1970), one obtains
$$t_{\mathrm{KN}}=3.2\times 10^3\frac{ϵ_{\mathrm{TeV}}R_{13}^2}{\mathrm{ln}(30.6ϵ_{\mathrm{TeV}})1.4}\mathrm{s}.$$
(4)
For PSR B1259$``$63, with its environment of a strong stellar wind and a strong radiative flow from the companion, we must in addition consider the adiabatic losses of the accelerated pulsar particles. This has already been noted by Tavani & Arons (1997). Indeed, the plasma flow beyond the pulsar wind termination shock differs strongly from the spherically symmetric configuration of an isolated pulsar in a static pressure environment. Rather than being decelerated in a spherically symmetric fashion while compressing the post-shock magnetic field, the actual flow will stagnate at the sub-stellar point at a distance $`l10^{12}`$ cm and expand relativistically into the downstream direction, which roughly points away from the Be star. The surface of separation between the interior flow of the pulsar particles and the exterior stellar wind flow is indicated as “Wind interface” in Fig. 6.
The expansion of the relativistic, shocked pulsar wind gas will proceed with a flow velocity of magnitude $`|\underset{¯}{u}|`$, roughly equal to the relativistic magnetosonic speed $`\epsilon v_{\mathrm{ms},}`$, where $`v_{\mathrm{ms},}c\sqrt{2/3}`$, over a distance $`\mathrm{\Delta }l`$, where $`\mathrm{\Delta }l`$ is a few $`\times l`$. The factor $`\epsilon `$ ranges between $`\sqrt{2}/2`$ and 1, if the direction of the postshock MHD flow ranges between a perpendicular and a parallel orientation with respect to the magnetic field direction, respectively. Assuming the pitch angle scattering mean free path of the accelerated pulsar wind particles to be small compared to $`l`$, their adiabatic loss time is then given by
$$t_{\mathrm{ad}}=\frac{3\epsilon }{\mathrm{div}\underset{¯}{u}}\epsilon \times 350\frac{\mathrm{\Delta }l}{2\times 10^{12}\mathrm{cm}}\mathrm{s},$$
(5)
with $`0.71<\epsilon <1`$, depending on the details of the post-shock MHD flow.
In this picture the adiabatic losses proceed about as fast as the synchrotron losses if $`B1`$ G (compare with Eq. (2)) – while both are faster than the inverse Compton losses – and the energy spectrum of the radiating electrons is close to their source spectrum produced during acceleration at the termination shock. We shall consider neither electron Bremsstrahlung nor $`\gamma `$-ray production in interactions of relativistic protons with the ambient gas via decay of secondary $`\pi ^0`$-mesons (see e.g. Kawachi et al., 2004) here, since the accelerated pulsar wind particles are hydromagnetically constrained to the magnetic field lines of the pulsar outflow and will hardly interact individually with the stellar wind particles, except in a thin boundary layer at the Wind interface far downstream.
The missing quantity in our cooling considerations is the value of $`B`$. We shall use the contemporaneous INTEGRAL results and our H.E.S.S. fluxes to estimate it as follows: Since there is little doubt that the detected TeV $`\gamma `$-rays are produced in the Klein-Nishina regime, we may conclude that they are emitted by electrons of the same energy, $`ϵ_eE_\gamma `$. This allows to set up a direct relation between $`\gamma `$-rays and X-rays produced by the same electrons,
$$E_\mathrm{X}>20B_\mathrm{G}E_{\mathrm{TeV}}^2\mathrm{keV}.$$
(6)
The Compton losses with $`E_{\mathrm{TeV}}`$ take place in a spatial region that is larger than that of synchrotron losses, and therefore adiabatic losses have already occurred there.
An experimental estimate of $`B`$ is possible by comparing the energy fluxes of the detected X- and $`\gamma `$-rays. The post-periastron data obtained in March by RXTE and INTEGRAL in the X-ray band from 1 keV to 100 keV (see Shaw et al., 2004) show a relatively flat spectral energy distribution $`\nu F_\nu E^2\mathrm{d}N/\mathrm{d}E`$ at the level of $`3\times 10^{11}\mathrm{erg}/\mathrm{cm}^2\mathrm{s}`$. The VHE $`\gamma `$-ray flux for the same period was $`2\times 10^{12}\mathrm{erg}/\mathrm{s}`$. This implies that in the TeV energy regime the synchrotron losses of electrons proceed an order of magnitude faster than the inverse Compton losses. Therefore, from the comparison of the synchrotron cooling time, given by Eq. (2), with the Klein-Nishina cooling time given by Eq. (4), and taking into account that two weeks after the periastron the separation between the pulsar and the companion star was approximately $`1.5\times 10^{13}\mathrm{cm}`$, we find
$$B1E_{\mathrm{TeV}}^1\mathrm{G}.$$
(7)
Our experimentally determined fields of the order of 1 G agree rather well with the theoretical estimate of the field derived from the MHD treatment of the pulsar wind (Tavani & Arons, 1997; Ball et al., 1999), and the value deduced from the observed unpulsed radio emission near periastron (Connors et al., 2002).
### 3.2 Spectral and Temporal Characteristics of the TeV Radiation
Let us assume that the accelerated electrons enter the $`\gamma `$-ray production region with a rate $`Q(ϵ)ϵ^{\alpha _0}`$. As a consequence, in the case of dominant synchrotron or inverse Compton losses in the Thompson regime, a steeper electron spectrum is established in the $`\gamma `$-ray production region with power-law index $`\alpha =\alpha _0+1`$, while dominating adiabatic (or Bremsstrahlung losses) will not change the initial electron spectrum, i.e. $`\alpha =\alpha _0`$, both for $`\alpha _0>1`$.
Let us conversely assume that the radiating electrons have a power law distribution $`\mathrm{d}N/\mathrm{d}ϵϵ^\alpha `$. Then the inverse Compton scattering in the Klein-Nishina regime leads to a $`\gamma `$-ray spectrum (Blumenthal & Gould, 1970)
$$\frac{\mathrm{d}N}{\mathrm{d}E}E^{(\alpha +1)}\left(\mathrm{ln}\frac{k_\mathrm{B}TE}{m_\mathrm{e}^2c^4}+C\right).$$
(8)
Using the H.E.S.S. results, the photon index of the radiating electrons can therefore be estimated from the observed $`\gamma `$-ray spectrum $`\mathrm{d}N/\mathrm{d}EE^{\mathrm{\Gamma }_\gamma }`$ with $`\mathrm{\Gamma }_\gamma =2.7\pm 0.2_{\mathrm{stat}}`$ by equating it with the expected Klein-Nishina spectrum of Eq. (8). A more detailed numerical evaluation of the IC $`\gamma `$-ray spectrum shows that in our energy range the average photon spectral index can be approximated as $`\mathrm{\Gamma }_\gamma \alpha +0.5`$. Therefore we find $`\alpha 2.2`$. In the scenario in which adiabatic losses dominate, this is, within errors, compatible with an acceleration process giving $`\alpha _0=2`$. If radiation losses dominate, this requires an injection spectrum significantly harder than $`\alpha _0=2`$.
The relevant photon spectral index range $`1.6<\mathrm{\Gamma }_X<1.8`$ in the $`1E_X10`$ keV energy range from the 1994 ASCA observations (Hirayama et al., 1996), although not coeval, would also indicate a rather hard spectrum of the radiating electrons $`2.2\alpha 2.6`$ which would again be difficult to reconcile with dominant synchrotron cooling. This 1994 synchrotron photon spectral index is consistent with a preliminary spectral fit with $`\mathrm{\Gamma }_X=1.82\pm 0.34`$ derived from the coeval INTEGRAL observations (Shaw, 2004). In this paper we shall ultimately rely on the H.E.S.S. spectrum.
Within the scheme of dominant adiabatic cooling, the strong variability of the VHE $`\gamma `$-ray emission might be explainable by a varying spatial confinement of the accelerated pulsar wind particles by the kinetic and thermal pressure of the stellar mass outflow. In Figure 6 the variations of the measured VHE $`\gamma `$-ray flux are illustrated within the context of the orbital parameters and environment of the binary system with respect to the line of sight. The enhanced mass outflow in the equatorial disk will result in a more compact emission region – and therefore weaker adiabatic cooling – during the pulsar disk passages compared to the other regions of the orbit, in particular the periastron phase, where the expected stronger adiabatic cooling could lead to a minimum of the $`\gamma `$-ray flux. A second flow aspect would be a Doppler modulation due to the relativistic bulk flow velocity away from the stagnation point of the shocked pulsar wind. Considering Fig. 6, this could again lead to a flux minimum during periastron. However, a detailed consideration of this overall physical picture is beyond the scope of the present paper.
A minimum in the X-ray flux, observed for previous periastron passages, has been interpreted as being due to increased IC losses (Tavani & Arons, 1997). However, H.E.S.S. measurements indicate a minimum in the IC flux around periastron, which would rule out such an interpretation for this periastron passage. Unfortunately, due to the full moon and bad weather conditions we were not able to take data during the periastron passage of the pulsar.
The predicted pre-periastron spectrum from Kirk et al. (1999) is compared to the corresponding H.E.S.S. data in Fig. 7. The model curves, taken from Kirk et al. (1999) (Figs. 2,6), were computed for the two scenarios of dominant radiative losses and dominant adiabatic losses, and matched the archival X-ray data for electron injection spectra with $`\alpha _0=1.4`$ and $`\alpha _0=2.4`$, respectively. They agree well with the TeV data, assuming a magnetic field strength of $`B=0.32`$G supporting the estimate of the order of 1 G obtained from the H.E.S.S. data. Despite the spectral agreement, the models fail to describe the TeV light curve as can be seen in Fig. 8 (dashed and dotted lines). In particular, neither the minimum around nor the high flux states before and after periastron are described by the model. This is perhaps not surprising, because the model assumed that the ratio between the energy densities of the magnetic field and the photon field does not change throughout the full orbit, and that the effects related to the pulsar passage through the stellar disk can be neglected.
Interestingly, the observed light curve seems to be qualitatively similar to the prediction made by the model of Kawachi et al. (2004) which is also shown in Fig. 8 (solid line). In this model, hadronic interactions and $`\pi ^0`$ production in the misaligned stellar disk plays a dominant role in the $`\gamma `$-ray production mechanism. However, present data do not allow safe conclusions concerning the interpretation of the TeV light curve. Studies of the most interesting part of the pulsar orbit have to be postponed until multi-wavelength observations during periastron become possible.
In summary, the detection of VHE $`\gamma `$-ray emission from the binary pulsar PSR B1259$``$63 by H.E.S.S. provides the first model-independent evidence of particle acceleration in this object. The results clearly demonstrate the power of $`\gamma `$-ray observations for the study of the properties and the nature of high energy processes in this unique cosmic accelerator. Further observations at VHE $`\gamma `$-ray energies are necessary to derive spectra on a daily basis also during periastron passage.
###### Acknowledgements.
The support of the Namibian authorities and of the University of Namibia in facilitating the construction and operation of H.E.S.S. is gratefully acknowledged, as is the support by the German Ministry for Education and Research (BMBF), the Max Planck Society, the French Ministry for Research, the CNRS-IN2P3 and the Astroparticle Interdisciplinary Programme of the CNRS, the U.K. Particle Physics and Astronomy Research Council (PPARC), the IPNP of the Charles University, the South African Department of Science and Technology and National Research Foundation, and by the University of Namibia. We appreciate the excellent work of the technical support staff in Berlin, Durham, Hamburg, Heidelberg, Palaiseau, Paris, Saclay, and in Namibia in the construction and operation of the equipment. We also thank S. Shaw for informations on the photon spectral index from the INTEGRAL X-ray observations.
|
warning/0506/math0506104.html
|
ar5iv
|
text
|
# Untitled Document
Factorisation of Lie Resolvents
R. M. BRYANT roger.bryant@manchester.ac.uk
School of Mathematics, University of Manchester, PO Box 88, Manchester M60 1QD, UK
M. SCHOCKER Research of second author supported by Deutsche Forschungsgemeinschaft (DFG Scho-799). m.schocker@swansea.ac.uk
Department of Mathematics, University of Wales Swansea, Singleton Park, Swansea SA2 8PP, UK
Abstract. Let $`G`$ be a group, $`F`$ a field of prime characteristic $`p`$ and $`V`$ a finite-dimensional $`FG`$-module. Let $`L(V)`$ denote the free Lie algebra on $`V`$, regarded as an $`FG`$-module, and, for each positive integer $`r`$, let $`L^r(V)`$ be the $`r`$th homogeneous component of $`L(V)`$, called the $`r`$th Lie power of $`V`$. In a previous paper we obtained a decomposition of $`L^r(V)`$ as a direct sum of modules of the form $`L^s(W)`$, where $`s`$ is a power of $`p`$. Here we derive some consequences. First we obtain a similar result for restricted Lie powers of $`V`$. Then we consider the ‘Lie resolvents’ $`\mathrm{\Phi }^r`$: certain functions on the Green ring of $`FG`$ which determine Lie powers up to isomorphism. For $`k`$ not divisible by $`p`$, we obtain the factorisation $`\mathrm{\Phi }^{p^mk}=\mathrm{\Phi }^{p^m}\mathrm{\Phi }^k`$, separating out the key case of $`p`$-power degree. Finally we study certain functions on power series over the Green ring, denoted by $`𝐒^{}`$ and $`𝐋^{}`$, which encode symmetric powers and Lie powers, respectively. In characteristic $`0`$, $`𝐋^{}`$ is the inverse of $`𝐒^{}`$. In characteristic $`p`$, the composite $`𝐋^{}𝐒^{}`$ maps any $`p`$-typical power series to a $`p`$-typical power series.
Keywords: free Lie algebra, Lie power, Adams operation, Lie resolvent
Mathematics Subject Classification (2000): 17B01, 20C07, 20C20
1. Introduction
Let $`G`$ be a group and $`F`$ a field. For any finite-dimensional $`FG`$-module $`V`$ let $`L(V)`$ be the free Lie algebra on $`V`$. Then $`L(V)`$ may be regarded as an $`FG`$-module on which each element of $`G`$ acts as a Lie algebra automorphism. Furthermore, each homogeneous component $`L^r(V)`$ is a finite-dimensional submodule, called the $`r`$th Lie power of $`V`$. We regard $`L(V)`$ as an $`FG`$-submodule of the free associative algebra $`T(V)`$. Thus $`L^r(V)`$ is a submodule of the $`r`$th homogeneous component $`T^r(V)`$.
When $`F`$ has prime characteristic $`p`$ we may also form the free restricted Lie algebra $`R(V)`$ which again may be regarded as an $`FG`$-submodule of $`T(V)`$. The homogeneous component $`R^r(V)`$ is called the $`r`$th restricted Lie power of $`V`$.
A general problem is to describe the modules $`L^r(V)`$ and $`R^r(V)`$ up to isomorphism. We refer to and the papers cited there for a discussion of progress on $`L^r(V)`$. Although some use has been made of $`R(V)`$ in studying $`L(V)`$, results so far on the module structure of $`R^r(V)`$ are rather sparse: but see .
The main result of , called the ‘Decomposition Theorem’, reduces the study of arbitrary Lie powers in characteristic $`p`$ to the study of Lie powers of $`p`$-power degree. It states that, for each positive integer $`r`$, there is a submodule $`B_r`$ of $`L^r(V)`$ such that $`B_r`$ is a direct summand of $`T^r(V)`$ and, for $`k`$ not divisible by $`p`$ and $`m0`$,
$$L^{p^mk}(V)=L^{p^m}(B_k)L^{p^{m1}}(B_{pk})\mathrm{}L^1(B_{p^mk}).$$
In this paper we derive some consequences of the Decomposition Theorem.
Our first result, obtained in §2, is an analogous result for $`R(V)`$. With the modules $`B_r`$ and $`k`$ and $`m`$ as before, we find that
$$R^{p^mk}(V)=R^{p^m}(B_k)R^{p^{m1}}(B_{pk})\mathrm{}R^1(B_{p^mk}).$$
In the remainder of the paper we turn to applications in the Green ring $`R_{FG}`$. This is the ring spanned by the isomorphism classes of finite-dimensional $`FG`$-modules with addition and multiplication coming from direct sums and tensor products, respectively. In §3 we describe the general framework which may be used to study Lie powers and symmetric powers of finite-dimensional $`FG`$-modules. This is mainly a summary of material contained in . In particular we describe the relevance of the Adams operations and the Lie resolvents. The latter are $``$-linear maps $`\mathrm{\Phi }^r:R_{FG}R_{FG}`$. Knowledge of these maps is essentially equivalent to knowledge of the isomorphism classes of Lie powers of finite-dimensional $`FG`$-modules. We also introduce certain functions $`𝐋^{}`$ and $`𝐒^{}`$ defined on formal power series in an indeterminate $`t`$ with coefficients from $`R_{FG}`$. Here $`𝐋^{}`$ encodes Lie powers and $`𝐒^{}`$ encodes symmetric powers. Furthermore, $`𝐋^{}`$ and $`𝐒^{}`$ have properties like those of the logarithm function and exponential function, respectively.
In §4 we obtain the main result of the paper, called the ‘Factorisation Theorem’. It gives a ‘factorisation’ of Lie resolvents under composition: for every non-negative integer $`m`$ and every positive integer $`k`$ not divisible by $`p`$, $`\mathrm{\Phi }^{p^mk}=\mathrm{\Phi }^{p^m}\mathrm{\Phi }^k`$. The Lie resolvents $`\mathrm{\Phi }^k`$ for $`k`$ not divisible by $`p`$ are comparatively well understood, so the result can be interpreted as a reduction to the case of $`p`$-power degree. The Factorisation Theorem was conjectured in and proved in in the special case $`m=1`$. It is interesting that the Witt polynomials (as used to define the operations on Witt vectors) arise here in connection with the Factorisation Theorem.
Finally, in §5, we describe relations between $`𝐋^{}`$ and $`𝐒^{}`$. In characteristic $`p`$ we use the Factorisation Theorem to show that the composite function $`𝐋^{}𝐒^{}`$ takes any power series of the form $`Vt`$ to a power series where the coefficient of $`t^r`$ is $`0`$ unless $`r`$ is a power of $`p`$. A power series of the latter form is called ‘$`p`$-typical’. It follows that $`𝐋^{}𝐒^{}`$ maps every $`p`$-typical power series to a $`p`$-typical power series. In characteristic $`0`$ we find that $`𝐋^{}`$ is the inverse of $`𝐒^{}`$. This is closely related to results of Joyal and Reutenauer .
2. Decomposition of Lie powers and restricted Lie powers
Let $`V`$ be a finite-dimensional $`FG`$-module, where $`F`$ is a field and $`G`$ is a group. We write $`T(V)`$ for the free associative algebra freely generated by any $`F`$-basis of $`V`$. Thus $`T(V)`$ has an $`F`$-space decomposition $`T(V)=_{r0}T^r(V)`$, where, for each $`r`$, $`T^r(V)`$ is the $`r`$th homogeneous component of $`T(V)`$ and $`T^1(V)`$ is identified with $`V`$. The action of $`G`$ on $`V`$ extends to $`T(V)`$ so that $`G`$ acts by algebra automorphisms. Thus $`T(V)`$ becomes an $`FG`$-module, and each $`T^r(V)`$ is a finite-dimensional submodule.
The algebra $`T(V)`$ may be made into a Lie algebra by defining $`[a,b]=abba`$ for all $`a,bT(V)`$. If $`F`$ has prime characteristic $`p`$ then $`T(V)`$ may be made into a restricted Lie algebra by taking the additional powering operation $`aa^p`$. The Lie subalgebra of $`T(V)`$ generated by $`V`$ is denoted by $`L(V)`$ and (when $`F`$ has characteristic $`p`$) the restricted Lie subalgebra of $`T(V)`$ generated by $`V`$ is denoted by $`R(V)`$. As is well known, $`L(V)`$ is a free Lie algebra and $`R(V)`$ is a free restricted Lie algebra, both freely generated by any basis of $`V`$ (see, for example, \[13, Sections 1.2, 1.6.3 and 2.5.2\]). For $`r1`$ we write $`L^r(V)=T^r(V)L(V)`$ and $`R^r(V)=T^r(V)R(V)`$. Thus $`L(V)=_{r1}L^r(V)`$ and $`R(V)=_{r1}R^r(V)`$. Here $`L^r(V)`$ and $`R^r(V)`$ are $`FG`$-submodules of $`T^r(V)`$, called, respectively, the $`r`$th Lie power and $`r`$th restricted Lie power of $`V`$.
As observed in \[4, §2\], if $`B`$ is a submodule of $`T^r(V)`$, the associative subalgebra of $`T(V)`$ generated by $`B`$ may be identified with $`T(B)`$. The Lie subalgebra of $`T(V)`$ generated by $`B`$ is then the free Lie algebra $`L(B)`$ and the restricted Lie subalgebra of $`T(V)`$ generated by $`B`$ is the free restricted Lie algebra $`R(B)`$. We assume such identifications in the next two theorems. The first is \[4, Theorem 4.4\].
Theorem 2.1 (Decomposition Theorem) Let $`F`$ be a field of prime characteristic $`p`$. Let $`G`$ be a group and $`V`$ a finite-dimensional $`FG`$-module. For each positive integer $`r`$ there is a submodule $`B_r`$ of $`L^r(V)`$ such that $`B_r`$ is a direct summand of $`T^r(V)`$ and, for $`k`$ not divisible by $`p`$ and $`m0`$,
$$L^{p^mk}(V)=L^{p^m}(B_k)L^{p^{m1}}(B_{pk})\mathrm{}L^1(B_{p^mk}).$$
(2.1)
We use this to prove a similar result for restricted Lie powers.
Theorem 2.2 In the notation of Theorem 2.1,
$$R^{p^mk}(V)=R^{p^m}(B_k)R^{p^{m1}}(B_{pk})\mathrm{}R^1(B_{p^mk}).$$
Proof: For any subset $`𝒮`$ of $`L(V)`$ and $`i0`$ we write $`𝒮^{[p^i]}`$ for the set $`\{s^{p^i}:s𝒮\}`$. We shall use the fact that if $`𝒳`$ is any basis of $`L(V)`$ then the elements $`x^{p^i}`$ with $`x𝒳`$ and $`i0`$ are distinct and form a basis of $`R(V)`$: they are linearly independent by the Poincaré–Birkhoff–Witt Theorem and it is easy to see that they span $`R(V)`$.
For each non-negative integer $`i`$ and each positive integer $`k`$ not divisible by $`p`$, let $`𝒳(i,k)`$ be a basis of $`L^{p^ik}(V)`$. Hence $`_{i,k}𝒳(i,k)`$ is a basis of $`L(V)`$. Here, and in the remainder of the proof, all unions are unions of disjoint sets. Furthermore, write
$$\widehat{𝒳}(i,k)=𝒳(0,k)^{[p^i]}𝒳(1,k)^{[p^{i1}]}\mathrm{}𝒳(i,k)^{[1]}.$$
(2.2)
Considerations of degree in the basis of $`R(V)`$ obtained from $`_{i,k}𝒳(i,k)`$ show that $`\widehat{𝒳}(i,k)`$ is a basis of $`R^{p^ik}(V)`$.
For each triple $`(i,j,k)`$ where $`i`$ and $`j`$ are non-negative integers and $`k`$ is a positive integer not divisible by $`p`$, let $`𝒴(i,j,k)`$ be any basis of $`L^{p^i}(B_{p^jk})`$ and write
$$\widehat{𝒴}(i,j,k)=𝒴(0,j,k)^{[p^i]}𝒴(1,j,k)^{[p^{i1}]}\mathrm{}𝒴(i,j,k)^{[1]}.$$
(2.3)
Thus $`\widehat{𝒴}(i,j,k)`$ is a basis of $`R^{p^i}(B_{p^jk})`$.
By the Decomposition Theorem, we can choose the basis $`𝒳(i,k)`$ of $`L^{p^ik}(V)`$ to satisfy
$$𝒳(i,k)=𝒴(i,0,k)𝒴(i1,1,k)\mathrm{}𝒴(0,i,k).$$
(2.4)
Let $`m`$ be a non-negative integer. It is easily verified from (2.2) and (2.4) that $`\widehat{𝒳}(m,k)`$ is the union of the sets $`𝒴(i,j,k)^{[p^r]}`$ with $`i+j+r=m`$. Hence, by (2.3),
$$\widehat{𝒳}(m,k)=\widehat{𝒴}(m,0,k)\widehat{𝒴}(m1,1,k)\mathrm{}\widehat{𝒴}(0,m,k).$$
Since $`\widehat{𝒳}(m,k)`$ spans $`R^{p^mk}(V)`$ and $`\widehat{𝒴}(i,mi,k)`$ spans $`R^{p^i}(B_{p^{mi}k})`$, we obtain the required result. $`\mathrm{}`$
3. Symmetric powers and Lie powers
The underlying ideas that we use are described in . However, we shall make the treatment here as self-contained as possible. We also formulate some of the ideas in a slightly new way, in order to emphasise the analogies between symmetric powers and Lie powers. Let $`G`$ be a group and $`F`$ a field. Let $`\mathrm{Mod}(FG)`$ be the class of all finite-dimensional $`FG`$-modules and let $`R_{FG}`$ be the associated Green ring (or representation ring), as defined in §1.
The $``$-algebra $`_{}R_{FG}`$ will be denoted by $`\mathrm{\Gamma }_{FG}`$ and we regard $`R_{FG}`$ as a subset of $`\mathrm{\Gamma }_{FG}`$. (In , $``$ was used instead of $``$. But, for what is needed here, $``$ works as well as $``$.)
Let $`\mathrm{\Pi }`$ denote the power series ring over $`\mathrm{\Gamma }_{FG}`$ in an indeterminate $`t`$; that is, $`\mathrm{\Pi }=\mathrm{\Gamma }_{FG}[[t]]`$. Thus $`t\mathrm{\Pi }`$ and $`1+t\mathrm{\Pi }`$ are the subsets consisting of all power series with constant term $`0`$ and $`1`$, respectively. For $`ft\mathrm{\Pi }`$ and $`g1+t\mathrm{\Pi }`$ we may define $`\mathrm{exp}(f)1+t\mathrm{\Pi }`$ and $`\mathrm{log}(g)t\mathrm{\Pi }`$ by
$$\mathrm{exp}(f)=1+f+f^2/2!+f^3/3!+\mathrm{},$$
$$\mathrm{log}(g)=(g1)(g1)^2/2+(g1)^3/3\mathrm{}.$$
Thus we have mutually inverse functions
$$\mathrm{exp}:t\mathrm{\Pi }1+t\mathrm{\Pi },\mathrm{log}:1+t\mathrm{\Pi }t\mathrm{\Pi }.$$
If $`V\mathrm{Mod}(FG)`$ we also write $`V`$ for the element of $`R_{FG}`$ or $`\mathrm{\Gamma }_{FG}`$ determined by $`V`$. Thus, for example, $`T^r(V)`$ as an element of $`R_{FG}`$ denotes the isomorphism class of the module $`T^r(V)`$ and it is equal to $`V^r`$.
We start with a discussion of symmetric powers and Adams operations. Let $`V\mathrm{Mod}(FG)`$. We write $`S(V)`$ for the polynomial algebra (free commutative associative algebra) over $`F`$ freely generated by any basis of $`V`$. The action of $`G`$ on $`V`$ extends to $`S(V)`$ so that $`G`$ acts by algebra automorphisms. For $`r0`$, the $`r`$th homogeneous component $`S^r(V)`$ is an $`FG`$-submodule of $`S(V)`$ called the $`r`$th symmetric power of $`V`$. Here $`S^1(V)`$ is identified with $`V`$.
For $`V\mathrm{Mod}(FG)`$, let $`S(V,t)1+t\mathrm{\Pi }`$ be defined by
$$S(V,t)=1+S^1(V)t+S^2(V)t^2+\mathrm{}.$$
It is well known and easy to verify that for $`U,V\mathrm{Mod}(FG)`$ we have
$$S(UV,t)=S(U,t)S(V,t).$$
It follows that there is a $``$-linear function $`\psi :\mathrm{\Gamma }_{FG}t\mathrm{\Pi }`$ satisfying
$$\psi (V)=\mathrm{log}(S(V,t))$$
(3.1)
for all $`V\mathrm{Mod}(FG)`$. Hence we may define a function
$$𝐒^+:t\mathrm{\Pi }t\mathrm{\Pi }$$
by
$$𝐒^+(A_1t+A_2t^2+A_3t^3+\mathrm{})=\psi (A_1)+\psi (A_2)_{tt^2}+\psi (A_3)_{tt^3}+\mathrm{},$$
(3.2)
for all $`A_1,A_2,\mathrm{}\mathrm{\Gamma }_{FG}`$, where the subscript $`tt^r`$ denotes the operation of replacing a power series $`X_1t+X_2t^2+\mathrm{}`$ by $`X_1t^r+X_2t^{2r}+\mathrm{}`$. The properties of $`𝐒^+`$ given in the following lemma are readily obtained from the definition.
Lemma 3.1 The function $`𝐒^+`$ is $``$-linear and $`𝐒^+(t^r\mathrm{\Pi })t^r\mathrm{\Pi }`$, for all $`r1`$. For all $`ft\mathrm{\Pi }`$ and all $`r1`$, $`𝐒^+(f_{tt^r})=𝐒^+(f)_{tt^r}`$. If $`f_it^i\mathrm{\Pi }`$ for $`i=1,2,\mathrm{}`$, then $`𝐒^+(f_i)=𝐒^+(f_i)`$.
We now define a function
$$𝐒^{}:t\mathrm{\Pi }1+t\mathrm{\Pi }$$
as the composite
$$𝐒^{}=\mathrm{exp}𝐒^+.$$
(3.3)
It is clear from (3.1) and (3.2) that $`𝐒^{}(Vt)=S(V,t)`$ for all $`V\mathrm{Mod}(FG)`$. Also, since $`𝐒^+`$ is additive,
$$𝐒^{}(f+g)=𝐒^{}(f)𝐒^{}(g)$$
for all $`f,gt\mathrm{\Pi }`$.
For $`A\mathrm{\Gamma }_{FG}`$ we may define $`\psi ^1(A),\psi ^2(A),\mathrm{}\mathrm{\Gamma }_{FG}`$ by the equation
$$𝐒^+(At)=\psi (A)=\psi ^1(A)t+\frac{1}{2}\psi ^2(A)t^2+\frac{1}{3}\psi ^3(A)t^3+\mathrm{}.$$
(3.4)
Thus we obtain $``$-linear functions $`\psi ^r:\mathrm{\Gamma }_{FG}\mathrm{\Gamma }_{FG}`$. These functions were denoted by $`\psi _S^r`$ in \[1, §5\]: they are the Adams operations on $`\mathrm{\Gamma }_{FG}`$ determined by symmetric powers. (Adams operations, different in general, can also be defined using exterior powers.) By \[1, §5\], if $`V\mathrm{Mod}(FG)`$ then $`\psi ^r(V)R_{FG}`$. Thus $`\psi ^r`$ restricts to a $``$-linear function $`\psi ^r:R_{FG}R_{FG}`$.
If $`V\mathrm{Mod}(FG)`$ then, by (3.1) and (3.4),
$$\mathrm{log}(S(V,t))=\psi ^1(V)t+\frac{1}{2}\psi ^2(V)t^2+\mathrm{}.$$
This equation shows that symmetric powers may be expressed in terms of Adams operations and vice versa.
Lemma 3.2 If $`r`$ and $`s`$ are positive integers such that $`r`$ is not divisible by the characteristic of $`F`$ then $`\psi ^r`$ is an algebra endomorphism of $`\mathrm{\Gamma }_{FG}`$ and
$$\psi ^{rs}=\psi ^r\psi ^s.$$
In particular, if $`F`$ has prime characteristic $`p`$ then
$$\psi ^{p^mk}=\psi ^k\psi ^{p^m}$$
for all non-negative integers $`m`$ and all positive integers $`k`$ not divisible by $`p`$.
Proof: See \[1, Theorem 5.4\]. $`\mathrm{}`$
All the results stated above for symmetric powers have analogues for exterior powers. However, we shall not need exterior powers for the applications in this paper.
We now turn to Lie powers and Lie resolvents. We shall summarise the general theory described in and formulate some of it in a slightly different way. The theory is based on a function
$$_{FG}:t\mathrm{\Pi }t\mathrm{\Pi }$$
called the Lie module function. When $`F`$ and $`G`$ are understood we write $``$ instead of $`_{FG}`$. The properties of this function are described in \[1, §2 and §3\]. In particular, by \[1, Lemma 3.1\],
$$(Vt^r)=L^1(V)t^r+L^2(V)t^{2r}+\mathrm{},$$
(3.5)
for all $`V\mathrm{Mod}(FG)`$ and $`r1`$, and
$$(f)+(g)=(f+gfg),$$
(3.6)
for all $`f,gt\mathrm{\Pi }`$.
There are advantages in considering a modification of $``$, namely the function
$$𝐋^{}:1+t\mathrm{\Pi }t\mathrm{\Pi }$$
defined by
$$𝐋^{}(f)=(1f)$$
(3.7)
for all $`f1+t\mathrm{\Pi }`$. It follows from (3.5), (3.6) and (3.7) that
$$𝐋^{}(1Vt)=L^1(V)t+L^2(V)t^2+\mathrm{},$$
for all $`V\mathrm{Mod}(FG)`$, and
$$𝐋^{}(fg)=𝐋^{}(f)+𝐋^{}(g),$$
for all $`f,g1+t\mathrm{\Pi }`$.
We may define a function
$$𝐋^+:t\mathrm{\Pi }t\mathrm{\Pi }$$
by $`𝐋^+=𝐋^{}\mathrm{exp}`$. Thus $`𝐋^+`$ is additive and
$$𝐋^{}=𝐋^+\mathrm{log}.$$
(3.8)
Let $`\mathrm{Exp}:t\mathrm{\Pi }t\mathrm{\Pi }`$ be defined by $`\mathrm{Exp}(f)=1\mathrm{exp}(f)`$ for all $`ft\mathrm{\Pi }`$. Since
$$𝐋^+(f)=𝐋^+(f)=𝐋^{}(\mathrm{exp}(f))=(1\mathrm{exp}(f)),$$
we obtain
$$𝐋^+=\mathrm{Exp}.$$
(3.9)
Additional properties of $`𝐋^+`$ are given in \[1, §3\], where $`𝐋^+`$ is denoted by $`_{FG}`$. These properties yield the following result.
Lemma 3.3 The function $`𝐋^+`$ has the properties of $`𝐒^+`$ stated in Lemma 3.1.
In particular $`𝐋^+`$ is $``$-linear.
For $`A\mathrm{\Gamma }_{FG}`$ we may define $`\mathrm{\Phi }^1(A),\mathrm{\Phi }^2(A),\mathrm{}\mathrm{\Gamma }_{FG}`$ by the equation
$$𝐋^+(At)=\mathrm{\Phi }^1(A)t+\frac{1}{2}\mathrm{\Phi }^2(A)t^2+\frac{1}{3}\mathrm{\Phi }^3(A)t^3+\mathrm{}.$$
(3.10)
Thus we obtain $``$-linear functions $`\mathrm{\Phi }^r:\mathrm{\Gamma }_{FG}\mathrm{\Gamma }_{FG}`$. These functions were considered in , where they were written $`\mathrm{\Phi }_{FG}^r`$ and called the Lie resolvents for $`G`$ over $`F`$. They restrict to $``$-linear functions $`\mathrm{\Phi }^r:R_{FG}R_{FG}`$. Lie powers may be expressed in terms of Lie resolvents and vice versa by means of the following expressions given in \[1, Corollary 3.3\]: for all $`V\mathrm{Mod}(FG)`$ and every positive integer $`r`$,
$$L^r(V)=\frac{1}{r}\underset{dr}{}\mathrm{\Phi }^d(V^{r/d})$$
(3.11)
and
$$\mathrm{\Phi }^r(V)=\underset{dr}{}\mu (r/d)dL^d(V^{r/d}),$$
(3.12)
where $`\mu `$ denotes the Möbius function.
We conclude this section with two further lemmas.
Lemma 3.4 If $`r`$ is not divisible by the characteristic of $`F`$, then
$$\mathrm{\Phi }^r=\mu (r)\psi ^r.$$
Proof: See \[1, Corollary 6.2\]. $`\mathrm{}`$
Lemma 3.5 Let $`ft\mathrm{\Pi }`$, where $`f=A_1t+A_2t^2+\mathrm{}`$. Then
$$𝐋^+(f)=\underset{r1}{}\left(\underset{dr}{}\frac{1}{d}\mathrm{\Phi }^d(A_{r/d})\right)t^r$$
and
$$𝐒^+(f)=\underset{r1}{}\left(\underset{dr}{}\frac{1}{d}\psi ^d(A_{r/d})\right)t^r.$$
Proof: By (3.10) and (3.4), and in the terminology of \[1, §2\], the functions $`\frac{1}{r}\mathrm{\Phi }^r`$ and $`\frac{1}{r}\psi ^r`$ are the ‘components’ of $`𝐋^+`$ and $`𝐒^+`$, respectively. Hence the result follows from \[1, (2.5)\]. $`\mathrm{}`$
4. The Factorisation Theorem
Let $`F`$ be an infinite field and $`n`$ a positive integer. We write $`G(n)=\mathrm{GL}(n,F)`$. We require some facts about polynomial modules and their characters. Most of these have already been collected in \[1, §5\], so we use this as a convenient reference: but see also . As in \[1, §5\], let $`R_{FG(n)}^{\mathrm{poly}}`$ denote the subring of the Green ring $`R_{FG(n)}`$ spanned by all the isomorphism classes of finite-dimensional polynomial modules. It is clear from (3.12) that, for each $`r`$, the Lie resolvent $`\mathrm{\Phi }^r`$ restricts to a map $`\mathrm{\Phi }^r:R_{FG(n)}^{\mathrm{poly}}R_{FG(n)}^{\mathrm{poly}}`$.
Let $`t_1,\mathrm{},t_n`$ be indeterminates and let $`\mathrm{\Delta }`$ be the subring of $`[t_1,\mathrm{},t_n]`$ consisting of all symmetric polynomials. For each positive integer $`r`$, the endomorphism of $`[t_1,\mathrm{},t_n]`$ satisfying $`t_it_i^r`$ for all $`i`$ restricts to an endomorphism of $`\mathrm{\Delta }`$, which we denote by $`\chi ^r`$. As explained in \[1, §5\], there is a ring homomorphism $`\mathrm{ch}:R_{FG(n)}^{\mathrm{poly}}\mathrm{\Delta }`$ such that, for every finite-dimensional polynomial $`FG(n)`$-module $`U`$, $`\mathrm{ch}(U)`$ is the formal character of $`U`$.
As stated in \[6, §3.2\], if $`V`$ is the natural $`FG(n)`$-module then
$$\mathrm{ch}(L^r(V))=\frac{1}{r}\underset{dr}{}\mu (d)(t_1^d+\mathrm{}+t_n^d)^{r/d}.$$
(4.1)
(The right-hand side is formally an element of $`_{}\mathrm{\Delta }`$, but is found to belong to $`\mathrm{\Delta }`$.) Suppose that $`U`$ is a finite-dimensional polynomial $`FG(n)`$-module. Then we may write $`\mathrm{ch}(U)=w_1+\mathrm{}+w_m`$ where $`m=dimU`$ and $`w_1,\mathrm{},w_m`$ are monomials in $`t_1,\mathrm{},t_n`$. We may choose a basis of $`U`$ consisting of elements from weight spaces. Then every diagonal element of $`G(n)`$ is represented on $`U`$ by a diagonal element of $`\mathrm{GL}(m,F)`$. By (4.1) with $`m`$ instead of $`n`$, we obtain
$$\mathrm{ch}(L^r(U))=\frac{1}{r}\underset{dr}{}\mu (d)(w_1^d+\mathrm{}+w_m^d)^{r/d}.$$
Hence
$$\mathrm{ch}(L^r(U))=\frac{1}{r}\underset{dr}{}\mu (d)\chi ^d(\mathrm{ch}(U^{r/d})).$$
(4.2)
Lemma 4.1 Let $`U`$ be a finite-dimensional polynomial $`FG(n)`$-module.
(i) For all $`r1`$, $`\mathrm{ch}(\mathrm{\Phi }^r(U))=\mu (r)\chi ^r(\mathrm{ch}(U))`$.
(ii) If $`(r,s)=1`$ then
$$\mathrm{ch}(\mathrm{\Phi }^{rs}(U))=\mathrm{ch}(\mathrm{\Phi }^r(\mathrm{\Phi }^s(U))).$$
Proof: By (3.11),
$$\mathrm{ch}(L^r(U))=\frac{1}{r}\underset{dr}{}\mathrm{ch}(\mathrm{\Phi }^d(U^{r/d})).$$
Comparing this with (4.2) and using induction on $`r`$ gives (i). Hence, for all $`WR_{FG(n)}^{\mathrm{poly}}`$,
$$\mathrm{ch}(\mathrm{\Phi }^r(W))=\mu (r)\chi ^r(\mathrm{ch}(W)).$$
(4.3)
Therefore
$`\mathrm{ch}(\mathrm{\Phi }^r(\mathrm{\Phi }^s(U)))`$ $`=`$ $`\mu (r)\chi ^r(\mathrm{ch}(\mathrm{\Phi }^s(U)))`$
$`=`$ $`\mu (r)\chi ^r(\mu (s)\chi ^s(\mathrm{ch}(U))).`$
However, $`\chi ^r\chi ^s=\chi ^{rs}`$ and, if $`r`$ and $`s`$ are coprime, $`\mu (r)\mu (s)=\mu (rs)`$. Therefore
$$\mathrm{ch}(\mathrm{\Phi }^r(\mathrm{\Phi }^s(U)))=\mu (rs)\chi ^{rs}(\mathrm{ch}(U)).$$
Hence, by (4.3), $`\mathrm{ch}(\mathrm{\Phi }^r(\mathrm{\Phi }^s(U)))=\mathrm{ch}(\mathrm{\Phi }^{rs}(U))`$, as required for (ii). $`\mathrm{}`$
We recall the Decomposition Theorem (Theorem 2.1). The submodules $`B_r`$ are not uniquely determined by $`V`$, as easy examples show. However, it follows from equations (2.1), by induction, that the $`B_r`$ are uniquely determined up to isomorphism. Thus they give uniquely determined elements of $`R_{FG}`$. We shall deduce the following result from the Decomposition Theorem.
Theorem 4.2 (Factorisation Theorem) Let $`F`$ be a field of prime characteristic $`p`$. Let $`G`$ be a group. For every non-negative integer $`m`$ and every positive integer $`k`$ not divisible by $`p`$,
$$\mathrm{\Phi }^{p^mk}=\mathrm{\Phi }^{p^m}\mathrm{\Phi }^k$$
(4.4)
and, for every finite-dimensional $`FG`$-module $`V`$, the elements $`B_r`$ of $`R_{FG}`$ obtained from the Decomposition Theorem satisfy
$$p^mB_{p^mk}+p^{m1}B_{p^{m1}k}^p+\mathrm{}+pB_{pk}^{p^{m1}}+B_k^{p^m}=L^k(V^{p^m}).$$
(4.5)
Proof: When we wish to show the dependence on $`V`$ we write $`B_r`$ as $`B^r(V)`$, and when we wish to show the role of $`F`$ and $`G`$ in $`\mathrm{\Phi }^r`$ we write it as $`\mathrm{\Phi }_{FG}^r`$. We first reduce to the case where $`F`$ is infinite. Let $`E`$ be an infinite extension field of $`F`$. Assume that the theorem holds with $`E`$ in place of $`F`$. Each $`FG`$-module $`V`$ determines an $`EG`$-module $`E_FV`$ by extension of scalars. Thus we obtain a ring homomorphism $`\iota :R_{FG}R_{EG}`$. It follows from the Noether–Deuring Theorem (see \[5, (29.7)\]) that $`\iota `$ is an embedding. By \[3, Lemma 2.4\], for all $`r1`$ and all $`V\mathrm{Mod}(FG)`$, we have $`L^r(\iota (V))=\iota (L^r(V))`$ and $`\mathrm{\Phi }_{EG}^r\iota =\iota \mathrm{\Phi }_{FG}^r`$.
By (4.4) over $`E`$, $`\mathrm{\Phi }_{EG}^{p^mk}\iota =\mathrm{\Phi }_{EG}^{p^m}\mathrm{\Phi }_{EG}^k\iota `$ and hence $`\iota \mathrm{\Phi }_{FG}^{p^mk}=\iota \mathrm{\Phi }_{FG}^{p^m}\mathrm{\Phi }_{FG}^k`$. Since $`\iota `$ is an embedding we obtain (4.4) over $`F`$. Let $`V\mathrm{Mod}(FG)`$. Applying $`\iota `$ to equations (2.1) over $`F`$ we find that $`\iota (B^r(V))=B^r(\iota (V))`$ for all $`r`$. By (4.5) over $`E`$ for the module $`\iota (V)`$,
$$p^mB^{p^mk}(\iota (V))+p^{m1}(B^{p^{m1}k}(\iota (V)))^p+\mathrm{}+(B^k(\iota (V)))^{p^m}=L^k(\iota (V)^{p^m}).$$
Thus
$$\iota (p^mB^{p^mk}(V)+\mathrm{}+(B^k(V))^{p^m})=\iota (L^k(V^{p^m})).$$
Since $`\iota `$ is an embedding we obtain equation (4.5) over $`F`$.
Hence it is enough to prove the theorem over $`E`$. In other words we may assume that $`F`$ is infinite.
We use induction on $`m`$ and $`k`$. If $`m=0`$ the result is clear because $`\mathrm{\Phi }^1`$ is the identity map and, for all $`V`$, $`B^k(V)=L^k(V)`$, by (2.1). Hence we may assume that $`m1`$ and that the result holds for $`p^id`$ with $`d`$ not divisible by $`p`$ if $`i<m`$ or $`i=m`$ and $`d<k`$.
Since the functions $`\mathrm{\Phi }^r`$ are linear, in order to prove (4.4) it suffices to prove
$$\mathrm{\Phi }^{p^mk}(V)=(\mathrm{\Phi }^{p^m}\mathrm{\Phi }^k)(V)$$
(4.6)
for all $`V\mathrm{Mod}(FG)`$.
Let $`V\mathrm{Mod}(FG)`$ and $`n=dimV`$. By choice of a basis of $`V`$, the representation of $`G`$ on $`V`$ gives a homomorphism $`\theta :GG(n)`$. Furthermore, $`\theta `$ induces a ring homomorphism $`\theta ^{}:R_{FG(n)}R_{FG}`$. We can regard $`V`$ as the natural module for $`G(n)`$, in which case we write it as $`V(n)`$. Thus $`\theta ^{}(V(n))=V`$. We now show that it is enough to prove the required results for $`G(n)`$ and $`V(n)`$. Suppose that (4.6) and (4.5) hold for $`G(n)`$ and $`V(n)`$. By \[3, Lemmas 2.2 and 2.3\], for all $`r1`$ and all $`U\mathrm{Mod}(FG(n))`$, we have $`L^r(\theta ^{}(U))=\theta ^{}(L^r(U))`$ and $`\theta ^{}\mathrm{\Phi }_{FG(n)}^r=\mathrm{\Phi }_{FG}^r\theta ^{}`$.
Thus, by applying $`\theta ^{}`$ to (4.6) for $`G(n)`$ and $`V(n)`$, we get (4.6) for $`G`$ and $`V`$. By applying $`\theta ^{}`$ to equations (2.1) for $`G(n)`$ and $`V(n)`$ we get $`\theta ^{}(B^r(V(n)))=B^r(V)`$ for all $`r`$. Hence, by applying $`\theta ^{}`$ to (4.5) for $`G(n)`$ and $`V(n)`$, we get (4.5) for $`G`$ and $`V`$.
Thus it remains to prove (4.6) and (4.5) for $`G(n)`$ and $`V(n)`$. To simplify the notation, write $`V`$ for $`V(n)`$ and $`B_r`$ for $`B^r(V(n))`$.
As in §3, let $`\mathrm{\Pi }=\mathrm{\Gamma }_{FG(n)}[[t]]`$. Let $`\mathrm{Log}:t\mathrm{\Pi }t\mathrm{\Pi }`$ be defined by
$$\mathrm{Log}(f)=\mathrm{log}(1f)=f+f^2/2+f^3/3+\mathrm{}$$
for all $`ft\mathrm{\Pi }`$. Thus $`\mathrm{Log}`$ is the inverse of the function $`\mathrm{Exp}`$ defined in §3.
For $`ft\mathrm{\Pi }`$ and $`r1`$, we write $`f_{(r)}`$ for the coefficient of $`t^r`$ in $`f`$; thus $`f_{(r)}\mathrm{\Gamma }_{FG(n)}`$ and $`f=_{r1}f_{(r)}t^r`$.
Let $`Q^{}=_{r0}\mathrm{Log}(B_{p^rk}t^{p^r})`$, $`Q^{\prime \prime }=_{i0}p^iL^k(V^{p^i})t^{p^i}`$ and $`Q=Q^{}Q^{\prime \prime }`$. Then it is easily calculated that
$$Q_{(p^i)}=\frac{1}{p^i}B_k^{p^i}+\frac{1}{p^{i1}}B_{pk}^{p^{i1}}+\mathrm{}+B_{p^ik}\frac{1}{p^i}L^k(V^{p^i}).$$
By the inductive hypothesis, $`Q_{(p^i)}=0`$ for $`i<m`$. Hence, by Lemma 3.5, we obtain
$`𝐋^+(Q)_{(p^m)}`$ $`=`$ $`\mathrm{\Phi }^1(Q_{(p^m)})=Q_{(p^m)}`$
$`=`$ $`{\displaystyle \frac{1}{p^m}}B_k^{p^m}+{\displaystyle \frac{1}{p^{m1}}}B_{pk}^{p^{m1}}+\mathrm{}+B_{p^mk}{\displaystyle \frac{1}{p^m}}L^k(V^{p^m}).`$
Since $`𝐋^+`$ is additive and $`𝐋^+\mathrm{Log}=`$, by (3.9), we have
$$𝐋^+(Q^{})=\underset{r0}{}(B_{p^rk}t^{p^r}).$$
Thus, by (3.5),
$$𝐋^+(Q^{})_{(p^m)}=L^{p^m}(B_k)+L^{p^{m1}}(B_{pk})+\mathrm{}+L^1(B_{p^mk}).$$
Therefore, by the Decomposition Theorem,
$$𝐋^+(Q^{})_{(p^m)}=L^{p^mk}(V).$$
By Lemma 3.5,
$$𝐋^+(Q^{\prime \prime })_{(p^m)}=\frac{1}{p^m}\underset{dp^m}{}\mathrm{\Phi }^d(L^k(V^{p^m/d})).$$
Hence, by (3.11),
$`𝐋^+(Q^{\prime \prime })_{(p^m)}`$ $`=`$ $`{\displaystyle \frac{1}{p^m}}{\displaystyle \underset{dp^m}{}}\mathrm{\Phi }^d\left({\displaystyle \frac{1}{k}}{\displaystyle \underset{ek}{}}\mathrm{\Phi }^e(V^{p^mk/de})\right)`$
$`=`$ $`{\displaystyle \frac{1}{p^mk}}{\displaystyle \underset{dp^m,ek}{}}(\mathrm{\Phi }^d\mathrm{\Phi }^e)(V^{p^mk/de}).`$
We write
$$\mathrm{\Phi }^{p^m}\mathrm{\Phi }^k=\mathrm{\Phi }^{p^mk}+(\mathrm{\Phi }^{p^m}\mathrm{\Phi }^k\mathrm{\Phi }^{p^mk})$$
and use the inductive hypothesis to write $`\mathrm{\Phi }^d\mathrm{\Phi }^e=\mathrm{\Phi }^{de}`$ for $`de<p^mk`$. Thus
$`𝐋^+(Q^{\prime \prime })_{(p^m)}`$ $`=`$ $`{\displaystyle \frac{1}{p^mk}}{\displaystyle \underset{sp^mk}{}}\mathrm{\Phi }^s(V^{p^mk/s})+{\displaystyle \frac{1}{p^mk}}(\mathrm{\Phi }^{p^m}(\mathrm{\Phi }^k(V))\mathrm{\Phi }^{p^mk}(V))`$
$`=`$ $`L^{p^mk}(V)+{\displaystyle \frac{1}{p^mk}}(\mathrm{\Phi }^{p^m}(\mathrm{\Phi }^k(V))\mathrm{\Phi }^{p^mk}(V)).`$
Since $`𝐋^+(Q)_{(p^m)}=𝐋^+(Q^{})_{(p^m)}𝐋^+(Q^{\prime \prime })_{(p^m)}`$, we obtain
$$\frac{1}{p^m}B_k^{p^m}+\frac{1}{p^{m1}}B_{pk}^{p^{m1}}+\mathrm{}+B_{p^mk}\frac{1}{p^m}L^k(V^{p^m})=\frac{1}{p^mk}(\mathrm{\Phi }^{p^mk}(V)\mathrm{\Phi }^{p^m}(\mathrm{\Phi }^k(V))).$$
Therefore
$$\frac{1}{k}(\mathrm{\Phi }^{p^mk}(V)\mathrm{\Phi }^{p^m}(\mathrm{\Phi }^k(V)))=W,$$
where
$$W=p^mB_{p^mk}+p^{m1}B_{p^{m1}k}^p+\mathrm{}+pB_{pk}^{p^{m1}}+B_k^{p^m}L^k(V^{p^m}).$$
By Lemma 4.1, $`\mathrm{ch}(\mathrm{\Phi }^{p^mk}(V)\mathrm{\Phi }^{p^m}(\mathrm{\Phi }^k(V)))=0`$. Thus $`\mathrm{ch}(W)=0`$. By the Decomposition Theorem, the modules $`B_{p^mk}`$, $`B_{p^{m1}k}^p`$, …, $`B_k^{p^m}`$ are isomorphic to direct summands of the tensor power $`V^{p^mk}`$. Hence they are tilting modules: see \[1, §5\]. Since $`pk`$, $`L^k(V^{p^m})`$ is also a direct summand of $`V^{p^mk}`$ (see, for example, \[6, §3.1\]). Thus it too is a tilting module. However, tilting modules are determined up to isomorphism by their formal characters (see, for example, ). Since $`\mathrm{ch}(W)=0`$ it follows that $`W=0`$. Thus $`\mathrm{\Phi }^{p^mk}(V)\mathrm{\Phi }^{p^m}(\mathrm{\Phi }^k(V))=0`$. This gives (4.5) and (4.6). $`\mathrm{}`$
Note that equations (4.5) give a recursive description, within $`\mathrm{\Gamma }_{FG}`$, of $`B_{p^mk}`$ as a polynomial in $`L^k(V)`$, $`L^k(V^p)`$, …, $`L^k(V^{p^m})`$. The polynomials in $`B_k,B_{pk},B_{p^2k},\mathrm{}`$ occurring on the left-hand side of (4.5) may be recognised as the Witt polynomials associated to the prime $`p`$: see \[16, Chapter II, §6\] or \[11, Chapter 3\]. Thus, in the terminology often used in relation to Witt vectors, $`L^k(V)`$, $`L^k(V^p)`$, $`L^k(V^{p^2})`$, … are the ‘ghost’ components of $`(B_k,B_{pk},B_{p^2k},\mathrm{})`$. However, we have not yet been able to deduce from this more explicit information on the modules $`B_{p^mk}`$.
Of particular interest is the case where $`V`$ is the natural module for $`\mathrm{GL}(n,F)`$. Then, since the modules $`B_k,B_{pk},\mathrm{}`$ and $`L^k(V),L^k(V^p),\mathrm{}`$ are direct summands of tensor powers of $`V`$, they are determined, up to isomorphism, by their formal characters. We may therefore translate equations (4.5) into equations in symmetric functions. Closely related equations have been considered by Reutenauer and Scharf and Thibon .
Statement (4.4) in the Factorisation Theorem is analogous to the property of Adams operations given by the second part of Lemma 3.2. The Factorisation Theorem and Lemma 3.4 reduce the study of Lie resolvents to the study of Adams operations and $`p`$-power Lie resolvents.
5. Symmetric powers and Lie powers revisited
Let $`F`$ be a field and $`G`$ a group. For each positive integer $`r`$ we may define a function $`\rho ^r:\mathrm{\Gamma }_{FG}\mathrm{\Gamma }_{FG}`$ by
$$\rho ^r=\frac{1}{r}\underset{dr}{}\mathrm{\Phi }^d\psi ^{r/d}.$$
(5.1)
Thus, since $`\psi ^1`$ is the identity map, the Lie resolvents $`\mathrm{\Phi }^r`$ may be expressed recursively in terms of the functions $`\rho ^r`$ and the Adams operations. Hence, if we assume the Adams operations, knowledge of the functions $`\rho ^r`$ is equivalent to knowledge of the Lie resolvents.
Theorem 5.1 Let $`G`$ be a group and $`F`$ a field of prime characteristic $`p`$. Then $`\rho ^r=0`$ unless $`r`$ is a power of $`p`$.
Proof: Write $`r=p^mk`$ where $`m0`$ and $`k`$ is not divisible by $`p`$. By (5.1) and the Factorisation Theorem,
$`\rho ^r`$ $`=`$ $`{\displaystyle \frac{1}{r}}{\displaystyle \underset{dr}{}}\mathrm{\Phi }^d\psi ^{r/d}`$
$`=`$ $`{\displaystyle \frac{1}{r}}{\displaystyle \underset{0im}{}}{\displaystyle \underset{ek}{}}\mathrm{\Phi }^{p^ie}\psi ^{p^{mi}k/e}`$
$`=`$ $`{\displaystyle \frac{1}{r}}{\displaystyle \underset{0im}{}}{\displaystyle \underset{ek}{}}\mathrm{\Phi }^{p^i}\mathrm{\Phi }^e\psi ^{p^{mi}k/e}.`$
Hence, by Lemmas 3.4 and 3.2,
$`\rho ^r`$ $`=`$ $`{\displaystyle \frac{1}{r}}{\displaystyle \underset{0im}{}}{\displaystyle \underset{ek}{}}\mu (e)(\mathrm{\Phi }^{p^i}\psi ^e\psi ^{p^{mi}k/e})`$
$`=`$ $`{\displaystyle \frac{1}{r}}{\displaystyle \underset{0im}{}}{\displaystyle \underset{ek}{}}\mu (e)(\mathrm{\Phi }^{p^i}\psi ^{p^{mi}k}).`$
If $`r`$ is not a power of $`p`$ then $`k>1`$ and so $`_{ek}\mu (e)=0`$, which gives $`\rho ^r=0`$. $`\mathrm{}`$
Lemma 5.2 We have $`𝐋^{}𝐒^{}=𝐋^+𝐒^+`$ and, for all $`A\mathrm{\Gamma }_{FG}`$,
$$(𝐋^{}𝐒^{})(At)=\rho ^1(A)t+\rho ^2(A)t^2+\mathrm{}.$$
Proof: This result is essentially the same as \[2, Lemma 4.1\], but, for completeness, we give a proof.
By (3.3) and (3.8), $`𝐋^{}𝐒^{}=𝐋^+𝐒^+`$. Also, by (3.4) and Lemma 3.5,
$`𝐋^+(𝐒^+(At))`$ $`=`$ $`𝐋^+(\psi ^1(A)t+{\displaystyle \frac{1}{2}}\psi ^2(A)t^2+\mathrm{})`$
$`=`$ $`{\displaystyle \underset{r1}{}}\left({\displaystyle \underset{dr}{}}{\displaystyle \frac{1}{d}}\mathrm{\Phi }^d((d/r)\psi ^{r/d}(A))\right)t^r.`$
Hence, by the linearity of $`\mathrm{\Phi }^d`$ and (5.1),
$$(𝐋^{}𝐒^{})(At)=(𝐋^+𝐒^+)(At)=\underset{r1}{}\rho ^r(A)t^r.$$
$`\mathrm{}`$
Recall that $`\mathrm{\Pi }=\mathrm{\Gamma }_{FG}[[t]]`$. An element of $`t\mathrm{\Pi }`$ will be called $`p`$-typical if, for every positive integer $`r`$, the coefficient of $`t^r`$ is zero unless $`r`$ is a power of $`p`$.
Theorem 5.3 Let $`G`$ be a group and $`F`$ a field of prime characteristic $`p`$. Then $`(𝐋^{}𝐒^{})(f)`$ is $`p`$-typical for every $`p`$-typical element $`f`$ of $`t\mathrm{\Pi }`$.
Proof: By Lemma 5.2, $`𝐋^{}𝐒^{}=𝐋^+𝐒^+`$. By Theorem 5.1 and Lemma 5.2, $`(𝐋^+𝐒^+)(At)`$ is $`p`$-typical for all $`A\mathrm{\Gamma }_{FG}`$. By Lemmas 3.1 and 3.3, $`𝐋^+𝐒^+`$ has the properties of $`𝐒^+`$ stated in Lemma 3.1. It follows that $`(𝐋^+𝐒^+)(f)`$ is $`p`$-typical for every $`p`$-typical element $`f`$ of $`t\mathrm{\Pi }`$. $`\mathrm{}`$
The functions $`\rho ^r`$ seem to have some importance in the study of Lie powers. As we have already seen, they may be used instead of the Lie resolvents, but, on the available evidence, their properties seem to be smoother. Theorem 5.1 shows that $`\rho ^r=0`$ unless $`r`$ is a power of $`p`$. It is remarkable also that, in the cases which have so far been calculated, even the functions $`\rho ^1`$, $`\rho ^p`$, $`\rho ^{p^2}`$, … are well behaved. Of course, $`\rho ^1`$ is the identity function. It follows from \[2, Corollary 4.5 and Lemma 4.6\] that if $`G`$ is cyclic of order $`p`$ (and $`F`$ has characteristic $`p`$) then $`\rho ^{p^m}=0`$ for all $`m>1`$. In fact, from the results in and it is not difficult to deduce the same fact for every finite group $`G`$ such that $`p^2|G|`$. It is interesting to speculate on how far this generalises to other groups: perhaps, if $`G`$ is finite, we have $`\rho ^{p^m}=0`$ for all sufficiently large $`m`$.
We have used the Factorisation Theorem in the proof of Theorem 5.1 to obtain the fact that $`\rho ^r=0`$ unless $`r`$ is a power of $`p`$. However, by \[2, Lemma 5.1 (ii)\], this fact itself yields the Factorisation Theorem: hence it is ‘equivalent’ to the Factorisation Theorem.
We conclude with a few remarks about the easier case where $`F`$ has characteristic $`0`$. In this case, by Lemma 3.4, $`\mathrm{\Phi }^r=\mu (r)\psi ^r`$ for all $`r`$. By a simplified version of the calculation of Theorem 5.1 we obtain $`\rho ^r=0`$ for all $`r>1`$. Thus, by Lemma 5.2, $`(𝐋^+𝐒^+)(At)=At`$ for all $`A\mathrm{\Gamma }_{FG}`$. By the properties of $`𝐒^+`$ and $`𝐋^+`$ it follows that $`𝐋^+𝐒^+`$ is the identity function on $`t\mathrm{\Pi }`$. A similar calculation shows that $`𝐒^+𝐋^+`$ is the identity function. However, $`𝐋^{}𝐒^{}=𝐋^+𝐒^+`$ and
$$𝐒^{}𝐋^{}=\mathrm{exp}𝐒^+𝐋^+\mathrm{log}.$$
Thus we have the following result, closely related to results of Joyal and Reutenauer .
Theorem 5.4 Let $`G`$ be a group and $`F`$ a field of characteristic $`0`$. Then $`𝐋^{}𝐒^{}`$ is the identity function on $`t\mathrm{\Pi }`$ and $`𝐒^{}𝐋^{}`$ is the identity function on $`1+t\mathrm{\Pi }`$.
Recall that, by (3.7), $`𝐋^{}(f)=(1f)`$ for all $`f1+t\mathrm{\Pi }`$. Since $`𝐒^{}𝐋^{}`$ is the identity function in characteristic $`0`$, it follows easily that
$$𝐒^{}((g))=(1g)^1=1+g+g^2+\mathrm{}$$
for all $`gt\mathrm{\Pi }`$. This may be regarded as a version of the Poincaré–Birkhoff–Witt Theorem (by taking $`g=Vt`$) and it is similar to a result of Joyal \[12, Chapter 4, Proposition 1\] (see also \[14, Lemma 3.2\]). Similarly, the fact that $`𝐋^{}𝐒^{}`$ is the identity function is a version of a result of Reutenauer \[14, Theorem 3.1\].
References
* R. M. Bryant, “Free Lie algebras and Adams operations”, J. London Math. Soc. (2) 68 (2003), 355–370.
* R. M. Bryant, “Modular Lie representations of groups of prime order”, Math. Z. 246 (2004), 603–617.
* R. M. Bryant, “Modular Lie representations of finite groups”, J. Austral. Math. Soc. 77 (2004), 401–423.
* R. M. Bryant and M. Schocker, “The decomposition of Lie powers”, preprint arXiv:math.RT/0505325.
* C. W. Curtis and I. Reiner, Representation Theory of Finite Groups and Associative Algebras, Wiley–Interscience, New York, 1962.
* S. Donkin and K. Erdmann, “Tilting modules, symmetric functions, and the module structure of the free Lie algebra”, J. Algebra 203 (1998), 69–90.
* K. Erdmann and M. Schocker, “Modular Lie powers and the Solomon descent algebra”, preprint arXiv:math.RT/0408211.
* J. A. Green, Polynomial Representations of $`\mathrm{GL}_n`$, Lecture Notes in Mathematics 830, Springer, Berlin, 1980.
* S. Guilfoyle, On Lie Powers of Modules for Cyclic Groups Ph. D. thesis, Manchester, 2000.
* S. Guilfoyle and R. Stöhr, “Invariant bases for free Lie algebras”, J. Algebra 204 (1998), 337–346.
* M. Hazewinkel, Formal Groups and Applications, Academic Press, New York, 1978.
* A. Joyal, “Foncteurs analytiques et espèces de structures”, in Combinatoire Énumérative, edited by G. Labelle and P. Leroux, Lecture Notes in Mathematics 1234, Springer, Berlin, 1986, pp. 126–159.
* C. Reutenauer, Free Lie Algebras, Clarendon Press, Oxford, 1993.
* C. Reutenauer, “On symmetric functions related to Witt vectors and the free Lie algebra”, Adv. Math. 110 (1995), 234–246.
* T. Scharf and J.-Y. Thibon, “On Witt vectors and symmetric functions”, Algebra Colloq. 3 (1996), 231–238.
* J.-P. Serre, Local Fields, Springer, New York, 1979.
* R. Stöhr, “Restricted Lazard elimination and modular Lie powers”, J. Austral. Math. Soc. 71 (2001), 259–277.
|
warning/0506/astro-ph0506152.html
|
ar5iv
|
text
|
# Nearby stars from the LSPM-north Proper Motion Catalog. I. Main Sequence Dwarfs and Giants Within 33 Parsecs of the Sun.
## 1. Introduction
Proper motion catalogs have historically been the primary sources for the identification of stellar systems in the vicinity of the Sun. While star distances are ultimately determined from measurements of annual parallactic motions, these require a substantial investment of time and effort. Parallax programs are thus most successful when they rely on input lists of objects already suspected to be nearby. Since a large proper motion is an indicator of proximity, high proper motion (HPM) stars have always been prime targets for parallax programs. It is a historical fact that the vast majority of the stars now known to be in the “Solar Neighborhood” ($`d<25`$ pc), and recorded in compendiums such as the Third Catalog of Nearby Stars (Gliese & Jahreiss, 1991) hereafter the CNS3, or the NASA NStars Database <sup>1</sup><sup>1</sup>1http://nstars.arc.nasa.gov, have first been identified as HPM stars.
Perhaps the most important feature of the nearby star census is the fact that it is dominated by low-luminosity red dwarfs, with absolute magnitudes $`9M_V16`$. The vast majority of the “Solar Neighborhood” red dwarfs are therefore not listed in the HIPPARCOS catalog (Perryman, 1997) which has a limiting magnitude $`V12`$. On the other hand, most nearby red dwarfs are expected to be detected on the photographic plates of the large Schmidt telescope surveys (e.g. POSS-I, POSS-II, SES, AAO) which typically reach $`V20`$. Because nearby red dwarfs have optical colors similar to those of distant giants or dust-reddened stars, they are most easily and reliably identified by their large proper motions. In fact, the main strength of proper motion surveys is that they can locate nearby stars independently of their physical appearance (color, luminosity).
One caveat is that a high proper motion is neither a necessary nor a sufficient condition for a star to be nearby. Nearby stars can have small proper motions if their motion vector is pointing toward, or away from the Sun (Garcia-Sanchez et al., 2001). Conversely, distant stars can have large proper motions if their transverse motion relative to the Sun is also large, such as is typically the case for stars on Galactic Halo orbits. Hence the dilemma: the completeness of a nearby star census based on proper motion surveys increases as the proper motion limit is set to smaller values, but decreasing the proper motion limit also greatly increases the number of distant stars that make it into the survey, which can be viewed as “contaminants” for parallax programs aimed at finding the nearest stars.
The earlier proper motion catalogs of Wolf (1919) and Ross (1939) had relatively high proper motion limits ($`\mu >0.3\mathrm{}`$ yr<sup>-1</sup>-0.5$`\mathrm{}`$ yr<sup>-1</sup>) and together contained a little over 2500 stars, a manageable size for parallax programs. However, later catalogs became increasingly larger, bringing the number of known high proper motion stars to sizes difficult to manage. The catalogs by Giclas, Burnham, & Thomas (1971) and Giclas, Burnham, & Thomas (1978) contain 11,000+ stars with $`\mu >0.2`$ yr<sup>-1</sup>, while the famous “NLTT catalog” of Luyten (1979b), the main reference in the field for 25 years, contains over 58,000 stars with $`\mu >0.18`$ yr<sup>-1</sup>, down to a magnitude $`V=20`$. These large catalogs provide several times more targets than can be handled by existing ground-based parallax programs (Monet et al., 1992; Henry et al., 2002; Dahn et al., 2002). To illustrate this, one need only consider the Yale catalog of trigonometric parallaxes (van Altena, Lee, & Hoffleit, 1995), which compiles all published ground-based parallax measurements prior to 1995, and which contains data for only 1501 stars fainter than $`V=12`$. Given these limitations, most efforts at finding nearby stars have focused on a subsample of the NLTT, the “LHS catalog” Luyten (1979a), which basically lists the 4470 NLTT stars with the largest proper motions ($`\mu 0.5\mathrm{}`$ yr<sup>-1</sup>).
It is however possible to mine large proper motion catalogs for nearby stars, by using secondary distance estimators such as photometric or spectroscopic distance moduli. Photometry and spectroscopy are more readily obtained for large samples of objects than parallax measurements, although they provide much less reliable distance estimates. But even approximate distance moduli can be used to trim a large proper motion sample down to a more manageable list of probable nearby stars. Luyten himself devoted substantial efforts to obtaining photographic magnitudes and colors for all the stars in his catalogs, though these were sometimes only approximate. While his $`m_{pg}`$,$`m_r`$ photographic magnitude system was used to identify some nearby objects (Gliese & Jahreiss, 1980), its usefulness in estimating distance moduli for the ubiquitous red dwarfs was found to be very limited in practice (Weis, 1984). More photometric/spectroscopic data was needed, and the follow-up VRI photometric survey of $`2000`$ NLTT stars with $`\mu <0.5\mathrm{}`$ yr<sup>-1</sup> (Weis, 1986, 1987, 1988) proved successful in identifying 295 new Solar Neighborhood candidates.
A major advance came with the availability of accurate infra-red JHK<sub>s</sub> photometry from the 2 Micron All-Sky Survey (2MASS). In a series of papers (“Meeting the Cool Neighbors. I.-IX.”), I. N. Reid and colleagues used data from the 2MASS Second Incremental Release to systematically mine the NLTT for nearby stars. In the first paper of their series, a combination of Luyten’s estimated photographic red magnitude ($`m_r`$) with accurate $`K_s`$ magnitudes from the 2MASS second incremental release was succesfully used to identify a subsample of candidate red dwarfs within 20 pc of the Sun (Reid & Cruz, 2002). Distance moduli were calculated based on the 2MASS magnitudes and follow-up BVRI photometry (Reid, Kilkenny & Cruz, 2002), and spectroscopy (Cruz & Reid, 2002; Reid et al., 2003a) of the NLTT candidates. Unfortunately, positions quoted in the NLTT catalog often contain large errors (Gould & Salim, 2003); this proved a major impediment, as 2MASS counterparts of NLTT stars sometimes could not be found, or were mismatched. These problems were mitigated by the use of the revised NLTT (or rNLTT) catalog of Salim & Gould (2003), which provide re-estimates of NLTT positions, and through the use of additional selection criteria (e.g. color-color cuts) to identify and reject mismatches (Reid et al., 2004).
With these successes, the nearby star census appears now to be only limited by the NLTT catalog itself. Apart from the issues of the accuracy of its photometry and astrometry, the NLTT has long been known to be significantly incomplete in some parts of the sky (Dawson, 1986). The nearby star census would benefit from a more complete proper motion survey, but also from an expansion to fainter magnitudes, and especially from an expansion to smaller proper motions, the NLTT itself being limited to stars with $`\mu >0.18\mathrm{}`$ yr<sup>-1</sup>.
This is why the new LSPM-north catalog, as a replacement of the old NLTT, promises to improve significantly on nearby stars surveys. The LSPM-north catalog (Lépine & Shara, 2005) is the product of a massive data mining of the Digitized Sky Surveys (DSS) with a specialized software (SUPERBLINK) that uses image subtraction algorithms to identify moving and variable objects. The LSPM-north lists 61,976 stars with proper motions $`\mu >0.15`$ down to V=21.0, and is estimated to be $`>99\%`$ complete down to $`V=19`$. Besides being marginally deeper, and significantly more complete than the NLTT, the LSPM is also much more accurate, with positions within $`1\mathrm{}`$ at the 2000.0 epoch, and optical photographic magnitudes to $`\pm 0.30.5`$mags obtained from the USNO-B1.0 catalog of Monet et al. (2003). The LSPM-north catalog also provides counterparts from the 2MASS All Sky Point Source Catalog (Cutri et al., 2003) when they exist. By being deeper, more accurate, and more complete, the LSPM-north catalog supersedes the NLTT in all respects (though only at northern declinations for now). The use of the 2MASS All Sky Point Source Catalog also offers the opportunity to expand significantly upon the Reid et al. analysis, which was restricted to the fraction of the sky covered by the 2MASS Second Incremental Release.
The first paper of this series presents our initial efforts at identifying nearby stars in the LSPM-north catalog, specifically the identification of new candidate main-sequence dwarfs within 33pc of the Sun. Data are found in the literature for all LSPM stars that have published trigonometric parallaxes or photometric/spectroscopic distance moduli. A calibration of the $`[M_V,VJ]`$ relationship in the $`V,J`$ system of the LSPM-north catalog provides the identification of 1672 new candidate nearby stars with distances estimated to be within 33pc of the Sun. Identification of nearby white dwarfs from the LSPM-north catalog will be detailed in the second paper of this series, while a subsequent paper will address the problem of identifying nearby metal-pool subdwarfs (Lépine, in preparation).
This paper is organized as follows: §2 describes the calibration of the $`M_V,VJ`$ color magnitude relationship for main sequence dwarfs, which is used to estimate photometric distance moduli of LSPM stars. §3 gives complete lists of nearby dwarfs listed in the LSPM catalog for which there exists published trigonometric parallaxes or photometric/spectroscopic distance moduli, and presents the selection of new candidate stars within 33pc. §4 analyzes the current completeness level of the nearby star census in terms of limiting magnitude and proper motion selection. The main results are summarized in the conclusion (§5).
## 2. Stellar distance estimates
### 2.1. Trigonometric parallax standards
Photometric distance moduli (m-M)<sub>V,J</sub> for all main sequence LSPM-north stars are estimated based on a color-magnitude relation in the (V,V-J) color-magnitude system used in the LSPM-north catalog. An absolute magnitude-color calibration $`M_V=M_V(VJ)`$ is calibrated with a subsample of 3,104 LSPM stars for which there exist reliable trigonometric parallax ($`\pi _{trig}`$) measurements. Most calibration stars are selected from the Hipparcos and Tycho catalog (Perryman, 1997), and from the Yale catalog of trigonometric parallaxes (van Altena, Lee, & Hoffleit, 1995) – hereafter YPC. Calibration stars are selected based on the following criteria: the star has a measured $`\pi _{trig}`$ with an estimated accuracy better than 10%, the star has $`\pi _{trig}>0.025\mathrm{}`$, placing it within 40pc of the Sun. the star does not have a companion within $`\rho =15\mathrm{}`$. The first constraint ensures that absolute magnitudes can be calculated from apparent magnitudes with an accuracy of $`\pm 0.1`$mag. The second constraint is used because our calibration is to be applied in determining distances for nearby stars, and we wish to compare objects of a given color class over a similar distance range in order to avoid or limit biases such as dust reddening or possible systematic errors in our USNO-B1.0 optical magnitudes. The third constraint excludes proper motion doubles with small separations, for which V and J magnitudes are often affected by large errors, particularly for the faint companions.
The Hipparcos and Tycho catalog provides accurate space-based parallaxes for 1530 LSPM stars. The Hipparcos parallaxes generally have errors under 5 mas. Unfortunately, these are limited to relatively bright ($`V<12`$) objects, which are only effective in mapping the brighter half of the main sequence (from spectral types OB to spectral type K). For the low-luminosity red dwarfs (spectral type M), one has to rely on ground-based parallaxes. The YPC catalog compiles most ground-based parallaxes published in the literature before 1995. These, however, are not as homogeneous or accurate as the Hipparcos parallaxes, and sometimes have errors as large as 40-50 mas. After applying the criteria above, we find reasonably accurate YPC parallaxes for a total of 695 additional LSPM stars.
Hipparcos parallaxes can also be used for a limited number of fainter stars that happen to be common proper motion companions of Hipparcos objects (and are thus physically related and at the same distance). Buy cross-correlating LSPM positions with the Hipparcos and Tycho catalog, we find 60 LSPM stars to be common proper motion companions of $`\pi _{trig}>0.025\mathrm{}`$ Hipparcos stars, with angular separations $`\rho >15\mathrm{}`$ between the components. The secondaries range in absolute magnitude from $`M_V=7.8`$ to $`M_V=14.8`$.
Despite these additions, the combined sample of Hipparcos and YPC stars remains deficient in objects at the bottom of the main sequence ($`M_V>15`$). Accurate parallaxes of very low-mass M and L dwarfs have however been published in Dahn et al. (2002) and in Reid et al. (2003b). These papers provide parallaxes for 13 faint LSPM stars, of which 11 have $`M_V>15`$. Finally, a search of the NStars database provides reliable parallaxes (compiled from a variety of sources) for 52 additional LSPM stars, including 16 with $`M_V>15`$. The final tally is 1,951 objects to be used as color-magnitude calibrators.
### 2.2. Photometric distance modulus calibration
Absolute magnitudes $`M_V`$ for stars in the calibration subsample are calculated from their apparent $`V`$ magnitudes and observed trigonometric parallaxes $`\pi _{trig}`$. These are plotted against their $`VJ`$ colors in Figure 1. The main sequence is clearly visible, as is the white dwarf sequence on the lower left of the diagram. A few stars are found in between, and are most probably subdwarfs. The top of the main sequence ($`M_V<10`$) is more densely populated than the bottom ($`M_V>10`$), which reflects the systematic availability of accurate $`\pi _{trig}`$ for bright stars (from the Hipparcos catalog). At this time, only limited samples of fainter stars have accurate (ground-based) parallaxes.
For the main sequence (MS) stars, we find a monotonic relationship between the $`VJ`$ color index and the absolute magnitude $`M_V`$. The relationship has several inflection points. We subdivide it into 5 segments. For each, we calculate the relationship from a linear regression of the data points (least squares fitting). We map the main sequence as follows:
$$M_V^{}=0.08+3.89(VJ)[0.7<VJ<1.5]$$
$$M_V^{}=2.78+2.09(VJ)[1.5<VJ<3.0]$$
$$M_V^{}=1.49+2.52(VJ)[3.0<VJ<4.0]$$
$$M_V^{}=2.17+2.35(VJ)[4.0<VJ<5.0]$$
$$M_V^{}=4.47+1.89(VJ)[5.0<VJ<9.0]$$
(1)
For any MS star, we can then calculate a photometric distance modulus $`(mM)_{V,J}`$ with:
$$(mM)_{V,J}=VM_V^{},$$
(2)
where $`M_V^{}=M_V^{}(VJ)`$ follows the relationships defined above. Photometric parallaxes $`\pi _{phot}`$ can then be estimated with:
$$\pi _{phot}=10^{\frac{(mM)_{V,J}}{5}1}.$$
(3)
### 2.3. Calibration accuracy
We estimate the accuracy of our photometric distance moduli by calculating $`(mM)_{V,J}`$ for all our calibration stars, and comparing with the trigonometric distance moduli $`(mM)_{trig}`$, where
$$(mM)_{trig}=5\mathrm{log}\pi _{trig}5.$$
(4)
For each star, we calculate the difference between the trigonometric and photometric distance moduli $`(mM)_{trig}(mM)_{V,J}`$. These are plotted in Figure 2 as a function of $`VJ`$ for our calibration stars. We find a dispersion that increases from $`\sigma 0.34`$ mag at $`VJ<3.0`$ to $`\sigma =0.69`$ mag at $`4.0<VJ<5.0`$, and down to $`\sigma =0.56`$ mag at $`5.0<VJ<6.0`$. The values for the scatter in the color-magnitude relationship are listed in Table 1, for given ranges in $`VJ`$. These provide standard errors for the $`(mM)_{V,J}`$ distance moduli estimates.
The smaller (larger) errors on $`\pi _{phot}`$ for the bluer (redder) stars is largely consistent with the expected accuracy of the $`V`$-band magnitudes in the LSPM. The bluer calibration stars happen to be among the brighter stars in the LSPM catalog ($`V<12`$) for which the $`V`$ magnitudes are quoted from the TYCHO-2 catalog Hog et al. (2000), and are accurate to $`\pm 0.1`$mag. Most of the redder calibration stars, on the other hand, are drawn from the group of fainter LSPM stars, which have their $`V`$ magnitudes estimated from photographic plate measurements, mostly based on magnitudes from the USNO-B1.0 catalog of Monet et al. (2003) which are only accurate to $`\pm 0.30.5`$mag.
One must note that at a given ($`VJ`$), the error in the distance modulus is generally not equivalent to the error in the $`V`$ magnitude. Rather, it depends on the slope $`b`$ of the color-magnitude relationship $`M_V^{}=a+b(VJ)`$. From equation 3, we see that a $`\delta V`$ error in $`V`$ propagates as $`(b1)\delta V`$. Hence for $`b>1`$, a steeper slope means a larger error on the photometric distance modulus for a given $`\delta V`$. Hence between $`4<(VJ)<5`$, the $`\pm 0.5`$ mag error on the photographic magnitudes is expected to yield an $`\pm 0.68`$ mag error on the distance modulus, which is consistent with our results.
In any case, the dispersion in $`(mM)_{trig}(mM)_{phot}`$ cannot be attributed only to errors in the $`V`$-band magnitude measurements. There is an intrinsic dispersion in the color-magnitude relationship caused by a variety of factors such as multiplicity, variability, and metallicity. This has been known historically as the “cosmic scatter”. This scatter is clear from the $`0.3`$ mag dispersion observed for $`(VJ)<3`$ stars, whose $`V`$ magnitudes are those determined from HIPPARCOS measurements, and are generally accurate to better than $`0.03`$ mag. Intrinsic factors are also probably responsible for the existence of the several outliers ($`>2\sigma `$ from the mean). These can be unresolved doubles, variables, or even metal-poor subdwarfs that are naturally expected to fall below the metal rich main sequence of the local Galactic disk stars.
### 2.4. Lutz-Kelker corrections
The direct calibration of absolute magnitudes using parallax measurements is generally susceptible to a type of systematic errors, known as the Lutz-Kelker bias. A nice review of the topic can be found in Sandage & Saha (2002). This bias generally results in absolute magnitudes being overestimated, and is dependent of the ratio of the parallax measurement errors over the measured parallaxes $`\sigma _\pi \pi ^1`$. Given that our calibrators were selected to have $`\sigma _\pi \pi ^1<0.1`$, and given the stars were selected based on proper motion, we need to determine whether the Lutz-Kelker is significant, and whether corrections should be applied.
Generally, the Lutz-Kelker bias occurs in a subsample of objects with measured parallaxes $`\pi _0`$ and affected by measurement errors $`\sigma _\pi `$, when the distribution of true parallaxes $`\pi `$ in the underlying population decreases monotonically with $`\pi `$ (as when the number of calibrators increases with distance). The measurement errors are symmetric about $`\pi _0`$, but the probability distribution $`P(\pi |\pi _0)`$ is not because of the asymmetry in the distribution of $`\pi `$ values around $`\pi _0`$. Simply put, because there are more distant stars than nearer ones, a star with measured parallax $`\pi _0`$ is more likely to be a distant star whose parallax is overestimated ($`\pi <\pi _0`$) than a nearby one whose parallax is underestimated ($`\pi >\pi _0`$). On average, the mean parallax of a group of stars will thus be overestimated, which means the derived absolute magnitude will be overestimated too. The classical case investigated by Lutz & Kelker (1973) is the one where the calibrators are uniformly distributed in space, for which the distribution in true parallax follows $`f(\pi )\pi ^4`$. An analytical derivation indicates a bias in the absolute magnitude calibration of +0.11 mag for $`\sigma _\pi \pi ^1=0.10`$, down to +0.02 mag for $`\sigma _\pi \pi ^1=0.05`$, and +0.01 mag for $`\sigma _\pi \pi ^1=0.025`$.
All our calibrators have $`\sigma _\pi \pi ^1<0.10`$, following our selection criteria. The HIPPARCOS stars, which populate the upper half of the main sequence ($`M_V<10`$), tend to have significantly more accurate parallax measurements, with $`\sigma _\pi \pi ^1=0.036`$. This value corresponds to a very small classical Lutz-Kelker bias of only +0.014 mag. For the calibrators that define the lower half of the main sequence ($`M_V>10`$), our sample has $`\sigma _\pi \pi ^10.048`$, which corresponds to a bias of +0.019 mag.
However, the above values most probably overestimate the actual Lutz-Kelker bias in our data. The reason is that our sample of calibrators is not drawn from the general supersample of stars in the Solar neighborhood (whose density is uniform) but rather from a restricted supersample that includes only stars with proper motions above a certain value. This proper motion limit is $`\mu =0.15\mathrm{}yr^1`$ for the HIPPARCOS objects (the lower limit of the LSPM-NORTH catalog). For the fainter calibration, it is much larger because faint parallax targets have historically been selected among subsamples of stars with very large proper motions (e.g. the LHS catalog, with a proper motion limit $`\mu =0.5\mathrm{}yr^1`$ ). The effect of drawing the calibrators from a proper-motion selected supersample tends to reduce the Lutz-Kelker bias, because the supersample will be increasingly incomplete at larger distances (see §4 below), and the distribution function $`f(\pi )`$ will be shallower than in the classical case. Estimates of the Lutz-Kelker bias for shallower distribution has been investigated by e.g. Hanson (1979), who showed that for power law distributions $`f(\pi )\pi ^n`$ with $`n<4`$, the magnitude of the Lutz-Kelker correction is reduced compared with the classical case ($`n=4`$).
Given the uncertainty in the underlying bias associated with the supersample (i.e. uncertainty in the form of $`f(\pi )`$), the application of any specific Lutz-Kelker correction to our calibration cannot be justified. In any case, the Lutz-Kelker bias is most likely to be very small ($`<0.02`$ mag). Our photometric distances may thus systematically underestimate the true distances of the candidate nearby stars, but only at the $`12\%`$ level. For any given star, this bias is negligible compared with e.g. the intrinsic “cosmic scatter” of $`0.3`$ mag, or to the $`0.5`$ mag photometric errors.
## 3. M dwarfs within 33 parsecs
Our goal is to identify the largest possible number of dwarf stars within the traditional confines of the Solar neighborhood ($`d<25`$ pc). Since our photometric distances are accurate only to $`\pm 35\%`$, an extension to 33pc ensures that the large majority of 25pc stars will fall into the sample. A limiting distance of 33pc is thus adopted. The LSPM-north catalog does allow one to identify objects at much larger distances, but the objective of this paper is to provide a list of high priority objects of a size that can be managed by ground-based parallax programs.
### 3.1. Previously known objects with parallax measurements
The LSPM-north contains thousands of new HPM stars, but lists all previously known HPM stars as well, including those from the NLTT and LHS catalogs. It is thus not surprising that a large number of LSPM entries are stars that have have long been suspected (or confirmed) to be nearby. Since the LSPM is virtually complete for all stars with $`\mu >0.15\mathrm{}`$ yr<sup>-1</sup> down to a magnitude $`V=19`$, the vast majority of known nearby stars from the north celestial hemisphere are expected to have counterparts in the LSPM-north. The only exceptions are known nearby stars with very small proper motions ($`\mu <0.15\mathrm{}`$ yr<sup>-1</sup>) or objects with very faint optical magnitudes ($`V>20`$). The latter group includes most known nearby L and T dwarfs (Kirkpatrick et al., 2000; Cruz et al., 2003); these are generally too faint in the optical bands to be detected on the POSS-I/POSS-II red plates, on which the LSPM-north is based. In any case, most of them are not stars but brown dwarfs.
We searched the literature for parallax measurements of LSPM sources which would place the star within 33 pc of the Sun. Our search turned up parallax measurements $`\pi _{trig}>0.03\mathrm{}`$ for 1676 LSPM stars. This excludes stars whose positions in the color-magnitude diagram suggest they are either white dwarfs or subdwarfs. The complete list of objects is provided in Table 2; most of the stars are main sequence objects (dwarfs) but there are a few subgiants and giants. Parallax data are primarily from the Hipparcos and Tycho Catalogue, including data for 1207 stars that are formal Hipparcos objects, and for 54 more stars that are close companions or common proper motion companions of Hipparcos objects. Another major source of parallax data is the YPC, which provides parallaxes for 361 additional LSPM stars. Additional $`\pi _{trig}`$ are found in the NStars database for 40 more LSPM stars. Parallax data are found in Dahn et al. (2002) for 13 faint, red LSPM stars. Finally, a parallax for LSR J1835+3259 is published in Reid et al. (2003b).
Table 2 gives the name under which the star is generally known along with the LSPM designation. The table quotes the LSPM position and proper motion at the 2000.0 epoch, and gives the $`V`$ magnitude and $`VJ`$ color, also from the LSPM. The trigonometric parallax is given along with its estimated error, and the bibliographical source. The distance modulus $`(mM)_{V,J}`$, as estimated from the calibration described in §2, is shown for comparison. Note that many stars do not have $`(mM)_{V,J}`$ estimates, because their blue color ($`VJ<0.7`$) falls outside the range of of calibration system. The final column in Table 2 gives the distance from the Sun (in parsecs) as derived from the trigonometric parallax.
Figure 3 compares the derived distance modulus $`(mM)_{V,J}`$ with the distance modulus calculated from the measured trigonometric parallax $`(mM)_{trig}=55\mathrm{log}\pi _{trig}`$. The expected locus of single stars is shown in Figure 3 (full line). The difference between the two estimates, $`(mM)_{V,J}(mM)_{trig}`$, has a mean value of +0.04 mag, and a dispersion 0.50 mag, after removal of the 3$`\sigma `$ outliers. This is consistent with our calibration of $`(mM)_{V,J}`$ using a subset of these stars (see §2 above). There appears to be a small number of stars that have $`(mM)_{V,J}(mM)_{trig}1.5`$ (see Figure 3). Such an underestimate of the distance, by photometric means, would be consistent with those stars being unresolved doubles. The expected locus of unresolved binaries with equal-luminosity components in the $`(mM)_{trig}`$,$`(mM)_{V,J}`$ diagram is drawn in Figure 3 (dashed line). There are also a few objects whose photometric distance appear to be significantly overestimated, with $`(mM)_{V,J}(mM)_{trig}>1.5`$. In all likelihood, these objects are metal-poor subdwarfs, for which the color-magnitude calibration of §2 is invalid.
### 3.2. Previously known objects with photometric/spectroscopic distance moduli
There exist several hundred high proper motion stars that do not have trigonometric parallaxes, but that have been previously cited in the literature as candidate nearby stars. We compile a list of the main sequence dwarfs (i.e. excluding known white dwarfs and subdwarfs) that have been cited as candidate nearby dwarfs with estimated distance within 33 pc, and that have a counterpart in the LSPM-north. The list, comprising 783 stars, is provided in Table 3. Some 468 stars have published photometric distance estimates, while another 315 have spectroscopic distance estimates.
Some 359 stars have photometric distances quoted in the CNS3 but no formal trigonometric parallax. Most of these stars have their distance modulus calculated from the photometric survey of Weis (1986, 1987, 1988). Most of the other nearby star candidates are from the study of the NLTT catalog conducted by Reid et al. We find photometric distance estimates for 56 stars in Reid, Kilkenny & Cruz (2002) and for 29 more in Reid et al. (2004). Reid et al. (2003a) give spectroscopic distance estimates for 28 stars that have counterparts in the LSPM-north, and Reid et al. (2003b) give spectroscopic distances for 169 more, and Reid et al. (2004) for 91. Spectroscopic distance estimates for high proper motion discovered with SUPERBLINK were also obtained for 104 stars with $`\mu >0.5\mathrm{}`$ yr<sup>-1</sup> by Lépine, Rich, & Shara (2003). The list comprises 19 stars that are $`d<33`$pc candidates; these are included in Table3.
There now also exists spectroscopic/photometric distance moduli for numerous cool and ultra-cool dwarfs discovered with the 2MASS and DENIS surveys. Photometric distance moduli are given for 52 ultra-cool 2MASS dwarfs in Gizis et al. (2000); we find matches in that list for 15 of the reddest LSPM stars. Spectroscopic distance moduli are calculated for 251 late M dwarfs and L dwarfs in Cruz et al. (2003). The majority of these objects are either in the southern sky or are too faint in the optical to be included in the LSPM catalog, but we nevertheless find matches to 5 additional red LSPM stars. Two nearby ultra-cool dwarfs discovered in the DENIS infrared survey are also found among the reddest LSPM entries: the L4 dwarf DENIS-P J104842.8+011158 has a spectroscopic distance modulus estimated by Kendall et al. (2004), while the DENIS-P J031225.1+002158 has a photometric distance modulus cited in Phan-Bao et al. (2001). A photometric distance for LP 584-4 (DENIS J000206.1+011536) is also given in Phan-Bao et al. (2003). The LSPM-north catalog also includes the very nearby star LHS 2090 ($`d=6`$pc), whose photometric distance was first reported by Scholz, Meusinger, & Jahreiss (2001). A spectrophotometric parallax for the very high proper motion ($`\mu =5.0\mathrm{}`$ yr<sup>-1</sup>) star SO 025300.5+165258 is given in Teegarden et al. (2003). Finally, a photometric/spectroscopic distance modulus for LSR J0602+3910, the brightest L dwarf in the sky (V=20.1), is given by Salim et al. (2003).
Figure 4 compares the photometric and spectroscopic distance moduli cited in the literature to the distance modulus $`(mM)_{V,J}`$ calculated from the V magnitude and V-J color cited in the LSPM catalog. The published $`(mM)_{phot}`$ agree very well with the LSPM distance modulus estimate. Values for $`(mM)_{V,J}(mM)_{phot}`$ have a mean of 0.03 mag and a dispersion of 0.43 mag, after removal of 3$`\sigma `$ outliers. The good match between the two distance moduli probably only attests to the fact that the two calibrations are essentially based on the same sample of objects with measured $`\pi _{trig}`$.
On the other hand, there is a slight disagreement between the published spectroscopic distance moduli and $`(mM)_{V,J}`$, with $`(mM)_{V,J}(mM)_{spec}`$ having a mean value of +0.15 mag. The excellent agreement observed between $`(mM)_{V,J}`$ and both $`(mM)_{trig}`$ and $`(mM)_{phot}`$ argues against any systematic error in $`(mM)_{V,J}`$. This means that the $`(mM)_{spec}`$ underestimate distances by $`7`$% on average. Stars with $`(mM)_{spec}`$ are overwhelmingly calibrated with the spectral-indices/absolute-magnitude relationships defined by Reid et al. (2003a). The 0.15 mag systematic error possibly results from the limited number of stars used to calibrate the relationships. Note that the dispersion of 0.47 mag is only marginally larger than the dispersion on $`(mM)_{V,J}(mM)_{phot}`$, which suggests that $`(mM)_{spec}`$ is, in principle, just about as accurate as $`(mM)_{phot}`$, provided the zero point is accurately calibrated.
### 3.3. New candidate nearby stars
Using the color-magnitude calibration in §2, it is straightforward to identify new candidate nearby red dwarfs from the LSPM-north catalog. Photometric distance moduli are simply estimated from $`V`$ and $`VJ`$ (Equations 1,2). The only caution is that since the relationship is valid only for metal-rich (disk) MS stars, one must avoid applying it to other types of objects. While the bulk of the stars in the LSPM-north catalog are disk MS stars, the catalog also includes significant numbers of metal-poor halo subdwarfs and white dwarfs, along with smaller numbers of giants and subgiants.
Relative to the dwarf MS stars, giant stars are overluminous for a given $`VJ`$ color. Applying the dwarf color-magnitude relationship thus yields underestimated distances. Giant stars are quite rare in the Solar Neighborhood, but they can significantly contaminate samples of color-selected nearby star candidates since they tend to be detected over a much larger volume. Fortunately, proper motion selection filters out the majority of more distant objects: the 0.15$`\mathrm{}`$ yr<sup>-1</sup> limit of the LSPM-north catalog eliminates most disk stars beyond $`100`$pc (assuming disk stars to have transverse velocities $`v_t`$70km/s), and most halo stars beyond $`600`$pc (assuming $`v_t`$400km/s). Giant stars that made it into the LSPM-north catalog are bright enough to be HIPPARCOS objects, and since there already is good parallaxes for them, no confusion is possible. The more problematic objects might be the subgiants, which are only moderately overluminous relative to the MS stars. The LSPM-north potentially contains subgiant stars that are too faint ($`V>11`$) to be HIPPARCOS objects. However, these are expected to be relatively blue ($`VJ<2.5`$). According to our calibration (Eq.1), main sequence stars with $`VJ<2.5`$ have absolute magnitudes $`M_V<8.0`$, which means they have apparent magnitude $`V<10.6`$ at 33pc. Since the current goal is to identify MS stars within 33pc of the Sun, the search can be restricted to LSPM stars with $`VJ>2.5`$. Any bluer object within 33pc should have already been found and measured by HIPPARCOS. This constraint also effectively eliminates subgiants as a source of contaminants.
White dwarf stars, on the other hand, are underluminous for any given $`VJ`$ color. As can be seen from Fig.1, the white dwarf sequence falls $``$5 magnitudes below the main sequence in the standard color-magnitude diagram. Assuming the color-magnitude relationship for MS stars (Eq.1) were applied to a white dwarf, the object would have its distance overestimated by a factor of 10. Restricting the search to stars within 33pc of the Sun virtually eliminates the possibility of a contamination of the sample with white dwarfs, because to make it into the 33pc sample of candidate stars, the white dwarf would actually have to be within 3.3pc of the Sun. At this time, there is no known isolated white dwarf star within 3.3pc of the Sun. The probability of white dwarfs “contaminating” the 33 pc sample is very small.
Subdwarf stars are also underluminous in $`M_V`$ for a given $`VJ`$ color, although not as much as are the white dwarfs. Most nearby subdwarfs are local member of the Galactic halo population. One way to separate disc dwarfs from halo subdwarfs is to use an optical-to-infrared reduced proper motion diagram (Salim & Gould, 2002). Because the reduced proper motion is proportional to both the absolute magnitude of a star and its transverse velocity, halo subdwarfs, with their larger systemic velocities, fall well below the disc dwarf sequence. But while this helps in separating halo dwarfs from disc subdwarfs, the method only works, statistically, on large groups of objects, and is unreliable for individual stars. In any case, halo subdwarfs are rare in the Solar Neighborhood (about 1 star in 200); if they are found in significant numbers in proper motion catalogs it is because of their larger mean velocity relative to the Sun. Again, restricting the search to MS stars within 33 pc of the Sun minimizes contamination of the nearby stars sample. The $`0.15\mathrm{}`$ yr<sup>-1</sup> limit includes all stars at 33 pc with transverse velocities $`v_{trans}>23.3`$km s<sup>-1</sup>, which still overwhelmingly consists of disk stars. Increasing the distance would, however, increase the ratio of subdwarfs to dwarfs in the census. For example, LSPM stars at 100 pc have $`v_{trans}>70.5`$ km s<sup>-1</sup> and exclude most of the disk dwarfs, leaving a much larger fraction of halo subdwarfs.
Our nearby MS stars selection box, in the $`V,VJ`$ diagram, is depicted in Figure 5, overlaid on the distribution of LSPM-north stars. The selection box marks the locus of MS stars within 3 pc and 33 pc from the Sun. Note how the selection box avoids regions on the graph that are mainly populated with giants, white dwarfs, and subdwarfs, as explained above. Any star that falls in the box, and that is not a known nearby star (i.e. it is not listed in Tables 2-3) is flagged as a new nearby star candidate.
Table 4 provides the complete list of new candidate nearby stars. The lists includes 1672 stars with $`8.20<V<20.63`$ and $`2.57<VJ<7.95`$. A total of 730 stars are objects listed in the literature or in previous proper motion catalogs, but which have never before been formally identified as nearby star contenders. The other 942 candidates are high proper motion stars identified with SUPERBLINK, and first listed in the LSPM-north catalog.
### 3.4. New nearby star candidates of particular interest
#### 3.4.1 HD 29271
Our color-magnitude calibration puts this bright star at only $`6.3\pm 1.9`$ pc from the Sun. While it is a $`V=8.2`$ magnitude star, HD 29271 is not a HIPPARCOS star, and there is no published parallax for it. The star is listed with a spectral type K0 in the Henry Draper Catalog, and is also listed in the TYCHO-2 catalog with TYCHO magnitudes $`B_T=9.98`$ and $`V_T=8.35`$. It is unclear whether this star is an actual nearby K dwarf or a more distant K giant. With $`JH=0.53`$ and $`JK_s=0.34`$ (from 2MASS), this object has very unusual IR colors which are consistent with neither a K dwarf or a K giant. Because it has a proper motion of only 0.165$`\mathrm{}`$ yr<sup>-1</sup>, it is more likely to be a more distant object, although it is kept on the list of candidate nearby stars at this time, pending clarification of its status.
#### 3.4.2 G 141-36
We estimate for this Giclas object a photometric distance for only $`7.6\pm 2.4`$ pc. That no follow-up observation has been performed on this object since its identification comes as a surprise, since it does have a large proper motion $`\mu =0.447\mathrm{}`$ yr<sup>-1</sup>, making it an obvious candidate for a nearby star search. This object should be a high priority target for parallax programs.
#### 3.4.3 LSPM J1735+2634
This faint ($`V=19.2`$) high proper motion star was first identified by SUPERBLINK, and is one of the new stars listed in the LSPM-north catalog. With $`VJ=7.9`$ and $`JK_s=1.09`$, the object is most probably a very low-mass star, of spectral type M7 or later. The photometric distance modulus places it at a distance of $`9.2\pm 2.9`$ pc from the Sun.
#### 3.4.4 LSPM J1314+1320
This moderately faint ($`V=15.9`$) star has a photometric distance estimate of only $`9.7\pm 3.0`$ pc. It is another one of the new high proper motion stars found with SUPERBLINK, and also one of the nearest. As with the one above, it should be a high priority target for parallax programs.
#### 3.4.5 LSPM J1935+3746
This $`V=11.6`$ star has a photometric distance estimate of only $`9.9\pm 3.3`$ pc. It is also a newly discovered high proper motion star. Its moderately small proper motion (0.17$`\mathrm{}`$ yr<sup>-1</sup>) and location at low galactic latitudes (b=+8.33) would have made it difficult to find in previous proper motion surveys. Even more interesting, this star was previously found to be a counterpart of the ROSAT point source RX J1935.4+3746 by Motch et al. (1998) who determined a spectral type M4Ve. The spectral type is consistent with the star being at a distance $`10`$ pc from the Sun.
## 4. Completeness of the northern nearby dwarf census
A histogram of the distribution of nearby and candidate nearby MS stars as a function of distance is presented in Figure 6. Stars that were known before this study are indicated by the shaded area. The present study makes a very significant contribution for stars beyond 20 pc; only few new additions are made within 15 pc. Assuming the current census to be complete up to a distance of 12 pc, we estimate the local stellar density of MS stars to be 0.77 stars pc<sup>-3</sup>. The expected distribution of objects as a function of distance, for a uniform distribution of stars in the local volume, is shown in Figure 6 for comparison.
The current census clearly shows signs of increasing incompleteness beyond 15pc. This is despite the fact that the present study is contributing a significant number of additions beyond this limit, and especially beyond 25pc, where the new candidates account for more than half of the known nearby objects.
The source of the incompleteness lies not in the limiting magnitude of the LSPM-north catalog. The LSPM-north is estimated to 99.5% complete down to $`V=19`$, with a limiting magnitude $`V=21`$ (Lépine & Shara, 2005). The vast majority of the MS disk dwarfs have absolute magnitudes $`M_V<16`$, and at 33 pc are all bright enough to be in the LSPM-north. Figure 7 plots the distribution of stars as a function of absolute magnitude and distance. It is clear that the bulk of the red dwarf stars are located between $`10<M_V<15`$. For stars within 33 pc, the majority are too faint to be in the HIPPARCOS catalog, but they are well within the magnitude limit of the LSPM-north. The 33 pc census built from the LSPM catalog is expected to be deficient only in stars that lie at the very bottom of the main sequence ($`M_V>18`$).
The main source of incompleteness in the 33 pc census remains the high proper motion selection. At 33 pc, the $`\mu >0.15\mathrm{}`$ yr<sup>-1</sup> limit of the LSPM selects for stars with transverse velocities $`v_{trans}>23.5`$ km s<sup>-1</sup>, which excludes a significant number of disk stars with low space motions relative to the Sun. Figure 8 plots the distribution of nearby and candidate nearby stars as a function of distance and transverse velocity $`V_{trans}`$ in km s<sup>-1</sup>, where the latter is calculated from the proper motion and estimated distance ($`V_{trans}=4.74\mu d`$). Northern nearby stars that are in the LSPM catalog are represented by black dots. The incompleteness due to proper motion selection is clear. The very abrupt drop across the $`\mu =0.15\mathrm{}`$ yr<sup>-1</sup> points to the existence of a significant population of $`d<33`$pc stars with proper motions $`\mu <0.15\mathrm{}`$ yr<sup>-1</sup>. The detection ranges of catalogs with various proper motion limits are noted in Figure 8.
The HIPPARCOS catalog lists 265 stars in the northern sky with $`d<33`$ pc and $`\mu <0.15\mathrm{}`$ yr<sup>-1</sup>. The YPC lists 37 more $`d<33`$ pc stars which are not in the LSPM-north either, because their proper motion is too small. These 302 low proper motion nearby stars are plotted as open circles in Figure 8. It is interesting to note that the census of known nearby stars with small proper motions is disproportionately biased toward HIPPARCOS ($`V<12`$) objects, whereas the census of nearby stars with large proper motions is dominated by non-HIPPARCOS ($`V>12`$) stars. This is consistent with the fact that ground-based parallax programs (which the YPC is a compilation of) have been largely dependent on lists of candidates selected on the basis of large proper motions.
The proper motion limit results in an increasing incompleteness in the nearby star census as a function of distance. It is possible to estimate how many low-velocity stars are missing at larger distances by mapping the distribution of low-velocity stars that actually get detected at smaller distances. Figure 9 compares the distribution as a function of transverse velocity ($`v_{trans}`$) for stars within 33 pc that are actually detected by the LSPM-north, with a estimate of the expected distribution of objects. The estimate is based on a correction of the actual number of stars found within 33 pc at a given $`v_{trans}`$ to the effective volume sampled by the LSPM-north for stars having that value of $`v_{trans}`$. No correction is necessary for velocity bins $`>23.5`$ km s<sup>-1</sup>, since the $`0.15\mathrm{}`$ yr<sup>-1</sup> limit of the LSPM-north is sensitive to star with such velocities out to at least 33 pc. For bins with $`v_{trans}<23.5`$ km s<sup>-1</sup>, the bins are normalized by a factor $`(d_{max}/33)3`$, where $`d_{max}=v_{trans}/4.74/0.15`$ is the maximum distance a star with transverse velocity $`v_{trans}`$ can be detected by a proper motion survey with $`\mu =0.15\mathrm{}`$ yr<sup>-1</sup> limit. The “predicted” velocity distribution at small $`v_{trans}`$ (see Figure 9) shows fluctuations because of small number statistics, but it is clear that at least a thousand stars must be missing from the census because of their small proper motion.
A simple estimate suggests that there should be $`6,500`$ stars within 33 pc of the Sun. The current census includes only 4434 stars, the 4,131 ones listed in the LSPM-North, plus the 302 known nearby low proper motion objects (see above). The current, overall completeness of the 33pc census is thus only $`68\%`$. This indicates that $`2,000`$ northern, nuclear-burning stars remain to be found within a distance of about 33 pc, and that these have proper motions $`\mu <0.15\mathrm{}`$ yr<sup>-1</sup>. The situation in the 25pc volume is slightly better, with $`2,800`$ nuclear burning stars expected in the northen sky, and 2,304 currently located (including 114 with $`\mu <0.15\mathrm{}`$ yr<sup>-1</sup>). The current completeness of the 25pc census, the traditional extent of the “Solar Neighborhood”, is thus $`82\%`$. The 500 or so missing stars are expected to have proper motions below the limit of the LSPM-North.
It is clear that much can be gained by expanding our proper motion survey to reach smaller proper motion limits. Figure 10 plots the expected rate of detection as a function of distance, for surveys with lower proper motion limits $`\mu =0.5\mathrm{}`$ yr<sup>-1</sup>, 0.2$`\mathrm{}`$ yr<sup>-1</sup>, 0.1$`\mathrm{}`$ yr<sup>-1</sup>, and 0.05$`\mathrm{}`$ yr<sup>-1</sup>. These detection rates are based on the transverse velocity distribution shown in Figure 9. One sees that to achieve a $`90\%`$ detection rate at a distance of 25 pc would require to assemble a complete survey of stars with $`\mu >0.05\mathrm{}`$ yr<sup>-1</sup>.
## 5. Conclusions
This paper has presented a list of nearby and candidate nearby main sequence stars, subgiants, and giants within 33 pc of the Sun. The list is based on an analysis of the LSPM-north catalog of stars with proper motions $`\mu >0.15\mathrm{}`$ yr<sup>-1</sup>. The census contains a total of 4221 stars in the half of the sky north of the J2000 celestial equator. The list includes 1676 previously known nearby stars with measured trigonometric parallaxes, 783 previously suspected nearby objects, and 1762 new candidate nearby stars.
A relationship is determined between the absolute V magnitude ($`M_V`$) and the $`VJ`$ color index. This relationship can be used to calculate photometric distance moduli $`(mM)_{V,VJ}`$ in the $`V,VJ`$ magnitude system. This is convenient for estimating distances of all the stars in the LSPM-north (for which $`V`$ and $`J`$ magnitude estimates exist) to the extent that the relationship is applied to main sequence(H-burning) stars of near-Solar metallicity. For single stars, the relationship is accurate to $`\pm 0.50.6`$ mag, which means that it provides distance estimates accurate to $`\pm 2530\%`$. The relationship breaks down for white dwarfs and subdwarfs, whose cases will be addressed in upcoming papers of this series. It is found, however, that the sample of main sequence dwarfs selected based on the color-magnitude relationship should not be significantly contaminated by mis-identified white dwarfs and subdwarfs, as long as the search is limited to stars within 33 pc of the Sun.
The main purpose of the present list of candidate objects is to provide targets for upcoming parallax programs. A fraction of the 783 previously suspected nearby MS stars are probably already on parallax programs. The list of 1762 new candidate nearby stars should be examined to sort out the most promising candidates to be added to the parallax programs. Note that current distance estimates based on photometry are only approximate, and should be used with caution.
A large spectroscopic follow-up program is under way, which will soon provide spectral types and radial velocities for most of the new candidate nearby stars presented here. Results will be presented in upcoming papers in this series. While these will also yield spectroscopic distance moduli, these are not expected to be significantly more accurate than the photometric distance moduli presented here. Accurate distances will ultimately come from parallax measurements.
The census of nuclear-burning stars within 33 pc of the Sun, based on the present study, appears to be only $`68\%`$ for the northern sky. The census of nuclear-burning stars within 25 pc, the traditional confines of the so-called “Solar neighborhood” is estimated to be $`82\%`$ complete. It is revealed that the main source of incompleteness lies in the low proper motion limit ($`\mu >0.15\mathrm{}`$ yr<sup>-1</sup>) of the LSPM-north catalog. Expanding our survey to reach a smaller proper motion limit appears to be the best way to significantly increase the completeness of the nearby star census. An extension of the SUPERBLINK survey down to a proper motion limit of $`\mu =0.10\mathrm{}`$ yr<sup>-1</sup> is now being completed, and a second extension down to $`\mu =0.05\mathrm{}`$ yr<sup>-1</sup> is being planned. An analysis of the $`\mu =0.05\mathrm{}`$ yr<sup>-1</sup> catalog should ultimately yield the identification of $`>90\%`$ of the nuclear burning stars up to and at a distance of 25 pc.
Acknowledgments This work has been made possible by the generous support of Hilary Lifsitz and of the American Museum of Natural History. The author is indebted to Michael M. Shara (AMNH) and R. Michael Rich (UCLA) for their constant support and encouragement.
|
warning/0506/hep-ph0506070.html
|
ar5iv
|
text
|
# Lepton flavor violating signals of the neutral top-pion in future lepton colliders
## Abstract
The presence of the top-pions $`\pi _t^{0,\pm }`$ in the low-energy spectrum is an inevitable feature of the topcolor scenario. Taking into account the constraints of the present experimental limit of the lepton flavor violating($`LFV`$) process $`\mu e\gamma `$ on the free parameters of topcolor-assisted techicolor(TC2) models, we study the contributions of the neutral top-pion $`\pi _t^0`$ to the $`LFV`$ processes $`\mu ^+\mu ^{}\tau \mu `$ ( or $`\tau e`$), $`\gamma \gamma \tau \mu `$ (or $`\tau e`$), $`e^+e^{}\tau \mu `$, and $`e\gamma e\pi _t^0e\tau \mu (e)`$ via the flavor changing ($`FC`$) couplings $`\pi _t^0l_il_j`$ and discuss the possibility of searching for the $`LFV`$ signals via these processes in future lepton colliders.
PACS number: 12.60.Cn, 13.35.Dx,14.70.Pw
I. Introduction
The solar neutrino experiments and the atmospheric neutrino experiments confirmed by reactor and accelerator experiments have made one believe that neutrinos are massive and oscillate in flavors, and provide the only direct observation of physics that cannot be accommodated within the standard model($`SM`$), which can be seen as the first experimental clue for the existence of new physics beyond the $`SM`$. Thus, the $`SM`$ requires some modification to account for the pattern of neutrino mixing, in which the lepton flavor violating($`LFV`$) processes are allowed. The observation of the $`LFV`$ signals in present or future high-energy experiments would be a clear signature of new physics beyond the $`SM`$.
Many kinds of popular specific models beyond the $`SM`$ predict the presence of new particles, such as new gauge bosons and new scalars, which can naturally lead to the tree-level $`LFV`$ couplings. In general, these new particles could enhance branching ratios for some $`LFV`$ processes, and perhaps bringing them into the observable threshold of the present and next generations of collider experiments. Furthermore, nonobservability of these $`LFV`$ processes can lead to strong constraints on the nature of new physics. Thus, studying the possible $`LFV`$ signals of new particles in various high-energy colliders is very interesting and needed.
To completely avoid the problems arising from the elementary Higgs field in the $`SM`$, various kinds of dynamical electroweak symmetry breaking ($`EWSB`$) models have been proposed, and among which the topcolor scenario is attractive because it can explain the large top quark mass and provide possible $`EWSB`$ mechanism. Almost all of these kind of models propose that the scale of the gauge groups should be flavor nonuniversal. When one writes the nonuniversal interactions in the masseigen basis, it can induce the tree -level flavor changing$`(FC)`$ couplings, which can generate rich phenomenology.
The presence of the physical top-pions $`\pi _t^{0,\pm }`$ in the low-energy spectrum is an inevitable feature of the topcolor scenario, regardless of the dynamics responsible for $`EWSB`$ and other quark masses. One of the most interesting features of the top-pions $`\pi _t^{0,\pm }`$ is that they have large Yukawa couplings to the third-generation quarks and can induce the tree-level $`FC`$ couplings in quark sector and in the lepton sector. The obviously phenomenological implication of this feature is that the top-pions can be significant produced via some $`FC`$ processes, and their possible signatures might be observed in future hadron colliders and lepton colliders. Furthermore, the top-pions can generate large corrections to some observables related with the $`FC`$ couplings and give characteristic signatures at various high-energy colliders. On the other hand, the top-pions can generate significant contributions to the $`LFV`$ processes, such as $`l_il_j\gamma `$, $`l_il_jl_kl_l`$ and $`Zl_il_j`$ . The branching ratios for some of these processes can be enhanced to their current or future experimental bounds, which can give severe constraints on the free parameters of topcolor scenario.
Among the various $`LFV`$ processes that have been considered in the literature, the most fruitful ones are the radiative decays $`\tau e\gamma `$, $`\tau \mu \gamma `$, $`\tau e\eta `$, and $`\mu e\gamma `$, since their branching ratios are tested with high precision. These processes usually provide the restrictive experimental bounds on the free parameters of the popular specific models beyond the $`SM`$. For example, the contributions of the Higgs bosons to these processes have been extensively studied in Refs.. Taking into account the constraints of the experimental upper limits of these $`LFV`$ processes on the free parameters, the $`LFV`$ decays of the Higgs bosons and possible signals in the future high-energy collider experiments have been studied in Ref..
In this paper, we will study the $`LFV`$ signals of the neutral top-point $`\pi _t^0`$ at the future various lepton colliders in the context of the topcolor-assisted technicolor $`(TC2)`$ models with free parameters being compatible with the most relevant data of $`\mu `$ radiative decay $`\mu e\gamma `$. In this work, we consider the contributions of the neutral top-pion $`\pi _t^0`$ to the $`LFV`$ processes $`\mu ^+\mu ^{}\tau \mu `$ ( or $`\tau e`$ ), $`\gamma \gamma \tau \mu `$ (or $`\tau e`$ ), $`e^+e^{}\tau \mu `$, and $`e\gamma e\pi _t^0e\tau \mu (e)`$ via the $`FC`$ couplings $`\pi _t^0l_il_j`$ and discuss possibility of directly searching for the $`LFV`$ signals in the future lepton colliders via these processes. For completeness, we also include an estimation of the contributions of the new gauge boson $`Z^{}`$ predicted by $`TC2`$ models to these processes and compare them with those of the neutral top-pion $`\pi _t^0`$.
This paper is organized as follows. Section II contains a short summary of the relevant coupling to ordinary particles and decay modes of neutral top-pion $`\pi _t^0`$ in $`TC2`$ models. Discussions of the constraints coming from the $`LFV`$ $`\tau `$ and $`\mu `$ decays, the muon anomalous magnetic moment $`a_\mu `$, and other observables on the relevant free parameters are also given in this section. Section III, IV and V are devoted to the computation of the produce cross sections generated by $`\pi _t^0`$ exchange for the $`LFV`$ processes $`\mu ^+\mu ^{}\tau \mu (\tau e),\gamma \gamma \tau \mu (\tau e),e^+e^{}\tau \mu (\tau e)`$, and $`e\gamma e\tau \mu (e\tau e)`$, respectively. Some phenomenological analyses are also included. Our conclusions are given in Sec. VI.
II. The $`LFV`$ coupling of $`\pi _t^0`$ to ordinary particles
For $`TC2`$ models , technicolor$`(TC)`$ interactions play a main role in breaking the electroweak symmetry. Topcolor interactions make small contributions to $`EWSB`$ and give rise to the main part of the top quark mass, $`(1\epsilon )m_t`$, with the parameter $`\epsilon <<1`$, Thus, there is the following relation:
$$\nu _\pi ^2+F_t^2=\nu _w^2,$$
(1)
where $`\nu _\pi `$ represents the contributions of $`TC`$ interactions to $`EWSB`$, $`\nu _w=\nu /\sqrt{2}`$174GeV. Here $`F_t`$50GeV is the physical top-pion decay constant. This means that the masses of the $`SM`$ gauge bosons $`W`$ and $`Z`$ are given by absorbing the linear combination of the top-pions and technipions. The orthogonal combination of the top-pion and technipions remains unabsorbed and physical. However, the absorbed Goldstone boson linear combination is mostly the technipions, while the physical combination is mostly the top-pions, which are usually called physical top-pions$`(\pi _t^{0,\pm })`$.
For $`TC2`$ models, the underlying interactions, topcolor interactions, are nonuniversal and therefore do not possess Glashow-Iliopoulos-Maiani($`GIM`$) mechanism. The nonuniversal gauge interactions result in the new $`FC`$ coupling vertices when one writes the interactions in the mass eigenbasis. Thus, the top-pions can induce the new $`FC`$ coupling vertices. The couplings of the neutral top-pion $`\pi _t^0`$ to ordinary fermions, which are related to our calculation, can be written as :
$`{\displaystyle \frac{m_t}{\sqrt{2}F_t}}{\displaystyle \frac{\sqrt{\nu _w^2F_t^2}}{\nu _w}}[K_{UR}^{tt}K_{UL}^{tt^{}}\overline{t}\gamma ^5t\pi _t^0+{\displaystyle \frac{m_bm_b^{}}{m_t}}\overline{b}\gamma ^5b\pi _t^0+K_{UR}^{tc}K_{UL}^{tt^{}}\overline{t_L}c_R\pi _t^0]`$
$`+{\displaystyle \frac{m_l}{\sqrt{2}\nu _w}}\overline{l}\gamma ^5l\pi _t^0+{\displaystyle \frac{m_\tau }{\sqrt{2}\nu _w}}K_{\tau i}\overline{\tau }\gamma ^5l_i\pi _t^0,`$ (2)
where $`m_b^{}0.1\epsilon m_t`$ is the part of the bottom-quark mass generated by extended technicolor$`(ETC)`$. $`K_{UL}`$ and $`K_{UR}`$ are rotation matrices that diagonalize the up-quark mass matrix $`M_U`$, i.e. $`K_{UL}^+M_UK_{UR}=M_U^{dia}`$. To yield a realistic form of the Cabibbo-Kobayashi-Maskawa($`CKM`$) matrix $`V`$, it has been shown that their values can be taken as :
$$K_{UL}^{tt}1,K_{UR}^{tt}=1\epsilon ,K_{UR}^{tc}\sqrt{2\epsilon \epsilon ^2}.$$
(3)
In the following calculation, we will take $`K_{UR}^{tc}=\sqrt{2\epsilon \epsilon ^2}`$ and take $`\epsilon `$ as a free parameter, which is assumed to be in the range of 0.01-0.1. $`l=\tau ,\mu `$ or $`e`$, $`l_i`$(i=1,2) is the first(second)generation lepton $`e`$($`\mu `$), and $`k_{\tau i}`$ is the flavor mixing factor between the third-and the first-or second- generation leptons. Certainly, there is also the $`FC`$ scalar coupling $`\pi _t^0\mu e`$. However, the topcolor interactions only contact with the third-generation fermions, and thus, the flavor mixing between the first- and second-generation fermions is very small, which can be ignored.
The limits on the top-pion mass $`m_{\pi _t}`$ might be obtained via studying its effects on various experimental observables. For example, considering the couplings of the neutral top-pion $`\pi _t^0`$ to bottom quark through instanton effects, Ref. has shown that the process $`bs\gamma `$, $`B\overline{B}`$ mixing and $`D\overline{D}`$ mixing demand that the top-pions are likely to be light, with masses of the order of a few hundred $`GeV`$. Since the negative top-pion corrections to the $`Zb\overline{b}`$ branching ratio $`R_b`$ become smaller when the top-pion is heavier, the electroweak precision measurement data of $`R_b`$ give rise to a certain lower bound on the top-pion mass. It was shown that the top-pion mass should not be lighter than the order of $`1TeV`$ to make TC2 models consistent with the electroweak precision measurement data. Reference restudied this problem and found that the top-pion mass $`m_{\pi _t}`$ is allowed to be in the range of a few hundred $`GeV`$ depending on the models. Thus, the value of the top-pion mass $`m_{\pi _t}`$ remains subject to large uncertainty. In general, the top-pion mass $`m_{\pi _t}`$ is allowed to be in the range of a few hundred $`GeV`$ depending on the models. In this paper, we will assume 150$`GeVm_{\pi _t}350GeV`$. In this case, the possible decay modes of $`\pi _t^0`$ are $`b\overline{b},\overline{t}c,\overline{f}f(f`$ is the first-or second-generation fermions, or the third-generation bottom quark or leptons), $`gg,\gamma \gamma `$ and $`\tau l`$, ($`l=\mu `$ or $`e`$). The total decay width which has been calculated in Ref.. However, the modes of decay into leptons are not included in this reference. If we take into account these decay modes, then the branching ratio $`Br(\pi _t^0\tau \mu `$) is in the range of $`7.2\times 10^3k_{\tau \mu }^22.5\times 10^5k_{\tau \mu }^2`$ for 150$`GeVm_{\pi _t}`$350$`GeV`$ and 0.01$`\epsilon 0.1`$.
Since the topcolor interactions only contact with the third- generation fermions and the flavor mixing between the first- and second-generation fermions is very small, the $`B`$ system observables cannot give significant constraints on the flavor mixing factor $`k_{\tau i}`$.
It is well known that the precision measurement of the muon anomalous magnetic moment $`a_\mu `$ is a sensitive test for the new physics beyond the $`SM`$. Comparing the new measurement value of $`a_\mu `$ with the present $`SM`$ prediction, there remains a tantalizing discrepancy:
$$a_\mu ^{exp}a_\mu ^{SM}=(24.5\pm 9)\times 10^{10}.$$
(4)
If we assume that the observed deviation, as shown in $`Eq.`$(4), comes from the contributions of new particles predicted by $`TC2`$ models, then we might obtain a constraint on $`TC2`$ models using this deviation. However, we find that the contributions of $`TC2`$ models to $`a_\mu `$ come mainly from the $`ETC`$ gauge bosons and nonuniversal gauge boson $`Z^{}`$; the contributions of $`\pi _t^0`$ are smaller than $`1\times 10^{13}`$ i.e. $`\delta a_\mu ^{\pi _t^0}<1\times 10^{13}`$. Thus, the recently measurement value of $`a_\mu `$ can not give significant constraints on the free parameters, which are related to the top-pion, such as $`\epsilon ,m_{\pi _t}`$, and $`k_{\tau _i}`$.
The neutral top-pion $`\pi _t^0`$ can produce significant contributions to the $`LFV`$ processes $`l_il_j\gamma `$, $`l_il_jl_kl_l`$, and $`Zl_il_j`$ via the $`FC`$ couplings $`\pi _t^0\tau \mu `$ and $`\pi _t^0\tau e`$, and can enhance the branching ratios of these processes by several orders of magnitude. For the processes $`l_il_j\gamma `$, the contributions come from the on-shell photon penguin diagrams, while the contributions come from both the on-shell photon penguin diagrams and the tree-level diagrams for the processes $`\tau l_il_jl_k`$. Reference has shown that, in all of the parameter space of $`TC2`$ models, the branching ratios of the processes $`\tau l\gamma `$, $`l_il_jl_kl_l`$, and $`Zl_il_j`$ are far below the experimental upper bound on these processes, except for the process $`\mu 3e`$, which might approach the observable threshold of near-future experiments. The present experimental limit of the process $`\mu e\gamma `$ can give severe constraints on the free parameters of $`TC2`$ models. If we assume $`k=k_{\tau \mu }=k_{\tau e}`$ and $`150GeV350GeV`$, then there must be $`k0.16`$. Taking into account this constraint, we will study possible $`LFV`$ signals of the neutral top-pion $`\pi _t^0`$ at future various lepton colliders in the following sections.
III. LFV signals of the neutral top-pion $`\pi _t^0`$ in the future muon collders
A muon collider is an excellent tool to study the properties of a heavy scalar or pseudoscalar and potential new physics effects. It has been shown that a large number of Higgs bosons or new particles, such as technihadron and technipions, can be produced through s-channel resonance processes. The $`FC`$ scalar couplings might be tested at future muon colliders. Thus, the future muon collider opens up the possibility of studying the $`LFV`$ signals mediated by the new scalars.
The neutral top-pion $`\pi _t^0`$ has $`FC`$ couplings to the leptons at tree level and thus can contribute to the processes $`\mu ^+\mu ^{}\tau \mu (\tau e)`$ via $`\pi _t^0`$ exchange in the s-channel. In spite of the fact that the coupling $`\pi _t^0\mu ^+\mu ^{}`$, being proportional to $`m_\mu /v`$, is very small, if the muon collider with the center-of-mass($`c.m.`$) energy $`\sqrt{s}`$ runs on $`\pi _t^0`$ resonance ($`\sqrt{s}=m_{\pi _t}`$), then the neutral top-pion $`\pi _t^0`$ may be produced at an appreciable rate. Thus, the neutral top-pion might give observable $`LFV`$ signals at future muon collider experiments.
After averaging over initial polarization and integrating over the scattering angle, it is straightforward to obtain the unpolarized cross section of the process $`\mu ^+\mu ^{}\tau \mu (\tau e)`$ mediated by $`\pi _t^0`$ exchange. The s-channel resonance cross section for this process can be approximately written as:
$$\sigma (\tau \mu (e))\frac{4\pi }{m_{\pi _t}}Br(\pi _t^0\mu ^+\mu ^{})Br(\pi _t^0\tau \mu (e)),$$
(5)
where $`Br(\pi _t^0\mu ^+\mu ^{})`$ and $`Br(\pi _t^0\tau \mu (e))`$ represent the branching ratios of $`\pi _t^0`$ decaying to $`\mu ^+\mu ^{}`$ and $`\tau \mu (e)`$, respectively. Observably, there is $`\sigma (\tau \mu )\sigma (\tau e)`$ for $`k_{\tau \mu }=k_{\tau e}`$ and neglecting the final state masses, which is different from that generated by the Higgs bosons. In that case, compared the production cross section $`\sigma (\tau \mu )`$, the cross section $`\sigma (\tau e)`$ is suppressed by a factor of $`m_e/m_\mu 1/200`$.
In $`Fig.`$1 we show the resonance production cross section $`\sigma (\tau \mu )`$ as a function of the top-pion mass $`m_{\pi _t}`$ for $`\sqrt{s}=m_{\pi _t}`$ and three values of the mixing parameter $`k`$. Since the cross section $`\sigma (\tau \mu )`$ is not sensitive to the free parameter $`\epsilon `$, we have taken $`\epsilon =0.05`$ in $`Fig.`$1. One can see from $`Fig.`$1 that the $`\sigma (\tau \mu )`$ decreases with $`m_{\pi _t}`$ increasing and the mixing parameter $`k`$ decreasing. For $`m_{\pi _t}m_t+m_c`$, the $`FC`$ channel $`\pi _t^0\overline{t}c`$ opens up and the branching ratios $`Br(\pi _t^0\mu ^+\mu ^{})`$ and $`Br(\pi _t^0\tau \mu )`$ drop substantially, and thus the cross sections $`\sigma (\tau \mu )`$ and $`\sigma (\tau e)`$ drop considerably. The values of $`\sigma (\tau \mu )`$ are in the ranges of $`1.54\times 10^23.4\times 10^2fb`$ and $`5.5\times 10^21.5\times 10^4fb`$ for $`k0.16`$, $`150GeVm_{\pi _t}<180GeV`$, and $`180GeVm_{\pi _t}350GeV`$, respectively. If we assume a future muon collider runing with the $`c.m.`$ energy $`\sqrt{s}=200500GeV`$ and a yearly integrated luminosity of $`=20fb^1`$, then there will be several and up to thousand $`\tau \mu `$(or $`\tau e`$) events to be generated a year for $`m_{\pi _t}<180GeV`$.
For the $`LFV`$ processes $`\mu ^+\mu ^{}\tau \mu (`$ or $`\tau e)`$, the final leptons always emerge back to back and carrying a constant energy which is one half of the $`c.m.`$ energy $`\sqrt{s}`$. The main background would arise from the process $`\mu ^+\mu ^{}\overline{\tau }\mu \overline{\nu _\mu }\nu _\tau `$ or $`\overline{\tau }e\overline{\nu _e}\nu _\tau `$. It has been shown that the contributions to the background come mainly from the diagrams with Higgs boson exchange. The background cross section can only reach a peak around $`m_H=130GeV`$ and then drops quickly out of this range. Thus, as long as $`150GeVm_{\pi _t}m_t+m_c`$, the $`LFV`$ signals of the neutral top pion $`\pi _t^0`$ should be observed via the resonance processes $`\mu ^+\mu ^{}\tau \mu `$(or $`\tau e`$) in the future muon colliders.
$`TC2`$ models also predict the existence of a nonuniversal $`U(1)`$ gauge boson $`Z^{}`$, which can lead to the tree-level $`FC`$ couplings $`Z^{}\tau \mu `$ and $`Z^{}\tau e`$. Thus, the nonuniversal gauge boson $`Z^{}`$ has contributions to the $`LFV`$ processes $`\mu ^+\mu ^{}\tau \mu (`$ or $`\tau e)`$ via the s-channel $`Z^{}`$ exchange. At leading order, the unpolarized cross section $`\sigma ^{}(\tau \mu )`$ generated by $`Z^{}`$ exchange can be written as:
$$\sigma ^{^{}}(\tau \mu )=\frac{25\pi ^2\alpha ^2\lambda _{\tau \mu }^2}{12C_W^6K_1}\frac{s}{(sM_Z^{}^2)^2+M_Z^{}^2\mathrm{\Gamma }_Z^{}^2},$$
(6)
where $`C_W=\mathrm{cos}\theta _W`$, $`\theta _W`$ is the Weinberg angle, $`\lambda _{\tau \mu }`$ is the coupling constant of the $`FC`$ vertex $`Z^{}\tau \mu `$. $`K_1`$ is the mixing parameter between the nonuniversal gauge boson $`Z^{}`$ and the $`SM`$ gauge boson $`Z`$. Using $`Eq.(6)`$, we can easily give the numerical results. However, considering the constraints of the electroweak precision measurement data on the $`Z^{}`$ mass $`M_Z^{}(M_Z^{}>1TeV)`$, the cross section $`\sigma ^{}(\tau \mu )`$ is smaller than $`10^4fb`$ in all of the parameter space at the future muon colliders with $`\sqrt{s}=300500GeV`$, which cannot produce observable signals. Furthermore, even if the nonuniversal gauge boson $`Z^{}`$ can produce observable $`LFV`$ signals, we can use polarized beams to separate the contributions from $`\pi _t^0`$ exchange from those from $`Z^{}`$ exchange.
IV. The neutral top-pion $`\pi _t^0`$ and the $`LFV`$ processes $`\gamma \gamma \pi _t^0\tau \mu (\tau e)`$
It is widely believed that hadron colliders, such as Tevatron and future $`LHC`$, can directly probe possible new physics beyond the $`SM`$ up to a few $`TeV`$, while the future high-energy linear $`e^+e^{}`$ collider $`(ILC)`$ is also required to complement the probe of the new particles with detailed measurement. A future $`ILC`$ will offer an excellent opportunity to study production and decay of the new particles with percent-level-precision. A unique feature of the future $`ILC`$ is that it can be transformed to $`\gamma \gamma `$ colliders with the photon beams generated by the Compton backward scattering of the initial electron and laser beams. Their effective luminosity and energy are expected to be comparable to those of the $`ILC`$. In some scenarios, they are the best instrument for the discovery of signals of new physics.
Similar to the contributions of the neutral top-pion $`\pi _t^0`$ to top-charm associated production at $`\gamma \gamma `$ colliders, the contributions of $`\pi _t^0`$ to the $`LFV`$ processes $`\gamma \gamma \tau \mu (\tau e)`$ via the $`FC`$ couplings $`\pi _t^0l_il_j`$ proceed through the self-energy diagrams, vertex diagrams and the s-channel triangle diagram. We have explicitly calculated the contributions of these Feynman diagrams and found that the dominant contributions come from the s-channel triangle diagram, whereas the remaining diagrams give negligible contributions. Our numerical results are shown in $`Fig.2`$, in which we plot the effective cross section $`\sigma (s)=\sigma (e^+e^{}\gamma \gamma \tau \mu )`$ for the $`LFV`$ process $`\gamma \gamma \tau \mu `$ as a function of the top-pion mass $`m_{\pi _t}`$ for the $`c.m.`$ energy $`\sqrt{s}=500GeV`$, the flavor mixing factor $`k=0.1`$, and three values of the parameter $`\epsilon `$. From this figure, one can see that the cross section $`\sigma (s)`$ for the process $`e^+e^{}\gamma \gamma \tau \mu `$ (or $`\tau e`$) is relatively insensitive to the parameter $`\epsilon `$. As the top-pion mass $`m_{\pi _t}`$ increases, the cross section increases monotonically. For $`k=0.1`$ and $`200GeVm_{\pi _t}350GeV`$, the value of the cross section $`\sigma (e^+e^{}\gamma \gamma \tau \mu )`$ is smaller than $`1\times 10^3fb`$. Certainly, if we assume $`150GeVm_{\pi _t}<180GeV`$, then the cross section increases rapidly and its value can reach $`8.3\times 10^2fb`$.
The flavor mixing parameter $`k`$ controls the strength of the $`FC`$ coupling $`\pi _t^0\tau l(l=\mu `$ or $`e)`$ and further determines the $`LFV`$ signals of the neutral top-pion $`\pi _t^0`$. Its value is severely constrained by the current experimental upper bound on the $`LFV`$ process $`\mu e\gamma `$. There is $`k0.16`$ for $`m_{\pi _t}350GeV`$. Taking into account this constraint, we plot the effective cross section $`\sigma (s)`$ as a function of $`k`$ for $`\sqrt{s}=500GeV,\epsilon =0.05`$, and three values of the top-pion mass $`m_{\pi _t}`$ in $`Fig.`$3.
From the above discussion, we can see that, if we assume $`200GeVm_{\pi _t}<350GeV`$, then the effective cross section $`\sigma (s)`$ is smaller than $`1\times 10^3fb`$ in most of the parameter space consistent with the constraint from the $`LFV`$ process $`\mu e\gamma `$. However, for $`150GeVm_{\pi _t}<180GeV`$, $`\sigma (s)`$ is in the range of $`1\times 10^2fb8\times 10^2fb`$. In this case, there will be several tens $`\tau \mu (e)`$ events to be generated a year at the future $`ILC`$ with $`\sqrt{s}=500GeV`$ and the yearly integrated $`_{int}=340fb^1`$, which might be observed in future $`ILC`$ experiments.
The neutral top-pion $`\pi _t^0`$ can also contribute to the $`LFV`$ processes $`e^+e^{}\tau \mu (\tau e)`$ via the $`FC`$ couplings $`\pi _t^0\tau \mu (e)`$. The relevant Feynman diagrams are shown in $`Fig.`$4. Since the s-channel process $`e^+e^{}\gamma ^{},Z^{}\tau \mu `$ is suppressed by the propagators of the intermediate photon or Z-boson in the future $`ILC`$ with high center-of-mass energy, the cross section of the process $`e^+e^{}\tau \mu `$ generated by the neutral top-pion $`\pi _t^0`$ should be smaller than that of the process $`e^+e^{}\gamma \gamma \tau \mu `$. We have confirmed this expectation through explicit calculation. Our numerical results show that the cross section for the process $`e^+e^{}\tau \mu `$ at the $`ILC`$ with $`\sqrt{s}=500GeV`$ is smaller than $`1\times 10^7fb`$ in all of the parameter space of $`TC2`$ models, which can not be detected in future $`ILC`$ experiments.
The production cross section of the process $`e^+e^{}Z^{}\tau \mu (\tau e)`$ is approximately equal to that of the process $`\mu ^+\mu ^{}Z^{}\tau \mu (\tau e)`$, which is smaller than $`1\times 10^4fb`$ in all of the parameter space of the $`TC2`$ models.
V. The neutral top-pion $`\pi _t^0`$ and the $`LFV`$ process $`e^{}\gamma e^{}\tau \mu `$
A future $`ILC`$ can also operate in $`e^{}\gamma `$ collisions, where the $`\gamma `$ beam is generated by the Compton backward scattering of the incident position and laser beam and its energy and luminosity can reach the same order of magnitude of the corresponding position beam. The $`e^{}\gamma `$ collisions can produce particles which are kinematically not accessible at the $`e^+e^{}`$ collisions at the same colliders. Thus, the $`e^{}\gamma `$ collisions are well suite for studying the production and decays of new particles. In this section, we will discuss the contributions of the neutral top-pion $`\pi _t^0`$ to the $`LFV`$ process $`e^{}\gamma e^{}\tau \mu `$ and see whether the $`LFV`$ signals of $`\pi _t^0`$ can be detected via $`e^{}\gamma `$ collisions in the future $`ILC`$.
In the $`TC2`$ models, $`e^{}\tau \mu `$ production in $`e^{}\gamma `$ collisions proceeds through the process $`e^{}\gamma e^{}\gamma ^{}\gamma e^{}\pi _t^0e^{}\tau \mu `$, in which the $`\gamma `$ beam is generated by the backward Compton scatting of incident positron and laser beam and the $`\gamma ^{}`$ beam is radiated from the $`e^{}`$ beam. The relevant Feynman diagram is shown in $`Fig.`$5. In our calculation, we use the $`Weizs\ddot{a}clcerWilliams`$ approximation and treat the virtual photon $`\gamma ^{}`$ coming from the $`e^{}`$ beam as a real photon. In this case, the effective cross section of the $`LFV`$ process $`e^{}\gamma e\tau \mu `$ in the $`ILC`$ experiments can be obtained by folding the cross section of the subprocess $`\gamma ^{}\gamma \tau \mu `$ with the backscattered photon distribution function $`F_{\gamma /e}(x)`$ and the function $`P_{\gamma /e}(x,E_e)`$, which is the probability of finding a photon with a fraction $`x`$ of energy $`E_e`$ in an ultrarelativistic electron.
Our numerical results are shown in $`Fig.`$6, in which we plot the cross section $`\sigma (s)`$ for the $`LFV`$ process $`e^+e^{}e^{}\gamma e^{}\gamma \gamma ^{}e^{}\tau \mu `$ as a function of $`m_{\pi _t}`$ for $`\sqrt{s}=500GeV`$ and three values of the mixing parameter $`k`$. Since the cross section $`\sigma (s)`$ is not sensitive to the parameter $`\epsilon `$, we have assumed $`\epsilon =0.05`$ in $`Fig.`$6. One can see from $`Fig.`$6 that, for $`180GeVm_{\pi _t}350GeV`$, the value of the cross section $`\sigma (s)`$ is smaller than $`4\times 10^5fb`$. Similar to the case for the process $`e^+e^{}\gamma \gamma \tau \mu `$, the production cross section of this process for $`150GeVm_{\pi _t}<180GeV`$ is significantly larger than that for $`180GeVm_{\pi _t}350GeV`$. For $`150GeVm_{\pi _t}<180GeV`$, the effective cross section of the process $`e^{}\gamma e^{}\tau \mu `$ can reach $`4.1\times 10^3fb`$. Even in this case, i.e. $`150GeVm_{\pi _t}<180GeV`$, the neuatral top-pion $`\pi _t^0`$ cannot produce observable signals via the $`LFV`$ process $`e^{}\gamma e^{}\tau \mu `$ in future lepton collider experiments.
VI. Conclusions
The individual lepton numbers $`L_e`$, $`L_\mu `$, and $`L_\tau `$ are automatically conserved and the tree-level $`LFV`$ processes are absent in the $`SM`$, due to unitary of the leptonic analog of $`CKM`$ mixing matrix and the masslessness of the three neutrinos. However, the neutrino oscillation data provide very strong evidence for mixing and oscillation of the flavor neutrinos, which imply that the separated lepton numbers are not conserved. Thus, any observation of the effects for the $`LFV`$ processes would be a clear signature of new physics. This fact and the improvement of the relevant experimental measurements have brought considerable attention to study these processes in the context of specific popular models beyond the $`SM`$ and see whether the $`LFV`$ effects can be tested in future high-energy experiments.
The topcolor scenario is one of the important candidates for the mechanism of $`EWSB`$. The presence of physical top-pions in the low-energy spectrum is a common feature of topcolor models. Since topcolor interactions are assumed to couple preferentially to the third generation and thus do not possess $`GIM`$ mechanism, the physical top-pions have large Yukawa couplings to the third family fermions and can induce the $`FC`$ scalar couplings, which might give significant contributions to the $`FC`$ processes. The effects of the top-pion on these processes are governed by its mass $`m_{\pi _t}`$ and the relevant flavor mixing factors, which might produce observable signals at future high-energy experiments.
In this paper, we have calculated the contributions of the neutral top-pion $`\pi _t^0`$ predicted by $`TC2`$ models to the $`LFV`$ processes $`\mu ^+\mu ^{}\tau \mu (\tau e),\gamma \gamma \tau \mu (\tau e),e^+e^{}\tau \mu (\tau e)`$, and $`e\gamma e\tau \mu (e\tau e)`$, and discussed its possible $`LFV`$ signals in the future lepton colliders. We find that the value of the cross sections for these $`LFV`$ processes can indeed be enhanced by several orders of magnitude. With reasonable values of the free parameter, some of these processes may be within the observable threshold of near-future lepton collider experiments. For example, taking into account the constrains of the present experimental limit of the $`LFV`$ process $`\mu e\gamma `$ on the mixing factor $`k`$ and assuming $`150GeVm_{\pi _t}<180GeV`$, we find that the cross sections of the $`LFV`$ processes $`\mu ^+\mu ^{}\tau \mu `$ and $`e^+e^{}\gamma \gamma \tau \mu `$ can reach $`1.5\times 10^2fb`$ and $`8.3\times 10^2fb`$, respectively. However, for $`m_{\pi _t}200GeV`$, the cross sections of all of the $`LFV`$ processes are very small, which cannot be detected in future experiments. Thus, we expect that the light top-pions predicted by topcolor scenario might produce observable $`LFV`$ signals in future lepton collider experiments.
ACKNOWLEDGMENTS
This work was supported in part by Program for New Century Excellent Talents in University(NCET-04-0209), the National Natural Science Foundation of China under the Grants No.90203005 and No.10475037, and the Natural Science Foundation of the Liaoning Scientific Committee(20032101).
|
warning/0506/cond-mat0506472.html
|
ar5iv
|
text
|
# Dynamical stabilization of matter-wave solitons revisited
## I Introduction
The nonlinear Schrodinger equation (NLSE) appears in many models of mathematical physics and has numerous applications. The one-dimensional NLSE is famous due to its integrability and soliton solutions. The two-dimensional and three-dimensional versions do not have such properties and are much less explored.
In the last decade dynamics of BECs has attracted enormous amount of interest which in turn is causing a renewed growth of interest in the NLSE, since it is known that NLSE (often called the Gross-Pitaevskii (GP) equation in that context) describes the dynamics of BEC at zero temperature very well RefB1 .
While early analytical studies of BECs were concentrated on (quasi-)one-dimensional systems, (quasi-)2D and 3D systems are more important for real experiments. In 2D and 3D systems analytical treatment of NLSE is very difficult and one has to use approximate methods.
One of the very interesting and complicated phenomena being studied recently is stabilization of BEC by the oscillating scattering length in two and three dimensions.
In 1D geometry, bright solitons can be stable without trapping potential if nonlinearity is attractive and sufficiently strong. In NLSE with attractive interaction (corresponding to BEC with negative scattering length) in 2D free space, kinetic energy can balance interaction energy at certain critical value of nonlinearity $`g_{cr}`$, but the resulting solution (Townes soliton) is unstable. That is, if nonlinearity is either increased or decreased (and kept fixed afterwards), the solution either expands or collapses correspondingly. It was shown by several authors that stabilized solutions are possible with the oscillating scattering length. The oscillations of the scattering length lead to creation of pulsating condensate, i.e. some kind of breather solution. One can draw an analogy with Kapitza pendulum (a pendulum with a rapidly oscillating pivot), where unstable equilibria of unperturbed system is stabilized by means of fast modulation. This idea was already applied to stabilization of beams in nonlinear media Towers . Among many other applications in related fields, the atom wire trap suggested in Ref. Hau should be mentioned. In Refs. SU ; AB the novel application of this stabilization mechanism to BEC physics was presented which in turn encouraged several other works on that subject Abdullaev1 ; Abdullaev2 ; Adhikari ; Garcia ; vektor .
We consider here the problem of stabilization of BEC in 2D free space by means of rapid oscillations of the scattering length in a greater detail (the third dimension is assumed to be excluded from the dynamics, say, due to a tight confinement). The system is described by the GP equation:
$$i\frac{\psi }{t}=\frac{1}{2}^2\psi +\frac{\omega _r^2(t)}{2}r^2\psi +g(t)|\psi |^2\psi ,$$
(1)
where $`r^2=x^2+y^2`$ and $`g(t)=(8\pi m\omega _z/\mathrm{})^{1/2}Na(t)`$ describes the strength of the two-body interaction. The interaction $`g(t)`$ is rapidly oscillating: $`g(t)=g_0+g_1\mathrm{sin}(\mathrm{\Omega }t)`$, while the confinement trap described by $`\omega _r(t)`$ is slowly turned off. Refs. SU ; AB ; Abdullaev1 ; Abdullaev2 suggest it is possible to obtain a dynamically stabilized bright soliton in free space in such a way. Interactions between such objects were very recently studied in Ref. vektor . This is a very interesting phenomenon not only in the context of BECs but also from a broader scope of nonlinear physics.
Such kind of stabilization in 3D has also been reported Adhikari . The latter finding is, however, in some disagreement with other investigations on this topic (for example, Ref. Abdullaev1 ). In Ref. 3D it was shown that the scattering length modulation may indeed provide for the stabilization in 3D, but only in combination with a quasi-1D periodic potential. So 3D geometry might need additional careful examination. In the present paper we concentrate on quasi-2D case only, where also not everything is clear yet. Unlike conventional 1D solitons, higher-dimensional solitonic objects may decay. Therefore, it is interesting to investigate the following question: is there indeed a novel genuine breather solution behind the phenomenon of stabilization? As we show in this paper, it turns out that the phenomenon does not fit into simple models being suggested earlier. For theoretical description of the process, several methods were used by different groups of authors: variational approximation based on the Gaussian anzatz Abdullaev1 ; SU , direct averaging of the GP equation Abdullaev1 , a method based on modulated Townes soliton Abdullaev1 , and the method of moments Garcia . Surprisingly, we find all the methods are not very satisfactory even for qualitative predictions. In brief, the direct averaging of the GP equation has the disadvantage of ommiting terms which are of the same order as those responsible for creation of the effective potential, while three other methods, although very different, all rely on the unwarranted assumption of parabolic dependence of the phase of the stabilized wavefunction on $`r`$: arg $`\psi =\alpha (t)+\beta (t)r^2`$. We find that the behavior of the exact numerical wavefunction is, however, completely different (see Fig. 2). The above-mentioned parabolic approximation (PA) of the phase factor is very popular because it is appealingly simple and indeed often appears in solutions of the time-dependent GP equation Kagan . Usually it comes from self-similar time evolution of the condensate density, for example in 3D the following dynamics of the condensate density is possible $`\rho (x,y,z)=[\lambda _1(t))\lambda _2(t)\lambda _3(t)]^1\rho (x/\lambda _1(t),y/\lambda _2(t),z/\lambda _3(t))`$ , where coefficients $`\lambda _i`$ are coupled by nonlinear differential equations. It is the important finding of the present paper that in our problem a stabilized wavefunction does not have such parabolic phase factor and does not fit into self-similar patterns implied by the above-mentioned methods. This qualitative difference between the exact numerical solution and all theoretical models considered so far was not mentioned earlier. Besides, we noticed presence of steady outgoing flux of atoms in numerical stabilized solutions. So, even numerically there is no 2D soliton so far, but some slowly decaying object instead. Section 2 reviews the abovementioned theoretical methods. In Section 3 we give some results obtained using the variational approximation with non-gaussian trial functions, including ”supergaussian anzatz”. It is shown that a better accuracy can be obtained for predicting critical nonlinearity $`g_{cr}`$, but we were not able to determine accurately such dynamical properties as the frequency of slow oscillations. Additionally, we checked the supergaussian anzatz for another problem: determination of critical number of attractive BEC in a parabolic well, and found it to be much more accurate than the usual gaussian anzatz. This example also demonstrates that the stabilization mechanism is essentially more complicated than that assumed by the present (PA-based) methods, because predictions of the supergaussian anzatz for dynamical properties of the stabilized solution are much less accurate than in static problems.
In Section 4 numerical results are presented and compared with predictions of the theoretical methods discussed in Sections 2 and 3. Configuration of stabilized solution is discussed and dynamics of some integral quantities of the solution is investigated.
In Section 5 concluding remarks are given. We mention the relation between the BEC stabilization problem and stabilization of optical solitons in a layered medium with sign-alternating Kerr nonlinearity.
## II Several approximate methods to study the problem: PA-based methods (Gaussian variational approximation, the modulated Townes soliton, the method of moments), and the direct averaging of the GP equation.
### II.1 PA-based methods
#### II.1.1 Gaussian variational approximation
The variational approach based on the Gaussian approximation (GA) is one of the most often used in studying dynamics of the GP equation. In actual calculations this approximation however often gives a large error as compared to exact numerical results Metens ; Abdullaev2 . For example, in Ref. Metens the Gaussian approximation in dynamics of attractive BEC was compared to exact numerical solution of the GP equation. It was found that in estimating the critical number $`𝒩_c`$ of the condensate (the maximal number of condensed particles in a trap before collapse occurs) the Gaussian approximation gives a 17% error, and similar values of discrepancy for other dynamical quantities (as a useful test, in the Appendix we provide corresponding results obtained with a supergaussian variational ansatz). However, it seems that in this example GA enables to reproduce important features of the system at least qualitatively. The GA was also used in many other treatments of the GP equation using a variational technique. In particular, it was applied to the problem of BEC stabilization by the oscillating scattering length. The Lagrangian density corresponding to the GP equation (1) is
$$L[\psi ]=\frac{i}{2}\left(\frac{\psi }{t}\psi ^{}\frac{\psi ^{}}{t}\psi \right)\frac{1}{2}\left|\frac{\psi }{r}\right|^2\frac{1}{2}g(t)|\psi |^4.$$
(2)
The normalization condition for the wavefunction is $`2\pi _0^{\mathrm{}}|\psi |^2r𝑑r=1`$.
In Ref. SU , a variational method with the following Gaussian anzatz was used,
$$\psi (r,t)=\frac{1}{\sqrt{\pi }R(t)}\mathrm{exp}\left[\frac{r^2}{2R^2(t)}+i\frac{\dot{R}(t)}{2R(t)}r^2\right],$$
(3)
where $`R(t)`$ is the variational parameter that characterizes the size of the condensate, and the phase factor of the wavefunction describes the mass current SU ; AB ; Cirac .
After substitution of expression (3) into the Lagrangian density (2) one obtains the effective Lagrangian $`L=2\pi _0^{\mathrm{}}rL[\psi ]𝑑r`$ and the corresponding Euler-Lagrange equations of motion. One can obtain then the equation of motion for $`R(t)`$ as
$$\ddot{R}(t)=\frac{1}{R^3(t)}+\frac{g_0+g_1\mathrm{sin}\mathrm{\Omega }t}{2\pi R^3(t)}.$$
(4)
So the gist of the model is to represent the 2D BEC as a classical nonlinear pendulum with modulated parameters. It is important that other one-parameter PA-based anzatzes also give the same nonlinear pendulum ($`\ddot{R}=(a+b\mathrm{sin}\mathrm{\Omega }t)/R^3`$, where $`a,b`$ depend on the parameters $`g_1,g_0,\mathrm{\Omega }`$), but with different functional dependence of a,b on the parameters.
The authors of Ref. SU use then the Kapitza averaging method to study behavior of the system with the rapidly oscillating scattering length. They assume the dynamics of $`R`$ can be separated into a slow part $`R_0`$ and a small rapidly oscillating component $`\rho `$: $`R=R_0(t)+\rho (\mathrm{\Omega }t)`$. From the equations of motion for $`R_0`$ and $`\rho `$ one extracts the effective potential for the slow variable $`U(R_0)\frac{A_2}{R_0^2}+\frac{A_6}{R_0^6}`$ and determines its minimum
$$R_{min}=\left(\frac{3}{4\pi (g_0+2\pi )}\right)^{1/4}\left(\frac{g_1}{\mathrm{\Omega }}\right)^{1/2}.$$
(5)
From the expression for the effective potential for $`R_0`$ they obtained dependence of the monopole moment $`<r>`$ and the breathing-mode frequency $`\omega _{br}`$ on parameters $`g_1,\mathrm{\Omega }`$. The frequency of small oscillations (breathing mode) around the minimum is given by SU
$$\omega _{br}^2=\frac{8\mathrm{\Omega }^2}{3g_1^2}(g_0+2\pi )^2.$$
(6)
Their numerical calculations were done for $`g_0=2\pi `$. One can see that theoretical predictions (5) and (6) based on the Gaussian approximation can catch $`(g_1/\mathrm{\Omega })^{1/2}`$ dependence of the monopole moment $`<r>`$ and $`(\mathrm{\Omega }/g_1)`$ dependence of the breathing-mode frequency $`\omega _{br}`$ but cannot determine the corresponding coefficients of proportionality, of which the one in (5) becomes infinity while the one in (6) becomes zero for $`g_0=2\pi `$, the value actually used in the numerical calculations. On the other hand, from numerical calculations they were able to determine the coefficients as $`1.06`$ and $`0.32`$ correspondingly (see Fig. 2 of Ref. SU ). It was also determined in Ref. SU that in order to stabilize the bright soliton, $`|g_0|`$ must exceed the critical value of collapse $`|g_{cr}|`$. Their numerical estimate for $`|g_{cr}|`$ is $``$ 5.8 while theoretical estimate based on Gaussian approximation is $`2\pi 6.28`$. The $`2\pi `$ estimate in fact corresponds to fitting the so-called Townes soliton by a Gaussian trial function as will be discussed below.
Inspired by the idea of comparing a numerical solution with simple model nonlinear pendulum, one may ask if it is possible to obtain a better accord with the numerical experiments using different ansatzes. We study this question in Section 3, and it seems that only the stationary Townes soliton can be fit accurately, but not the stabilized breather solutions.
#### II.1.2 Modulated Townes soliton
A method based on modulated Townes soliton used in Ref. Abdullaev1 ; Garcia should be mentioned. The Townes soliton is a stationary solution to the 2D NLS equation with constant nonlinearity $`g_{cr}`$. In our notations $`|g_{cr}|1.862\pi 5.85`$. This solution is unstable: if $`|g|`$ is slightly increased or decreased, the solution will start to collapse or expand correspondingly. If the value of $`g`$ is close to $`g_{cr}`$, one may search for a solution of the problem with fast oscillating $`g`$ in the form of a modulated Townes soliton, as described in Refs. Abdullaev1 ; Garcia . A solution is sought in the form of
$`\mathrm{\Psi }(r,t)`$ $``$ $`[a(t)]^1R_T[r/a(t)]e^{iS},`$
$`S`$ $`=`$ $`\sigma (t)+{\displaystyle \frac{r^2\dot{a}}{4a}},\dot{\sigma }=a^2,`$ (7)
where $`R_T`$ represents amplitude of the Townes soliton. Then, starting from the approximation (7), one can derive the evolution equation for $`a(t)`$ and so determine the dynamics of the system. Note that the approach is also PA-based. It is inevitable if we are to use one-parameter self-similar trial function in the form of $`|\psi (r,t)|=Af(r/a,t)`$.
#### II.1.3 The method of moments
Another PA-based method we would like to mention here is the method of moments Garcia . One introduces integral quantities $`I_1,I_2,I_3,..`$ as
$`I_1`$ $`=`$ $`{\displaystyle _0^{\mathrm{}}}|\psi |^2𝑑𝐫,I_2={\displaystyle _0^{\mathrm{}}}r^2|\psi |^2𝑑𝐫,`$
$`I_3`$ $`=`$ $`i{\displaystyle _0^{\mathrm{}}}\left(\psi {\displaystyle \frac{\psi }{r}}\psi ^{}{\displaystyle \frac{\psi }{r}}\right)r𝑑𝐫,`$ (8)
$`I_4`$ $`=`$ $`{\displaystyle \frac{1}{2}}{\displaystyle _0^{\mathrm{}}}\left(|\psi |^2+{\displaystyle \frac{n}{2}}g(t)|\psi |^4\right)𝑑𝐫,`$
$`I_5`$ $`=`$ $`{\displaystyle \frac{n}{4}}{\displaystyle _0^{\mathrm{}}}|\psi |^4𝑑𝐫,`$
where $`n=2,3`$ is the dimension of the problem. In 2D, $`d𝐫=2\pi rdr`$, and in 3D $`d𝐫=4\pi r^2dr`$.
For all t, we have $`I_1=1`$. For the remaining $`I_i`$ one can write down the dynamical equations of motion as Garcia :
$`\dot{I_2}=I_3,\dot{I_3}=I_4,\dot{I_4}=g{\displaystyle \frac{n2}{n}}I_5+\dot{g}I_5,`$ (9)
$`\dot{I_5}={\displaystyle \frac{n\pi 2^n}{8}}{\displaystyle _0^{\mathrm{}}}{\displaystyle \frac{|\psi |^4}{r}}{\displaystyle \frac{\mathrm{arg}\psi }{r}}r^{n1}𝑑r`$ (10)
The system of equations for the momenta is not closed because of $`I_5`$, and one should make some approximation in order to close it. In Ref. Garcia it was assumed that
$$\mathrm{arg}\psi =\frac{I_3r^2}{4I_2},$$
(11)
i.e. the phase factor is proportional to $`r^2`$ (so that again it is a PA-based method) and the coefficient of proportionality is given by the ratio of $`I_3`$ and $`I_2`$. Then the system (9) poses dynamical invariants Garcia :
$`Q_1`$ $`=`$ $`2(I_4gI_5)I_2{\displaystyle \frac{1}{4}}I_3^2,`$ (12)
$`Q_2`$ $`=`$ $`2I_2^{n/2}I_5.`$ (13)
With the help of these invariants, the system becomes
$$\ddot{I}_2\frac{1}{2I_2}\left(\dot{I}_2\right)^2=2\left(\frac{Q_1}{I_2}+g\frac{Q_2}{I_2^{n/2}}\right).$$
(14)
Introducing $`X(t)=\sqrt{I_2(t)}`$ one obtains Garcia
$$\ddot{X}=\frac{Q_1}{X^3}+g(t)\frac{Q_2}{X^{n+1}}.$$
(15)
The equation is analogous to that obtained by other PA-based methods. One can investigate the obtained equation (15) using various methods of nonlinear dynamics. The simplest Kapitza averaging method can be used again, but of course it is better to use rigorous averaging technique since modern averaging methods are available AKN which have been extensively used already in plasma physics, hydrodynamics, classical mechanics It1 . The authors of Ref. Garcia fulfilled rigorous analysis of model Eq. 15 using results of Ref. Ref\_Garcia . It is important to have in mind that the relation between the exact dynamics of the full system and that of the model (15) of the method of moments remains unclear, therefore one cannot determine sufficient conditions for stabilization, etc. In Ref. Garcia it was noticed that the correspondence between numerical simulation of full 2D GP equation and dynamics of the model system (15) is not good. As it is seen from Fig. 3 of Ref. Garcia , neither the frequency of slow oscillations nor the position of the minimum of the effective potential is predicted correctly. Nevertheless, we found that in numerical stabilized solutions magnitudes of $`Q_1`$ and $`Q_2`$ are often well-conserved, i.e. they oscillate about some mean value (see Section 4).
### II.2 Direct averaging of the GP equation
Ref. Abdullaev1 also explores the Gaussian variational approximation. Beside that, a very promising method of directly averaging the GP equation was investigated. It is based on an analogous method used for the one-dimensional NLSE with periodically managed dispersion (in the context of optical solitons) Kath . In Ref. Abdullaev1 the solution is sought as an expansion in powers of $`1/\mathrm{\Omega }`$ (in our notation):
$$\psi (r,t)=A(r,T_k)+\mathrm{\Omega }^1u_1(A,\zeta )+\mathrm{\Omega }^2u_2(A,\zeta )+\mathrm{}.,$$
(16)
with $`<u_k>=0`$, where $`<\mathrm{}>`$ stands for the average over the period of the rapid modulation, $`T_k\mathrm{\Omega }^kt`$ are the slow temporal variables ($`k=0,1,2,\mathrm{}`$), while the fast time is $`\zeta =\mathrm{\Omega }t`$. Then, for the first and second corrections the following formulas were obtained:
$`u_1`$ $`=`$ $`i[\mu _1<\mu _1>]|A|^2A,`$
$`\mu _1`$ $``$ $`{\displaystyle _0^\zeta }[g(\tau )<g_1>]𝑑\tau ,`$ (17)
$`u_2`$ $`=`$ $`[\mu _2<\mu _2>][2i|A|^2A_t+iA^2A_t^{}+\mathrm{\Delta }(|A|^2A)]`$
$``$ $`|A|^4A({\displaystyle \frac{1}{2}}[(\mu _1<\mu _1>)^22M]+`$
$`+`$ $`<g>(\mu _2<\mu _2>))],`$
$`\mu _2`$ $`=`$ $`{\displaystyle _0^\zeta }(\mu _1<\mu _1>)𝑑s,M={\displaystyle \frac{1}{2}}(<\mu _1^2><\mu _1>^2).`$
Using these results, the following equation was obtained for the slowly varying field $`A(r,T_0)`$, derived up to the order of $`\mathrm{\Omega }^2`$:
$`i{\displaystyle \frac{A}{t}}`$ $`=`$ $`\mathrm{\Delta }A+|A|^2A+2M\left({\displaystyle \frac{g_1}{\mathrm{\Omega }}}\right)^2[|A|^6A`$
$`3|A|^4\mathrm{\Delta }A`$ $`+`$ $`2|A|^2\mathrm{\Delta }(|A|^2A)+A^2\mathrm{\Delta }(|A|^2A^{})].`$
The above equation was represented in the quasi-Hamiltonian form
$`[1`$ $`+`$ $`6M({\displaystyle \frac{g_1}{\mathrm{\Omega }}}^2|A|^4)]{\displaystyle \frac{A}{t}}={\displaystyle \frac{\delta H_q}{\delta A^{}}},`$
$`H_q`$ $`=`$ $`{\displaystyle }dV[|A|^22M\left({\displaystyle \frac{g_1}{\mathrm{\Omega }}}\right)^2|A|^8`$ (19)
$``$ $`{\displaystyle \frac{1}{2}}|A|^4+4M{\displaystyle \frac{g_1}{\mathrm{\Omega }}}|(|A|^2A)|^2)^2].`$
However, some contribution was missed while deriving Eq. (LABEL:slow). Let us take into account the third correction $`u_3(A,\zeta )`$:
$`\psi (r,t)`$ $`=`$ $`A(r,T_k)+\mathrm{\Omega }^1u_1(A,\zeta )+\mathrm{\Omega }^2u_2(A,\zeta )`$ (20)
$`+`$ $`\mathrm{\Omega }^3u_3(A,\zeta )+\mathrm{}.`$
Then, up to terms of order $`\mathrm{\Omega }^2`$ it changes nothing in r.h.s of Eq.(LABEL:slow) (spatial part), but it adds to l.h.s. of Eq. (LABEL:slow) an undetermined term $`\mathrm{\Omega }^2u_3/\zeta `$ . This term has the same order $`\mathrm{\Omega }^2`$ as the terms from the second correction. So we do not get here a consistent equation for the slow field $`A`$ because we do not have a closed set of equations for the second-order corrections (third order correction becomes second order correction after differentiating in time), and so the quasi-Hamiltonian (19) contains an undetermined error of the second order in $`\mathrm{\Omega }^1`$. The influence of the contribution is not very clear but require additional investigation. Nevertheless, formally the omitted terms have the same order as those responsible for the creation of the effective potential. Having in mind how many difficulties arise in averaging of systems of ordinary differential equations AKN , the rigorous direct averaging of the GP equation constitutes a very interesting and challenging open problem, since in principle it could reveal a true periodic solutions in such oscillating objects.
## III Variational approximation with non-Gaussian ansatzes
Here we try to investigate the system more accurately using some non-Gaussian ansatzes and see if it is possible to get more accurate theoretical estimates. One may be interested in three dynamical quantities of the system: the value of critical nonlinearity $`g_{cr}`$, slow frequency of breathing oscillations of the stabilized soliton $`\omega _{br}`$, and minimum of the effective potential $`R_{min}`$ about which the expectation value of the monopole moment $`<r>`$ oscillates slowly.
Table 1 summarizes results of variational predictions for the critical nonlinearity $`g_{cr}`$ and frequency of small breathing oscillations using several different ansatzes. Note that the phase dependence of a one-parameter trial function is not important for calculating $`g_{cr}`$. It is understood that if we choose a trial wavefunction with its amplitude in the form of $`|\psi (r,t)|=Af[r/a(t)]`$, then we need to use a phase factor with quadratic $`r`$ dependence in order for the ansatz to be self-consistent (i.e., the mass current generated by the changing parameter would be incorporated in the phase factor of an ansatz). On the other hand, since amplitude part of the trial function is just an approximation, one may try to use other forms of phase factor with the same functional form of the amplitude.
When predicting the frequency of breathing oscillations from the corresponding effective potential, it is easy to obtain the result for small amplitude linear breathing oscillations (given in Table 1), but in actual stabilized solutions amplitudes of breathing oscillations are not so small.
It is possible to take into account anharmonicity of breathing oscillations. As was mentioned earlier, all PA-based anzatzes produce the nonlinear pendulum $`\ddot{R}+(a+b\mathrm{sin}\mathrm{\Omega }t)/R^3`$, with a corresponding effective potential having $`R_{min}=\left(\frac{3b^2}{2\mathrm{\Omega }^2a}\right)^{1/4}`$ , $`\omega _{br}=\sqrt{\frac{8}{3}}\mathrm{\Omega }|a/b|,`$ where $`\omega _{br}`$ is the frequency of the small amplitude breathing oscillations (near the bottom of the effective potential). For larger breathing oscillations the (anharmonic) breathing frequency will be amplitude-dependent: $`\omega _{br}^{anh}=2\pi \left(\sqrt{\frac{2}{h}}\left[\frac{x_3}{\sqrt{x_2x_3}}\text{K}(k)+\frac{x_2}{x_1}\sqrt{x_2x_3}\text{E}(k)\right]\right)^1`$, with $`k=\sqrt{\frac{x_2x_1}{x_2x_3}},`$ where $`x_1=R_1^2`$, $`x_2=R_2^2`$ ($`R_1,R_2`$ being the turning points), $`x_3`$ is the third root of the equation $`h=\frac{a}{2x}+\frac{b}{4\mathrm{\Omega }^2x^3}`$. The magnitudes of $`x_1,x_2,x_3,h`$ can be determined from numerically obtained breathing oscillations (but results depend on the choice of a particular anzatz). Even this improvement is not helpful, simply because the parabolic approximation is not valid.
Finding $`g_{cr}`$ only might be considered as an approximation to the stationary Townes soliton by a trial function so that the mass current term equals zero and that a phase factor may be skipped from the calculations. It is known that the Townes soliton $`\psi _t=e^{it}R_T(r,t)`$ at large $`r`$ has asymptotic behavior for its amplitude in the form $`R_Te^r/\sqrt{r}`$ . So that Gaussian ansatz is not very good for finding $`g_{cr}`$ just because it is decaying too fast at large $`r`$. The supergaussian trial function provide a better approximation, namely $`g_{cr}=\pi 2^{\frac{1}{\text{ln}2}}\text{ln}2`$ which corresponds to the supergaussian wavefunction with $`\eta =\eta _T=2\text{ln}2<2`$. Previously the supergaussian ansatz was used to fit stationary solutions of some nonlinear problems including NLS equation in the context of BECs Pramana . The superposition of two Gaussians in the form $`A\mathrm{exp}(\frac{r^2}{2R^2})\text{Cosh}(\gamma \frac{r^2}{2R^2})`$ also enables one to obtain some improvement: $`g_{cr}5.883`$. The Secanth ansatz
$$\psi =\frac{A}{\mathrm{cosh}(r/R)}\mathrm{exp}[iS(\dot{R},R)r^2]$$
works better, with only one parameter it overcomes the above-mentioned two-parameter trial functions. A very good approximation is provided by the simplest ansatz among all considered:
$$\psi =\frac{1}{3R\sqrt{\pi }}\left(1+\frac{r}{2R}\right)\mathrm{exp}\left\{\frac{r}{2R}+iS(\dot{R},R)r^2\right\}.$$
(21)
It fits the Townes soliton adequately both at the origin and asymptotically at infinite $`r`$ ( a pre-exponential multiplier is not so important as the exponential factor ). The pre-exponential factor is needed in order to fulfill the boundary condition in the origin $`lim_{r0}\frac{1}{r}\psi _r<\mathrm{}`$ . Note that in the supergaussian ansatz the former condition is not fulfilled, otherwise (if one included it in a similar way) the result would be better at the cost of more bulky calculations. The accuracy of the prediction implies that ansatz (21) provides a very good approximation to the Townes soliton at fixed $`R`$, and could approximately represent the modulated Townes soliton when $`R`$ is time-dependent and the phase factor with parabolic $`r`$dependence is used in accordance with the continuity condition.
After obtaining estimates for $`g_{cr}`$, one can use the above-mentioned ansatzes in order to find an effective potential, its minimum and frequency of the breathing oscillations of the monopole moment about this minimum in the same way as it was done for the Gaussian ansatz. We checked the Sech ansatz and the supergaussian with quadratic phase dependence. In the supergaussian ansatz the parameter $`\eta `$ was fixed at the value of its ”Townes soliton-like” solution $`\eta =\eta _T=2\text{ln}2`$. In such a way the variational approximation with supergaussian ansatz resembles method of modulated Townes soliton. However, we find that such trial function seriously underestimate minimum of the effective potential (i.e. the mean value about which the monopole moment oscillates). Nevertheless, the result of the Gaussian ansatz is even worse since for $`g_0=2\pi `$ it gives the diverging expression for $`R_{min}`$ and zero for frequency of slow breathing oscillations $`\omega _{br}`$, as mentioned in Section 1 and SU . A natural idea for remedy is to use two-parameter trial functions to reproduce the non-parabolic phase factor dependence on $`r`$. In the supergaussian ansatz it can be done by considering $`\eta `$ as a dynamical (time-dependent) parameter. The problem is that it is difficult to obtain the self-consistent expression for the phase factor. We also try the supergaussian ansatz with fixed $`\eta `$ and with non-quadratic phase dependence (which is unfortunately not self-consistent trial function) $`\psi (r,t)=A\mathrm{exp}\left[\frac{(a+ib)r^{\eta _T}}{2}\right],`$ where $`A,a,b,`$ and $`\eta `$ are all functions of time, parameter $`\eta `$ is fixed at the value of its Townes soliton-like solution $`\eta =\eta _T=2\text{ln}2`$. We find that such modification drastically changes dynamical parameters of the system. Still, the resulting model is the same classical nonlinear pendulum as in the Gaussian approximation, but with different parameters. The rigorous way to employ the two-parameter supergaussian ansatz is to let $`\eta `$ be a dynamical variable and construct a phase factor fulfilling continuity condition for the trial function. One could then obtain the two-dimensional effective potential within the same Kapitza approach.
As a useful test of applicability of the supergaussian anzatz, we determine the critical number of attractive BEC in the 3D parabolic trap studied in Ref. Metens . Their numerical result was $`N_{cr}=1258.5`$, while the gaussian approximation yields $`N_{cr}^G=1467.7`$. We found the supergaussian prediction to be very accurate $`N_{cr}^{SG}=1236.1`$.
## IV Numerical results
Numerical calculations reveal the fact that stabilized solutions do not have parabolic phase factors in contradiction to all the methods considered in Section 2 (except the method of direct averaging). The calculations were done using explicit finite difference schemes. We use explicit finite differences of second and forth order for spatial derivatives and 4-th order Runge-Kutta method for time propagation. We use meshes varying from 2000 to 10000 points, timesteps $`\mathrm{\Delta }t=0.00010.0004`$, and spatial steps $`\mathrm{\Delta }r=0.020.04`$. In addition, we found that it is very important to use absorbing (imaginary) potential at the edge of the mesh, in accordance with the conclusions of Ref. Garcia . Without such an adsorbing potential, a wave reflected from the edge sometimes destroys the otherwise stable solution.
Following SU , initially we start with a Gaussian wavepacket in a parabolic trap. Then the trap was slowly turned off while the oscillating nonlinearity was slowly turned on in a way similar to Ref. SU . In Figure 1 one can see indeed the creation of a stabilized soliton. In Figure 1b and 1d oscillations of amplitude of the wavefunction at the origin are shown. It decays very slowly. In fact, this is in accord with the calculations of Ref. SU : after a careful examination of the corresponding figures in that paper one notices the same behavior. Monopole moment grows very slowly (Figs. 1a,c). We checked that in the case when the trap is not turned off completely, the norm is conserved during the same long time with a high accuracy (of order $`10^8`$), so decay is certainly not due to numerical errors.
In Fig. 2 configuration of the quasi-stabilized wavefunction is shown. One can see the smooth core pulse profile, tiny oscillations in the tail, and an outgoing cylindrical wave leaking from the core pulse. In Fig. 2e the behavior of the phase factor is shown. It is seen to differ from parabolic with $`r`$ considerably.
Figs. 2f,g shows the slow decay of the norm of the solution due to the flux of atoms from the core to infinity. We made a series of numerical experiments with different parameters. We found that the behavior of the matter-wave pulse is often unpredictable. When the Gaussian approximation predicts stabilization, in the corresponding numerical solution it does not necessarily occur. Neither can the method of moments give reliable predictions for the stabilization. We checked the latter method carefully. As it was mentioned already in Sections 1 and 2, the method relies on the crucial approximation of Eq. (11 ). It is due to this approximation one obtains the existence of dynamical invariants $`Q_1`$ and $`Q_2`$ (see Eq. 13). As a result, dynamics is determined by Eq. (15). Returning back to Figure 2, we see a snapshot of the phase factor, arg $`\psi `$, of a stabilized solution. It clearly demonstrates that none of the PA-based methods reproduce the dynamics of the system adequately. Only at small $`r`$ the parabolic law is fulfilled, while the deviation from this quadratic dependence is very strong even at $`r1`$, where the amplitude of the solution is not small at all (and is sufficient to drastically influence the dynamics of the system). Snapshots at other moments produce similar results: the phase of the solution is changing with time but remains very far from being parabolic in $`r`$. It is easy to check that dynamical properties of the system within a variational approximation are very sensitive to $`r`$ dependence in the phase factor of a trial function. To check the dynamics further, we calculated time evolution of the ”invariants” $`Q_1`$ and $`Q_2`$ in the stabilized solution. They are constants in the model but not in the exact numerical solution. We found that in the numerical quasi-stabilized solution these magnitudes oscillate around some mean value. Actually, it was already found in Ref. Garcia that the method of moments does not work for Gaussian initial data, still it is interesting to trace dynamics of relevant magnitudes. The time evolution of $`Q_1`$ and $`Q_2`$, and other magnitudes related to the method of moments are shown in Figures 4, and 5. It is seen that the magnitudes of $`Q_1`$ and $`Q_2`$ related to a stabilized soliton undergo slow oscillations.
When calculating values of $`Q_1`$ and $`Q_2`$, and other properties of the quasi-stabilized solution it is necessary to stop integration at some reasonable value of $`r=r_{max}`$ (we take $`r_{max}=20`$ where the amplitude of the wavefunction becomes very small (of order $`10^4`$ in our case). In that way we separate the properties of the quasi-stabilized soliton from that of the tail which, although has very small amplitude, can carry large moments $`I_2,I_3`$ and would give large contribution to $`Q_1`$ and $`Q_2`$ (so that in the corresponding figures we presented these quantities for the core soliton and the whole solution (including tail) separately).
Similar features can be seen in Fig. 6 where calculations with $`g_0=7.0`$ are presented. Several snapshots of the phase factor at different moments are presented in order to demonstrate that non-quadratic behavior of the phase factor is typical. Time evolution on very long time is traced. We find that sometimes magnitudes of $`Q_1`$ and $`Q_2`$ of stabilized solutions are almost conserved (undergoing small oscillations about its mean value) despite the strongly non-quadratic behavior of the phase factor. It suggests that the method of moments developed in Garcia might provide useful perspective for studying the problem and it would be fruitful to extend it taking into account non-parabolicity of the phase factor.
## V Concluding remarks
Despite there are many publications dedicated to the stabilization of a trapless BEC by the rapidly oscillating scattering length, it seems that the strong non-parabolic behavior of the phase of the stabilized wavefunction has not been brought to attention yet. It should be noted that the role of deviation of the phase profile of NLSE solutions from the parabolic shape was addressed previously in the contexts of solitons in optical fibers in Refs. nonparabolic1 ; nonparabolic2 .
Despite that several independent methods were used previously, we have seen that three of the four theoretical methods used rely on the unwarranted parabolic approximation, while the fourth method (direct averaging of GP equation) is, strictly speaking, incorrect, despite its inspiring motivation (in the sense that the omitted terms has the same order as those responsible for the creation of the effective potential ). Besides, we find that there is no evidence presently for stabilization in a strict sense. It seems that the numerical examples presented so far deal with quasi-stable solutions which slowly decays due to the leaking of atoms from the core pulse as an outgoing cylindrical wave. It means that even from a numerical point of view there are no evidence for true 2D solitons (breathers) yet.
It should be mentioned also that the phenomenon of BEC stabilization has its counterpart in nonlinear optics. As was studied in Ref. Towers , in the periodically alternating Kerr media the stabilization of beams is possible. Mathematically, one deals with a similar NLSE. Instead of the time-dependence of the scattering length of BEC one has dependence of the media nonlinearity coefficient on the coordinate z along which a beam propagates:
$$iu_z+\frac{1}{2}_{tr}^2u+\gamma (z)|u|^2u=0,$$
(22)
where the diffraction operator $`_{tr}^2`$ acts on the transverse coordinate $`x`$ and $`y`$. Nonlinearity coefficient $`\gamma (z)`$ jumps between constant values $`\gamma _\pm `$ of opposite signs inside the layers of widths $`L_\pm `$. The analysis of this problem was done using variational approximation based on a natural Sech ansatz $`U=A(z)\mathrm{exp}[ib(z)r^2+i\varphi (z)]\text{Sech}[r/w(z)]`$. However, behavior of the phase factor was not checked *aposteriori* . We see that it would be useful to investigate the problem of (2+1)-dimensional solitons in a layered medium with sign-alternating Kerr nonlinearity in a greater detail because behavior of the phase factor of the numerical solution has not been reported yet. Interplay between the phenomenon of stabilization in Kerr media and BEC was addressed also in Ref. Adhikari2 in the context of stabilization of (3+1)- dimensional optical solitons and BEC in periodic optical-lattice potential (without addressing the issue of validity of the parabolic approximation).
Returning back to the BEC stabilization, we note that the two main difficulties should be resolved in the future: the non-trivial behavior of the argument of the stabilized wavefunction, and the possibility to stop the leak of atoms from the tail of the solution.
Using several non-Gaussian variational functions, we were able to determine accurately one of the magnitudes characterizing the stabilization phenomena: critical nonlinearity $`g_{cr}`$, but not other dynamical properties such as the frequency of slow oscillations.
## VI Acknowledgements
Alexander Itin (A.I.) acknowledges support by the JSPS fellowship P04315. A.I. thanks Professor Masahito Ueda and Professor S.V. Dmitriev for helpful discussions.
This work was supported in part by Grants-in-Aid for Scientific Research No. 15540381 and 16-04315 from the Ministry of Education, Culture, Sports, Science and Technology, Japan.
|
warning/0506/cond-mat0506573.html
|
ar5iv
|
text
|
# Antiferromagnetism and singlet formation in underdoped high-Tc cuprates: Implications for superconducting pairing
## I Introduction
Since the discovery of high temperature superconducting cuprates, Bednorz1986 it has been established that the strong Coulomb repulsion between electrons plays an essential role in the physics there RMP and the magnetic mechanism of the superconductivity has been studied intensively. There are two streams of thoughts; one is the antiferromagnetic (AF) spin fluctuation exchange based on the (nearly) AF ordered state, while the other is the resonating valence bond (RVB) mechanism with the focus being put on the spin singlet formation by the kinetic exchange interaction $`J`$. The representative of the former is the spin-bag theory Schrieffer1989 ; Schrieffer1989long where the doped carriers into the spin density wave (SDW) state form small hole pockets, and exchange the AF spin fluctuation to result in the $`d_{xy}`$ superconductivity. The $`z`$-component of the spin fluctuation, i.e., $`\chi ^{zz}`$, gives the dominant contribution. The RVB picture, on the other hand, puts more weight on the spin singlet formation and takes into account the order parameter defined on the bond but usually does not consider the antiferromagnetic long range ordering (AFLRO). Anderson1987 These two scenarios have been studied rather separately thus far, and the relation between them remains unclear. Hsu was the first to take into account both the AFLRO and the RVB correlation at half-filling in terms of the Gutzwiller approximation. Hsu1990 His picture is that the AFLRO occurs on top of the $`d`$-wave RVB or equivalently the flux state. It is also supported by the variational Monte Carlo study on Heisenberg model, which shows that the Gutzwiller-projected wavefunction $`\mathrm{\Psi }_{\mathrm{RVB}+\mathrm{SDW}}`$ starting from the coexisting $`d`$-wave RVB and SDW mean field state gives an excellent agreement with the exact diagonalization concerning the ground state energy and staggered moment. tjhole The variational wavefunction projecting the simple SDW state, on the other hand, gives higher energy. These results mean that both aspects, i.e., SDW and RVB, coexist in the Mott insulator. At finite hole doping concentration $`\mathrm{x}`$, the AFLRO is rapidly suppressed and the superconductivity emerges. Here an important question still remains, namely short range AF fluctuation dominates or the RVB correlation is more important. Because there is no AFLRO and no distinction between $`\chi ^{zz}`$ and $`\chi ^\pm `$, this question might appear an academic one with the reality being somewhere inbetween. Well-known results from the variational Monte Carlo studies are that the variational wavefunction $`\mathrm{\Psi }_{\mathrm{RVB}+\mathrm{SDW}}`$ has the lowest energy therefore RVB (superconductivity) and SDW coexist also at small doping level. ogatahimeda From the viewpoint of the particle-hole SU(2) symmetry, this means that the degeneracy, i.e., the SU(2) gauge symmetry, between $`d`$-wave pairing and $`\pi `$-flux state is lifted at finite $`\mathrm{x}`$, and the former has lower energy.
However this accepted view has been challenged by recent studies based on the exact diagonalization tjexact and variational Monte Carlo method. tjhole These works found that the energetically most favorable state with small pockets at small hole doping levels is not superconducting. This nonsuperconducting state is consistent with the doped SDW+$`\pi `$-flux state with the hole pockets near $`𝒌=(\pi /2,\pm \pi /2)`$. This means that the mean field picture is not so reliable at small $`\mathrm{x}`$ because the phase fluctuation of the superconductivity is huge at small $`\mathrm{x}`$ where the charge density, which is the canonical conjugate operator to the phase, is suppressed. Once the superconductivity is destroyed by this quantum fluctuation, the only order surviving is the AFLRO, and the system remains nonsuperconducting. Considering that the AFLRO state at $`\mathrm{x}=0`$ is well described by the SDW+($`d`$-wave RVB or $`\pi `$-flux) state, the relevant mean field state at small but finite $`\mathrm{x}`$ is SDW+$`\pi `$-flux state with small hole pockets.
From the experimental side, recent ARPES datas on Na-CCOC found the small ”Fermi arc” near $`𝒌=(\pi /2,\pm \pi /2)`$, and the $`𝒌`$-dependence of the ”pseudo-gap” is quite different from that of $`(\mathrm{cos}k_x\mathrm{cos}k_y)`$ expected for the $`d_{x^2y^2}`$ pairing. Shen-PRB2003 ; Shen-JPSJ2003 This result strongly suggests that the pseudo-gap is distinct from the superconducting gap, and the superconductivity comes from some other interaction(s) different from $`J`$. Although only the ”arc” is observed experimentally, the Fermi surface can not terminate at some $`𝒌`$-points inside of the Brillouin zone (BZ). Considering that the Fermi surface disappears with a large pseudogap at the anti-nodal direction, i.e., near $`𝒌=(\pi ,0)`$ and $`(0,\pi )`$, it is natural to assume that the small hole pocket is formed, along half of which the intensity is small and/or too broad to be observed experimentally. note Then the question of the pairing symmetry arises because $`d_{x^2y^2}`$ requires the nodes at the small hole pockets, which usually reduces the condensation energy and is energetically unfavorable. The natural symmetry appears to be $`d_{xy}`$ without the nodes at the hole pockets, as has been claimed by the original spin-bag scenario. Schrieffer1989 ; Schrieffer1989long
In this paper, we study the superconductivity of the doped SDW+$`\pi `$-flux state for small $`\mathrm{x}`$. This state includes both the RVB correlation and the AFLRO. There are two important effects of this RVB correlation; one is to introduce the parity anomaly to the nodal Dirac fermions at $`𝒌=(\pi /2,\pm \pi /2)`$ and the other is to enhance the transverse spin-spin correlation $`\chi ^\pm `$ compared with the longitudinal $`\chi ^{zz}`$. These two aspects might have crutial influence on the superconductivity in the underdoped region. We have employed the $`1/N`$-expansion, or random-phase approximation (RPA), to derive the effective interaction between the quasi-particles along the small hole pockets and the pairing force derived from it.
The plan of this paper follows. In Section II, we discuss the model, formulation, and its mean field treatment. The Gaussian (second order) fluctuation around the mean field saddle point is treated and the effective interactions between the quasi-particles are studied in Section III. Section IV is devoted to discussion and conclusions.
## II Model and Mean Field Theory
### II.1 Formulation of the microscopic model
The microscopic model of cuprates we consider for the study is the well-known $`tJ`$ model. Anderson1987 ; Rice1988 The second and third nearest neighbor hopping terms are taken into account, Fukuyama1993 as they are necessary to describe the ARPES experiments measurements. ARPES-1998 We use the slave bosons representation Barnes1976 ; Coleman1984 of the electronic operators
$`c_{i,\sigma }^{}=f_{i,\sigma }^{}b_i,`$ (1)
with the constraint
$`b_i^{}b_i+{\displaystyle \underset{\sigma =,}{}}f_{i,\sigma }^{}f_{i,\sigma }=1.`$ (2)
In terms of this formalism, the double occupancy of each site is excluded. AndersontSL ; Fukuyama1988 ; Lavagna1994 The operators $`f^{}`$, $`f`$ are the fermionic ones while $`b^{}`$, $`b`$ are bosonic ones (slave bosons). KivRokSet ; AndersonPRL1990 In all what follows we will exclusively work at zero temperature, and all the slave bosons will be assumed to be condensed. (In actual calculation, we take an extremely low temperature by technical reason. The results, however, are saturated and can be regarded as those at zero temperature.)
We consider the following Hamiltonian on a square lattice with the lattice constant put to be unity and containing $`\mathrm{N}_\mathrm{s}`$ sites
$`=`$ $``$ $`\left({\displaystyle \underset{i,j}{}}t_{ij}+{\displaystyle \underset{i,j^{^{}}}{}}t_{ij}^{^{}}+{\displaystyle \underset{i,j^{^{\prime \prime }}}{}}t_{ij}^{^{\prime \prime }}\right)`$ (3)
$`\times {\displaystyle \underset{\sigma =,}{}}[b_ib_j^{}f_{i,\sigma }^{}f_{j,\sigma }+b_jb_i^{}f_{j,\sigma }^{}f_{i,\sigma }]`$
$`+`$ $`J{\displaystyle \underset{i,j}{}}𝐒_i𝐒_j,`$
where $``$, $`^{^{}}`$, $`^{^{\prime \prime }}`$ denote nearest neighbor, second nearest neighbor and third nearest neighbor pair, respectively. We assume that the hopping contribution is uniform: $`t_{ij}=t`$ , $`t_{ij}^{^{}}=t^{^{}}`$ , $`t_{ij}^{^{\prime \prime }}=t^{^{\prime \prime }}`$. For $`t^{}<0,t^{\prime \prime }>0`$ the previous Hamiltonian deals with hole doped (p-type) cuprates, while the case $`t^{}>0,t^{\prime \prime }<0`$ concerns electron doped (n-type) cuprates. The unity of energy is: $`t0.4\text{eV}=1`$, and we will only study the hole doped case. Following Ref. ARPES-1998, we assume $`t^{^{}}=0.3`$, $`t^{^{\prime \prime }}=0.2`$ for the extended hopping terms, and $`J=0.3`$ for the Heisenberg interaction. We have checked that slight variations in the numerical values of these parameters do not induce any significant modification of our results, as discussed in Section III.3.
We construct the mean field theory to study this Hamiltonian. It has been recognized that the validity of the mean field theory is much less trivial in the theory with constraint. The simple comparison of energy does not offer the criterion because the mean field theory violates the constraint, and hence can lower the energy in the physically forbidden Hilbert space. Therefore the choice of the mean field theory requires a physical intuition, and we employ the nonsuperconducting saddle point according to the reasons explained in the Introduction. In order to study simultaneously the influences of antiferromagnetism and flux in the resonating valence bond state, we introduce the two mean field parameters. The first one is a staggered magnetization Schrieffer1989 ; Schrieffer1989long
$`m_i^{s_m}=𝐒_i^{s_m},𝐒_i^{s_m}={\displaystyle \frac{1}{2}}{\displaystyle \underset{\sigma ,\sigma ^{^{}}}{}}f_{i,\sigma }^{}\stackrel{~}{\sigma }_{\sigma ,\sigma ^{^{}}}^{s_m}f_{i,\sigma ^{^{}}},`$ (4)
with $`s_m=x`$, $`y`$, $`z`$, and $`\stackrel{~}{\sigma }^{s_m}`$ being the Pauli matrices. The second one is the flux phase parameter Affleck1989 ; Marston1989
$`\chi _{ij}={\displaystyle \underset{\sigma =,}{}}f_{i,\sigma }^{}f_{j,\sigma },\chi =|\chi _{ij}|,`$
$`\chi _{i,i+x}=\chi \mathrm{exp}\left[+\mathrm{i}{\displaystyle \frac{\varphi }{4}}(1)^i\right],`$
$`\chi _{i,i+y}=\chi \mathrm{exp}\left[\mathrm{i}{\displaystyle \frac{\varphi }{4}}(1)^i\right].`$ (5)
Fig. 1 shows the pattern of the flux and the staggered magnetization on the square lattice.
We rewrite the model Hamiltonian (3) as
$$=^t+^m+^\chi ,$$
(6)
with
$`^t`$ $`=\left({\displaystyle \underset{i,j}{}}t{\displaystyle \underset{i,j^{^{}}}{}}t^{^{}}{\displaystyle \underset{i,j^{^{\prime \prime }}}{}}t^{^{\prime \prime }}\right)`$
$`\times {\displaystyle \underset{\sigma }{}}[b_ib_j^{}f_{i,\sigma }^{}f_{j,\sigma }+b_jb_i^{}f_{j,\sigma }^{}f_{i,\sigma }],`$
$`^m`$ $`=\alpha J{\displaystyle \underset{i,j}{}}𝐒_i𝐒_j,`$ (7)
$`^\chi `$ $`=(1\alpha )J{\displaystyle \underset{i,j}{}}𝐒_i𝐒_j.`$
The mean field Hamiltonian for each parameter is given by
$`_{MF}^m=`$ $`\alpha J`$ $`{\displaystyle \underset{i,j}{}}{\displaystyle \underset{s_m=x,y,z}{}}[m_j^{s_m}𝐒_i^{s_m}+m_i^{s_m}𝐒_j^{s_m}`$
$`m_i^{s_m}m_j^{s_m}],`$
$`_{MF}^\chi =`$ $``$ $`{\displaystyle \frac{(1\alpha )J}{2}}{\displaystyle \underset{i,j}{}}{\displaystyle \underset{\sigma }{}}\left[\chi _{ij}f_{j,\sigma }^{}f_{i,\sigma }+\chi _{ij}^{}f_{i,\sigma }^{}f_{j,\sigma }\right]`$
$`+`$ $`{\displaystyle \frac{(1\alpha )J}{2}}{\displaystyle \underset{i,j}{}}\chi _{ij}\chi _{ij}^{}.`$
In the previous expressions a parameter $`\alpha `$ has been introduced to divide the Heisenberg exchange interaction into AF part and flux part. Here, the value of $`\alpha `$ is determined so as to reproduce the optimized result by the Gutzwiller variational method at half-filling. Hsu1990 Hence $`\alpha `$ should be regarded as a variational parameter, but we will keep it constant ($`\alpha =0.301127`$, as explained in Section II.3) independently of the doping. We work in a path integral formalism by using the Stratonovitch - Hubbard transformation. Strato ; Hub The partition function is written
$`𝒵={\displaystyle 𝒟\overline{\mathrm{\Psi }}𝒟\mathrm{\Psi }𝒟\chi 𝒟m𝒟\lambda 𝒟b\mathrm{exp}\left(_0^\beta 𝑑\tau (\tau )\right)},`$ (8)
with $`\beta =1/k_BT`$ the thermal factor and the Lagrangian
$`(\tau )=`$ $``$ $`\left({\displaystyle \underset{i,j}{}}t+{\displaystyle \underset{i,j^{^{}}}{}}t^{^{}}+{\displaystyle \underset{i,j^{^{\prime \prime }}}{}}t^{^{\prime \prime }}\right){\displaystyle \underset{\sigma }{}}\left[b_i(\tau )b_j^{}(\tau )\overline{\mathrm{\Psi }}_{i,\sigma }(\tau )\mathrm{\Psi }_{j,\sigma }(\tau )+b_j(\tau )b_i^{}(\tau )\overline{\mathrm{\Psi }}_{j,\sigma }(\tau )\mathrm{\Psi }_{i,\sigma }(\tau )\right]`$ (9)
$`+`$ $`{\displaystyle \underset{i,\sigma }{}}\overline{\mathrm{\Psi }}_{i,\sigma }(\tau )\left[_\tau \mathrm{i}\lambda _i(\tau )\mu \right]\mathrm{\Psi }_{i,\sigma }(\tau )+{\displaystyle \underset{i}{}}b_i^{}(\tau )[_\tau \mathrm{i}\lambda _i(\tau )]b_i(\tau )`$
$`+`$ $`\alpha J{\displaystyle \underset{i,j}{}}{\displaystyle \underset{\sigma ,\sigma ^{^{}}}{}}{\displaystyle \underset{s=x,y,z}{}}\left[m_j^s(\tau )\overline{\mathrm{\Psi }}_{i,\sigma }(\tau )\stackrel{~}{\sigma }_{\sigma ,\sigma ^{^{}}}^s\mathrm{\Psi }_{i,\sigma ^{^{}}}(\tau )+m_i^s(\tau )\overline{\mathrm{\Psi }}_{j,\sigma }(\tau )\stackrel{~}{\sigma }_{\sigma ,\sigma ^{^{}}}^s\mathrm{\Psi }_{j,\sigma ^{^{}}}(\tau )\right]`$
$``$ $`{\displaystyle \frac{(1\alpha )J}{2}}{\displaystyle \underset{i,j}{}}{\displaystyle \underset{\sigma }{}}\left[\chi _{ij}(\tau )\overline{\mathrm{\Psi }}_{j,\sigma }(\tau )\mathrm{\Psi }_{i,\sigma }(\tau )+\chi _{ij}^{}(\tau )\overline{\mathrm{\Psi }}_{i,\sigma }(\tau )\mathrm{\Psi }_{j,\sigma }(\tau )\right]`$
$``$ $`\alpha J{\displaystyle \underset{i,j}{}}\left[m_i^x(\tau )m_j^x(\tau )+m_i^y(\tau )m_j^y(\tau )+m_i^z(\tau )m_j^z(\tau )\right]+{\displaystyle \frac{(1\alpha )J}{2}}{\displaystyle \underset{i,j}{}}\chi _{ij}(\tau )\chi _{ij}^{}(\tau ).`$
In Eq. (9) $`\overline{\mathrm{\Psi }}_{i,\sigma }`$, $`\mathrm{\Psi }_{i,\sigma }`$ are the Grassmann variables associated with the $`f_{i,\sigma }^{}`$, $`f_{i,\sigma }`$ operators, respectively. The Lagrange multipliers $`(\mathrm{i}\lambda _i)`$ assure that the constraint of no double occupancy (2) is satisfied. The chemical potential $`\mu `$ associated with the fermions controls the electron density $`n_f`$, and hence the hole dopping level $`\mathrm{x}`$.
### II.2 Saddle point hypothesis and derivation of the mean field Hamiltonian
To identify the saddle point solution we consider the following assumptions. As we work at zero temperature, all the holons are supposed to be condensed, i.e., $`(b_i)_0=b_0`$ with $`(b_0)^2=\mathrm{x}`$ where $`\mathrm{x}`$ is the hole doping concentration. Therefore the $`f`$ operators can be simply viewed as the renormalized electron operators. The Lagrange multipliers are considered independent of the sites, which signifies that the constraint is imposed on average at the mean field level
$`(\mathrm{i}\lambda _i)_0=\lambda _0.`$ (10)
For the antiferromagnetic part we impose an AF order polarized in $`z`$-direction
$`(m_i^x)_0=0,(m_i^y)_0=0,(m_i^z)_0=(1)^im.`$ (11)
Concerning the flux contribution we impose at half-filling ($`\mathrm{x}=0`$) a $`\pi `$-flux scheme ($`\varphi =\pi `$), which is energetically the most favorable case for a square lattice RiWieg1989
$`(\chi _{i,i+x})_0=(\chi _i^x)_0=\chi \mathrm{exp}\left[+\mathrm{i}{\displaystyle \frac{\pi }{4}}(1)^i\right],`$
$`(\chi _{i,i+y})_0=(\chi _i^y)_0=\chi \mathrm{exp}\left[\mathrm{i}{\displaystyle \frac{\pi }{4}}(1)^i\right].`$ (12)
To write the mean field Hamiltonian $`_{MF}`$, we note that the wavevector of the AF and $`\pi `$-flux is $`𝑸=(\pi ,\pi )`$. Therefore the summation over the momentum is done in the reduced magnetic first BZ. The borders of the magnetic (or reduced) BZ are given by the Fermi surface of the Hubbard model at half filling. The reduced BZ is characterized by the nesting property under the vector $`𝑸=(\pi ,\pi )`$. $`_{MF}`$ can be expressed in an appropriate spinor basis of the momentum space
$`\left[\begin{array}{c}f_{𝒌,\sigma }\\ f_{𝒌+𝑸,\sigma }\end{array}\right].`$ (15)
We obtain the mean field $`tJ`$ Hamiltonian written in a matrix form
$`_{MF}={\displaystyle \underset{𝒌}{}}{\displaystyle {}_{}{}^{}\underset{\sigma }{}}\left[f_{𝒌,\sigma }^{}f_{𝒌+𝑸,\sigma }^{}\right]\left[\begin{array}{cc}\xi _𝒌& \mathrm{\Delta }_{𝒌,\sigma }^m\mathrm{\Delta }_{𝒌,\sigma }^\chi \\ (\mathrm{\Delta }_{𝒌,\sigma }^m)^{}(\mathrm{\Delta }_{𝒌,\sigma }^\chi )^{}& \xi _{𝒌+𝑸}\end{array}\right]\left[\begin{array}{c}f_{𝒌,\sigma }\\ f_{𝒌+𝑸,\sigma }\end{array}\right].`$ (20)
The summation $`_𝒌^{}`$ in $`𝒌`$-space is over the reduced magnetic BZ, and we have defined the following energies
$`ϵ_𝒌`$ $`=`$ $`2t[\mathrm{cos}(k_x)+\mathrm{cos}(k_y)]4t^{^{}}[\mathrm{cos}(k_x).\mathrm{cos}(k_y)]`$
$`2t^{^{\prime \prime }}\left[\mathrm{cos}(2k_x)+\mathrm{cos}(2k_y)\right],`$
$`\xi _𝒌`$ $`=`$ $`ϵ_𝒌\mathrm{x}(\lambda _0+\mu )`$ (21)
$`(1\alpha )J\chi \mathrm{cos}(\varphi /4)\left[\mathrm{cos}(k_x)+\mathrm{cos}(k_y)\right],`$
$`\xi _{𝒌+𝑸}`$ $`=`$ $`ϵ_{𝒌+𝑸}\mathrm{x}(\lambda _0+\mu )`$ (22)
$`+(1\alpha )J\chi \mathrm{cos}(\varphi /4)\left[\mathrm{cos}(k_x)+\mathrm{cos}(k_y)\right].`$
The order parameters associated with the staggered magnetization and $`\varphi `$-flux are respectively
$`\mathrm{\Delta }_{𝒌,\sigma }^m`$ $`=`$ $`2\alpha Jm\sigma ,`$ (23)
$`\mathrm{\Delta }_{𝒌,\sigma }^\chi `$ $`=`$ $`\text{i}(1\alpha )J\chi \mathrm{sin}(\varphi /4)\left[\mathrm{cos}(k_x)\mathrm{cos}(k_y)\right],`$
for $`\sigma =\pm 1`$. $`_{MF}`$ can be diagonalized as
$`_{MF}={\displaystyle \underset{𝒌}{}}{\displaystyle {}_{}{}^{}\underset{\sigma }{}}\left[E_𝒌^{up}\gamma _{1𝒌,\sigma }^{}\gamma _{1𝒌,\sigma }+E_𝒌^{low}\gamma _{2𝒌,\sigma }^{}\gamma _{2𝒌,\sigma }\right],`$
by a Bogoliubov-Valatin unitary transformation Bogo-trans ; Val-trans
$`\left[\begin{array}{c}f_{𝒌,\sigma }\\ f_{𝒌+𝑸,\sigma }\end{array}\right]=\left[\begin{array}{cc}u_{𝒌,\sigma }& v_𝒌\\ v_𝒌& u_{𝒌,\sigma }^{}\end{array}\right]\left[\begin{array}{c}\gamma _{1𝒌,\sigma }\\ \gamma _{2𝒌,\sigma }\end{array}\right].`$ (30)
In the previous expression $`\gamma _{1𝒌,\sigma }^{}`$, $`\gamma _{1𝒌,\sigma }`$ ($`\gamma _{2𝒌,\sigma }^{}`$, $`\gamma _{2𝒌,\sigma }`$ ) are the creation and annihilation operators of the upper (lower) Hubbard band, respectively, with the corresponding energy eigenvalues (see Fig. 2)
$`E_𝒌^{up}`$ $`=`$ $`+{\displaystyle \frac{1}{2}}\sqrt{\left[\xi _𝒌\xi _{𝒌+𝑸}\right]^2+4(|\mathrm{\Delta }_{𝒌,\sigma }^m|^2+|\mathrm{\Delta }_{𝒌,\sigma }^\chi |^2)}`$ (31)
$`+{\displaystyle \frac{\xi _𝒌+\xi _{𝒌+𝑸}}{2}},`$
$`E_𝒌^{low}`$ $`=`$ $`{\displaystyle \frac{1}{2}}\sqrt{\left[\xi _𝒌\xi _{𝒌+𝑸}\right]^2+4(|\mathrm{\Delta }_{𝒌,\sigma }^m|^2+|\mathrm{\Delta }_{𝒌,\sigma }^\chi |^2)}`$ (32)
$`+{\displaystyle \frac{\xi _𝒌+\xi _{𝒌+𝑸}}{2}}.`$
The elements of the unitary matrix are given by
$`u_{𝒌,\sigma }=\mathrm{cos}(\theta _𝒌).\mathrm{e}^{\mathrm{i}\varphi _{𝒌,\sigma }},v_𝒌=\mathrm{sin}(\theta _𝒌),`$ (33)
with the trigonometric factors
$`\mathrm{cos}(\theta _𝒌)`$
$`=`$ $`\sqrt{{\displaystyle \frac{1}{2}}\left(1+{\displaystyle \frac{\xi _𝒌\xi _{𝒌+𝑸}}{\sqrt{\left(\xi _𝒌\xi _{𝒌+𝑸}\right)^2+4(|\mathrm{\Delta }_{𝒌,\sigma }^m|^2+|\mathrm{\Delta }_{𝒌,\sigma }^\chi |^2)}}}\right)},`$
$`\mathrm{sin}(\theta _𝒌)`$
$`=`$ $`\sqrt{{\displaystyle \frac{1}{2}}\left(1{\displaystyle \frac{\xi _𝒌\xi _{𝒌+𝑸}}{\sqrt{\left(\xi _𝒌\xi _{𝒌+𝑸}\right)^2+4(|\mathrm{\Delta }_{𝒌,\sigma }^m|^2+|\mathrm{\Delta }_{𝒌,\sigma }^\chi |^2)}}}\right)},`$
$`\mathrm{cos}(\varphi _{𝒌,\sigma })={\displaystyle \frac{\mathrm{\Delta }_{𝒌,\sigma }^m}{\sqrt{|\mathrm{\Delta }_{𝒌,\sigma }^m|^2+|\mathrm{\Delta }_{𝒌,\sigma }^\chi |^2)}}},`$
$`\mathrm{sin}(\varphi _{𝒌,\sigma })={\displaystyle \frac{\text{i}\mathrm{\Delta }_{𝒌,\sigma }^\chi }{\sqrt{|\mathrm{\Delta }_{𝒌,\sigma }^m|^2+|\mathrm{\Delta }_{𝒌,\sigma }^\chi |^2)}}}.`$
It is noted here that the unitary transformation contains the complex phase factor $`\mathrm{e}^{\mathrm{i}\varphi _{𝒌,\sigma }}`$, which becomes important when we solve the BCS equation in Section III. In the field-theory terminology, this offers an example of ”parity anomaly” in (2+1)D. QFTPAnomaly
### II.3 Saddle point solution and mean field equations
In the path integral language, the saddle point action is
$`𝒮_0={\displaystyle \underset{𝒌}{}}{\displaystyle {}_{}{}^{}\underset{\sigma }{}}{\displaystyle \underset{\text{i}\omega _n}{}}`$ $`\left[\overline{\mathrm{\Psi }}_{𝒌,\sigma }(\mathrm{i}\omega _n)\overline{\mathrm{\Psi }}_{𝒌+𝑸,\sigma }(\mathrm{i}\omega _n)\right]`$ (36)
$`\times `$ $`\stackrel{~}{𝒢}_0^1(𝒌,\sigma ,\mathrm{i}\omega _n)`$
$`\times `$ $`\left[\begin{array}{c}\mathrm{\Psi }_{𝒌,\sigma }(\mathrm{i}\omega _n)\\ \mathrm{\Psi }_{𝒌+𝑸,\sigma }(\mathrm{i}\omega _n)\end{array}\right],`$
with $`\text{i}\omega _n`$ the fermionic Matsubara frequencies and $`\stackrel{~}{𝒢}_0`$ the free Green’s functions’ matrix
$`\stackrel{~}{𝒢}_0(𝒌,\sigma ,\mathrm{i}\omega _n)`$ (39)
$`=`$ $`{\displaystyle \frac{1}{\omega _n^2+\mathrm{i}\omega _n(\xi _𝒌+\xi _{𝒌+𝑸})\xi _𝒌\xi _{𝒌+𝑸}+|\mathrm{\Delta }_{𝒌,\sigma }^m+\mathrm{\Delta }_{𝒌,\sigma }^\chi |^2}}`$
$`\times \left[\begin{array}{cc}(\mathrm{i}\omega _n\xi _{𝒌+𝑸})& \mathrm{\Delta }_{𝒌,\sigma }^m+\mathrm{\Delta }_{𝒌,\sigma }^\chi \\ \mathrm{\Delta }_{𝒌,\sigma }^m\mathrm{\Delta }_{𝒌,\sigma }^\chi & (\mathrm{i}\omega _n\xi _𝒌)\end{array}\right].`$
Now we expand the action up to the second order with respect to the deviation from the saddle point solution. At the first order we can derive the self-consistent mean field equations. The details are given in Appendix A, and we give here the final results
$`1{\displaystyle \frac{4}{\mathrm{N}_\mathrm{s}}}\alpha J{\displaystyle \underset{𝒌}{}}^{}`$ $`\{`$ $`{\displaystyle \frac{1}{E_𝒌^{up}E_𝒌^{low}}}\}=0,`$ (40)
$`{\displaystyle \frac{1}{\mathrm{N}_\mathrm{s}}}{\displaystyle \underset{𝒌}{}}{}_{}{}^{}\{{\displaystyle \frac{\left[\mathrm{cos}(k_x)+\mathrm{cos}(k_y)\right](\xi _{𝒌+𝑸}\xi _𝒌)\mathrm{cos}(\varphi /4)}{E_𝒌^{up}E_𝒌^{low}}}`$ (41)
$`{\displaystyle \frac{2\text{i}\left[\mathrm{cos}(k_x)\mathrm{cos}(k_y)\right]\mathrm{\Delta }_{𝒌,\sigma }^\chi \mathrm{sin}(\varphi /4)}{E_𝒌^{up}E_𝒌^{low}}}\}=\chi ,`$
$`{\displaystyle \frac{1}{\mathrm{N}_\mathrm{s}}}{\displaystyle \underset{𝒌}{}}{}_{}{}^{}\left\{{\displaystyle \frac{\left[\mathrm{cos}(k_x)+\mathrm{cos}(k_y)\right](\xi _{𝒌+𝑸}\xi _𝒌)}{E_𝒌^{up}E_𝒌^{low}}}\right\}`$ (42)
$`=\chi \mathrm{cos}(\varphi /4),`$
$`{\displaystyle \frac{4}{\mathrm{N}_\mathrm{s}}}{\displaystyle \underset{𝒌}{}}{}_{}{}^{}\{\stackrel{~}{t}_𝒌{\displaystyle \frac{(\xi _{𝒌+𝑸}\xi _𝒌)}{E_𝒌^{up}E_𝒌^{low}}}`$ (43)
$`+(\stackrel{~}{t}_𝒌^{^{}}+\stackrel{~}{t}_𝒌^{^{\prime \prime }}){\displaystyle \frac{(\xi _{𝒌+𝑸}+\xi _𝒌2E_𝒌^{low})}{E_𝒌^{up}E_𝒌^{low}}}\}=\lambda _0,`$
where the quantities $`\xi _𝒌`$, $`\xi _{𝒌\mathbf{+}𝑸}`$, $`\mathrm{\Delta }_{𝒌,\sigma }^\chi `$, $`E_𝒌^{up}`$, $`E_𝒌^{low}`$, $`\stackrel{~}{t}_𝒌`$, $`\stackrel{~}{t}_𝒌^{^{}}`$ and $`\stackrel{~}{t}_𝒌^{^{\prime \prime }}`$ are given in Eqs. (21), (22), (23), (31), (32), (137), (138) and (139), respectively.
We adjust the electron number $`n_f`$ by the chemical potential $`\mu `$. At zero temperature the upper Hubbard band is empty, while the lower band is partially filled as
$`{\displaystyle \frac{1}{\mathrm{N}_\mathrm{s}}}{\displaystyle \underset{𝒌}{}}{\displaystyle \frac{1}{\mathrm{exp}\left[\beta E_𝒌^{low}\right]+1}}=1\mathrm{x},`$ (44)
where ($`_𝒌`$) is extended over the first BZ.
We have solved numerically these equations by discretizing the reduced BZ in 2 millions of points, providing us a precision on the obtained values better than $`10^5`$. In particular at half-filling ($`\mathrm{x}=0`$) we have found
$`\alpha =0.301127,`$ (45)
by imposing the relation: $`m=0.5\chi `$ previously obtained by Hsu. Hsu1990 Away from half-filling this relation between $`m`$ and $`\chi `$ will be changed but the value of $`\alpha `$ (45) is assumed to be the same on all the range of doping. This assumption means that the weight assigned to each of the decoupling terms of the Heisenberg exchange interaction is kept constant as a function of the doping. This assumption does not change the essential features of our results presented below.
We present the numerically obtained values of the mean field parameters as a function of the doping $`\mathrm{x}`$ in Figs. 3 and 4. We see that the Néel state disappears at a very small value of the doping, typically 1.5 %, which is in agreement with the experimental phase diagram of hole doped cuprates. This is in sharp contrast to the previous studies of the $`tJ`$ model, which found that the AF state could remain until doping values of around 15 %. Fukuyama1992 The flux $`\varphi `$ decreases relatively linearly from $`\pi `$ at half-filling, and reaches $`\varphi 2`$ for a doping of 10 %; this result is similar to the data obtained by Hsu et al.. HsuAffl-1991
We also show the shape of the Fermi surface for three doping cases: $`\mathrm{x}=0.01,0.05,0.1`$ in Fig. 5, where the Fermi surface is composed of arcs which delimit the hole pockets located around the four nodes $`𝒌=(\pm \pi /2,\pm \pi /2)`$.
In the next section we will consider the quantum fluctuation around this mean field solution. The second order Gaussian fluctuation corresponds to the RPA. We will calculate the “$`\stackrel{`}{a}laBCS`$BCS ; Schrief-book pairing potential in terms of the exchange of these fluctuations, to see the effects of the coexistence of antiferromagnetism and the staggered flux on the pairing force in underdoped region.
## III Gaussian Fluctuation and BCS Pairing Interaction between Quasi-Particles
### III.1 Second order action and correlation functions
In this Section we expand the action with respect to the fluctuations of the mean field parameters up to second order starting from the saddle point solution for the $`tJ`$ model (20). By analogy with a diagrammatic language such a treatment is equivalent to taking into account all the bubbles associated with the fluctuating bosonic fields. Nagaosa\_books This approach allows us to calculate the different correlation functions (or susceptibilities, both being linked via the fluctuation-dissipation theorem Kubo\_book ) between those fields.
Integrating over the fermionic Grassman variables and expanding the action with respect to the fluctuations of the bosonic modes up to the second order, we obtain
$`𝒮_2={\displaystyle _0^\beta }d\tau \{{\displaystyle \underset{i}{}}(b_0)^2[2\mathrm{i}\delta \lambda _i\delta b_i+\lambda _0(\delta b_i)^2]\alpha J{\displaystyle \underset{i,j}{}}{\displaystyle \underset{s_m=x,y,z}{}}[\delta m_i^{s_m}\delta m_j^{s_m}]+{\displaystyle \frac{(1\alpha )}{2}}J{\displaystyle \underset{i,j}{}}[\delta \chi _{ij}\delta \chi _{ij}^{}]`$
$`({\displaystyle \underset{i,j}{}}t+{\displaystyle \underset{i,j^{^{}}}{}}t^{^{}}+{\displaystyle \underset{i,j^{^{\prime \prime }}}{}}t^{^{\prime \prime }})(b_0)^2\times {\displaystyle \underset{\sigma }{}}\delta b_i\delta b_j[\overline{\mathrm{\Psi }}_{i,\sigma }\mathrm{\Psi }_{j,\sigma }+\overline{\mathrm{\Psi }}_{j,\sigma }\mathrm{\Psi }_{i,\sigma }]\}+{\displaystyle \frac{1}{2}}\mathrm{Tr}[\stackrel{~}{𝒢}_0\stackrel{~}{𝒱}_1\stackrel{~}{𝒢}_0\stackrel{~}{𝒱}_1].`$ (46)
Evaluating the bubbles $`\frac{1}{2}\mathrm{Tr}\left[\stackrel{~}{𝒢}_0\stackrel{~}{𝒱}_1\stackrel{~}{𝒢}_0\stackrel{~}{𝒱}_1\right]`$ as explained in Appendix B, the quadratic action is given by
$`𝒮_2`$ $`=`$ $`{\displaystyle \underset{𝒒}{}}{\displaystyle {}_{}{}^{}\underset{𝒒_\mathrm{𝟏},𝒒_\mathrm{𝟐}=𝒒,𝒒+𝑸}{}}{\displaystyle \underset{\mathrm{i}\omega _{\mathrm{}}}{}}{\displaystyle \underset{i,j=1}{\overset{9}{}}}\delta X_i(𝒒_\mathrm{𝟏},\mathrm{i}\omega _{\mathrm{}})\left(_{i,j}(𝒒,𝒒_\mathrm{𝟏},𝒒_\mathrm{𝟐},\mathrm{i}\omega _{\mathrm{}})+{\displaystyle \frac{1}{2}}\mathrm{\Pi }_{i,j}(𝒒,𝒒_\mathrm{𝟏},𝒒_\mathrm{𝟐},\mathrm{i}\omega _{\mathrm{}})\right)\delta X_j(𝒒_\mathrm{𝟐},\mathrm{i}\omega _{\mathrm{}}),`$ (47)
where the first order fluctuations of the bosonic fields $`\delta X_i`$ are defined in Eq. (168), and the matrix elements $`_{i,j}`$ and $`\mathrm{\Pi }_{i,j}`$ are detailed in Eqs. (169) and (215), respectively. In the previous formula, $`\mathrm{i}\omega _{\mathrm{}}`$ denotes the bosonic Matsubara frequencies.
In a path-integral formalism, the second order term of the effective action gives easy access to the different correlation functions between bosonic fields renormalized at a RPA level Nagaosa\_books
$`\stackrel{~}{𝒞}(𝒒,𝒒_\mathrm{𝟏},𝒒_\mathrm{𝟐},\mathrm{i}\omega _{\mathrm{}})=\left[𝒞_{i,j}(𝒒,𝒒_\mathrm{𝟏},𝒒_\mathrm{𝟐},\mathrm{i}\omega _{\mathrm{}})\right]_{1i,j9}=\left[X_i(𝒒_\mathrm{𝟏},\mathrm{i}\omega _{\mathrm{}})X_j(𝒒_\mathrm{𝟐},\mathrm{i}\omega _{\mathrm{}})_{RPA}\right]_{1i,j9},`$
$`\left[X_i(𝒒_\mathrm{𝟏},\mathrm{i}\omega _{\mathrm{}})X_j(𝒒_\mathrm{𝟐},\mathrm{i}\omega _{\mathrm{}})_{RPA}\right]_{1i,j9}=\left(\stackrel{~}{}(𝒒,𝒒_\mathrm{𝟏},𝒒_\mathrm{𝟐},\mathrm{i}\omega _{\mathrm{}})+{\displaystyle \frac{1}{2}}\stackrel{~}{\mathrm{\Pi }}(𝒒,𝒒_\mathrm{𝟏},𝒒_\mathrm{𝟐},\mathrm{i}\omega _{\mathrm{}})\right)^1.`$ (48)
The behaviour of the obtained transverse spin-spin correlation function $`\chi ^\pm `$ is in agreement with well-known results concerning the Heisenberg antiferromagnets, as discussed in Appendix D.
### III.2 Derivation of the effective action
We can now explicitly calculate the ”$`\stackrel{`}{a}laBCS`$BCS ; Schrief-book pairing potential exchanging all the collective modes described in the previous section. Following Schrieffer $`\mathrm{𝑒𝑡}\mathrm{𝑎𝑙}.`$, Schrieffer1989 ; Schrieffer1989long we adapt their approach to the framework of the $`tJ`$ model. We neglect the retardation effects specific to the Eliashberg theory Schrief-book ; Eliashberg and build a pairing potential by assuming the static limit ($`\mathrm{i}\omega _{\mathrm{}}=0`$). Therefore the frequency dependence will not be mentioned from now on. This is justified in the weak coupling region, where the adiabatic approximation $`k_BT_c,\mathrm{\Delta }_{SC}<<\mathrm{}\omega _D`$ is satisfied ( $`T_c`$: transition temperature, $`\mathrm{\Delta }_{SC}`$: superconducting order parameter, $`\omega _D`$: frequency of the exchanging bosons). This approach is not sufficient to describe the superconducting state in underdoped cuprates where the features of the strong coupling effect are observed experimentally. Nernst However, as will be shown below, the magnetic mechanism based on the generalized spin-bag theory gives only a very small pairing force or is pair breaking. Therefore this weak coupling approximation is justified a posteriori, even though it does not describe the real cuprates.
We start with the linear interaction between the fermions and the bosonic fields (143)
$`𝒮_1^{fe}={\displaystyle \underset{𝒌,𝒒}{}}{\displaystyle {}_{}{}^{}\underset{\sigma ^{^{}},\sigma }{}}`$ $`\left[\overline{\mathrm{\Psi }}_{𝒌+𝒒,\sigma ^{^{}}}\overline{\mathrm{\Psi }}_{𝒌+𝒒+𝑸,\sigma ^{^{}}}\right]`$ (51)
$`\times `$ $`\stackrel{~}{𝒱}_1(𝒌+𝒒,\sigma ^{^{}};𝒌,\sigma )`$
$`\times `$ $`\left[\begin{array}{c}\mathrm{\Psi }_{𝒌,\sigma }\\ \mathrm{\Psi }_{𝒌+𝑸,\sigma }\end{array}\right].`$
Using the notations defined in Appendix B, the interaction matrix $`\stackrel{~}{𝒱}_1`$ (A) can be rewritten as
$`\stackrel{~}{𝒱}_1(𝒌+𝒒,\sigma ^{^{}};𝒌,\sigma )`$
$`=`$ $`{\displaystyle \underset{𝒒_\mathrm{𝟏}=𝒒,𝒒+𝑸}{}}{\displaystyle \underset{i,=1}{\overset{9}{}}}\left({\displaystyle \underset{a^{^{}}=1,2}{}}c_{i,a^{^{}}}^{\sigma ^{^{}},\sigma }(𝒌,𝒒_\mathrm{𝟏})\stackrel{~}{s}_{i,a^{^{}}}(𝒒_\mathrm{𝟏})\right)`$
$`\times \delta X_i(𝒒_\mathrm{𝟏}),`$
where $`\delta X_i`$, $`c_{i,a^{^{}}}`$, $`\stackrel{~}{s}_{i,a^{^{}}}`$ are given by Eqs. (168), (181), (192), and (213), respectively.
We remember that $`\overline{\mathrm{\Psi }}`$, $`\mathrm{\Psi }`$ are the Grassmann variables associated with the $`f^{}`$, $`f`$ spinon operators. As we are interested in the interactions between two $`\gamma _2`$ fermions of the lower Hubbard band, we introduce $`\overline{\mathrm{\Phi }}`$, $`\mathrm{\Phi }`$ the Grassmann variables associated with the $`\gamma _2^{}`$, $`\gamma _2`$ operators. By neglecting the contribution of the upper Hubbard band we obtain from the diagonalization of the mean-field Hamiltonian (30)
$`\mathrm{\Psi }_{𝒌,\sigma }`$ $`=`$ $`\mathrm{sin}(\theta _𝒌)\mathrm{\Phi }_{𝒌,\sigma },`$
$`\mathrm{\Psi }_{𝒌+𝑸,\sigma }`$ $`=`$ $`\mathrm{e}^{\mathrm{i}\varphi _{𝒌,\sigma }}\mathrm{cos}(\theta _𝒌)\mathrm{\Phi }_{𝒌,\sigma }.`$ (53)
The first order action related to the $`\gamma _2`$ operators is given by
$`𝒮_1^{fe}`$ $`=`$ $`{\displaystyle \underset{𝒌,𝒒}{}}{\displaystyle {}_{}{}^{}\underset{\sigma ^{^{}},\sigma }{}}\overline{\mathrm{\Phi }}_{𝒌+𝒒,\sigma ^{^{}}}`$ (57)
$`\times \left[\mathrm{sin}(\theta _{𝒌+𝒒})\mathrm{e}^{\mathrm{i}\varphi _{𝒌+𝒒,\sigma ^{^{}}}}\mathrm{cos}(\theta _{𝒌+𝒒})\right]`$
$`\times \{{\displaystyle \underset{𝒒_\mathrm{𝟏},=𝒒,𝒒+𝑸}{}}{\displaystyle \underset{i,=1}{\overset{9}{}}}\left({\displaystyle \underset{a^{^{}}=1,2}{}}c_{i,a^{^{}}}^{\sigma ^{^{}},\sigma }(𝒌,𝒒_\mathrm{𝟏})\stackrel{~}{s}_{i,a^{^{}}}(𝒒_\mathrm{𝟏})\right)`$
$`\times \delta X_i(𝒒_\mathrm{𝟏})\}`$
$`\times \left[\begin{array}{c}\mathrm{sin}(\theta _𝒌)\\ \mathrm{e}^{\mathrm{i}\varphi _{𝒌,\sigma }}\mathrm{cos}(\theta _𝒌)\end{array}\right]`$
$`\times \mathrm{\Phi }_{𝒌,\sigma }.`$
To incorporate the interaction effects at a RPA level, we build an effective action by adding to $`𝒮_1^{fe}`$ the second order in $`\delta X_i`$ contribution $`𝒮_2`$ , i.e. using Eqs. (47) and (48),
$`𝒮_2`$ $`=`$ $`{\displaystyle \underset{𝒒}{}}{\displaystyle {}_{}{}^{}\underset{𝒒_\mathrm{𝟏},𝒒_\mathrm{𝟐}=𝒒,𝒒+𝑸}{}}{\displaystyle \underset{i,j=1}{\overset{9}{}}}\delta X_i(𝒒_\mathrm{𝟏})`$
$`\times 𝒞_{i,j}^1(𝒒,𝒒_\mathrm{𝟏},𝒒_\mathrm{𝟐})\delta X_j(𝒒_\mathrm{𝟐}),`$
then the effective action $`𝒮^{eff}`$ is given by
$`𝒮^{eff}=𝒮_1^{fe}+𝒮_2.`$ (59)
In order to obtain the pairing potential, we integrate out the bosonic fields $`\delta X_i`$ in $`𝒮^{eff}`$. For convenience we define $`\stackrel{}{\varphi }(𝒒_\mathrm{𝟏})`$ for $`𝒒_\mathrm{𝟏}=𝒒,𝒒+𝑸`$ as
$`\stackrel{}{\varphi }(𝒒_\mathrm{𝟏})=\left[\varphi _i(𝒒_\mathrm{𝟏})\right]_{1i9},`$
$`\varphi _i(𝒒_\mathrm{𝟏})={\displaystyle \frac{1}{2}}{\displaystyle \underset{𝒌_\mathrm{𝟏}}{}}{\displaystyle {}_{}{}^{}\underset{\sigma _1^{^{}},\sigma _1}{}}`$ $`\overline{\mathrm{\Phi }}_{𝒌_\mathrm{𝟏}+𝒒,\sigma _1^{^{}}}\left[\mathrm{sin}(\theta _{𝒌_\mathrm{𝟏}+𝒒})\mathrm{e}^{\mathrm{i}\varphi _{𝒌_\mathrm{𝟏}+𝒒,\sigma _1^{^{}}}}\mathrm{cos}(\theta _{𝒌_\mathrm{𝟏}+𝒒})\right]\left({\displaystyle \underset{a_1^{^{}}=1,2}{}}c_{i,a_1^{^{}}}^{\sigma _1^{^{}},\sigma _1}(𝒌_\mathrm{𝟏},𝒒_\mathrm{𝟏})\stackrel{~}{s}_{i,a_1^{^{}}}(𝒒_\mathrm{𝟏})\right)`$ (62)
$`\times `$ $`\left[\begin{array}{c}\mathrm{sin}(\theta _{𝒌_\mathrm{𝟏}})\\ \mathrm{e}^{\mathrm{i}\varphi _{𝒌_\mathrm{𝟏},\sigma _1}}\mathrm{cos}(\theta _{𝒌_\mathrm{𝟏}})\end{array}\right]\mathrm{\Phi }_{𝒌_\mathrm{𝟏},\sigma _1}.`$
After integrating out the bosonic fields we have
$`𝒮^{eff}={\displaystyle \underset{𝒒}{}}{\displaystyle {}_{}{}^{}\underset{𝒒_\mathrm{𝟏},𝒒_\mathrm{𝟐}=𝒒,𝒒+𝑸}{}}\stackrel{}{\varphi }(𝒒_\mathrm{𝟐})\stackrel{~}{𝒞}(𝒒,𝒒_\mathrm{𝟏},𝒒_\mathrm{𝟐})\stackrel{}{\varphi }(𝒒_\mathrm{𝟏}),`$
or equivalently from Eq. (62)
$`𝒮^{eff}`$ $`=`$ $`{\displaystyle \frac{1}{4}}{\displaystyle \underset{𝒒}{}}{\displaystyle {}_{}{}^{}\underset{𝒒_\mathrm{𝟏},𝒒_\mathrm{𝟐}=𝒒,𝒒+𝑸}{}}{\displaystyle \underset{i,j=1}{\overset{9}{}}}{\displaystyle \underset{𝒌_\mathrm{𝟏},𝒌_\mathrm{𝟐}}{}}{\displaystyle {}_{}{}^{}\underset{\sigma _1^{^{}},\sigma _1,\sigma _2^{^{}},\sigma _2}{}}\overline{\mathrm{\Phi }}_{𝒌_\mathrm{𝟐}𝒒,\sigma _2^{^{}}}\mathrm{\Phi }_{𝒌_\mathrm{𝟐},\sigma _2}\overline{\mathrm{\Phi }}_{𝒌_\mathrm{𝟏}+𝒒,\sigma _1^{^{}}}\mathrm{\Phi }_{𝒌_\mathrm{𝟏},\sigma _1}`$ (68)
$`\times \left\{\left[\mathrm{sin}(\theta _{𝒌_\mathrm{𝟐}𝒒})\mathrm{e}^{\mathrm{i}\varphi _{𝒌_\mathrm{𝟐}𝒒,\sigma _2^{^{}}}}\mathrm{cos}(\theta _{𝒌_\mathrm{𝟐}𝒒})\right]\left({\displaystyle \underset{a_2^{^{}}=1,2}{}}c_{i,a_2^{^{}}}^{\sigma _2^{^{}},\sigma _2}(𝒌_\mathrm{𝟐},𝒒_\mathrm{𝟐})\stackrel{~}{s}_{i,a_2^{^{}}}(𝒒_\mathrm{𝟐})\right)\left[\begin{array}{c}\mathrm{sin}(\theta _{𝒌_\mathrm{𝟐}})\\ \mathrm{e}^{\mathrm{i}\varphi _{𝒌_\mathrm{𝟐},\sigma _2}}\mathrm{cos}(\theta _{𝒌_\mathrm{𝟐}})\end{array}\right]\right\}`$
$`\times 𝒞_{i,j}(𝒒,𝒒_\mathrm{𝟏},𝒒_\mathrm{𝟐})`$
$`\times \left\{\left[\mathrm{sin}(\theta _{𝒌_\mathrm{𝟏}+𝒒})\mathrm{e}^{\mathrm{i}\varphi _{𝒌_\mathrm{𝟏}+𝒒,\sigma _1^{^{}}}}\mathrm{cos}(\theta _{𝒌_\mathrm{𝟏}+𝒒})\right]\left({\displaystyle \underset{a_1^{^{}}=1,2}{}}c_{j,a_1^{^{}}}^{\sigma _1^{^{}},\sigma _1}(𝒌_\mathrm{𝟏},𝒒_\mathrm{𝟏})\stackrel{~}{s}_{j,a_1^{^{}}}(𝒒_\mathrm{𝟏})\right)\left[\begin{array}{c}\mathrm{sin}(\theta _{𝒌_\mathrm{𝟏}})\\ \mathrm{e}^{\mathrm{i}\varphi _{𝒌_\mathrm{𝟏},\sigma _1}}\mathrm{cos}(\theta _{𝒌_\mathrm{𝟏}})\end{array}\right]\right\}.`$
### III.3 Calculation of the BCS pairing interaction
To have a BCS form in the previous effective action (68) we impose the following relations
$``$ $`𝒌_\mathrm{𝟏}=𝒌,𝒌_\mathrm{𝟐}=𝒌,𝒒=𝒌^{^{\mathbf{}}}𝒌,`$ (72)
$`\sigma _1^{^{}}=,\sigma _2^{^{}}=,\sigma _2=,\sigma _1=.`$
$`𝒌_\mathrm{𝟏}=𝒌,𝒌_\mathrm{𝟐}=𝒌,𝒒=𝒌𝒌^{^{\mathbf{}}},`$
$`\sigma _1^{^{}}=,\sigma _2^{^{}}=,\sigma _2=,\sigma _1=.`$
$`𝒌_\mathrm{𝟏}=𝒌,𝒌_\mathrm{𝟐}=𝒌,𝒒=𝒌^{^{\mathbf{}}}+𝒌,`$
$`\sigma _1^{^{}}=,\sigma _2^{^{}}=,\sigma _2=,\sigma _1=.`$
$`𝒌_\mathrm{𝟏}=𝒌,𝒌_\mathrm{𝟐}=𝒌,𝒒=𝒌^{^{\mathbf{}}}𝒌,`$
$`\sigma _1^{^{}}=,\sigma _2^{^{}}=,\sigma _2=,\sigma _1=.`$
The configurations (72) and (72) are related to the $`zz`$ contribution, while the forms (72) and (72) give the $`\pm `$ contribution. Therefore the BCS effective action is
$`𝒮_{BCS}^{eff}={\displaystyle \underset{𝒌,𝒌^{^{\mathbf{}}}}{}}^{}`$ $`V_{BCS}(𝒌,𝒌^{^{\mathbf{}}})`$ (73)
$`\times `$ $`\overline{\mathrm{\Phi }}_{𝒌^{^{\mathbf{}}},}\overline{\mathrm{\Phi }}_{𝒌^{^{\mathbf{}}},}\mathrm{\Phi }_{𝒌,}\mathrm{\Phi }_{𝒌,}.`$
where
$$V_{BCS}(𝒌,𝒌^{^{\mathbf{}}})=V_{BCS}^{zz}(𝒌,𝒌^{^{\mathbf{}}})+V_{BCS}^\pm (𝒌,𝒌^{^{\mathbf{}}})$$
(74)
The $`zz`$ BCS - type pairing potential $`V_{BCS}^{zz}`$ is obtained by putting the configurations (72) and (72) in the effective action (68)
$`V_{BCS}^{zz}(𝒌,𝒌^{^{\mathbf{}}})={\displaystyle \underset{𝒒_\mathrm{𝟏},𝒒_\mathrm{𝟐}=𝒌^{^{\mathbf{}}}𝒌,𝒌^{^{\mathbf{}}}𝒌+𝑸}{}}{\displaystyle \underset{i,j=1}{\overset{9}{}}}\left({\displaystyle \frac{1}{4}}\right)`$
$`\times \{\left[\mathrm{sin}(\theta _𝒌^{^{}})\mathrm{e}^{\mathrm{i}\varphi _{𝒌^{^{}},}}\mathrm{cos}(\theta _𝒌^{^{}})\right]\left({\displaystyle \underset{a_2^{^{}}=1,2}{}}c_{i,a_2^{^{}}}^,(𝒌,𝒒_\mathrm{𝟐})\stackrel{~}{s}_{i,a_2^{^{}}}(𝒒_\mathrm{𝟐})\right)\left[\begin{array}{c}\mathrm{sin}(\theta _𝒌)\\ \mathrm{e}^{\mathrm{i}\varphi _{𝒌,}}\mathrm{cos}(\theta _𝒌)\end{array}\right]𝒞_{i,j}(𝒌^{^{\mathbf{}}}𝒌,𝒒_\mathrm{𝟏},𝒒_\mathrm{𝟐})`$ (77)
$`\times \left[\mathrm{sin}(\theta _𝒌^{^{\mathbf{}}})\mathrm{e}^{\mathrm{i}\varphi _{𝒌^{^{\mathbf{}}},}}\mathrm{cos}(\theta _𝒌^{^{\mathbf{}}})\right]\left({\displaystyle \underset{a_1^{^{}}=1,2}{}}c_{j,a_1^{^{}}}^,(𝒌,𝒒_\mathrm{𝟏})\stackrel{~}{s}_{j,a_1^{^{}}}(𝒒_\mathrm{𝟏})\right)\left[\begin{array}{c}\mathrm{sin}(\theta _𝒌)\\ \mathrm{e}^{\mathrm{i}\varphi _{𝒌,}}\mathrm{cos}(\theta _𝒌)\end{array}\right]`$ (80)
$`+\left[\mathrm{sin}(\theta _𝒌^{^{\mathbf{}}})\mathrm{e}^{\mathrm{i}\varphi _{𝒌^{^{\mathbf{}}},}}\mathrm{cos}(\theta _𝒌^{^{\mathbf{}}})\right]\left({\displaystyle \underset{a_2^{^{}}=1,2}{}}c_{i,a_2^{^{}}}^,(𝒌,𝒒_\mathrm{𝟐})\stackrel{~}{s}_{i,a_2^{^{}}}(𝒒_\mathrm{𝟐})\right)\left[\begin{array}{c}\mathrm{sin}(\theta _𝒌)\\ \mathrm{e}^{\mathrm{i}\varphi _{𝒌,}}\mathrm{cos}(\theta _𝒌)\end{array}\right]𝒞_{i,j}(𝒌𝒌^{^{}},𝒒_\mathrm{𝟏},𝒒_\mathrm{𝟐})`$ (83)
$`\times \left[\mathrm{sin}(\theta _𝒌^{^{}})\mathrm{e}^{\mathrm{i}\varphi _{𝒌^{^{}},}}\mathrm{cos}(\theta _𝒌^{^{}})\right]\left({\displaystyle \underset{a_1^{^{}}=1,2}{}}c_{j,a_1^{^{}}}^,(𝒌,𝒒_\mathrm{𝟏})\stackrel{~}{s}_{j,a_1^{^{}}}(𝒒_\mathrm{𝟏})\right)\left[\begin{array}{c}\mathrm{sin}(\theta _𝒌)\\ \mathrm{e}^{\mathrm{i}\varphi _{𝒌,}}\mathrm{cos}(\theta _𝒌)\end{array}\right]\}.`$ (86)
By imposing the configurations (72) and (72) in the effective action, we get the $`\pm `$ BCS - type pairing potential
$`V_{BCS}^\pm (𝒌,𝒌^{^{\mathbf{}}})={\displaystyle \underset{𝒒_\mathrm{𝟏},𝒒_\mathrm{𝟐}=𝒌^{^{\mathbf{}}}+𝒌,𝒌^{^{\mathbf{}}}+𝒌+𝑸}{}}{\displaystyle \underset{i,j=1}{\overset{9}{}}}\left(+{\displaystyle \frac{1}{4}}\right)`$
$`\times \{\left[\mathrm{sin}(\theta _𝒌^{^{}})\mathrm{e}^{\mathrm{i}\varphi _{𝒌^{^{}},}}\mathrm{cos}(\theta _𝒌^{^{}})\right]\left({\displaystyle \underset{a_2^{^{}}=1,2}{}}c_{i,a_2^{^{}}}^,(𝒌,𝒒_\mathrm{𝟐})\stackrel{~}{s}_{i,a_2^{^{}}}(𝒒_\mathrm{𝟐})\right)\left[\begin{array}{c}\mathrm{sin}(\theta _𝒌)\\ \mathrm{e}^{\mathrm{i}\varphi _{𝒌,}}\mathrm{cos}(\theta _𝒌)\end{array}\right]𝒞_{i,j}(𝒌^{^{\mathbf{}}}+𝒌,𝒒_\mathrm{𝟏},𝒒_\mathrm{𝟐})`$ (89)
$`\times \left[\mathrm{sin}(\theta _𝒌^{^{\mathbf{}}})\mathrm{e}^{\mathrm{i}\varphi _{𝒌^{^{\mathbf{}}},}}\mathrm{cos}(\theta _𝒌^{^{\mathbf{}}})\right]\left({\displaystyle \underset{a_1^{^{}}=1,2}{}}c_{j,a_1^{^{}}}^,(𝒌,𝒒_\mathrm{𝟏})\stackrel{~}{s}_{j,a_1^{^{}}}(𝒒_\mathrm{𝟏})\right)\left[\begin{array}{c}\mathrm{sin}(\theta _𝒌)\\ \mathrm{e}^{\mathrm{i}\varphi _{𝒌,}}\mathrm{cos}(\theta _𝒌)\end{array}\right]`$ (92)
$`+\left[\mathrm{sin}(\theta _𝒌^{^{\mathbf{}}})\mathrm{e}^{\mathrm{i}\varphi _{𝒌^{^{\mathbf{}}},}}\mathrm{cos}(\theta _𝒌^{^{\mathbf{}}})\right]\left({\displaystyle \underset{a_2^{^{}}=1,2}{}}c_{i,a_2^{^{}}}^,(𝒌,𝒒_\mathrm{𝟐})\stackrel{~}{s}_{i,a_2^{^{}}}(𝒒_\mathrm{𝟐})\right)\left[\begin{array}{c}\mathrm{sin}(\theta _𝒌)\\ \mathrm{e}^{\mathrm{i}\varphi _{𝒌,}}\mathrm{cos}(\theta _𝒌)\end{array}\right]𝒞_{i,j}(𝒌𝒌^{^{}},𝒒_\mathrm{𝟏},𝒒_\mathrm{𝟐})`$ (95)
$`\times \left[\mathrm{sin}(\theta _𝒌^{^{}})\mathrm{e}^{\mathrm{i}\varphi _{𝒌^{^{}},}}\mathrm{cos}(\theta _𝒌^{^{}})\right]\left({\displaystyle \underset{a_1^{^{}}=1,2}{}}c_{j,a_1^{^{}}}^,(𝒌,𝒒_\mathrm{𝟏})\stackrel{~}{s}_{j,a_1^{^{}}}(𝒒_\mathrm{𝟏})\right)\left[\begin{array}{c}\mathrm{sin}(\theta _𝒌)\\ \mathrm{e}^{\mathrm{i}\varphi _{𝒌,}}\mathrm{cos}(\theta _𝒌)\end{array}\right]\}.`$ (98)
The correlation functions and pairing interactions have been numerically calculated, the results of which are presented in Figs. 6 and 7. These figures show the effective interactions between two $`\gamma _2`$ fermions, one located at the fixed position $`𝒌=(\pi /2,\pi /2)`$, while the other one is at $`𝒌^{^{\mathbf{}}}`$ moving inside the magnetic BZ. The intensities of the effective interactions are plotted as a function of $`𝒌^{^{\mathbf{}}}`$: for example the values given for $`𝒌^{^{\mathbf{}}}=(\pi /2,\pi /2)`$ correspond to a wavevector $`𝒒_𝒅=𝒌^{^{\mathbf{}}}𝒌=(\pi ,\pi )`$. As we assume the static limit ($`\mathrm{i}\omega _{\mathrm{}}=0`$), the pairing potentials take real values. The dark color region means the repulsive interaction ($`V(𝒌,𝒌^{^{\mathbf{}}})>0`$), while the light region displays the attractive interaction ($`V(𝒌,𝒌^{^{\mathbf{}}})<0`$). An interesting feature we can observe in Figs. 6 and 7 is that the total pairing potential $`V_{BCS}`$ is almost completely driven by the $`zz`$ (i.e. longitudinal or ”spin-bag”) contribution, without any major modication introduced by the $`\pm `$ (i.e. transverse or ”RVB”) correlation. As mentioned in Section II.1, the results displayed in Figs. 6 and 7 have been obtained by assuming $`t^{^{}}=0.3`$, $`t^{^{\prime \prime }}=0.2`$, and $`J=0.3`$. We have also considered several additional sets of numerical values for the parameters in the range: $`0.35t^{^{}}0.3`$, $`0.2t^{^{\prime \prime }}0.25`$, $`0.3J0.35`$. We do not have observed any significant modification of the results. The robustness of these results with respect to slight variations of the parameter values can be explained as followed. In the present approach, the fermion spectrum contains two Dirac fermions and associated small hole pockets at the nodal points. The appearance of Dirac fermions means that a topologically nontrivial feature is introduced, which does not exist in the previous spin bag theory. Such a property gives qualitatively specific characteristics to the system which are independent of slight variations of the parameter values, and therefore can explain the robustness of the results. In the next section, we will discuss the implications of the interaction $`V_{BCS}`$ on the possible superconductivity in this model.
## IV Discussion and Conclusions
In previous sections, we have calculated the effective interactions between the two holes in the background of the AF and RVB orders. There are four half-pockets of holes in the reduced first BZ as shown in Fig. 7, or the two hole pockets in the shifted BZ. These small hole pockets around $`𝒌=(\pm \pi /2,\pm \pi /2)`$ are different from those of the simple SDW state as assumed in the original spin-bag theory. Schrieffer1989 ; Schrieffer1989long Namely the description of the antiferromagnetic state includes the singlet formation via the RVB order parameter in addition to the Néel order, and correspondingly the wavefunctions of the doped holes are distinct from the SDW state. Already there opens the gap near $`𝒌=(\pm \pi ,0),(0,\pm \pi )`$ due to the staggered flux, and the superconducting pairings near these points are irrelevant. However the pairing force is still dominated by the $`zz`$-component of the spin susceptibility, and is repulsive for the momentum transfer $`𝒒_𝒅=\pm (\pi ,\pi )`$. When the superconducting gaps at $`𝒌=(\pm \pi ,0),(0,\pm \pi )`$ are dominant, this leads to the $`d_{x^2y^2}`$ pairing as discussed by several authors moriya since the sign of the order parameter is opposite between $`𝒌=(\pm \pi ,0)`$ and $`𝒌=(0,\pm \pi )`$. However in the present case, only the states near $`𝒌=(\pm \pi /2,\pm \pi /2)`$ could contribute to the pairing. One of the main predictions of the spin-bag theory Schrieffer1989 ; Schrieffer1989long is that the superconductivity occurs through the pairing of holes in the small pockets, which are delimited by the Fermi surface as we can see in Fig. 5. The present study shows that the total pairing potential $`V_{BCS}`$ is clearly repulsive, or pair breaking, on all the range of doping for a wavevector $`𝒒_𝒅`$ in the $`(0,0)`$, $`(\pi ,0)`$ and $`(0,\pi )`$ areas of the magnetic BZ. An attractive behaviour is also found when the momentum transfer is around $`(\pi ,\pi )`$. It corresponds to relatively large values of $`𝒒_𝒅`$. Namely, the repulsive interaction (dark contribution) near $`𝒌=(\pi /2,\pi /2)`$ is dominating over the attractive (light) region with large $`𝒒_𝒅`$. Therefore as long as one considers the pairing form which is constant over each of the small hole pocket, there occurs no superconducting instability because the intra-pocket pair breaking force is larger than the inter-pocket force independently of the relative sign of the order parameters between different pockets.
On the other hand, one can consider the pairing with the nodes in the small hole pockets as in the case of $`d_{x^2y^2}`$ pairing realized in real cuprates. Tsuei-RMP2000 ; Tsuei-Nat1997 In this case, however, most of the interaction cancels within the small pocket due to the sign change of the order parameter. We could not determine the sign of the residual pairing interaction, but we can safely conclude that it is very weak even though pair creating. Therefore we conclude that the doped staggered flux state is stable against the superconducting instability when only the magnetic mechanism is considered. This is in accordance with the recent exact diagonalization studies on the $`tJ`$ model at small doping. tjhole Furthermore this suggests that the other mechanism of superconductivity is active in cuprates at least for small hole doping region. Also a related work has been done by Singh and Tešanović, SiResa-PRB1990 where the quantum correction to the spin-bag mean field theory is studied up to one loop level and also the doped case is considered. This is along the same line as the present study, and they also obtained the repulsive interaction between two holes. The new aspect introduced in our study is the RVB correlation represented by the flux order parameter. Combining these two works, the pair breaking nature of the magnetic interactions seems to be rather robust in the low hole doping limit.
There are two possible routes leading to the superconductivity. One possibility is that the starting mean field ansatz becomes inappropriate for optimal doping case. According to the SU(2) formulation of the RVB state, the staggered flux state is the quantum mechanical mixture of the flux state we discussed above and the $`d_{x^2y^2}`$ pairing state, and the latter one is more and more weighted as the doping proceeds. Therefore our formulation is valid only for the small doping region and can not capture the crossover to the superconducting state since it is based on the perturbative method around the AF + flux state. On the other hand, starting from the d-wave superconducting state in the SU(2) formalism, the staggered flux fluctuation has been studied. KimLee-AnnPhys1999 ; HonLee-PRL2003 This is another language describing the instability towards the AF ordering, which is represented by the chiral symmetry breaking in the gauge theory. It is also found the self-energy due to the staggered flux fluctuation leads to the quasi-particle damping strongly anisotropic in the momentum space. HonLee-PRL2003 These works are the approach from the superconducting side, and are complementary to ours.
Taking the view that the magnetic interaction is pair breaking in the low hole doping limit, we should look for other forces which are active in the cuprates. One of the most promising candidates for this is the electron-phonon interaction, NagaosaPhonons ; ZeyPhon which already manifests itself in the angle-resolved photoemission spectroscopy. elph However, much more work is needed to establish this scenario. Especially the interplay between the strong correlation and the electron-phonon interaction remains an important issue to be studied.
In conclusion, we have studied the extended spin-bag scenario taking into account the staggered flux resonating valence bond order in addition to the antiferromagnetic order. The pairing potential between the quasi-particles has been calculated at the Gaussian level, and it is found that the magnetically mediated interaction is pair breaking or very weak at the small hole doping limit, suggesting the other mechanisms such as electron-phonon interaction are active.
###### Acknowledgements.
D. B. is thankful to Mireille Lavagna, Andrés Jerez and Roland Zeyher for extremely fruitful discussions.
## Appendix A Derivation of the self-consistent equations
We allow the different fields to fluctuate at the first order around their saddle point values (10), (11) and (12)
$`b_i=b_0(1+\delta b_i);`$
$`\mathrm{i}\lambda _i=\lambda _0+\mathrm{i}\delta \lambda _i;`$
$`m_i^{s_m}=(m_i^{s_m})_0+\delta m_i^{s_m},s_m=x,y,z;`$
$`\chi _{i,i+s_\chi }=\chi _i^{s_\chi }=(\chi _i^{s_\chi })_0+\delta \chi _i^{s_\chi },`$
$`\delta \chi _i^{s_\chi }=\delta ^{^{}}\chi _i^{s_\chi }+\mathrm{i}\delta ^{^{\prime \prime }}\chi _i^{s_\chi },s_\chi =x,y.`$
By expanding the fluctuations at the first order we obtain
$`𝒮_1={\displaystyle _0^\beta }d\tau \{`$ $``$ $`{\displaystyle \underset{i}{}}(b_0)^2(\mathrm{i}\delta \lambda _i+2\lambda _0\delta b_i)\alpha J{\displaystyle \underset{i,j}{}}\left[(1)^im\delta m_j^z+(1)^jm\delta m_i^z\right]`$ (99)
$`+`$ $`{\displaystyle \frac{(1\alpha )}{2}}J{\displaystyle \underset{i,j}{}}\left[(\chi _{ij})_0\delta \chi _{ij}^{}+(\chi _{ij})_0^{}\delta \chi _{ij}\right]{\displaystyle \underset{i,\sigma }{}}\left[\mathrm{i}\delta \lambda _i\overline{\mathrm{\Psi }}_{i,\sigma }\mathrm{\Psi }_{i,\sigma }\right]`$
$``$ $`\left({\displaystyle \underset{i,j}{}}t+{\displaystyle \underset{i,j^{^{}}}{}}t^{^{}}+{\displaystyle \underset{i,j^{^{\prime \prime }}}{}}t^{^{\prime \prime }}\right){\displaystyle \underset{\sigma }{}}(b_0)^2(\delta b_i+\delta b_j)\left[\overline{\mathrm{\Psi }}_{i,\sigma }\mathrm{\Psi }_{j,\sigma }+\overline{\mathrm{\Psi }}_{j,\sigma }\mathrm{\Psi }_{i,\sigma }\right]`$
$`+`$ $`{\displaystyle \frac{\alpha }{2}}J{\displaystyle \underset{i,j}{}}{\displaystyle \underset{\sigma ,\sigma ^{^{}}}{}}{\displaystyle \underset{s_m=x,y,z}{}}\left[\delta m_j^{s_m}(\overline{\mathrm{\Psi }}_{i,\sigma }\stackrel{~}{\sigma }_{\sigma ,\sigma ^{^{}}}^{s_m}\mathrm{\Psi }_{i,\sigma ^{^{}}})+\delta m_i^{s_m}(\overline{\mathrm{\Psi }}_{j,\sigma }\stackrel{~}{\sigma }_{\sigma ,\sigma ^{^{}}}^{s_m}\mathrm{\Psi }_{j,\sigma ^{^{}}})\right]`$
$``$ $`{\displaystyle \frac{(1\alpha )}{2}}J{\displaystyle \underset{i,j}{}}{\displaystyle \underset{\sigma }{}}[\delta \chi _{i,j}\overline{\mathrm{\Psi }}_{j,\sigma }\mathrm{\Psi }_{i,\sigma }+\delta \chi _{i,j}^{}\overline{\mathrm{\Psi }}_{i,\sigma }\mathrm{\Psi }_{j,\sigma }]\}.`$
In the preceding formula, like in those which will follow in the appendices, the imaginary time dependence of the Grassmann variables and first order bosonic fluctuations follows the Lagrangian (9) and is implicit. For clarity in the following calculations we decompose $`𝒮_1`$ into two parts
$`𝒮_1=𝒮_1^{bo}+𝒮_1^{fe}.`$ (100)
$`𝒮_1^{bo}`$ contains terms with exclusively auxiliary bosonic fields and no Grassmann variables. It corresponds to the three first terms of Eq. (99)
$`𝒮_1^{bo}={\displaystyle _0^\beta }d\tau \{{\displaystyle \underset{i}{}}(b_0)^2(\mathrm{i}\delta \lambda _i+2\lambda _0\delta b_i)`$ (101)
$`\alpha J{\displaystyle \underset{i,j}{}}\left[(1)^im\delta m_j^z+(1)^jm\delta m_i^z\right]`$
$`+{\displaystyle \frac{(1\alpha )}{2}}J{\displaystyle \underset{i,j}{}}[(\chi _{ij})_0\delta \chi _{ij}^{}+(\chi _{ij})_0^{}\delta \chi _{ij}]\}.`$
$`𝒮_1^{fe}`$ includes all the terms containing fermionic variables $`\overline{\mathrm{\Psi }}`$, $`\mathrm{\Psi }`$. It takes the four last terms of Eq. (99)
$`𝒮_1^{fe}={\displaystyle _0^\beta }d\tau \{{\displaystyle \underset{i,\sigma }{}}[\mathrm{i}\delta \lambda _i\overline{\mathrm{\Psi }}_{i,\sigma }\mathrm{\Psi }_{i,\sigma }]`$ (102)
$`\left({\displaystyle \underset{i,j}{}}t+{\displaystyle \underset{i,j^{^{}}}{}}t^{^{}}+{\displaystyle \underset{i,j^{^{\prime \prime }}}{}}t^{^{\prime \prime }}\right)(b_0)^2`$
$`\times {\displaystyle \underset{\sigma }{}}(\delta b_i+\delta b_j)[\overline{\mathrm{\Psi }}_{i,\sigma }\mathrm{\Psi }_{j,\sigma }+\overline{\mathrm{\Psi }}_{j,\sigma }\mathrm{\Psi }_{i,\sigma }]`$
$`+{\displaystyle \frac{\alpha }{2}}J{\displaystyle \underset{i,j}{}}{\displaystyle \underset{\sigma ,\sigma ^{^{}}}{}}{\displaystyle \underset{s_m=x,y,z}{}}[\delta m_j^{s_m}(\overline{\mathrm{\Psi }}_{i,\sigma }\stackrel{~}{\sigma }_{\sigma ,\sigma ^{^{}}}^{s_m}\mathrm{\Psi }_{i,\sigma ^{^{}}})`$
$`+\delta m_i^{s_m}(\overline{\mathrm{\Psi }}_{j,\sigma }\stackrel{~}{\sigma }_{\sigma ,\sigma ^{^{}}}^{s_m}\mathrm{\Psi }_{j,\sigma ^{^{}}})]`$
$`{\displaystyle \frac{(1\alpha )}{2}}J{\displaystyle \underset{i,j,\sigma }{}}[\delta \chi _{i,j}\overline{\mathrm{\Psi }}_{j,\sigma }\mathrm{\Psi }_{i,\sigma }+\delta \chi _{i,j}^{}\overline{\mathrm{\Psi }}_{i,\sigma }\mathrm{\Psi }_{j,\sigma }]\}.`$
We start by focusing on the bosonic part $`𝒮_1^{bo}`$ of the action $`𝒮_1`$ expanded at the first order in fluctuations. By passing in the space of moments and Matsubara frequencies we obtain
$`𝒮_1^{bo}=\sqrt{\beta \mathrm{N}_\mathrm{s}}[(b_0)^2\mathrm{i}\delta \lambda _{𝒒=0}(\mathrm{i}\omega _{\mathrm{}}=0)`$ (103)
$`2(b_0)^2\lambda _0\delta b_{𝒒=0}(\mathrm{i}\omega _{\mathrm{}}=0)`$
$`+4\alpha Jm\delta m_{𝒒=𝑸}^z(\mathrm{i}\omega _{\mathrm{}}=0)]`$
$`+(1\alpha )J{\displaystyle \underset{𝒒,\mathrm{i}\omega _{\mathrm{}}}{}}[\chi _𝒒^y(\mathrm{i}\omega _{\mathrm{}})\delta \chi _𝒒^x(\mathrm{i}\omega _{\mathrm{}})`$
$`+\chi _𝒒^x(\mathrm{i}\omega _{\mathrm{}})\delta \chi _𝒒^y(\mathrm{i}\omega _{\mathrm{}})],`$
where $`\mathrm{i}\omega _{\mathrm{}}`$ labels the bosonic Matsubara frequencies and $`_𝒒`$ is expanded over the first BZ.
We study now the fermionic contribution $`𝒮_1^{fe}`$ and decompose it into four contributions
$`𝒮_1^{fe}=𝒮_1^\lambda +𝒮_1^b+𝒮_1^m+𝒮_1^\chi .`$ (104)
By passing in the space of moments and Matsubara frequencies we obtain
$`𝒮_1^\lambda ={\displaystyle \frac{1}{\sqrt{\beta \mathrm{N}_\mathrm{s}}}}{\displaystyle \underset{𝒌,𝒒}{}}{\displaystyle {}_{}{}^{}\underset{\sigma ^{^{}},\sigma }{}}{\displaystyle \underset{\mathrm{i}\omega _m,\mathrm{i}\omega _n}{}}\sigma _{\sigma ^{^{}},\sigma }^0\left[\overline{\mathrm{\Psi }}_{𝒌+𝒒,\sigma ^{^{}}}(\mathrm{i}\omega _m)\overline{\mathrm{\Psi }}_{𝒌+𝒒+𝑸,\sigma ^{^{}}}(\mathrm{i}\omega _m)\right]`$ (105)
$`\times \left[\begin{array}{cc}\mathrm{i}\delta \lambda _𝒒(\mathrm{i}\omega _m\mathrm{i}\omega _n)& \mathrm{i}\delta \lambda _{𝒒+𝑸}(\mathrm{i}\omega _m\mathrm{i}\omega _n)\\ \mathrm{i}\delta \lambda _{𝒒+𝑸}(\mathrm{i}\omega _m\mathrm{i}\omega _n)& \mathrm{i}\delta \lambda _𝒒(\mathrm{i}\omega _m\mathrm{i}\omega _n)\end{array}\right]\left[\begin{array}{c}\mathrm{\Psi }_{𝒌,\sigma }(\mathrm{i}\omega _n)\\ \mathrm{\Psi }_{𝒌+𝑸,\sigma }(\mathrm{i}\omega _n)\end{array}\right],`$ (110)
$`𝒮_1^b`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{\beta \mathrm{N}_\mathrm{s}}}}{\displaystyle \underset{𝒌,𝒒}{}}{\displaystyle {}_{}{}^{}\underset{\sigma ^{^{}},\sigma }{}}{\displaystyle \underset{\mathrm{i}\omega _m,\mathrm{i}\omega _n}{}}\sigma _{\sigma ^{^{}},\sigma }^0\left[\overline{\mathrm{\Psi }}_{𝒌+𝒒,\sigma ^{^{}}}(\mathrm{i}\omega _m)\overline{\mathrm{\Psi }}_{𝒌+𝒒+𝑸,\sigma ^{^{}}}(\mathrm{i}\omega _m)\right]`$ (118)
$`\times \left[\begin{array}{c}2(b_0)^2(\stackrel{~}{t}_𝒌+\stackrel{~}{t}_{𝒌+𝒒}+\stackrel{~}{t}_𝒌^{^{}}+\stackrel{~}{t}_{𝒌+𝒒}^{^{}}+\stackrel{~}{t}_𝒌^{^{\prime \prime }}+\stackrel{~}{t}_{𝒌+𝒒}^{^{\prime \prime }})\delta b_𝒒(\mathrm{i}\omega _m\mathrm{i}\omega _n)\hfill \\ 2(b_0)^2(\stackrel{~}{t}_𝒌+\stackrel{~}{t}_{𝒌+𝒒}+\stackrel{~}{t}_𝒌^{^{}}+\stackrel{~}{t}_{𝒌+𝒒}^{^{}}+\stackrel{~}{t}_𝒌^{^{\prime \prime }}+\stackrel{~}{t}_{𝒌+𝒒}^{^{\prime \prime }})\delta b_{𝒒+𝑸}(\mathrm{i}\omega _m\mathrm{i}\omega _n)\hfill \\ 2(b_0)^2(\stackrel{~}{t}_𝒌\stackrel{~}{t}_{𝒌+𝒒}+\stackrel{~}{t}_𝒌^{^{}}+\stackrel{~}{t}_{𝒌+𝒒}^{^{}}+\stackrel{~}{t}_𝒌^{^{\prime \prime }}+\stackrel{~}{t}_{𝒌+𝒒}^{^{\prime \prime }})\delta b_{𝒒+𝑸}(\mathrm{i}\omega _m\mathrm{i}\omega _n)\hfill \\ 2(b_0)^2(\stackrel{~}{t}_𝒌\stackrel{~}{t}_{𝒌+𝒒}+\stackrel{~}{t}_𝒌^{^{}}+\stackrel{~}{t}_{𝒌+𝒒}^{^{}}+\stackrel{~}{t}_𝒌^{^{\prime \prime }}+\stackrel{~}{t}_{𝒌+𝒒}^{^{\prime \prime }})\delta b_𝒒(\mathrm{i}\omega _m\mathrm{i}\omega _n)\hfill \end{array}\right]\left[\begin{array}{c}\mathrm{\Psi }_{𝒌,\sigma }(\mathrm{i}\omega _n)\\ \mathrm{\Psi }_{𝒌+𝑸,\sigma }(\mathrm{i}\omega _n)\end{array}\right],`$
$`𝒮_1^m`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{\beta \mathrm{N}_\mathrm{s}}}}{\displaystyle \underset{𝒌,𝒒}{}}{\displaystyle {}_{}{}^{}\underset{\sigma ^{^{}},\sigma }{}}{\displaystyle \underset{\mathrm{i}\omega _m,\mathrm{i}\omega _n}{}}{\displaystyle \underset{s_m=x,y,z}{}}\sigma _{\sigma ^{^{}},\sigma }^{s_m}\left[\overline{\mathrm{\Psi }}_{𝒌+𝒒,\sigma ^{^{}}}(\mathrm{i}\omega _m)\overline{\mathrm{\Psi }}_{𝒌+𝒒+𝑸,\sigma ^{^{}}}(\mathrm{i}\omega _m)\right]`$ (124)
$`\times \left[\begin{array}{cc}\alpha \stackrel{~}{J}_𝒒\delta m_𝒒^{s_m}(\mathrm{i}\omega _m\mathrm{i}\omega _n)& \alpha \stackrel{~}{J}_𝒒\delta m_{𝒒+𝑸}^{s_m}(\mathrm{i}\omega _m\mathrm{i}\omega _n)\\ \alpha \stackrel{~}{J}_𝒒\delta m_{𝒒+𝑸}^{s_m}(\mathrm{i}\omega _m\mathrm{i}\omega _n)& \alpha \stackrel{~}{J}_𝒒\delta m_𝒒^{s_m}(\mathrm{i}\omega _m\mathrm{i}\omega _n)\end{array}\right]\left[\begin{array}{c}\mathrm{\Psi }_{𝒌,\sigma }(\mathrm{i}\omega _n)\\ \mathrm{\Psi }_{𝒌+𝑸,\sigma }(\mathrm{i}\omega _n)\end{array}\right],`$
$`𝒮_1^\chi `$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{\beta \mathrm{N}_\mathrm{s}}}}{\displaystyle \underset{𝒌,𝒒}{}}{\displaystyle {}_{}{}^{}\underset{\sigma ^{^{}},\sigma }{}}{\displaystyle \underset{\mathrm{i}\omega _m,\mathrm{i}\omega _n}{}}{\displaystyle \underset{s_\chi =x,y}{}}\sigma _{\sigma ^{^{}},\sigma }^0\left[\overline{\mathrm{\Psi }}_{𝒌+𝒒,\sigma ^{^{}}}(\mathrm{i}\omega _m)\overline{\mathrm{\Psi }}_{𝒌+𝒒+𝑸,\sigma ^{^{}}}(\mathrm{i}\omega _m)\right]`$ (136)
$`\times \left[\begin{array}{c}(1\alpha )J[\delta ^{^{}}\chi _𝒒^{s_\chi }(\mathrm{i}\omega _m\mathrm{i}\omega _n)\mathrm{cos}(k_{s_\chi }+q_{s_\chi }/2)\hfill \\ \delta ^{^{\prime \prime }}\chi _𝒒^{s_\chi }(\mathrm{i}\omega _m\mathrm{i}\omega _n)\mathrm{sin}(k_{s_\chi }+q_{s_\chi }/2)]\hfill \\ (1\alpha )J[\delta ^{^{}}\chi _{𝒒+𝑸}^{s_\chi }(\mathrm{i}\omega _m\mathrm{i}\omega _n)\mathrm{sin}(k_{s_\chi }+q_{s_\chi }/2)\hfill \\ +\delta ^{^{\prime \prime }}\chi _{𝒒+𝑸}^{s_\chi }(\mathrm{i}\omega _m\mathrm{i}\omega _n)\mathrm{cos}(k_{s_\chi }+q_{s_\chi }/2)]\hfill \\ (1\alpha )J[+\delta ^{^{}}\chi _{𝒒+𝑸}^{s_\chi }(\mathrm{i}\omega _m\mathrm{i}\omega _n)\mathrm{sin}(k_{s_\chi }+q_{s_\chi }/2)\hfill \\ \delta ^{^{\prime \prime }}\chi _{𝒒+𝑸}^{s_\chi }(\mathrm{i}\omega _m\mathrm{i}\omega _n)\mathrm{cos}(k_{s_\chi }+q_{s_\chi }/2)]\hfill \\ (1\alpha )J[+\delta ^{^{}}\chi _𝒒^{s_\chi }(\mathrm{i}\omega _m\mathrm{i}\omega _n)\mathrm{cos}(k_{s_\chi }+q_{s_\chi }/2)\hfill \\ +\delta ^{^{\prime \prime }}\chi _𝒒^{s_\chi }(\mathrm{i}\omega _m\mathrm{i}\omega _n)\mathrm{sin}(k_{s_\chi }+q_{s_\chi }/2)]\hfill \end{array}\right]\left[\begin{array}{c}\mathrm{\Psi }_{𝒌,\sigma }(\mathrm{i}\omega _n)\\ \mathrm{\Psi }_{𝒌+𝑸,\sigma }(\mathrm{i}\omega _n)\end{array}\right].`$
In the previous formulas $`\sigma _{\sigma ^{^{}},\sigma }^0`$ is an element of the $`(2\times 2)`$ identity matrix, $`\mathrm{i}\omega _m`$ is (like $`\mathrm{i}\omega _n`$) a fermionic Matsubara frequency and we have defined the following quantities
$`\stackrel{~}{t}_𝒌`$ $`=`$ $`t\left[\mathrm{cos}(k_x)+\mathrm{cos}(k_y)\right],`$ (137)
$`\stackrel{~}{t}_𝒌^{^{}}`$ $`=`$ $`t^{^{}}[2.\mathrm{cos}(k_x).\mathrm{cos}(k_y)],`$ (138)
$`\stackrel{~}{t}_𝒌^{^{\prime \prime }}`$ $`=`$ $`t^{^{\prime \prime }}\left[\mathrm{cos}(2k_x)+\mathrm{cos}(2k_y)\right],`$ (139)
$`\stackrel{~}{J}_𝒒`$ $`=`$ $`J\left[\mathrm{cos}(q_x)+\mathrm{cos}(q_y)\right].`$ (140)
By using the Pauli matrices, we can put the different contributions (105) - (136) together, and rewrite $`𝒮_1^{fe}`$ (104) as
$`𝒮_1^{fe}={\displaystyle \underset{𝒌,𝒒}{}}{\displaystyle {}_{}{}^{}\underset{\sigma ^{^{}},\sigma }{}}{\displaystyle \underset{\mathrm{i}\omega _m,\mathrm{i}\omega _n}{}}\left[\overline{\mathrm{\Psi }}_{𝒌+𝒒,\sigma ^{^{}}}(\mathrm{i}\omega _m)\overline{\mathrm{\Psi }}_{𝒌+𝒒+𝑸,\sigma ^{^{}}}(\mathrm{i}\omega _m)\right]\stackrel{~}{𝒱}_1(𝒌+𝒒,\sigma ^{^{}},\mathrm{i}\omega _m;𝒌,\sigma ,\mathrm{i}\omega _n)\left[\begin{array}{c}\mathrm{\Psi }_{𝒌,\sigma }(\mathrm{i}\omega _n)\\ \mathrm{\Psi }_{𝒌+𝑸,\sigma }(\mathrm{i}\omega _n)\end{array}\right],`$ (143)
with $`\stackrel{~}{𝒱}_1`$ the interaction matrix
$`\sqrt{\beta \mathrm{N}_\mathrm{s}}\stackrel{~}{𝒱}_1(𝒌+𝒒,\sigma ^{^{}},\mathrm{i}\omega _m;𝒌,\sigma ,\mathrm{i}\omega _n)`$
$`=`$ $`{\displaystyle \underset{s_\chi =x,y}{}}\{[(1\alpha )J\mathrm{cos}(k_{s_\chi }+q_{s_\chi }/2)\sigma _{\sigma ^{^{}},\sigma }^0\stackrel{~}{\sigma }^z]\delta ^{^{}}\chi _𝒒^{s_\chi }(\mathrm{i}\omega _m\mathrm{i}\omega _n)`$
$`+\left[(1\alpha )J\mathrm{sin}(k_{s_\chi }+q_{s_\chi }/2)\sigma _{\sigma ^{^{}},\sigma }^0\mathrm{i}\stackrel{~}{\sigma }^y\right]\delta ^{^{}}\chi _{𝒒+𝑸}^{s_\chi }(\mathrm{i}\omega _m\mathrm{i}\omega _n)`$
$`+\left[(1\alpha )J\mathrm{sin}(k_{s_\chi }+q_{s_\chi }/2)\sigma _{\sigma ^{^{}},\sigma }^0\stackrel{~}{\sigma }^z\right]\delta ^{^{\prime \prime }}\chi _𝒒^{s_\chi }(\mathrm{i}\omega _m\mathrm{i}\omega _n)`$
$`+[+(1\alpha )J\mathrm{cos}(k_{s_\chi }+q_{s_\chi }/2)\sigma _{\sigma ^{^{}},\sigma }^0\mathrm{i}\stackrel{~}{\sigma }^y]\delta ^{^{\prime \prime }}\chi _{𝒒+𝑸}^{s_\chi }(\mathrm{i}\omega _m\mathrm{i}\omega _n)\}`$
$`+{\displaystyle \underset{s_m=x,y,z}{}}\left\{\left[\alpha \stackrel{~}{J}_𝒒\sigma _{\sigma ^{^{}},\sigma }^{s_m}\stackrel{~}{\sigma }^0\right]\delta m_𝒒^{s_m}(\mathrm{i}\omega _m\mathrm{i}\omega _n)+\left[\alpha \stackrel{~}{J}_𝒒\sigma _{\sigma ^{^{}},\sigma }^{s_m}\stackrel{~}{\sigma }^x\right]\delta m_{𝒒+𝑸}^{s_m}(\mathrm{i}\omega _m\mathrm{i}\omega _n)\right\}`$
$`+\left\{\left[\sigma _{\sigma ^{^{}},\sigma }^0\stackrel{~}{\sigma }^0\right]\mathrm{i}\delta \lambda _𝒒(\mathrm{i}\omega _m\mathrm{i}\omega _n)+\left[\sigma _{\sigma ^{^{}},\sigma }^0\stackrel{~}{\sigma }^x\right]\mathrm{i}\delta \lambda _{𝒒+𝑸}(\mathrm{i}\omega _m\mathrm{i}\omega _n)\right\}`$
$`+\{[2(b_0)^2(\stackrel{~}{t}_𝒌^{^{}}+\stackrel{~}{t}_{𝒌+𝒒}^{^{}}+\stackrel{~}{t}_𝒌^{^{\prime \prime }}+\stackrel{~}{t}_{𝒌+𝒒}^{^{\prime \prime }})\sigma _{\sigma ^{^{}},\sigma }^0\stackrel{~}{\sigma }^02(b_0)^2(\stackrel{~}{t}_𝒌+\stackrel{~}{t}_{𝒌+𝒒})\sigma _{\sigma ^{^{}},\sigma }^0\stackrel{~}{\sigma }^z]\delta b_𝒒(\mathrm{i}\omega _m\mathrm{i}\omega _n)`$
$`+[2(b_0)^2(\stackrel{~}{t}_𝒌^{^{}}+\stackrel{~}{t}_{𝒌+𝒒}^{^{}}+\stackrel{~}{t}_𝒌^{^{\prime \prime }}+\stackrel{~}{t}_{𝒌+𝒒}^{^{\prime \prime }})\sigma _{\sigma ^{^{}},\sigma }^0\stackrel{~}{\sigma }^x+2(b_0)^2(\stackrel{~}{t}_𝒌\stackrel{~}{t}_{𝒌+𝒒})\sigma _{\sigma ^{^{}},\sigma }^0\mathrm{i}\stackrel{~}{\sigma }^y]\delta b_{𝒒+𝑸}(\mathrm{i}\omega _m\mathrm{i}\omega _n)\}.`$
Finally the mean field equations are obtained after having integrated out the fermionic fields by checking the saddle point property Nagaosa\_books
$`\mathrm{Tr}\left[\stackrel{~}{𝒢}_0\stackrel{~}{𝒱}_1\right]=0,`$ (145)
which gives by using the previously obtained expressions (39), (100), (103), (143) the set of coupled mean field equations (40) - (43).
## Appendix B Derivation of the Gaussian Action
We decompose the second order effective action $`𝒮_2`$ (46) into two parts
$`𝒮_2=𝒮_2^{bo}+𝒮_2^{bub}.`$ (146)
$`𝒮_2^{bo}`$ contains only contributions due to auxiliary bosonic fields. It includes all the terms of Eq. (46) with the exception of the trace
$`𝒮_2^{bo}`$ $`=`$ $`{\displaystyle _0^\beta }d\tau \{{\displaystyle \underset{i}{}}(b_0)^2[2\mathrm{i}\delta \lambda _i\delta b_i+\lambda _0(\delta b_i)^2]\alpha J{\displaystyle \underset{i,j}{}}{\displaystyle \underset{s_m=x,y,z}{}}[\delta m_i^{s_m}\delta m_j^{s_m}]+{\displaystyle \frac{(1\alpha )}{2}}J{\displaystyle \underset{i,j}{}}[\delta \chi _{ij}\delta \chi _{ij}^{}]`$ (147)
$`({\displaystyle \underset{i,j}{}}t+{\displaystyle \underset{i,j^{^{}}}{}}t^{^{}}+{\displaystyle \underset{i,j^{^{\prime \prime }}}{}}t^{^{\prime \prime }}){\displaystyle \underset{\sigma }{}}(b_0)^2\delta b_i\delta b_j[\overline{\mathrm{\Psi }}_{i,\sigma }\mathrm{\Psi }_{j,\sigma }+\overline{\mathrm{\Psi }}_{j,\sigma }\mathrm{\Psi }_{i,\sigma }]\}.`$
$`𝒮_2^{bub}`$ takes into account the “bubble” contributions
$`𝒮_2^{bub}={\displaystyle \frac{1}{2}}\mathrm{Tr}\left[\stackrel{~}{𝒢}_0\stackrel{~}{𝒱}_1\stackrel{~}{𝒢}_0\stackrel{~}{𝒱}_1\right].`$ (148)
We firstly evaluate $`𝒮_2^{bo}`$ (147) by passing in the space of moments and Matsubara frequencies
$`𝒮_2^{bo}=(b_0)^2{\displaystyle \underset{𝒒}{}}{\displaystyle {}_{}{}^{}\underset{\mathrm{i}\omega _{\mathrm{}}}{}}\{2\mathrm{i}[\delta \lambda _𝒒(\mathrm{i}\omega _{\mathrm{}})\delta b_𝒒(\mathrm{i}\omega _{\mathrm{}})+\delta \lambda _{𝒒+𝑸}(\mathrm{i}\omega _{\mathrm{}})\delta b_{𝒒𝑸}(\mathrm{i}\omega _{\mathrm{}})]`$
$`+\lambda _0[\delta b_𝒒(\mathrm{i}\omega _{\mathrm{}})\delta b_𝒒(\mathrm{i}\omega _{\mathrm{}})+\delta b_{𝒒+𝑸}(\mathrm{i}\omega _{\mathrm{}})\delta b_{𝒒𝑸}(\mathrm{i}\omega _{\mathrm{}})]\}`$
$`\alpha {\displaystyle \underset{𝒒}{}}{\displaystyle {}_{}{}^{}\underset{\mathrm{i}\omega _{\mathrm{}}}{}}\stackrel{~}{J}_𝒒{\displaystyle \underset{s_m=x,y,z}{}}\left[\delta m_𝒒^{s_m}(\mathrm{i}\omega _{\mathrm{}})\delta m_𝒒^{s_m}(\mathrm{i}\omega _{\mathrm{}})\delta m_{𝒒+𝑸}^{s_m}(\mathrm{i}\omega _{\mathrm{}})\delta m_{𝒒𝑸}^{s_m}(\mathrm{i}\omega _{\mathrm{}})\right]`$
$`+{\displaystyle \frac{(1\alpha )}{2}}J{\displaystyle \underset{𝒒}{}}{\displaystyle {}_{}{}^{}\underset{\mathrm{i}\omega _{\mathrm{}}}{}}{\displaystyle \underset{s_\chi =x,y}{}}[\delta ^{^{}}\chi _𝒒^{s_\chi }(\mathrm{i}\omega _{\mathrm{}})\delta ^{^{}}\chi _𝒒^{s_\chi }(\mathrm{i}\omega _{\mathrm{}})+\delta ^{^{\prime \prime }}\chi _𝒒^{s_\chi }(\mathrm{i}\omega _{\mathrm{}})\delta ^{^{\prime \prime }}\chi _𝒒^{s_\chi }(\mathrm{i}\omega _{\mathrm{}})`$
$`+\delta ^{^{}}\chi _{𝒒+𝑸}^{s_\chi }(\mathrm{i}\omega _{\mathrm{}})\delta ^{^{}}\chi _{𝒒𝑸}^{s_\chi }(\mathrm{i}\omega _{\mathrm{}})+\delta ^{^{\prime \prime }}\chi _{𝒒+𝑸}^{s_\chi }(\mathrm{i}\omega _{\mathrm{}})\delta ^{^{\prime \prime }}\chi _{𝒒𝑸}^{s_\chi }(\mathrm{i}\omega _{\mathrm{}})]`$
$`4(b_0)^2{\displaystyle \underset{𝒒}{}}{\displaystyle {}_{}{}^{}\underset{\mathrm{i}\omega _{\mathrm{}}}{}}[\{tl_1^x[\mathrm{cos}(q_x)+\mathrm{cos}(q_y)]+2t^{^{}}l_2^{x,y}\mathrm{cos}(q_x)\mathrm{cos}(q_y)+t^{^{\prime \prime }}l_2^{x,x}[\mathrm{cos}(2q_x)+\mathrm{cos}(2q_y)]\}`$
$`\times \delta b_𝒒(\mathrm{i}\omega _{\mathrm{}})\delta b_𝒒(\mathrm{i}\omega _{\mathrm{}})`$
$`+\left\{tl_1^x\left[\mathrm{cos}(q_x)+\mathrm{cos}(q_y)\right]+2t^{^{}}l_2^{x,y}\mathrm{cos}(q_x)\mathrm{cos}(q_y)+t^{^{\prime \prime }}l_2^{x,x}\left[\mathrm{cos}(2q_x)+\mathrm{cos}(2q_y)\right]\right\}`$
$`\times \delta b_{𝒒+𝑸}(\mathrm{i}\omega _{\mathrm{}})\delta b_{𝒒𝑸}(\mathrm{i}\omega _{\mathrm{}})],`$ (149)
where $`\stackrel{~}{J}_𝒒`$ is given by Eq. (140) and we have defined for $`s_t,s_t^{^{}}=x,y`$
$`l_1^{s_t}`$ $`=`$ $`\overline{\mathrm{\Psi }}_{i,\sigma }\mathrm{\Psi }_{i+s_t,\sigma }={\displaystyle \frac{1}{\mathrm{N}_\mathrm{s}}}{\displaystyle \underset{𝒌}{}}{}_{}{}^{}{\displaystyle \frac{(\xi _{𝒌+𝑸}\xi _𝒌)}{E_𝒌^{up}E_𝒌^{low}}}\mathrm{cos}(k_{s_t}),`$
$`l_2^{s_t,s_t^{^{}}}`$ $`=`$ $`\overline{\mathrm{\Psi }}_{i,\sigma }\mathrm{\Psi }_{i+s_t+s_t^{^{}},\sigma }={\displaystyle \frac{1}{\mathrm{N}_\mathrm{s}}}{\displaystyle \underset{𝒌}{}}{}_{}{}^{}{\displaystyle \frac{(\xi _{𝒌+𝑸}\xi _𝒌)}{E_𝒌^{up}E_𝒌^{low}}}\mathrm{cos}(k_{s_t}+k_{s_t^{^{}}}).`$
In order to write $`𝒮_2^{bub}`$ (148) and $`𝒮_2^{bo}`$ (149) in a compact way which will be useful in the following calculations we define the vector of nine components $`\delta \stackrel{}{X}`$ containing the first order fluctuations of the bosonic fields
$`\delta \stackrel{}{X}(𝒒,\mathrm{i}\omega _{\mathrm{}})=\left[\delta X_i(𝒒,\mathrm{i}\omega _{\mathrm{}})\right]_{1i9}=\left[\begin{array}{c}\delta ^{^{}}\chi _𝒒^x(\mathrm{i}\omega _{\mathrm{}})\\ \delta ^{^{}}\chi _𝒒^y(\mathrm{i}\omega _{\mathrm{}})\\ \delta ^{^{\prime \prime }}\chi _𝒒^x(\mathrm{i}\omega _{\mathrm{}})\\ \delta ^{^{\prime \prime }}\chi _𝒒^y(\mathrm{i}\omega _{\mathrm{}})\\ \delta m_𝒒^x(\mathrm{i}\omega _{\mathrm{}})\\ \delta m_𝒒^y(\mathrm{i}\omega _{\mathrm{}})\\ \delta m_𝒒^z(\mathrm{i}\omega _{\mathrm{}})\\ \delta \lambda _𝒒(\mathrm{i}\omega _{\mathrm{}})\\ \delta b_𝒒(\mathrm{i}\omega _{\mathrm{}})\end{array}\right],\delta \stackrel{}{X}(𝒒+𝑸,\mathrm{i}\omega _{\mathrm{}})=\left[\delta X_i(𝒒+𝑸,\mathrm{i}\omega _{\mathrm{}})\right]_{1i9}=\left[\begin{array}{c}\delta ^{^{}}\chi _{𝒒+𝑸}^x(\mathrm{i}\omega _{\mathrm{}})\\ \delta ^{^{}}\chi _{𝒒+𝑸}^y(\mathrm{i}\omega _{\mathrm{}})\\ \delta ^{^{\prime \prime }}\chi _{𝒒+𝑸}^x(\mathrm{i}\omega _{\mathrm{}})\\ \delta ^{^{\prime \prime }}\chi _{𝒒+𝑸}^y(\mathrm{i}\omega _{\mathrm{}})\\ \delta m_{𝒒+𝑸}^x(\mathrm{i}\omega _{\mathrm{}})\\ \delta m_{𝒒+𝑸}^y(\mathrm{i}\omega _{\mathrm{}})\\ \delta m_{𝒒+𝑸}^z(\mathrm{i}\omega _{\mathrm{}})\\ \delta \lambda _{𝒒+𝑸}(\mathrm{i}\omega _{\mathrm{}})\\ \delta b_{𝒒+𝑸}(\mathrm{i}\omega _{\mathrm{}})\end{array}\right].`$ (168)
Using Eq. (149) we define the matrix $`\stackrel{~}{}`$ so that
$`\stackrel{~}{}(𝒒,𝒒_\mathrm{𝟏},𝒒_\mathrm{𝟐},\mathrm{i}\omega _{\mathrm{}})=\left[_{i,j}(𝒒,𝒒_\mathrm{𝟏},𝒒_\mathrm{𝟐},\mathrm{i}\omega _{\mathrm{}})\right]_{1i,j9},`$
$`𝒮_2^{bo}={\displaystyle \underset{𝒒}{}}{\displaystyle {}_{}{}^{}\underset{𝒒_\mathrm{𝟏},𝒒_\mathrm{𝟐}=𝒒,𝒒+𝑸}{}}{\displaystyle \underset{\mathrm{i}\omega _{\mathrm{}}}{}}{\displaystyle \underset{i,j=1}{\overset{9}{}}}\delta X_i(𝒒_\mathrm{𝟏},\mathrm{i}\omega _{\mathrm{}})`$
$`\times _{i,j}(𝒒,𝒒_\mathrm{𝟏},𝒒_\mathrm{𝟐},\mathrm{i}\omega _{\mathrm{}})\delta X_j(𝒒_\mathrm{𝟐},\mathrm{i}\omega _{\mathrm{}}).`$ (169)
We now tackle the calculation of the “bubble” part $`𝒮_2^{bub}`$ (148). We have Nagaosa\_books
$`\mathrm{Tr}\left[\stackrel{~}{𝒢}_0\stackrel{~}{𝒱}_1\stackrel{~}{𝒢}_0\stackrel{~}{𝒱}_1\right]`$
$`=`$ $`{\displaystyle \underset{𝒌,𝒒}{}}{\displaystyle {}_{}{}^{}\underset{\sigma ,\sigma ^{^{}}}{}}{\displaystyle \underset{\mathrm{i}\omega _n,\mathrm{i}\omega _{\mathrm{}}}{}}\mathrm{tr}[\stackrel{~}{𝒢}_0(𝒌+𝒒,\sigma ^{^{}},\mathrm{i}\omega _n+\mathrm{i}\omega _{\mathrm{}})`$
$`\times \stackrel{~}{𝒱}_1(𝒌+𝒒,\sigma ^{^{}},\mathrm{i}\omega _n+\mathrm{i}\omega _{\mathrm{}};𝒌,\sigma ,\mathrm{i}\omega _n)`$
$`\times \stackrel{~}{𝒢}_0(𝒌,\sigma ,\mathrm{i}\omega _n)`$
$`\times \stackrel{~}{𝒱}_1(𝒌,\sigma ,\mathrm{i}\omega _n;𝒌+𝒒,\sigma ^{^{}},\mathrm{i}\omega _n+\mathrm{i}\omega _{\mathrm{}})],`$
where $`\mathrm{tr}`$ is related to the spinor basis (15). We define the vector $`\stackrel{}{c}^{\sigma ^{^{}},\sigma }`$ containing the coefficients associated with the first order fluctuations in the $`\stackrel{~}{𝒱}_1`$ matrix. As we can see in Eq. (A) $`\delta b_𝒒`$ (and $`\delta b_{𝒒+𝑸}`$) gives two separated contributions and not one like all the other auxiliary bosonic fields whose first order fluctuations are contained in the vector $`\delta \stackrel{}{X}`$ (168). Therefore $`\stackrel{}{c}^{\sigma ^{^{}},\sigma }`$ has ten components and not nine, which is indicated by the index $`a^{^{}}`$ being equal to one for all the fields except $`\delta b_𝒒`$ (and $`\delta b_{𝒒+𝑸}`$) for which it takes the values one and two
$`\stackrel{}{c}^{\sigma ^{^{}},\sigma }(𝒌,𝒒)=\left[c_{i,a^{^{}}}^{\sigma ^{^{}},\sigma }(𝒌,𝒒)\right]_{1i9}=\left[\begin{array}{c}(1\alpha )J\mathrm{cos}(k_x+q_x/2)\sigma _{\sigma ^{^{}},\sigma }^0\\ (1\alpha )J\mathrm{cos}(k_y+q_y/2)\sigma _{\sigma ^{^{}},\sigma }^0\\ (1\alpha )J\mathrm{sin}(k_x+q_x/2)\sigma _{\sigma ^{^{}},\sigma }^0\\ (1\alpha )J\mathrm{sin}(k_y+q_y/2)\sigma _{\sigma ^{^{}},\sigma }^0\\ \alpha \stackrel{~}{J}_𝒒\sigma _{\sigma ^{^{}},\sigma }^x\\ \alpha \stackrel{~}{J}_𝒒\sigma _{\sigma ^{^{}},\sigma }^y\\ \alpha \stackrel{~}{J}_𝒒\sigma _{\sigma ^{^{}},\sigma }^z\\ \mathrm{i}\sigma _{\sigma ^{^{}},\sigma }^0\\ 2(b_0)^2(\stackrel{~}{t}_𝒌^{^{}}+\stackrel{~}{t}_{𝒌+𝒒}^{^{}}+\stackrel{~}{t}_𝒌^{^{\prime \prime }}+\stackrel{~}{t}_{𝒌+𝒒}^{^{\prime \prime }})\sigma _{\sigma ^{^{}},\sigma }^0\\ 2(b_0)^2(\stackrel{~}{t}_𝒌+\stackrel{~}{t}_{𝒌+𝒒})\sigma _{\sigma ^{^{}},\sigma }^0\end{array}\right],`$ (181)
$`\stackrel{}{c}^{\sigma ^{^{}},\sigma }(𝒌,𝒒+𝑸)=\left[c_{i,a^{^{}}}^{\sigma ^{^{}},\sigma }(𝒌,𝒒+𝑸)\right]_{1i9}=\left[\begin{array}{c}\mathrm{i}(1\alpha )J\mathrm{sin}(k_x+q_x/2)\sigma _{\sigma ^{^{}},\sigma }^0\\ \mathrm{i}(1\alpha )J\mathrm{sin}(k_y+q_y/2)\sigma _{\sigma ^{^{}},\sigma }^0\\ \mathrm{i}(1\alpha )J\mathrm{cos}(k_x+q_x/2)\sigma _{\sigma ^{^{}},\sigma }^0\\ \mathrm{i}(1\alpha )J\mathrm{cos}(k_y+q_y/2)\sigma _{\sigma ^{^{}},\sigma }^0\\ \alpha \stackrel{~}{J}_𝒒\sigma _{\sigma ^{^{}},\sigma }^x\\ \alpha \stackrel{~}{J}_𝒒\sigma _{\sigma ^{^{}},\sigma }^y\\ \alpha \stackrel{~}{J}_𝒒\sigma _{\sigma ^{^{}},\sigma }^z\\ \mathrm{i}\sigma _{\sigma ^{^{}},\sigma }^0\\ 2(b_0)^2(\stackrel{~}{t}_𝒌^{^{}}+\stackrel{~}{t}_{𝒌+𝒒}^{^{}}+\stackrel{~}{t}_𝒌^{^{\prime \prime }}+\stackrel{~}{t}_{𝒌+𝒒}^{^{\prime \prime }})\sigma _{\sigma ^{^{}},\sigma }^0\\ 2\mathrm{i}(b_0)^2(\stackrel{~}{t}_𝒌\stackrel{~}{t}_{𝒌+𝒒})\sigma _{\sigma ^{^{}},\sigma }^0\end{array}\right],`$ (192)
where the terms $`\stackrel{~}{t}_𝒌`$, $`\stackrel{~}{t}_𝒌^{^{}}`$, $`\stackrel{~}{t}_𝒌^{^{\prime \prime }}`$ and $`\stackrel{~}{J}_𝒒`$ are defined in Eqs. (137), (138), (139) and (140), respectively. In the same manner we define the vector $`\stackrel{}{s}`$ to associate each fluctuation element $`\delta X_i`$ to its corresponding Pauli matrix appearing in the expression of the $`\stackrel{~}{𝒱}_1`$ matrix (A)
$`\stackrel{}{s}(𝒒)=\left[\stackrel{~}{s}_{i,a^{^{}}}(𝒒)\right]_{1i9}=\left[\begin{array}{c}\stackrel{~}{\sigma }^z\\ \stackrel{~}{\sigma }^z\\ \stackrel{~}{\sigma }^z\\ \stackrel{~}{\sigma }^z\\ \stackrel{~}{\sigma }^0\\ \stackrel{~}{\sigma }^0\\ \stackrel{~}{\sigma }^0\\ \stackrel{~}{\sigma }^0\\ \stackrel{~}{\sigma }^0\\ \stackrel{~}{\sigma }^z\end{array}\right],\stackrel{}{s}(𝒒+𝑸)=\left[\stackrel{~}{s}_{i,a^{^{}}}(𝒒+𝑸)\right]_{1i9}=\left[\begin{array}{c}\stackrel{~}{\sigma }^y\\ \stackrel{~}{\sigma }^y\\ \stackrel{~}{\sigma }^y\\ \stackrel{~}{\sigma }^y\\ \stackrel{~}{\sigma }^x\\ \stackrel{~}{\sigma }^x\\ \stackrel{~}{\sigma }^x\\ \stackrel{~}{\sigma }^x\\ \stackrel{~}{\sigma }^x\\ \stackrel{~}{\sigma }^y\end{array}\right].`$ (213)
The summation over the fermionic Matsubara frequencies $`\mathrm{i}\omega _n`$ is incorporated in
$`F_{\stackrel{~}{s}_{i,a^{^{}}}(𝒒_\mathrm{𝟏}),\stackrel{~}{s}_{j,a^{^{\prime \prime }}}(𝒒_\mathrm{𝟐})}^{\sigma ^{^{}},\sigma }(𝒌,𝒒,\mathrm{i}\omega _{\mathrm{}})={\displaystyle \frac{1}{\beta }}{\displaystyle \underset{\mathrm{i}\omega _n}{}}\mathrm{tr}\left[\stackrel{~}{𝒢}_0(𝒌+𝒒,\sigma ^{^{}},\mathrm{i}\omega _n+\mathrm{i}\omega _{\mathrm{}})\stackrel{~}{s}_{i,a^{^{}}}(𝒒_\mathrm{𝟏})\stackrel{~}{𝒢}_0(𝒌,\sigma ,\mathrm{i}\omega _n)\stackrel{~}{s}_{j,a^{^{\prime \prime }}}(𝒒_\mathrm{𝟐})\right],`$ (214)
with $`𝒒_\mathrm{𝟏},𝒒_\mathrm{𝟐}=𝒒,𝒒+𝑸`$. Considering the identity matrix and the three $`SU(2)`$ Pauli matrices we get from Eq. (214) sixteen different thermal integrals which are detailed in Appendix C.
Using the preceding expressions (168), (181), (192), (213), and the formulas associated with the $`F`$ \- integrals, we define the matrix $`\stackrel{~}{\mathrm{\Pi }}`$ which explicitly incorporates at a RPA (“bubble”) level the contributions of the different fields fluctuations to the correlation functions
$`\stackrel{~}{\mathrm{\Pi }}(𝒒,𝒒_\mathrm{𝟏},𝒒_\mathrm{𝟐},\mathrm{i}\omega _{\mathrm{}})=\left[\mathrm{\Pi }_{i,j}(𝒒,𝒒_\mathrm{𝟏},𝒒_\mathrm{𝟐},\mathrm{i}\omega _{\mathrm{}})\right]_{1i,j9},`$ (215)
with the matricial elements
$`\mathrm{\Pi }_{i,j}(𝒒,𝒒_\mathrm{𝟏},𝒒_\mathrm{𝟐},\mathrm{i}\omega _{\mathrm{}})`$
$`=`$ $`{\displaystyle \frac{1}{\mathrm{N}_\mathrm{s}}}{\displaystyle \underset{𝒌}{}}{\displaystyle {}_{}{}^{}\underset{\sigma ^{^{}},\sigma }{}}{\displaystyle \underset{a^{^{}},a^{^{\prime \prime }}=1}{\overset{2}{}}}c_{i,a^{^{}}}^{\sigma ^{^{}},\sigma }(𝒌,𝒒_\mathrm{𝟏})`$
$`\times F_{\stackrel{~}{s}_{i,a^{^{}}}(𝒒_\mathrm{𝟏}),\stackrel{~}{s}_{j,a^{^{\prime \prime }}}(𝒒_\mathrm{𝟐})}^{\sigma ^{^{}},\sigma }(𝒌,𝒒,\mathrm{i}\omega _{\mathrm{}})c_{j,a^{^{\prime \prime }}}^{\sigma ,\sigma ^{^{}}}(𝒌,𝒒_\mathrm{𝟐}).`$
From Eqs. (B), (214) and (215) we get
$`\mathrm{Tr}\left[\stackrel{~}{𝒢}_0\stackrel{~}{𝒱}_1\stackrel{~}{𝒢}_0\stackrel{~}{𝒱}_1\right]`$
$`=`$ $`{\displaystyle \underset{𝒒}{}}{\displaystyle {}_{}{}^{}\underset{𝒒_\mathrm{𝟏},𝒒_\mathrm{𝟐}=𝒒,𝒒+𝑸}{}}{\displaystyle \underset{\mathrm{i}\omega _{\mathrm{}}}{}}{\displaystyle \underset{i,j=1}{\overset{9}{}}}\delta X_i(𝒒_\mathrm{𝟏},\mathrm{i}\omega _{\mathrm{}})`$
$`\times \mathrm{\Pi }_{i,j}(𝒒,𝒒_\mathrm{𝟏},𝒒_\mathrm{𝟐},\mathrm{i}\omega _{\mathrm{}})\delta X_j(𝒒_\mathrm{𝟐},\mathrm{i}\omega _{\mathrm{}}).`$
With Eqs. (146), (169), (148) and (B) we finally obtain $`𝒮_2`$ given by Eq. (47).
## Appendix C Thermal integrals expressions
We give here the analytic expressions of the sixteen different $`F`$ integrals (214) useful to calculate at a RPA level the two (bosonic) fields correlation functions (48). We define
$`E_{u,l}^4=\left[E_{𝒌+𝒒}^{up}(\mathrm{i}\omega _n+\mathrm{i}\omega _{\mathrm{}})\right]\left[(\mathrm{i}\omega _n+\mathrm{i}\omega _{\mathrm{}})E_{𝒌+𝒒}^{low}\right]`$
$`\times \left[E_𝒌^{up}\mathrm{i}\omega _n\right]\left[\mathrm{i}\omega _nE_𝒌^{low}\right],`$
$`g_{0,0}=2(\mathrm{i}\omega _n)^2\mathrm{i}\omega _{\mathrm{}}(\xi _𝒌+\xi _{𝒌+𝑸})`$
$`+\mathrm{i}\omega _n\left[2\mathrm{i}\omega _{\mathrm{}}(\xi _𝒌+\xi _{𝒌+𝑸}+\xi _{𝒌+𝒒}+\xi _{𝒌+𝒒+𝑸})\right],`$
$`g_{0,x}=2\mathrm{i}\omega _n(\mathrm{\Delta }_{𝒌,\sigma }^m+\mathrm{\Delta }_{𝒌+𝒒,\sigma ^{^{}}}^m)2\mathrm{i}\omega _{\mathrm{}}\mathrm{\Delta }_{𝒌,\sigma }^m`$
$`+\mathrm{\Delta }_{𝒌,\sigma }^m(\xi _{𝒌+𝒒}+\xi _{𝒌+𝒒+𝑸})+\mathrm{\Delta }_{𝒌+𝒒,\sigma ^{^{}}}^m(\xi _𝒌+\xi _{𝒌+𝑸}),`$
$`g_{0,y}=2\mathrm{i}\omega _n(\mathrm{i}\mathrm{\Delta }_{𝒌,\sigma }^\chi +\mathrm{i}\mathrm{\Delta }_{𝒌+𝒒,\sigma ^{^{}}}^\chi )2\mathrm{i}\omega _{\mathrm{}}\mathrm{i}\mathrm{\Delta }_{𝒌,\sigma }^\chi `$
$`+\mathrm{i}\mathrm{\Delta }_{𝒌,\sigma }^\chi (\xi _{𝒌+𝒒}+\xi _{𝒌+𝒒+𝑸})+\mathrm{i}\mathrm{\Delta }_{𝒌+𝒒,\sigma ^{^{}}}^\chi (\xi _𝒌+\xi _{𝒌+𝑸}),`$
$`g_{0,z}=\mathrm{i}\omega _n(\xi _𝒌\xi _{𝒌+𝑸}+\xi _{𝒌+𝒒}\xi _{𝒌+𝒒+𝑸})`$
$`+\mathrm{i}\omega _{\mathrm{}}(\xi _𝒌\xi _{𝒌+𝑸})(\xi _𝒌\xi _{𝒌+𝒒}\xi _{𝒌+𝑸}\xi _{𝒌+𝒒+𝑸}),`$
$`g_{x,y}=\mathrm{i}\omega _n(\mathrm{i}\xi _𝒌+\mathrm{i}\xi _{𝒌+𝑸}+\mathrm{i}\xi _{𝒌+𝒒}\mathrm{i}\xi _{𝒌+𝒒+𝑸})`$
$`\mathrm{i}\omega _{\mathrm{}}(\mathrm{i}\xi _𝒌\mathrm{i}\xi _{𝒌+𝑸})+(\mathrm{i}\xi _𝒌\xi _{𝒌+𝒒+𝑸}\mathrm{i}\xi _{𝒌+𝑸}\xi _{𝒌+𝒒}),`$
$`g_{x,z}=2\mathrm{i}\omega _n(\mathrm{\Delta }_{𝒌,\sigma }^\chi \mathrm{\Delta }_{𝒌+𝒒,\sigma ^{^{}}}^\chi )+2\mathrm{i}\omega _{\mathrm{}}\mathrm{\Delta }_{𝒌,\sigma }^\chi `$
$`\mathrm{\Delta }_{𝒌,\sigma }^\chi (\xi _{𝒌+𝒒}+\xi _{𝒌+𝒒+𝑸})+\mathrm{\Delta }_{𝒌+𝒒,\sigma ^{^{}}}^\chi (\xi _𝒌+\xi _{𝒌+𝑸}),`$
$`g_{y,z}=2\mathrm{i}\omega _n(\mathrm{i}\mathrm{\Delta }_{𝒌,\sigma }^m\mathrm{i}\mathrm{\Delta }_{𝒌+𝒒,\sigma ^{^{}}}^m)+2\mathrm{i}\omega _{\mathrm{}}\mathrm{i}\mathrm{\Delta }_{𝒌,\sigma }^m`$
$`\mathrm{i}\mathrm{\Delta }_{𝒌,\sigma }^m(\xi _{𝒌+𝒒}+\xi _{𝒌+𝒒+𝑸})+\mathrm{i}\mathrm{\Delta }_{𝒌+𝒒,\sigma ^{^{}}}^m(\xi _𝒌+\xi _{𝒌+𝑸}),`$
and we obtain
$`F_{\stackrel{~}{\sigma }^0,\stackrel{~}{\sigma }^0}^{\sigma ^{^{}},\sigma }(𝒌,𝒒,\mathrm{i}\omega _{\mathrm{}})={\displaystyle \frac{1}{\beta }}{\displaystyle \underset{\mathrm{i}\omega _n}{}}{\displaystyle \frac{1}{E_{u,l}^4}}\{g_{0,0}`$
$`+(\xi _𝒌\xi _{𝒌+𝒒}+\xi _{𝒌+𝑸}\xi _{𝒌+𝒒+𝑸})`$
$`+2(\mathrm{\Delta }_{𝒌,\sigma }^m\mathrm{\Delta }_{𝒌+𝒒,\sigma ^{^{}}}^m\mathrm{\Delta }_{𝒌,\sigma }^\chi \mathrm{\Delta }_{𝒌+𝒒,\sigma ^{^{}}}^\chi )\},`$
$`F_{\stackrel{~}{\sigma }^x,\stackrel{~}{\sigma }^x}^{\sigma ^{^{}},\sigma }(𝒌,𝒒,\mathrm{i}\omega _{\mathrm{}})={\displaystyle \frac{1}{\beta }}{\displaystyle \underset{\mathrm{i}\omega _n}{}}{\displaystyle \frac{1}{E_{u,l}^4}}\{g_{0,0}`$
$`+(\xi _𝒌\xi _{𝒌+𝒒+𝑸}+\xi _{𝒌+𝑸}\xi _{𝒌+𝒒})`$
$`+2(\mathrm{\Delta }_{𝒌,\sigma }^m\mathrm{\Delta }_{𝒌+𝒒,\sigma ^{^{}}}^m+\mathrm{\Delta }_{𝒌,\sigma }^\chi \mathrm{\Delta }_{𝒌+𝒒,\sigma ^{^{}}}^\chi )\},`$
$`F_{\stackrel{~}{\sigma }^y,\stackrel{~}{\sigma }^y}^{\sigma ^{^{}},\sigma }(𝒌,𝒒,\mathrm{i}\omega _{\mathrm{}})={\displaystyle \frac{1}{\beta }}{\displaystyle \underset{\mathrm{i}\omega _n}{}}{\displaystyle \frac{1}{E_{u,l}^4}}\{g_{0,0}`$
$`+(\xi _𝒌\xi _{𝒌+𝒒+𝑸}+\xi _{𝒌+𝑸}\xi _{𝒌+𝒒})`$
$`2(\mathrm{\Delta }_{𝒌,\sigma }^m\mathrm{\Delta }_{𝒌+𝒒,\sigma ^{^{}}}^m+\mathrm{\Delta }_{𝒌,\sigma }^\chi \mathrm{\Delta }_{𝒌+𝒒,\sigma ^{^{}}}^\chi )\},`$
$`F_{\stackrel{~}{\sigma }^z,\stackrel{~}{\sigma }^z}^{\sigma ^{^{}},\sigma }(𝒌,𝒒,\mathrm{i}\omega _{\mathrm{}})={\displaystyle \frac{1}{\beta }}{\displaystyle \underset{\mathrm{i}\omega _n}{}}{\displaystyle \frac{1}{E_{u,l}^4}}\{g_{0,0}`$
$`+(\xi _𝒌\xi _{𝒌+𝒒}+\xi _{𝒌+𝑸}\xi _{𝒌+𝒒+𝑸})`$
$`2(\mathrm{\Delta }_{𝒌,\sigma }^m\mathrm{\Delta }_{𝒌+𝒒,\sigma ^{^{}}}^m\mathrm{\Delta }_{𝒌,\sigma }^\chi \mathrm{\Delta }_{𝒌+𝒒,\sigma ^{^{}}}^\chi )\},`$
$`F_{\stackrel{~}{\sigma }^0,\stackrel{~}{\sigma }^x}^{\sigma ^{^{}},\sigma }(𝒌,𝒒,\mathrm{i}\omega _{\mathrm{}})={\displaystyle \frac{1}{\beta }}{\displaystyle \underset{\mathrm{i}\omega _n}{}}{\displaystyle \frac{1}{E_{u,l}^4}}\{g_{0,x}`$
$`\mathrm{\Delta }_{𝒌,\sigma }^\chi (\xi _{𝒌+𝒒}\xi _{𝒌+𝒒+𝑸})+\mathrm{\Delta }_{𝒌+𝒒,\sigma ^{^{}}}^\chi (\xi _𝒌\xi _{𝒌+𝑸})\},`$
$`F_{\stackrel{~}{\sigma }^x,\stackrel{~}{\sigma }^0}^{\sigma ^{^{}},\sigma }(𝒌,𝒒,\mathrm{i}\omega _{\mathrm{}})={\displaystyle \frac{1}{\beta }}{\displaystyle \underset{\mathrm{i}\omega _n}{}}{\displaystyle \frac{1}{E_{u,l}^4}}\{g_{0,x}`$
$`+\mathrm{\Delta }_{𝒌,\sigma }^\chi (\xi _{𝒌+𝒒}\xi _{𝒌+𝒒+𝑸})\mathrm{\Delta }_{𝒌+𝒒,\sigma ^{^{}}}^\chi (\xi _𝒌\xi _{𝒌+𝑸})\},`$
$`F_{\stackrel{~}{\sigma }^0,\stackrel{~}{\sigma }^y}^{\sigma ^{^{}},\sigma }(𝒌,𝒒,\mathrm{i}\omega _{\mathrm{}})={\displaystyle \frac{1}{\beta }}{\displaystyle \underset{\mathrm{i}\omega _n}{}}{\displaystyle \frac{1}{E_{u,l}^4}}\{g_{0,y}`$
$`\mathrm{i}\mathrm{\Delta }_{𝒌,\sigma }^m(\xi _{𝒌+𝒒}\xi _{𝒌+𝒒+𝑸})+\mathrm{i}\mathrm{\Delta }_{𝒌+𝒒,\sigma ^{^{}}}^m(\xi _𝒌\xi _{𝒌+𝑸})\},`$
$`F_{\stackrel{~}{\sigma }^y,\stackrel{~}{\sigma }^0}^{\sigma ^{^{}},\sigma }(𝒌,𝒒,\mathrm{i}\omega _{\mathrm{}})={\displaystyle \frac{1}{\beta }}{\displaystyle \underset{\mathrm{i}\omega _n}{}}{\displaystyle \frac{1}{E_{u,l}^4}}\{g_{0,y}`$
$`+\mathrm{i}\mathrm{\Delta }_{𝒌,\sigma }^m(\xi _{𝒌+𝒒}\xi _{𝒌+𝒒+𝑸})\mathrm{i}\mathrm{\Delta }_{𝒌+𝒒,\sigma ^{^{}}}^m(\xi _𝒌\xi _{𝒌+𝑸})\},`$
$`F_{\stackrel{~}{\sigma }^0,\stackrel{~}{\sigma }^z}^{\sigma ^{^{}},\sigma }(𝒌,𝒒,\mathrm{i}\omega _{\mathrm{}})={\displaystyle \frac{1}{\beta }}{\displaystyle \underset{\mathrm{i}\omega _n}{}}{\displaystyle \frac{1}{E_{u,l}^4}}\{g_{0,z}`$
$`+2(\mathrm{\Delta }_{𝒌,\sigma }^m\mathrm{\Delta }_{𝒌+𝒒,\sigma ^{^{}}}^\chi \mathrm{\Delta }_{𝒌,\sigma }^\chi \mathrm{\Delta }_{𝒌+𝒒,\sigma ^{^{}}}^m)\},`$
$`F_{\stackrel{~}{\sigma }^z,\stackrel{~}{\sigma }^0}^{\sigma ^{^{}},\sigma }(𝒌,𝒒,\mathrm{i}\omega _{\mathrm{}})={\displaystyle \frac{1}{\beta }}{\displaystyle \underset{\mathrm{i}\omega _n}{}}{\displaystyle \frac{1}{E_{u,l}^4}}\{g_{0,z}`$
$`2(\mathrm{\Delta }_{𝒌,\sigma }^m\mathrm{\Delta }_{𝒌+𝒒,\sigma ^{^{}}}^\chi \mathrm{\Delta }_{𝒌,\sigma }^\chi \mathrm{\Delta }_{𝒌+𝒒,\sigma ^{^{}}}^m)\},`$
$`F_{\stackrel{~}{\sigma }^x,\stackrel{~}{\sigma }^y}^{\sigma ^{^{}},\sigma }(𝒌,𝒒,\mathrm{i}\omega _{\mathrm{}})={\displaystyle \frac{1}{\beta }}{\displaystyle \underset{\mathrm{i}\omega _n}{}}{\displaystyle \frac{1}{E_{u,l}^4}}\{g_{x,y}`$
$`+2\mathrm{i}(\mathrm{\Delta }_{𝒌,\sigma }^m\mathrm{\Delta }_{𝒌+𝒒,\sigma ^{^{}}}^\chi +\mathrm{\Delta }_{𝒌,\sigma }^\chi \mathrm{\Delta }_{𝒌+𝒒,\sigma ^{^{}}}^m)\},`$
$`F_{\stackrel{~}{\sigma }^y,\stackrel{~}{\sigma }^x}^{\sigma ^{^{}},\sigma }(𝒌,𝒒,\mathrm{i}\omega _{\mathrm{}})={\displaystyle \frac{1}{\beta }}{\displaystyle \underset{\mathrm{i}\omega _n}{}}{\displaystyle \frac{1}{E_{u,l}^4}}\{g_{x,y}`$
$`+2\mathrm{i}(\mathrm{\Delta }_{𝒌,\sigma }^m\mathrm{\Delta }_{𝒌+𝒒,\sigma ^{^{}}}^\chi +\mathrm{\Delta }_{𝒌,\sigma }^\chi \mathrm{\Delta }_{𝒌+𝒒,\sigma ^{^{}}}^m)\},`$
$`F_{\stackrel{~}{\sigma }^x,\stackrel{~}{\sigma }^z}^{\sigma ^{^{}},\sigma }(𝒌,𝒒,\mathrm{i}\omega _{\mathrm{}})={\displaystyle \frac{1}{\beta }}{\displaystyle \underset{\mathrm{i}\omega _n}{}}{\displaystyle \frac{1}{E_{u,l}^4}}\{g_{x,z}`$
$`\mathrm{\Delta }_{𝒌,\sigma }^m(\xi _{𝒌+𝒒}\xi _{𝒌+𝒒+𝑸})\mathrm{\Delta }_{𝒌+𝒒,\sigma ^{^{}}}^m(\xi _𝒌\xi _{𝒌+𝑸})\},`$
$`F_{\stackrel{~}{\sigma }^z,\stackrel{~}{\sigma }^x}^{\sigma ^{^{}},\sigma }(𝒌,𝒒,\mathrm{i}\omega _{\mathrm{}})={\displaystyle \frac{1}{\beta }}{\displaystyle \underset{\mathrm{i}\omega _n}{}}{\displaystyle \frac{1}{E_{u,l}^4}}\{g_{x,z}`$
$`\mathrm{\Delta }_{𝒌,\sigma }^m(\xi _{𝒌+𝒒}\xi _{𝒌+𝒒+𝑸})\mathrm{\Delta }_{𝒌+𝒒,\sigma ^{^{}}}^m(\xi _𝒌\xi _{𝒌+𝑸})\},`$
$`F_{\stackrel{~}{\sigma }^y,\stackrel{~}{\sigma }^z}^{\sigma ^{^{}},\sigma }(𝒌,𝒒,\mathrm{i}\omega _{\mathrm{}})={\displaystyle \frac{1}{\beta }}{\displaystyle \underset{\mathrm{i}\omega _n}{}}{\displaystyle \frac{1}{E_{u,l}^4}}\{g_{y,z}`$
$`\mathrm{i}\mathrm{\Delta }_{𝒌,\sigma }^\chi (\xi _{𝒌+𝒒}\xi _{𝒌+𝒒+𝑸})\mathrm{i}\mathrm{\Delta }_{𝒌+𝒒,\sigma ^{^{}}}^\chi (\xi _𝒌\xi _{𝒌+𝑸})\},`$
$`F_{\stackrel{~}{\sigma }^z,\stackrel{~}{\sigma }^y}^{\sigma ^{^{}},\sigma }(𝒌,𝒒,\mathrm{i}\omega _{\mathrm{}})={\displaystyle \frac{1}{\beta }}{\displaystyle \underset{\mathrm{i}\omega _n}{}}{\displaystyle \frac{1}{E_{u,l}^4}}\{g_{y,z}`$
$`\mathrm{i}\mathrm{\Delta }_{𝒌,\sigma }^\chi (\xi _{𝒌+𝒒}\xi _{𝒌+𝒒+𝑸})\mathrm{i}\mathrm{\Delta }_{𝒌+𝒒,\sigma ^{^{}}}^\chi (\xi _𝒌\xi _{𝒌+𝑸})\}.`$
## Appendix D Existence of a Goldstone mode
In this appendix we examine the contribution to the second order action $`𝒮_2`$ given by $`m_{𝒒+𝑸}^x`$ and $`m_{𝒒+𝑸}^y`$ in the half filled, uniform and static limit
$`\mathrm{x}=0,𝒒=\mathrm{𝟎},\mathrm{i}\omega _{\mathrm{}}=0.`$ (217)
We write it as a part of $`𝒮_2`$ (47)
$`𝒮_2^{m_{x,y}}=\left(_{5,5}(𝒒=\mathrm{𝟎},𝑸,𝑸,\mathrm{i}\omega _{\mathrm{}}=0)+{\displaystyle \frac{1}{2}}\mathrm{\Pi }_{5,5}(𝒒=\mathrm{𝟎},𝑸,𝑸,\mathrm{i}\omega _{\mathrm{}}=0)\right)\left[\delta m_𝑸^x(\mathrm{i}\omega _{\mathrm{}}=0)\right]^2`$
$`+\left(_{6,6}(𝒒=\mathrm{𝟎},𝑸,𝑸,\mathrm{i}\omega _{\mathrm{}}=0)+{\displaystyle \frac{1}{2}}\mathrm{\Pi }_{6,6}(𝒒=\mathrm{𝟎},𝑸,𝑸,\mathrm{i}\omega _{\mathrm{}}=0)\right)\left[\delta m_𝑸^y(\mathrm{i}\omega _{\mathrm{}}=0)\right]^2`$
$`=\alpha \stackrel{~}{J}_{𝒒=\mathrm{𝟎}}\left(\delta m_𝑸^x\right)^2+\alpha \stackrel{~}{J}_{𝒒=\mathrm{𝟎}}\left(\delta m_𝑸^y\right)^2+{\displaystyle \frac{1}{2\mathrm{N}_\mathrm{s}}}{\displaystyle \underset{𝒌}{}}{\displaystyle {}_{}{}^{}\underset{\sigma ,\sigma ^{^{}}}{}}\left[\alpha \stackrel{~}{J}_{𝒒=\mathrm{𝟎}}\sigma _{\sigma ^{^{}},\sigma }^x\right]F_{\stackrel{~}{\sigma }^x,\stackrel{~}{\sigma }^x}^{\sigma ^{^{}},\sigma }\left[\alpha \stackrel{~}{J}_{𝒒=\mathrm{𝟎}}\sigma _{\sigma ,\sigma ^{^{}}}^x\right]\left(\delta m_𝑸^x\right)^2`$
$`+{\displaystyle \frac{1}{2\mathrm{N}_\mathrm{s}}}{\displaystyle \underset{𝒌}{}}{\displaystyle {}_{}{}^{}\underset{\sigma ,\sigma ^{^{}}}{}}\left[\alpha \stackrel{~}{J}_{𝒒=\mathrm{𝟎}}\sigma _{\sigma ^{^{}},\sigma }^y\right]F_{\stackrel{~}{\sigma }^x,\stackrel{~}{\sigma }^x}^{\sigma ^{^{}},\sigma }\left[\alpha \stackrel{~}{J}_{𝒒=\mathrm{𝟎}}\sigma _{\sigma ,\sigma ^{^{}}}^y\right]\left(\delta m_𝑸^y\right)^2.`$ (218)
By using Eqs. (169) and (215) we get in the simplified limit (217)
$`𝒮_2^{m_{x,y}}`$ $`=`$ $`\left(2\alpha J{\displaystyle \frac{1}{\mathrm{N}_\mathrm{s}}}{\displaystyle \underset{𝒌}{}}{}_{}{}^{}{\displaystyle \frac{8\alpha ^2J^2}{E_𝒌^{up}E_𝒌^{low}}}\right)`$ (219)
$`\times \left[\left(\delta m_𝑸^x\right)^2+\left(\delta m_𝑸^y\right)^2\right].`$
The mean field equation (40) related to the magnetization gives
$`{\displaystyle \frac{1}{\mathrm{N}_\mathrm{s}}}{\displaystyle \underset{𝒌}{}}{}_{}{}^{}{\displaystyle \frac{8\alpha ^2J^2}{E_𝒌^{up}E_𝒌^{low}}}=2\alpha J,`$
then with Eq. (219) we have finally: $`𝒮_2^{m_{x,y}}=0`$.
The behaviour of the second order action concerning the $`x`$ and $`y`$ directions of the magnetization implies that the transverse spin-spin correlation function $`\chi ^\pm `$ contains a gapless pole. It is a consequence of the Goldstone theorem, Goldstone-Th which has to be applied in the present case because of the rotational symmetry breaking in spin space due to the imposed AF order. It is in agreement with the effective field theory of quantum antiferromagnets, which was built by Haldane for one-dimensional chains Haldane1983 and extended later to square lattice systems. AF2D The existence of this Goldstone mode has already been observed in the original spin-bag approach. Schrieffer1989 ; Schrieffer1989long
|
warning/0506/math0506189.html
|
ar5iv
|
text
|
# A Variational Principle in the Dual Pair of Reproducing Kernel Hilbert Spaces and an Application
## 1 Introduction
In this paper we will consider certain variational principle arising in the dual pair of reproducing kernel Hilbert spaces (abbreviated RKHS’s hereafter). Then we will find an application in showing the Gibbsianness of some determinantal point processes (in short DPP’s) in discrete spaces.
Let $`E`$ be any countable set, e.g., $`E=^d`$, the $`d`$-dimensional lattice space. Let $`_0:=l^2(E)`$ be the space of square summable functions (sequences) on $`E`$ with inner product
$$(f,g)_0:=\underset{xE}{}\overline{f(x)}g(x).$$
(1.1)
We denote the corresponding norm by $`_0`$. Let $`A`$ be a bounded positive definite operator on $`_0`$. We assume that the kernel space is trivial: $`\text{ker}A=\{0\}`$. Then the range, $`\text{ran}A`$, is dense in $`_0`$ and we introduce two new norms. First on $`_0`$ we define
$$f_{}^2:=(f,Af)_0,f_0.$$
(1.2)
Let $`_{}`$ be the closure of $`_0`$ w.r.t. this norm. Next we define a norm $`_+`$ on $`\text{ran}A`$ by
$$g_+^2:=(g,A^1g)_0,g\text{ran}A.$$
(1.3)
The closure of $`\text{ran}A`$ w.r.t. the norm $`_+`$ is denoted by $`_+`$. We then get a triple with inclusions:
$$_{}_0_+.$$
(1.4)
Let us denote by $`𝖡:=\{e_x:xE\}`$ the usual basis of $`_0`$, i.e., $`e_x_0`$ is the unit vector whose component is one at $`x`$ and zero at all other sites. Let $`A(x,y)`$, $`x,yE`$, be the representation of $`A`$ w.r.t. the basis $`𝖡`$. Then the space $`_+`$ is nothing but a RKHS with a reproducing kernel (abbreviated RK) $`A(x,y)`$, $`x,yE`$ (see Subsection 2.1). We allow $`0\text{spec}A`$, the spectrum of $`A`$. That is, the inverse $`A^1`$ of $`A`$ may be an unbounded operator on $`_0`$. But we will impose some conditions on $`A`$ so that the space $`_{}`$ is also a RKHS with a RK $`B(x,y)`$, $`x,yE`$. See the hypothesis (H) in Section 2. Informally saying, the function $`B(x,y)`$ is the kernel function of the inverse operator $`A^1`$:
$$B(x,y)=A^1(x,y),x,yE.$$
(1.5)
The variational principle we will address is the following. We notice first that the assumption of $`_{}`$ being a RKHS implies in particular that $`e_x_+`$ for all $`xE`$ (see Subsection 2.1). Let $`x_0E`$ be a fixed point and let $`\{x_0\}R_1R_2=E`$ be any partition of $`E`$. We define
$`\alpha `$ $`:=`$ $`\underset{\mathrm{\Lambda }E}{lim}\alpha _\mathrm{\Lambda };`$
$`\alpha _\mathrm{\Lambda }`$ $`:=`$ $`\underset{f\text{span}\{e_x:x\mathrm{\Lambda }R_1\}}{inf}e_{x_0}f_{}^2,`$ (1.6)
and similary
$`\beta `$ $`:=`$ $`\underset{\mathrm{\Lambda }E}{lim}\beta _\mathrm{\Lambda };`$
$`\beta _\mathrm{\Lambda }`$ $`:=`$ $`\underset{g\text{span}\{e_x:x\mathrm{\Lambda }R_2\}}{inf}e_{x_0}g_+^2,`$ (1.7)
where $`\mathrm{\Lambda }`$ increases to $`E`$ through finite subsets. We will show that the two numbers $`\alpha `$ and $`\beta `$ are the inverses to each other (Theorem 2.3):
$$\alpha \beta =1.$$
(1.8)
This result has been shown by Shirai and Takahashi in the case when $`A`$ is a strictly positive operator, and hence $`A^1`$ is also bounded. They applied this result to show the Gibbsianness of a DPP defined by the operator $`A(I+A)^1`$ (see Section 2 for the definition of DPP’s). In fact, the variational principle (1.8) will guarantee the existence of global Papangelou intensity. In other words, it will prove the existence of the limit of local Papangelou intensities as the local region increases to the whole space (Theorem 2.4). This proves the Gibbsianness of the DPP and we will give a proper interaction potential and show also the uniqueness of the Gibbs measure (Theorem 2.5). The interaction potential is actually given by the logarithm of the determinants of the submatrices of $`A`$:
$$V(\xi )=\mathrm{log}det(A(x_i,x_j))_{i,j=1}^n,$$
(1.9)
where $`\xi =\{x_1,\mathrm{},x_n\}E`$ is any finite configuration.
We remark here that the main idea in showing the Gibbsianness has been borrowed from . We should, however, point out that since the operators $`A`$ dealt with in are strictly positive, there is a severe restriction in applications. For example, if $`A`$ is a diagonal matrix with diagonal elements $`\alpha _x>0`$ that decrease to zero as $`x\mathrm{}`$ (we let $`E=`$ or $``$), then the DPP corresponding to the operator $`A(I+A)^1`$ is clearly a Gibbs measure. The system has the one-body interactions only and the potential energy is given by
$$V(\xi )=\underset{x\xi }{}\mathrm{log}\alpha _x.$$
(1.10)
Even this kind of simple example lies outside the regime of . This paper improves (in regard of Gibbsianness of DPP’s) in that our setting includes more general classes as well as the above example.
This paper is organized as follows. In Section 2, we introduce the basics of the RKHS’s (Subsection 2.1) and DPP’s (Subsection 2.2), and then give the main results (Subsection 2.3). Section 3 is devoted to the proof of variational principle, Theorem 2.3. In Section 4, we first prove the existence of the global Papangelou intensity, Theorem 2.4. Then we prove the Gibbsianness and its uniqueness, Theorem 2.5. In the Appendix, we provide with some examples.
## 2 Preliminaries and Main Results
In this Section we review some basics of RKHS’s and DPP’s. Then we state the main results of this paper.
### 2.1 Reproducing Kernel Hilbert Spaces
For our convenience, we start from a Hermitian positive definite bounded linear operator $`A`$ on the complex Hilbert space $`_0:=l^2(E)`$ equipped with an inner product
$$(f,g)_0:=\underset{xE}{}\overline{f(x)}g(x),f,g_0.$$
(2.1)
Here $`E`$ is any countable set. Throughout this paper we assume that the kernel space of $`A`$ is trivial:
$$\text{ker}A=\{0\}.$$
(2.2)
Then, since $`\overline{\text{ran}A}=(\text{ker}A^{})^{}=(\text{ker}A)^{}=_0`$, $`\text{ran}A`$ is dense in $`_0`$. As in the introduction, let $`𝖡=\{e_x:xE\}`$ be the usual basis of $`_0`$. Let $`A(x,y)`$, $`x,yE`$, be the matrix element of the operator $`A`$ w.r.t. the basis $`𝖡`$:
$$A(x,y):=(e_x,Ae_y)_0,x,yE.$$
(2.3)
On the dense subspace $`\text{ran}A`$, we define a new inner product as
$$(f,g)_+:=(f,A^1g)_0,f,g\text{ran}A.$$
(2.4)
Denote by $`_+`$ the resulting norm and let $`_+`$ be the completion of $`\text{ran}A`$ w.r.t. $`_+`$. We notice that $`_+`$ is a RKHS with kernel function $`A(x,y)`$, that is the following defining conditions are satisfied:
* For every $`xE`$, the function $`A(,x)`$ belongs to $`_+`$,
* The reproducing property: for every $`xE`$ and $`g_+`$,
$$g(x)=(A(,x),g)_+.$$
(2.5)
Let us now consider another Hilbert space $`_{}`$ which is the closure of $`_0`$ w.r.t. the norm $`_{}`$ induced by the inner product:
$$(f,g)_{}:=(f,Ag)_0,f,g_0.$$
(2.6)
It is important to notice that though $`_0`$ may be understood as a class of functions defined on the set $`E`$, the completed space $`_{}`$ may not be a space of functions defined on the same space $`E`$. This is so called a functional completion problem and will be discussed below. By the boundedness of $`A`$ we have the inclusions:
$$_{}_0_+.$$
(2.7)
We want to see $`_{}`$ also as a RKHS. First we define a dual pairing between the spaces $`_{}`$ and $`_+`$. For $`f_0`$ and $`g\text{ran}A`$, define
$$_{}f,g_+:=\underset{xE}{}\overline{f(x)}g(x).$$
(2.8)
We have then the bound $`|_{}f,g_+|f_{}g_+`$. Since $`_0`$ and $`\text{ran}A`$ are dense respectively in $`_{}`$ and $`_+`$, the dual pairing extends continuously to a bilinear form on $`_{}\times _+`$, for which we use the same notation $`{}_{}{}^{}f,g_{+}^{}`$, $`f_{}`$ and $`g_+`$, and the bound also continues to hold:
$$|_{}f,g_+|f_{}g_+,f_{},g_+.$$
(2.9)
For a convenience, we also define its conjugate bilinear form
$$_+g,f_{}:=\overline{{}_{}{}^{}f,g_{+}^{}},f_{},g_+.$$
(2.10)
Notice that for $`f_0`$, $`Af_+`$ and
$$Af_+^2=(Af,A^1Af)_0=f_{}^2.$$
(2.11)
Thus, $`A`$ extends to an isometry between $`_{}`$ and $`_+`$. We will denote the extension by the same $`A`$ and its inverse by $`A^1`$.
Let us now introduce the notion of functional completion of an incomplete class $`𝖥`$ of functions on $`E`$ which is a pre-Hilbert space. By this, as introduced in \[1, p 347\], we mean a completion of $`𝖥`$ by adjunction of functions on $`E`$ such that the evaluation map at any site $`yE`$ is a continuous function on the completed space. The following theorem proved by Aronszajn gives a necessary and sufficient condition for the functional completion.
###### Theorem 2.1 (Aronszajn)
Let $`𝖥`$ be a class of functions on $`E`$ forming a pre-Hilbert space. In order that there exists a functional completion of $`𝖥`$, it is necessary and sufficient that
* for every fixed $`yE`$, the linear functional $`f(y)`$ defined in $`𝖥`$ is continuous;
* for a Cauchy sequence $`\{f_n\}𝖥`$, the condition $`f_n(y)0`$ for every $`y`$ implies that $`f_n`$ itself converges to $`0`$ in norm.
If the functional completion is possible, it is unique.
In our setting, the incomplete class of functions is $`_0`$ equipped with the inner product $`(,)_{}`$. We shall demand $`_{}`$ to be functionally completed. We state all the conditions we need as a hypothesis:
(H) The Hermitian positive definite linear operator $`A`$ on $`_0`$ is bounded and satisfies (i) $`\text{ker}A=\{0\}`$; (ii) $`_{}`$ is functionally completed.
In the Appendix we will consider some examples of the operators $`A`$ that satisfy the conditions in (H).
Now $`_{}`$ being functionally completed, it satisfies, by definition, that for every $`yE`$, the functional $`f(y)`$ is continuous on $`_{}`$. Notice that by the dual pairing $`{}_{}{}^{},_{+}^{}`$, it is equivalent to saying that $`e_y_+`$ for any $`yE`$. In fact, it is not hard to check that the functional $`{}_{}{}^{},g_{+}^{}`$ on $`_{}`$ has norm $`g_+`$ for any $`g_+`$, and the functional $`{}_{}{}^{}f,_{+}^{}`$ on $`_+`$ has norm $`f_{}`$ for each $`f_{}`$. Moreover, by the isometries $`A:_{}_+`$ and its inverse $`A^1:_+_{}`$, it is easy to check that
$$_{},g_+=(,A^1g)_{}\text{ and }_{}f,_+=(Af,)_+,f_{},g_+.$$
(2.12)
That is, $`_+`$ and $`_{}`$ are respectively the dual spaces of each other via the dual pairing $`{}_{}{}^{},_{+}^{}`$. Now if $`e_y_+`$ for every $`yE`$, then obviously $`f(y)=_{}f,e_y_+`$ is continuous on $`_{}`$. On the other hand, suppose that the functional $`f(y)`$ is continuous on $`_{}`$ for every $`yE`$. Then, for each $`y`$, by the above observation, there is a unique element $`l_y_+`$ such that
$$f(y)=_{}f,l_y_+,f_{}.$$
(2.13)
Since finitely supported vectors $`f`$ are dense in $`_{}`$ and for those vectors $`f`$ we have $`{}_{}{}^{}f,l_y_{+}^{}=_x\overline{f(x)}l_y(x)`$, $`l_y`$ must be $`e_y`$.
Finally, we notice that since for any fixed $`yE`$ the functionals $`_{}ff(y)`$ and $`_+gg(y)`$ are continuous, respectively in $`_{}`$ and $`_+`$, it is obvious that
$$_{}f,g_+=\underset{xE}{}\overline{f(x)}g(x),\text{ if either }f\text{ or }g\text{ is locally supported}.$$
(2.14)
### 2.2 Determinantal Point Processes on Discrete Sets
Determinantal point processes, or fermion random point fields, are probability measures on the configuration space of, say, particles. The particles may move on the continuum spaces or on the discrete spaces. In this paper we will focus on the DPP’s on the discrete sets.
The basics of DPP’s including their definitions and basic properties can be found in several papers . We will review the definition of DPP’s mainly from the paper . Let $`E`$ be a countable set and let $`K`$ be a Hermitian positive definite bounded linear operator on the Hilbert space $`_0=l^2(E)`$. Let $`𝒳`$ be the configuration space on $`E`$, that is, $`𝒳`$ is the class of all subsets of $`E`$. We frequently understand a point $`\xi =(x_i)_{i=1,2,\mathrm{}}𝒳`$ as a configuration of particles located at the sites $`x_iE`$, $`i=1,2,\mathrm{}`$. The following theorem gives an existence theorem for DPP’s. We state it as appeared in .
###### Theorem 2.2
Let $`E`$ be a countable discrete space and $`K`$ be a Hermitian bounded operator on $`_0=l^2(E)`$. Assume that $`0KI`$. Then, there exists a unique probability Borel measure $`\mu `$ on $`𝒳`$ such that for any finite subset $`XE`$,
$$\mu (\{\xi 𝒳:\xi X\})=det(K(x,y))_{x,yX}.$$
(2.15)
The $`\sigma `$-algebra on $`𝒳`$ is induced from the product topology on $`\{0,1\}^E`$ (see Section 4). Here we remark that the left hand side of (2.15) is just the correlation function of the probability measure $`\mu `$, thus the theorem says that the correlation functions of DPP’s are given by the determinants of positive definite kernel functions.
The most useful feature in the theory of DPP’s is that there can be given an exact formula for the density functions of local marginals. For each subset $`\mathrm{\Lambda }E`$, let $`P_\mathrm{\Lambda }`$ denote the projection operator on $`_0`$ onto the space of vectors which have supports on the set $`\mathrm{\Lambda }`$. Let $`K_\mathrm{\Lambda }:=P_\mathrm{\Lambda }KP_\mathrm{\Lambda }`$ be the restriction of $`K`$ on the projection space. Given a configuration $`\xi 𝒳`$, we let $`\xi _\mathrm{\Lambda }`$ be the restriction of $`\xi `$ on the set $`\mathrm{\Lambda }`$, i.e.,
$$\xi _\mathrm{\Lambda }:=\xi \mathrm{\Lambda }.$$
(2.16)
For each finite subset $`\mathrm{\Lambda }E`$, assuming first that $`I_\mathrm{\Lambda }K_\mathrm{\Lambda }`$ is invertible, we define
$$A_{[\mathrm{\Lambda }]}:=K_\mathrm{\Lambda }(I_\mathrm{\Lambda }K_\mathrm{\Lambda })^1.$$
(2.17)
Then for the DPP $`\mu `$ corresponding to the operator $`K`$, the marginals are given by the formula: for each finite subset $`\mathrm{\Lambda }E`$ and fixed $`\xi 𝒳`$,
$$\mu (\{\zeta :\zeta _\mathrm{\Lambda }=\xi _\mathrm{\Lambda }\})=det(I_\mathrm{\Lambda }K_\mathrm{\Lambda })det(A_{[\mathrm{\Lambda }]}(x,y))_{x,y\xi _\mathrm{\Lambda }},$$
(2.18)
where $`A_{[\mathrm{\Lambda }]}(x,y)`$, $`x,y\mathrm{\Lambda }`$, denotes the matrix components of $`A_{[\mathrm{\Lambda }]}`$. Though in this paper we will confine ourselves to the case where $`A_{[\mathrm{\Lambda }]}`$ is well-defined as a bounded operator, we remark that the formula (2.18) is meaningful even if $`K_\mathrm{\Lambda }`$ has $`1`$ in its spectrum .
### 2.3 Results
First we will consider a variational principle for the positive definite operator $`A`$ introduced in Subsection 2.1. Since we are assuming that $`_{}`$ is functionally completed, for any $`yE`$ the functional $`f(y)`$ is continuous on $`_{}`$, or $`e_y_+`$. This condition, on the other hand, is equivalent to the one that $`_{}`$ is a RKHS \[1, p 343\]. Let $`B(x,y)`$ be the RK for $`_{}`$. From the reproducing property we see that $`B(x,y)`$ is the value of the function $`A^1e_y`$ at $`x`$ \[1, p 344\], that is
$$B(x,y)=_+e_x,A^1e_y_{},x,yE.$$
(2.19)
Let $`x_0E`$ be a fixed point and let $`R_1`$ and $`R_2`$ be any two subsets of $`E`$ such that $`E`$ is partitioned into three sets:
$$E=\{x_0\}R_1R_2.$$
(2.20)
For each $`\mathrm{\Delta }E`$, we let $`𝖥_{\text{loc},\mathrm{\Delta }}`$ be the local functions supported on $`\mathrm{\Delta }`$:
$$𝖥_{\text{loc},\mathrm{\Delta }}:=\text{the class of finite linear combinations of }\{e_x:x\mathrm{\Delta }\}.$$
(2.21)
In the sequel, we denote by $`\mathrm{\Lambda }E`$ that $`\mathrm{\Lambda }`$ is a finite subset of $`E`$. We are concerned with the following numbers. For each $`\mathrm{\Lambda }E`$, define
$$\alpha _\mathrm{\Lambda }:=\underset{f𝖥_{\text{loc},\mathrm{\Lambda }R_1}}{inf}e_{x_0}f_{}^2$$
(2.22)
and
$$\beta _\mathrm{\Lambda }:=\underset{g𝖥_{\text{loc},\mathrm{\Lambda }R_2}}{inf}e_{x_0}g_+^2.$$
(2.23)
Obviously, both of the sequences of nonnegative numbers $`\{\alpha _\mathrm{\Lambda }\}_{\mathrm{\Lambda }E}`$ and $`\{\beta _\mathrm{\Lambda }\}_{\mathrm{\Lambda }E}`$ decrease as $`\mathrm{\Lambda }`$ increases. We let
$$\alpha :=\underset{\mathrm{\Lambda }E}{lim}\alpha _\mathrm{\Lambda }\text{and}\beta :=\underset{\mathrm{\Lambda }E}{lim}\beta _\mathrm{\Lambda }.$$
(2.24)
One of the main result of this paper is the following:
###### Theorem 2.3
Let the operator $`A`$ satisfy the conditions in the hypothesis (H). Then the product of the numbers $`\alpha `$ and $`\beta `$ defined in (2.24) is one: $`\alpha \beta =1`$.
We remark that the result of the theorem was obtained by Shirai and Takahashi \[16, Theoem 6.3\] in the case that the bounded operator $`A`$ is strictly greater than $`0`$, i.e., $`0<cIA`$ for some positive constant $`c`$.
One of the main purpose of this paper is to apply the above result to show the Gibbsianness of some DPP’s. Let $`A`$ be an operator on $`_0`$ that satisfies the hypothesis (H). Let $`\mu `$ be the DPP corresponding to the operator $`K:=A(I+A)^1`$. Given a fixed point $`x_0E`$ and a configuration $`\xi 𝒳`$ with $`x_0\xi `$, and for each $`\mathrm{\Lambda }E`$, let $`\alpha _{[\mathrm{\Lambda }]}`$ be the conditional probability of finding a particle at the site $`x_0`$ given the particle configuration $`\xi _\mathrm{\Lambda }`$ in $`\mathrm{\Lambda }`$:
$$\alpha _{[\mathrm{\Lambda }]}:=\mu (x_0\xi _\mathrm{\Lambda }|\xi _\mathrm{\Lambda })=\frac{\mu _\mathrm{\Lambda }(x_0\xi _\mathrm{\Lambda })}{\mu _\mathrm{\Lambda }(\xi _\mathrm{\Lambda })},$$
(2.25)
where we have simplified the event $`\{\zeta 𝒳:\zeta _\mathrm{\Lambda }=\xi _\mathrm{\Lambda }\}\xi _\mathrm{\Lambda }`$, etc, and $`x_0\xi _\mathrm{\Lambda }=\{x_0\}\xi _\mathrm{\Lambda }`$. By (2.18), $`\alpha _{[\mathrm{\Lambda }]}`$ is computed via the ratio of determinants:
$$\alpha _{[\mathrm{\Lambda }]}=\frac{detA_{[\mathrm{\Lambda }]}(x_0\xi _\mathrm{\Lambda },x_0\xi _\mathrm{\Lambda })}{detA_{[\mathrm{\Lambda }]}(\xi _\mathrm{\Lambda },\xi _\mathrm{\Lambda })},$$
(2.26)
where $`A_{[\mathrm{\Lambda }]}=K_\mathrm{\Lambda }(I_\mathrm{\Lambda }K_\mathrm{\Lambda })^1`$ and $`A_{[\mathrm{\Lambda }]}(\xi _\mathrm{\Lambda },\xi _\mathrm{\Lambda })=(A_{[\mathrm{\Lambda }]}(x,y))_{x,y\xi _\mathrm{\Lambda }}`$. We are interested in the behavior of the sequence $`\{\alpha _{[\mathrm{\Lambda }]}\}`$ as $`\mathrm{\Lambda }`$ increases to $`E`$. The following theorem gives the answer.
###### Theorem 2.4
Let the operator $`A`$ satisfy the conditions in (H). Then
$$\underset{\mathrm{\Lambda }E}{lim}\alpha _{[\mathrm{\Lambda }]}=\alpha ,$$
(2.27)
where $`\alpha `$ is given in (2.22) and (2.24) with $`R_1=\xi `$ and $`R_2=E(\xi \{x_0\})`$.
A corollary to this theorem is that the DPP $`\mu `$ corresponding to the operator $`A(I+A)^1`$ is a Gibbs measure. We state this as a theorem.
###### Theorem 2.5
Let the operator $`A`$ satisfy the conditions in (H). Then the DPP $`\mu `$ corresponding to the operator $`A(I+A)^1`$ is a Gibbs measure. The interaction potential is given by the logarithm of determinants of submatrices of $`A`$: for any finite configuration $`\xi 𝒳`$, the interaction potential $`V(\xi )`$ is
$$V(\xi )=\mathrm{log}det(A(x,y))_{x,y\xi }.$$
(2.28)
Moreover, $`\mu `$ is the only Gibbs measure for the potential energy (2.28).
The above result also extends that obtained in \[16, Theorem 6.2\], where $`KA(I+A)^1`$ is assumed to have its spectrum in the open interval $`(0,1)`$. We also notice that the idea developed in refs. and , which concerns exclusively with continuum models, can be applied to discrete model and would get some result on the Gibbsianness of $`\mu `$. The result would look like the following (cf. \[4, Proposition 3.9\]): Let $`E^d`$ and suppose that (i) $`A`$ is of finite range in the sense that $`A(x,y)=0`$ if $`|xy|R`$ for some finite number $`R>0`$ and (ii) $`\mu `$ does not percolate. Then $`\mu `$ is a Gibbs measure corresponding to the potential in (2.28). Our result 2.5 is stronger than this, too.
## 3 Proof of the Variational Principle
In this Section we prove Theorem 2.3. The most important tool in the proof is the theory of restrictions and projections in the RKHS’s. In Subsection 3.1, we deal with the variational principle in the finite systems. In Subsection 3.2, we first introduce the restriction theory in the RKHS’s and then discuss the limit theorems of RK’s. The proof of Theorem 2.3 is given in Subsection 3.3.
### 3.1 Variational Principle in the Finite Systems
We discuss the variational principle for positive definite matrices on a finite set. Let $`\mathrm{\Lambda }E`$ be a finite set and let $`(C(x,y))_{x,y\mathrm{\Lambda }}`$ be a positive definite matrix with an inverse $`C^1`$. We define two norms on the class $`𝖥_\mathrm{\Lambda }`$ of functions on $`\mathrm{\Lambda }`$ as follows:
$$f_{}^2:=\underset{x,y\mathrm{\Lambda }}{}\overline{f(x)}C(x,y)f(y),f𝖥_\mathrm{\Lambda }$$
(3.1)
and
$$g_+^2:=\underset{x,y\mathrm{\Lambda }}{}\overline{g(x)}C^1(x,y)g(y),g𝖥_\mathrm{\Lambda }.$$
(3.2)
Suppose that $`\mathrm{\Lambda }=\{x_0\}\mathrm{\Lambda }_1\mathrm{\Lambda }_2`$ is a partition of $`\mathrm{\Lambda }`$ with disjoint sets $`\{x_0\}`$, $`\mathrm{\Lambda }_1`$, and $`\mathrm{\Lambda }_2`$. Similarly to (2.22)-(2.23) we define
$$a:=\underset{f𝖥_{\mathrm{\Lambda }_1}}{inf}e_{x_0}f_{}^2$$
(3.3)
and
$$b:=\underset{g𝖥_{\mathrm{\Lambda }_2}}{inf}e_{x_0}g_+^2.$$
(3.4)
In the above $`𝖥_{\mathrm{\Lambda }_i}`$ denotes the class of functions on $`\mathrm{\Lambda }_i`$, $`i=1,\mathrm{\hspace{0.17em}2}`$. Applying the method of finding extreme values of functions of several variables and using the elementary properties of determinants of finite matrices, we obtain the following results, which, as a matter of fact, take a role of recipe for the theory in the infinite systems (cf. \[16, Section 6\]). Below we denote by $`C(\mathrm{\Lambda }_1,\mathrm{\Lambda }_2)`$ the submatrix $`(C(x,y))_{x\mathrm{\Lambda }_1,y\mathrm{\Lambda }_2}`$ for any subsets $`\mathrm{\Lambda }_1`$ and $`\mathrm{\Lambda }_2`$ of $`\mathrm{\Lambda }`$. We also simplify $`\{x\}\mathrm{\Lambda }_1`$ by $`x\mathrm{\Lambda }_1`$ for $`x\mathrm{\Lambda }_1`$.
###### Proposition 3.1
Let $`(C(x,y))_{x,y\mathrm{\Lambda }}`$ be a Hermitian positive definite matrix on a finite set $`\mathrm{\Lambda }`$ with inverse $`C^1`$. Let $`\mathrm{\Lambda }=\{x_0\}\mathrm{\Lambda }_1\mathrm{\Lambda }_2`$ be a partition of $`\mathrm{\Lambda }`$ and let the norms $`_{}`$ and $`_+`$, and the numbers $`a`$ and $`b`$ be defined as in (3.1)-(3.4). Then the following results hold:
* The minimum values $`a`$ and $`b`$ are attained respectively at the unique vectors $`f_0=C(\mathrm{\Lambda }_1,\mathrm{\Lambda }_1)^1C(\mathrm{\Lambda }_1,x_0)`$ and $`g_0=(C^1(\mathrm{\Lambda }_2,\mathrm{\Lambda }_2))^1C^1(\mathrm{\Lambda }_2,x_0)`$:
$$a=e_{x_0}f_0_{}^2;b=e_{x_0}g_0_+^2.$$
(3.5)
* $`a`$ $`=`$ $`{\displaystyle \frac{detC(x_0\mathrm{\Lambda }_1,x_0\mathrm{\Lambda }_1)}{detC(\mathrm{\Lambda }_1,\mathrm{\Lambda }_1)}}=(C(x_0\mathrm{\Lambda }_1,x_0\mathrm{\Lambda }_1)^1(x_0,x_0))^1`$ (3.6)
$`=`$ $`C(x_0,x_0)C(x_0,\mathrm{\Lambda }_1)C(\mathrm{\Lambda }_1,\mathrm{\Lambda }_1)^1C(\mathrm{\Lambda }_1,x_0)`$
and similarly
$`b`$ $`=`$ $`{\displaystyle \frac{detC^1(x_0\mathrm{\Lambda }_2,x_0\mathrm{\Lambda }_2)}{detC^1(\mathrm{\Lambda }_2,\mathrm{\Lambda }_2)}}=((C^1(x_0\mathrm{\Lambda }_2,x_0\mathrm{\Lambda }_2))^1(x_0,x_0))^1`$ (3.7)
$`=`$ $`C^1(x_0,x_0)C^1(x_0,\mathrm{\Lambda }_2)(C^1(\mathrm{\Lambda }_2,\mathrm{\Lambda }_2))^1C^1(\mathrm{\Lambda }_2,x_0)`$
* $`ab=1`$.
### 3.2 Restrictions in RKHS’s and Limit Theorems of RK’s
In this Subsection, we discuss the restriction and projection theories in RKHS’s and the limit theorems of RK’s. These are crucial to characterize the values $`\alpha `$ and $`\beta `$ in (2.24) more concretely. The results we need have been already obtained in . For the readers’ convenience, however, we provide it here.
Let us begin with an introduction of the restriction theory in the RKHS’s. Suppose that $``$ is a RKHS with kernel $`K(x,y)`$, $`x,yE`$. $``$ might be $`_{}`$ or $`_+`$ of our concern. For each subset $`\mathrm{\Lambda }E`$, the function $`K_\mathrm{\Lambda }(x,y)`$, the restriction of $`K(x,y)`$ to $`\mathrm{\Lambda }`$, is still positive definite. The following theorem was proved by Aronszajn \[1, p 351\]:
###### Theorem 3.2
The function $`K(x,y)`$ restricted to a subset $`\mathrm{\Lambda }E`$ is the reproducing kernel of the class $`_\mathrm{\Lambda }`$ of all restrictions of functions of $``$ to the subset $`\mathrm{\Lambda }`$. For any such restriction $`f_\mathrm{\Lambda }_\mathrm{\Lambda }`$, the norm $`f_\mathrm{\Lambda }_\mathrm{\Lambda }`$ is the minimum of $`f`$ (the norm of $`f`$ in $``$) for all $`f`$ whose restrictions to $`\mathrm{\Lambda }`$ are $`f_\mathrm{\Lambda }`$.
When it is needed to designate the kernel, we use the notations $`_{\mathrm{\Lambda };K}`$ and $`_{\mathrm{\Lambda };K}`$ respectively for the restriction spaces and norms. The basic argument in Theorem 3.2 is the following. First let $`𝖥^0`$ be the class of functions that vanish on $`\mathrm{\Lambda }`$. This is a closed subspace and let $`𝖥^{}:=𝖥^0`$ be the orthogonal complement of $`𝖥^0`$. It is not hard to show that all the functions $`f`$ which have the same restriction $`f_\mathrm{\Lambda }`$ on $`\mathrm{\Lambda }`$ have a common projection $`f^{}`$ on $`𝖥^{}`$ and that the restriction of $`f^{}`$ to $`\mathrm{\Lambda }`$ is equal to $`f_\mathrm{\Lambda }`$. Clearly, among all these functions $`f`$, $`f^{}`$ is the one which has the smallest norm. We define
$$f_\mathrm{\Lambda }_\mathrm{\Lambda }:=f^{}.$$
(3.8)
The norm $`_\mathrm{\Lambda }`$ on $`_\mathrm{\Lambda }`$ defined this way is the one stated in the theorem. We refer to \[1, p 351\] for the details.
Next we discuss the limit theorems of RK’s. We will consider two kinds of limits.
A. The case of decreasing sequence. Let $`\{E_n\}`$ be an increasing sequence of sets with $`E=_{n=1}^{\mathrm{}}E_n`$. For each $`n=1,2,\mathrm{}`$, let $`𝖥_n`$ be a RKHS defined in $`E_n`$ with RK $`K_n(x,y)`$, $`x,yE_n`$. we denote the norm in the space $`𝖥_n`$ by $`_n`$, $`n1`$. For a function $`f_n𝖥_n`$ we will denote by $`f_{nm}`$, $`mn`$, the restriction of $`f_n`$ to the set $`E_mE_n`$. We shall suppose the following two conditions:
1. for every $`f_n𝖥_n`$ and every $`mn`$, $`f_{nm}𝖥_m`$;
2. for every $`f_n𝖥_n`$ and every $`mn`$, $`f_{nm}_mf_n_n`$.
From (A2) we see by \[1, Theorem II of Section 7\] that
$$K_{nm}K_m,m<n,$$
(3.9)
meaning that $`K_m(x,y)K_{nm}(x,y)`$, $`x,yE_m`$, is a positive definite function, where $`K_{nm}`$ is the restriction of $`K_n`$ to the set $`E_m`$. The following theorem appears in \[1, Theorem I, Section 9\]:
###### Theorem 3.3
Under the above assumptions on the classes $`𝖥_n`$, the kernels $`K_n`$ converge to a Kernel $`K_0(x,y)`$defined for all $`x,y`$ in $`E`$. $`K_0`$ is the RK of the class $`𝖥_0`$ of all functions $`f_0`$ defined in $`E`$ such that
1. their restrictions $`f_{0n}`$ in $`E_n`$ belong to $`𝖥_n`$, $`n=1,2,\mathrm{}`$;
2. $`lim_n\mathrm{}f_{0n}_n<\mathrm{}`$.
The norm of $`f_0𝖥_0`$ is given by $`f_0_0=lim_n\mathrm{}f_{0n}_n`$.
B. The case of increasing sequence. Let $`\{E_n\}`$ be a decreasing sequence of sets and $`R`$ be their intersection:
$$R=_{n=1}^{\mathrm{}}E_n.$$
(3.10)
As in the case A, let $`𝖥_n`$, $`n=1,2\mathrm{}`$, be the RKHS’s with corresponding kernel functions $`K_n(x,y)`$, $`x,yE_n`$, $`n1`$. As before, we define the restrictions $`f_{nm}`$ for $`f_n𝖥_n`$, but now $`m`$ has to be greater than $`n`$. We suppose that $`𝖥_n`$ form an increasing sequence and the norms $`_n`$ form a decreasing sequence satisfying the following two conditions:
1. for every $`f_n𝖥_n`$ and every $`mn`$, $`f_{nm}𝖥_m`$;
2. for every $`f_n𝖥_n`$ and every $`mn`$, $`f_{nm}_mf_n_n`$.
We then get for the restrictions $`K_{nm}`$ of $`K_n`$ the formula
$$K_{nm}K_m,\text{for }mn.$$
(3.11)
For each $`yR`$, $`\{K_m(y,y)\}`$ is an increasing sequence of positive numbers. Its limit may be infinite. We define, consequently,
$$R_0:=\text{ the set of }yR\text{ such that }K_0(y,y):=\underset{m\mathrm{}}{lim}K_m(y,y)<\mathrm{}.$$
(3.12)
Suppose that $`R_0`$ is not empty and let $`𝖥_0`$ be the class of all restrictions $`f_{n0}`$ of functions $`f_n𝖥_n`$ ($`n=1,2,\mathrm{}`$) to the set $`R_0`$. From (B2), the limit $`lim_k\mathrm{}f_{nk}_k`$ exists and we define a norm $`_0^{}`$ on $`𝖥_0`$ by<sup>2</sup><sup>2</sup>2The original definition in is such that $`f_{n0}_0:=lim_k\mathrm{}f_{nk}_k`$, but it seems that there is no way to guarantee that $`f_{n0}_0=g_{n0}_0`$ for different $`f_n`$ and $`g_n`$ in $`𝖥_n`$ with $`f_{n0}=g_{n0}`$. However, all the arguments in hold true even if the new norm $`_0^{}`$ in (3.13) is used. In particular, the Theorem 3.4 below holds.
$$f_0^{}:=inf\underset{k\mathrm{}}{lim}f_{nk}_k,f𝖥_0,$$
(3.13)
where the infimum is taken over all functions $`f_n𝖥_n`$, $`n1`$, whose restrictions to $`R_0`$ are $`f`$, i.e., $`f(y)=f_{n0}`$, $`yR_0`$, for some $`f_n𝖥_n`$. Now we construct a new space $`𝖥_0^{}`$ and norm $`_0^{}`$ on it. Let $`𝖥_0^{}`$ be the class of all functions $`f_0^{}`$ on $`R_0`$ such that there is a Cauchy sequence $`\{f_0^{(n)}\}𝖥_0`$ satisfying
$$f_0^{}(x)=\underset{n\mathrm{}}{lim}f_0^{(n)}(x),\text{for all }xR_0.$$
(3.14)
For those vectors $`f_0^{}`$ we define a norm
$$f_0^{}_0^{}:=\mathrm{min}\underset{n\mathrm{}}{lim}f_0^{(n)}_0^{},$$
(3.15)
the minimum being taken over all Cauchy sequences $`\{f_0^{(n)}\}𝖥_0`$ satisfying (3.14). There exists at least one Cauchy sequence for which the minimum is attained. Such sequences are called determining $`f_0^{}`$. The scalar product corresponding to $`_0^{}`$ is defined by
$$(f_0^{},g_0^{})_0^{}:=\underset{n\mathrm{}}{lim}(f_0^{(n)},g_0^{(n)})_0^{}$$
(3.16)
for any two Cauchy sequences $`\{f_0^{(n)}\}`$ and $`\{g_0^{(n)}\}`$ determining $`f_0^{}`$ and $`g_0^{}`$, respectively. We refer to \[1, Section 9\] for the details. The following theorem is in \[1, Theorem II, Section 9\]:
###### Theorem 3.4
In the setting of the case B, the restrictions $`K_{n0}(x,y)`$ for every fixed $`yR_0`$ form a Cauchy sequence in $`𝖥_0`$. They converge to a function $`K_0^{}(x,y)𝖥_0^{}`$ which is the RK of $`𝖥_0^{}`$.
As an application of Theorem 3.4, we prove the convergence of norms in the perturbed RKHS’s, which will be used in the proof of Theorem 2.3. Let $`A`$ be the operator of our concern satisfying the conditions in the hypothesis (H). For each $`\epsilon >0`$ we define new operators as follows:
$$A(\epsilon ):=A+\epsilon \text{and}B(\epsilon ):=A(\epsilon )^1,\epsilon >0.$$
(3.17)
Let $`RE`$ be any subset of $`E`$. Following Theorem 3.2, we let $`_{R;B}`$ be the norm of the RKHS $`_{R;B}`$ consisting of all restrictions of vectors in $`_{}`$ to the set $`R`$ and having a RK $`B_R(x,y)`$, $`x,yR`$, the restriction of $`B(x,y)`$ to the set $`R`$. Similarly, $`_{R;B(\epsilon )}`$ denotes the norm defined by replacing $`B`$ with $`B(\epsilon )`$. We want to prove the convergence $`f_{R;B(\epsilon )}f_{R;B}`$ for all $`fl^2(R)`$ as $`\epsilon 0`$. See Lemma 3.7. For that purpose we proceed as follows. Let $`_{R;B}^{}l^2(R)`$ be the dual space of $`_{R;B}`$: an element $`gl^2(R)`$ belongs to $`_{R;B}^{}`$ if and only if the (anti-)linear functional
$${}_{R;B}{}^{}f,g_{R;B}^{}:=\underset{xR}{}\overline{f(x)}g(x),fl^2(R),$$
(3.18)
is continuous w.r.t. $`_{R;B}`$-norm, i.e., there exists $`M(g)>0`$ such that
$$|{}_{R;B}{}^{}f,g_{R;B}^{}|M(g)f_{R;B},\text{for all }fl^2(R).$$
(3.19)
For each $`g_{R;B}^{}`$ we extend the functional of (3.18) to the whole space $`_{R;B}l^2(R)`$ and keep the dual pairing notation $`{}_{R;B}{}^{},_{R;B}^{}`$. We denote the norm in $`_{R;B}^{}`$ by $`_{R;B}^{}`$. As in the case of the dual pairing $`{}_{}{}^{},_{+}^{}`$ we see that for any $`f_{R;B}`$, $`(B_R)^1f_{R;B}^{}`$ and
$$(B_R)^1f_{R;B}^{}=f_{R;B}.$$
(3.20)
It is not hard to show that for any $`hl^2(R)`$,
$$(B_R)^1h_+\text{ and }(B_R)^1h_+(B_R)^1h_{R;B}^{}.$$
(3.21)
In fact, we have for any $`f_0=l^2(E)`$,
$`|{}_{}{}^{}f,(B_R)^1h_{+}^{}|`$ $`=`$ $`|{\displaystyle \underset{xR}{}}\overline{f(x)}(B_R)^1h(x)|`$ (3.22)
$`=`$ $`|{}_{R;B}{}^{}f_R,(B_R)^1h_{R;B}^{}|`$
$``$ $`f_R_{R;B}(B_R)^1h_{R;B}^{}`$
$``$ $`f_{}(B_R)^1h_{R;B}^{},`$
where $`f_R`$ is the restriction of $`f`$ to $`R`$. Since $`_0`$ is dense in $`_{}`$, (3.22) proves (3.21). Because $`l^2(R)`$ is dense in $`_{R;B}`$, (3.21) also shows that
$$_{R;B}^{}_+l^2(R).$$
(3.23)
###### Lemma 3.5
Let $`g_{R;B}^{}_+l^2(R)`$. Then for any $`f_{}`$ that vanishes on $`R`$, we have $`{}_{}{}^{}f,g_{+}^{}=0`$.
Proof. Denote by $`P_R`$ the restriction operator $`P_R:_{}_{R;B}`$ defined by $`P_Rf:=f_R`$ for all $`f_{}`$. Since $`f_R_{R;B}f_{}`$, the operator $`P_R`$ is bounded with norm less than or equal to $`1`$. Now let $`g`$ and $`f`$ be as in the statement of the lemma. Let $`\{f_n\}`$ be any sequence in $`_0`$ that converges to $`f`$ in $`_{}`$. Then since $`gl^2(R)`$, by using the continuity of the operator $`P_R`$ in $`_{}`$, we have
$`{}_{}{}^{}f,g_{+}^{}`$ $`=`$ $`\underset{n\mathrm{}}{lim}{}_{}{}^{}f_n,g_{+}^{}`$
$`=`$ $`\underset{n\mathrm{}}{lim}{\displaystyle \underset{xR}{}}\overline{f_n(x)}g(x)`$
$`=`$ $`\underset{n\mathrm{}}{lim}{}_{R;B}{}^{}P_Rf_n,g_{R;B}^{}`$
$`=`$ $`{}_{R;B}{}^{}P_Rf,g_{R;B}^{}`$
$`=`$ $`0,`$
because $`P_Rf=0`$. $`\mathrm{}`$
###### Lemma 3.6
For any $`hl^2(R)`$, $`(B_R)^1h_+=(B_R)^1h_{R;B}^{}`$.
Proof. By (3.21) it is enough to show that $`(B_R)^1h_{R;B}^{}(B_R)^1h_+`$. We have
$`((B_R)^1h_{R;B}^{})^2`$ $`=`$ $`h_{R;B}^2`$ (3.24)
$`=`$ $`{}_{R;B}{}^{}h,(B_R)^1h_{R;B}^{}`$
$`=`$ $`{}_{}{}^{}h,(B_R)^1h_{+}^{}.`$
Let $`h^{}_{}`$ be the element such that $`P_Rh^{}=h`$ and $`h^{}_{}=h_{R;B}`$ (see (3.8)). Notice that $`h^{}h_{}`$ vanishes on $`R`$ and $`(B_R)^1h_{R;B}^{}`$. Thus by (3.24) and Lemma 3.5 we have
$`h_{R;B}^2`$ $`=`$ $`{}_{}{}^{}h,(B_R)^1h_{+}^{}`$
$`=`$ $`{}_{}{}^{}h^{},(B_R)^1h_{+}^{}`$
$``$ $`h^{}_{}(B_R)^1h_+`$
$`=`$ $`h_{R;B}(B_R)^1h_+.`$
This, together with (3.24), proves that $`(B_R)^1h_{R;B}^{}(B_R)^1h_+`$. $`\mathrm{}`$
Recall the definition $`A(\epsilon )=A+\epsilon `$ and $`B(\epsilon )=A(\epsilon )^1`$ for $`\epsilon >0`$.
###### Lemma 3.7
Let $`RE`$ be any set. Then for any $`fl^2(R)`$,
$$\underset{\epsilon 0}{lim}f_{R;B(\epsilon )}=f_{R;B}.$$
(3.25)
Proof. First we show that
$$\underset{\epsilon 0}{lim}B(\epsilon )(x,y)=B(x,y),\text{for all }x,yE.$$
(3.26)
It is obvious that
$$B(\epsilon )B(\epsilon ^{})\text{for }0<\epsilon ^{}<\epsilon ,$$
(3.27)
in the sense defined in (3.9). Also, it holds trivially that
$$B(\epsilon )(y,y)B(y,y)<\mathrm{},yE.$$
(3.28)
Moreover, for each fixed $`\epsilon >0`$, since $`B(\epsilon )`$ is bounded and strictly positive, the norms $`_{;\epsilon }`$ ($`:=_{E;B(\epsilon )}`$) and $`_0`$ are equivalent on $`_0=l^2(E)`$. That is, as a set, $`_{;\epsilon }`$ ($`:=_{E;B(\epsilon )}`$) is the same as $`_0`$.
Now for each $`f_0`$, the norm $`f_{;\epsilon }`$ decreases as $`\epsilon `$ decreases. It is easy to check that
$$\underset{\epsilon 0}{lim}f_{;\epsilon }=f_{}.$$
(3.29)
In fact, for $`f_0`$,
$`\underset{\epsilon 0}{lim}f_{;\epsilon }^2`$ $`=`$ $`\underset{\epsilon 0}{lim}(f,A(\epsilon )f)_0`$
$`=`$ $`(f,Af)_0`$
$`=`$ $`f_{}^2.`$
Since the norm $`_{}`$ is the one for the RKHS with kernel $`B(x,y)`$, the equality (3.26) follows from Theorem 3.4 (see the remark on \[1, p 368\]).
Let us now prove (3.25). Obviously, for each $`fl^2(R)`$, $`f_{R;B(\epsilon )}`$ decreases as $`\epsilon `$ decreases and $`f_{R;B(\epsilon )}f_{R;B}`$ for all $`\epsilon >0`$. Thus the limit
$$f_0^{}:=\underset{\epsilon 0}{lim}f_{R;B(\epsilon )},fl^2(R),$$
(3.30)
defines a norm on $`l^2(R)`$. Now we have to show $`f_0^{}=f_{R;B}`$. Considering the dual norms it is equivalent to showing that
$$\underset{\epsilon 0}{lim}g_{R;B(\epsilon )}^{}=g_{R;B}^{}$$
(3.31)
for $`gl^2(R)`$ whenever the limit is finite. Thus suppose that $`gl^2(R)`$ and $`lim_{\epsilon 0}g_{R;B(\epsilon )}^{}`$ is finite. Since $`B(\epsilon )`$ is a strictly positive and bounded operator, we see that
$`(g_{R;B(\epsilon )}^{})^2`$ $`=`$ $`(g,B(\epsilon )_Rg)_0`$ (3.32)
$`=`$ $`(g,{\displaystyle \frac{B}{I+\epsilon B}}g)_0.`$
Now consider the form $``$ on $`_0`$ generated by the operator $`B`$:
$$(f,g):=(f,Bg)_0,f,g\text{ran}A.$$
(3.33)
Then the space $`_+`$ is nothing but the closure of $`\text{ran}A`$ w.r.t. this form norm. We denote the closure of the form $`(,\text{ran}A)`$ by $`(,D())`$. Using this notation, the last quantity in (3.32) becomes $`(\frac{1}{\sqrt{I+\epsilon B}}g,\frac{1}{\sqrt{I+\epsilon B}}g)`$. By the assumption, these values are bounded from above (as $`\epsilon `$ varies). On the other hand, $`\frac{1}{\sqrt{I+\epsilon B}}g`$ converges strongly to $`g`$ as $`\epsilon `$ goes to $`0`$. By \[9, Lemma 2.12\], we conclude that $`gD()`$, i.e., $`g_+`$, and
$`g_+^2=(g,g)`$ $``$ $`\underset{\epsilon 0}{lim\; inf}({\displaystyle \frac{1}{\sqrt{I+\epsilon B}}}g,{\displaystyle \frac{1}{\sqrt{I+\epsilon B}}}g)`$ (3.34)
$`=`$ $`\underset{\epsilon 0}{lim}(g_{R;B(\epsilon )})^2.`$
To finish the proof we notice that the space $`(B_R)^1(l^2(R))`$ is dense in $`_{R;B}^{}`$ because $`l^2(R)`$ is dense in $`_{R;B}`$ and the map $`(B_R)^1:(l^2(R),_{R;B})_{R;B}^{}`$ is an isometry. Therefore, it is enough to check (3.31) for those vectors $`g`$ of the form $`g=(B_R)^1h`$ for some $`hl^2(R)`$. We use Lemma 3.6 and (3.34) inserting $`(B_R)^1h`$ for $`g`$. Then we get
$$(B_R)^1h_{R;B}^{}=(B_R)^1h_+\underset{\epsilon 0}{lim}(B_R)^1h_{R;B(\epsilon )}^{}(B_R)^1h_{R;B}^{}.$$
(3.35)
The last inequality in the above comes from the fact that $`g_{R;B(\epsilon )}^{}g_{R;B}^{}`$ for all $`g_{R;B}^{}`$. Eq. (3.35) says that all the quantities there are equal to each other, and we have proven (3.31). $`\mathrm{}`$
### 3.3 Proof of Theorem 2.3
We are now ready to prove Theorem 2.3. It will be done in several steps.
Proof of Theorem 2.3. Step 1: General facts. Recall the notation $`𝖥_{\text{loc},\mathrm{\Lambda }}`$, the class of local functions supported on $`\mathrm{\Lambda }`$ for the subsets $`\mathrm{\Lambda }E`$. Since these spaces are closed in $`_{}`$, for each $`\mathrm{\Lambda }E`$ there exists a unique element $`f_{\mathrm{\Lambda }_1,0}𝖥_{\text{loc},\mathrm{\Lambda }_1}`$ such that
$$\alpha _\mathrm{\Lambda }=\underset{f𝖥_{\text{loc},\mathrm{\Lambda }_1}}{inf}e_{x_0}f_{}^2=e_{x_0}f_{\mathrm{\Lambda }_1,0}_{}^2,$$
(3.36)
where $`\mathrm{\Lambda }_1:=\mathrm{\Lambda }R_1`$. Let $`_{1,}`$ be the closure (in $`_{}`$) of $`_{\mathrm{\Lambda }E}𝖥_{\text{loc},\mathrm{\Lambda }_1}`$. Notice that any vector $`f_{1,}`$ vanishes on $`R_1^c`$, that is, it is supported on $`R_1`$. In fact, since for each $`\mathrm{\Lambda }E`$, $`𝖥_{\text{loc},\mathrm{\Lambda }_1}𝖥_{R_1^c}^0`$, the space of functions that vanish on $`R_1^c`$, and $`𝖥_{R_1^c}^0`$ is closed, we have $`_{1,}=\overline{_{\mathrm{\Lambda }E}𝖥_{\text{loc},\mathrm{\Lambda }_1}}𝖥_{R_1^c}^0`$. We also notice that for each $`\mathrm{\Lambda }E`$, $`f_{\mathrm{\Lambda }_1,0}`$ is the projection of $`e_{x_0}`$ onto the space $`𝖥_{\text{loc},\mathrm{\Lambda }_1}=𝖥_{\mathrm{\Lambda }_1^c}^0`$, and as $`\mathrm{\Lambda }`$ increases, $`f_{\mathrm{\Lambda }_1,0}`$ converges to the projection of $`e_{x_0}`$ onto the space $`_{1,}`$, we call it $`f_{R_1,0}`$:
$$\underset{\mathrm{\Lambda }E}{lim}f_{\mathrm{\Lambda }_1,0}=f_{R_1,0}(\text{in }_{}).$$
(3.37)
Let us now apply the (extended) operator $`A`$ to the vector $`e_{x_0}f_{R_1,0}`$. We claim that
$$A(e_{x_0}f_{R_1,0})=\alpha e_{x_0}+a_2_+,$$
(3.38)
where the vector $`a_2_+`$ is supported on $`R_2`$. In fact, let $`A(e_{x_0}f_{R_1,0})=a_0e_{x_0}+a_2_+`$ with $`a_2`$ being supported on $`E\{x_0\}`$. Since $`f_{R_1,0}`$ is the projection of $`e_{x_0}`$ onto the space $`_{1,}`$, we have
$$(e_{x_0}f_{R_1,0},f)_{}=0\text{for all }f𝖥_{\text{loc},R_1}.$$
(3.39)
Thus, we have for all $`f𝖥_{\text{loc},R_1}`$,
$`0`$ $`=`$ $`(e_{x_0}f_{R_1,0},f)_{}`$ (3.40)
$`=`$ $`{}_{+}{}^{}A(e_{x_0}f_{R_1,0}),f_{}^{}`$
$`=`$ $`{}_{+}{}^{}a_0e_{x_0}+a_2,f)_{}`$
$`=`$ $`{\displaystyle \underset{xR_1}{}}\overline{a_2(x)}f(x),`$
because $`f`$ is a local function supported on $`R_1`$ (see (2.14)). Since $`f𝖥_{\text{loc},R_1}`$ is arbitrary, the equation (3.40) proves that $`a_2`$ vanishes on $`R_1`$. Similarly, it is easily checked that
$`\alpha `$ $`=`$ $`e_{x_0}f_{R_1,0}_{}^2`$ (3.41)
$`=`$ $`{}_{}{}^{}e_{x_0}f_{R_1,0},A(e_{x_0}f_{R_1,0})_{+}^{}`$
$`=`$ $`\underset{\mathrm{\Lambda }E}{lim}{}_{}{}^{}e_{x_0}f_{\mathrm{\Lambda }_1,0},a_0e_{x_0}+a_2_{+}^{}`$
$`=`$ $`\underset{\mathrm{\Lambda }E}{lim}a_0=a_0.`$
We have shown (3.38). Let us now interchange the roles of $`A`$, $`_{}`$, $`R_1`$, and $`\mathrm{\Lambda }_1`$ by $`A^1`$, $`_+`$, $`R_2`$, and $`\mathrm{\Lambda }_2`$, respectively. Then we have for each $`\mathrm{\Lambda }E`$,
$$\beta _\mathrm{\Lambda }=\underset{g𝖥_{\text{loc},\mathrm{\Lambda }_2}}{inf}e_{x_0}g_+^2=e_{x_0}g_{\mathrm{\Lambda }_2,0}_+^2,$$
(3.42)
for a unique $`g_{\mathrm{\Lambda }_2,0}𝖥_{\text{loc},\mathrm{\Lambda }_2}`$, where $`\mathrm{\Lambda }_2:=R_2\mathrm{\Lambda }`$. Also, if we denote by $`_{2,+}`$ the closure of $`_{\mathrm{\Lambda }E}𝖥_{\text{loc},\mathrm{\Lambda }_2}`$ w.r.t. the $`_+`$-norm, there is a unique $`g_{R_2,0}_{2,+}`$ such that
$$\underset{\mathrm{\Lambda }E}{lim}g_{\mathrm{\Lambda }_2,0}=g_{R_2,0}(\text{in }_+)$$
(3.43)
and
$$\beta =e_{x_0}g_{R_2,0}_+^2.$$
(3.44)
Similarly to (3.39), we have
$$A^1(e_{x_0}g_{R_2,0})=\beta e_{x_0}+b_1_{},$$
(3.45)
where $`b_1`$ is supported on $`R_1`$. Now we have on the one hand
$`{}_{}{}^{}e_{x_0}f_{R_1,0},e_{x_0}g_{R_2,0}_{+}^{}`$ $`=`$ $`\underset{\mathrm{\Lambda }E}{lim}{}_{}{}^{}e_{x_0}f_{\mathrm{\Lambda }_1,0},e_{x_0}g_{\mathrm{\Lambda }_2,0}_{+}^{}`$
$`=`$ $`\underset{\mathrm{\Lambda }E}{lim}(e_{x_0}f_{\mathrm{\Lambda }_1,0},e_{x_0}g_{\mathrm{\Lambda }_2,0})_0`$
$`=`$ $`\underset{\mathrm{\Lambda }E}{lim}1=1.`$
On the other hand we have
$`1`$ $`=`$ $`{}_{}{}^{}e_{x_0}f_{R_1,0},e_{x_0}g_{R_2,0}_{+}^{}`$ (3.46)
$`=`$ $`{}_{+}{}^{}A(e_{x_0}f_{R_1,0}),A^1(e_{x_0}g_{R_2,0})_{}^{}`$
$`=`$ $`{}_{+}{}^{}\alpha e_{x_0}+a_2,\beta e_{x_0}+b_1_{}^{}`$
$`=`$ $`\alpha \beta +{}_{+}{}^{}a_2,b_1_{}^{}.`$
The proof is completed if we could show that $`{}_{+}{}^{}a_2,b_1_{}^{}=0`$. Notice that $`a_2`$ is supported on $`R_2`$ and $`b_1`$ on $`R_1`$, and $`R_1R_2=\mathrm{}`$. Thus it seems that $`{}_{+}{}^{}a_2,b_1_{}^{}=0`$, but we need to confirm it.
Step 2: The case when $`A`$ is strictly positive. Suppose that there exist $`c_1,c_2>0`$ such that $`c_1IAc_2I`$. In this case $`A`$ has a bounded inverse $`A^1`$ in $`_0=l^2(E)`$. The RK $`B(x,y)`$ for $`_{}`$ (see (2.19)) is given by
$$B(x,y)={}_{+}{}^{}e_x,A^1e_y_{}^{}=(e_x,A^1e_y)_0,x,yE.$$
(3.47)
Moreover, as for the elements, the inclusions in (2.7) now become the equalities and the dual pairings in (2.8) and (2.10) are just the inner product in the center space $`_0`$:
$${}_{}{}^{}f,g_{+}^{}=(f,g)_0=\overline{(g,f)_0}=\overline{{}_{+}{}^{}g,f_{}^{}},f,g_0=_{}=_+,$$
(3.48)
the equalities $`_0=_{}=_+`$ meaning that all the spaces have the same elements. We will, however, keep the pairing notations $`{}_{}{}^{},_{+}^{}`$ and $`{}_{+}{}^{},_{}^{}`$ for a convenience. Now let us come back to the equation (3.46). The dual pairing is just an inner product in $`_0`$ and the vector $`a_2`$ vanishes on $`R_1`$ and $`b_1`$ lives only on $`R_1`$. We therefore have
$${}_{+}{}^{}a_2,b_1_{}^{}=(a_2,b_1)_0=0.$$
(3.49)
From (3.46) and (3.49) we have $`\alpha \beta =1`$.
We now extend the formula in Proposition 3.1(b) to the infinite system. That is, we will show that if $`A`$ is strictly positive, then
$$\alpha =A(x_0,x_0)A(x_0,R_1)A(R_1,R_1)^1A(R_1,x_0).$$
(3.50)
Notice that the function $`A(,x_0)`$ is an element of the space $`_+`$ and for each $`\mathrm{\Delta }E`$ the function $`\mathrm{\Delta }yA(y,x_0)`$, which we denote by $`A(\mathrm{\Delta },x_0)`$, is the restriction of $`A(,x_0)`$ to the set $`\mathrm{\Delta }`$. Following Theorem 3.2, we denote this space by $`_{\mathrm{\Delta };A}`$ equipped with the norm $`_{\mathrm{\Delta };A}`$. Since $`A_\mathrm{\Delta }(x,y)`$, $`x,y\mathrm{\Delta }`$, is the RK for $`_{\mathrm{\Delta };A}`$, it is obvious that for each $`g_{\mathrm{\Delta };A}`$,
$$g_{\mathrm{\Delta };A}^2=(g,A(\mathrm{\Delta },\mathrm{\Delta })^1g)_0,$$
(3.51)
where $`(,)_0`$ is the usual inner product in $`l^2(\mathrm{\Delta })`$. Thus (3.50) is equivalent to saying that
$$\alpha =A(x_0,x_0)A(R_1,x_0)_{R_1;A}^2.$$
(3.52)
Now by Proposition 3.1(b) we see that for each $`\mathrm{\Lambda }E`$, putting $`\mathrm{\Lambda }_1=R_1\mathrm{\Lambda }`$,
$`\alpha _\mathrm{\Lambda }`$ $`=`$ $`A(x_0,x_0)A(x_0,\mathrm{\Lambda }_1)A(\mathrm{\Lambda }_1,\mathrm{\Lambda }_1)^1A(\mathrm{\Lambda }_1,x_0)`$ (3.53)
$`=`$ $`A(x_0,x_0)A(\mathrm{\Lambda }_1,x_0)_{\mathrm{\Lambda }_1;A}^2.`$
On the other hand, as $`\mathrm{\Lambda }`$ increases, we have by Theorem 3.3,
$$\underset{\mathrm{\Lambda }E}{lim}A(\mathrm{\Lambda }_1,x_0)_{\mathrm{\Lambda }_1;A}^2=A(R_1,x_0)_{R_1;A}^2.$$
(3.54)
From (2.24) and (3.52)-(3.54) we have shown (3.50).
Step 3: The case when one of $`R_1`$ and $`R_2`$ is finite. In this case either $`a_2`$ in (3.38) or $`b_1`$ in (3.45) is finitely supported. Moreover, since they have disjoint supports, by (2.14) we have
$${}_{+}{}^{}a_2,b_1_{}^{}=\underset{xE}{}\overline{a_2(x)}b_1(x)=0.$$
(3.55)
This, together with (3.46), proves the theorem. This observation, as a matter of fact, gives us more information. Notice that the number $`\beta `$ in (2.24) is not altered even if we considered the restriction of $`B`$ to the set $`\stackrel{~}{R_2}:=\{x_0\}R_2`$. Recall the notation $`_{\stackrel{~}{R_2};B}`$ for the norm in the RKHS $`_{\stackrel{~}{R_2};B}`$ consisting of all the restrictions of vectors in $`_{}`$ to the set $`\stackrel{~}{R_2}`$. $`_{\stackrel{~}{R_2};B}`$ has its RK $`B_{\stackrel{~}{R_2}}(x,y)`$, $`x,y\stackrel{~}{R_2}`$, the restriction of $`B`$ onto $`\stackrel{~}{R_2}`$. We consider $`\stackrel{~}{R_2}`$ being partitioned as $`\stackrel{~}{R_2}=\{e_{x_0}\}\mathrm{}R_2`$, and then apply the result in this step to get
$$\beta ^1=e_{x_0}_{\stackrel{~}{R_2};B}^2.$$
(3.56)
In passing, we note that $`e_{x_0}_{\stackrel{~}{R_2};B}^2=(e_{x_0},(B_{\stackrel{~}{R_2}})^1e_{x_0})_0`$, where $`(B_{\stackrel{~}{R_2}})^1`$ is the “inverse”of $`B_{\stackrel{~}{R_2}}`$ having the components
$$(B_{\stackrel{~}{R_2}})^1(x,y)=(e_x,e_y)_{\stackrel{~}{R_2};B},x,y\stackrel{~}{R_2}.$$
(3.57)
For each $`\epsilon >0`$, we introduce the strictly positive and bounded operators $`A(\epsilon ):=A+\epsilon `$ and $`B(\epsilon ):=A(\epsilon )^1`$ on $`_0`$. Let $`\alpha (\epsilon )`$ and $`\beta (\epsilon )`$ be the numbers defined as in (2.24) by replacing the operators $`A`$ and $`B`$ with $`A(\epsilon )`$ and $`B(\epsilon )`$, respectively. By the result in Step 2, we have
$$\alpha (\epsilon )\beta (\epsilon )=1,\epsilon >0.$$
(3.58)
On the other hand, by (3.56) we have
$$\beta (\epsilon )^1=e_{x_0}_{\stackrel{~}{R_2};B(\epsilon )}^2:=(e_{x_0},(B(\epsilon )_{\stackrel{~}{R_2}})^1e_{x_0})_0.$$
(3.59)
In Lemma 3.7, we have shown that
$$\underset{\epsilon 0}{lim}e_{x_0}_{\stackrel{~}{R_2};B(\epsilon )}^2=e_{x_0}_{\stackrel{~}{R_2};B}^2,$$
(3.60)
that is
$$\underset{\epsilon 0}{lim}\beta (\epsilon )^1=\beta ^1.$$
(3.61)
It is easy to check that
$$\underset{\epsilon 0}{lim}\alpha (\epsilon )=\alpha .$$
(3.62)
We thus get by (3.58), (3.61)-(3.62), $`\alpha \beta =1`$. The proof is completed. $`\mathrm{}`$
## 4 Proofs of Theorem 2.4 and Theorem 2.5
The proof of Theorem 2.4 will follow from the variational principle of Theorem 2.3 and the projection-inversion inequalities, which we now introduce. For a matrix $`A`$ on $`E`$, we denote by $`A_\mathrm{\Lambda }`$ for the submatrix, or projection of $`A`$ on the set $`\mathrm{\Lambda }E`$.
###### Lemma 4.1
Let $`A(x,y)`$ and $`B(x,y)`$, $`x,yE`$, be the RK’s respectively for $`_+`$ and $`_{}`$ in Section 2. Then, for any finite subsets $`\mathrm{\Lambda }\mathrm{\Delta }E`$, the following inequalities hold:
1. $`(A_\mathrm{\Lambda })^1((A_\mathrm{\Delta })^1)_\mathrm{\Lambda }B_\mathrm{\Lambda }`$;
2. $`(B_\mathrm{\Lambda })^1((B_\mathrm{\Delta })^1)_\mathrm{\Lambda }A_\mathrm{\Lambda }`$.
For a proof we need the projection-inversion lemma (see \[11, p 18\], \[16, Corollary 5.3\], and \[4, Lemma A.5\]):
###### Lemma 4.2
Let $`T`$ be any bounded positive definite operator with bounded inverse $`T^1`$. Then for any projection $`P`$,
$$P(PTP)^1PPT^1P.$$
(4.1)
Proof of Lemma 4.1. The first inequalities in (a) and (b) follow from Lemma 4.2. In order to prove the second inequalities it is enough to show $`(A_\mathrm{\Lambda })^1B_\mathrm{\Lambda }`$ and $`(B_\mathrm{\Lambda })^1A_\mathrm{\Lambda }`$, because $`(B_\mathrm{\Delta })_\mathrm{\Lambda }=B_\mathrm{\Lambda }`$ and $`(A_\mathrm{\Delta })_\mathrm{\Lambda }=A_\mathrm{\Lambda }`$ for $`\mathrm{\Lambda }\mathrm{\Delta }E`$. Moreover, since the matrices are positive definite, either one of the inequalities $`(A_\mathrm{\Lambda })^1B_\mathrm{\Lambda }`$ or $`(B_\mathrm{\Lambda })^1A_\mathrm{\Lambda }`$ implies the other. So, it is enough to prove $`(B_\mathrm{\Lambda })^1A_\mathrm{\Lambda }`$. Let $`_{\mathrm{\Lambda };B}`$ be the norm on the space $`_{\mathrm{\Lambda };B}`$ of all restrictions of functions of $`_{}`$ to the subset $`\mathrm{\Lambda }`$ given in Theorem 3.2. Since the function $`B_\mathrm{\Lambda }(x,y)`$, $`x,y\mathrm{\Lambda }`$, is the corresponding RK for $`_{\mathrm{\Lambda };B}`$, it is obvious that
$$f_\mathrm{\Lambda }_{\mathrm{\Lambda };B}^2=(f_\mathrm{\Lambda },(B_\mathrm{\Lambda })^1f_\mathrm{\Lambda })_0.$$
(4.2)
On the other hand, since $`f_\mathrm{\Lambda }_{\mathrm{\Lambda };B}`$ is the smallest number for all the values $`g_{}`$ such that $`g_\mathrm{\Lambda }=f_\mathrm{\Lambda }`$, we have the inequality
$$f_\mathrm{\Lambda }_{\mathrm{\Lambda };B}^2f_\mathrm{\Lambda }_{}^2=(f_\mathrm{\Lambda },A_\mathrm{\Lambda }f_\mathrm{\Lambda })_0.$$
(4.3)
Combining (4.2) and (4.3) we get the inequality $`(B_\mathrm{\Lambda })^1A_\mathrm{\Lambda }`$, and the proof is completed. $`\mathrm{}`$
###### Remark 4.3
Notice that $`B`$ is formally the inverse of $`A`$ (see (2.19)), and that the operator $`A`$ may have $`0`$ in its spectrum (it should then be a continuous spectrum). In that case, the operator $`B`$, considered on the space $`_0`$, is an unbounded operator. Therefore, Lemma 4.1 extends Lemma 4.2.
Next we discuss the order relations between the restriction operators and the interaction operators giving the local probability densities of DPP’s. For each finite set $`\mathrm{\Lambda }E`$, we let, as before,
$$A_\mathrm{\Lambda }:=P_\mathrm{\Lambda }AP_\mathrm{\Lambda }\text{ and }A_{[\mathrm{\Lambda }]}:=K_\mathrm{\Lambda }(IK_\mathrm{\Lambda })^1,$$
(4.4)
where $`P_\mathrm{\Lambda }`$ is the projection on $`_0=l^2(E)`$ onto $`l^2(\mathrm{\Lambda })`$ and $`K:=A(I+A)^1`$. We let
$$B_{[\mathrm{\Lambda }]}:=(A_{[\mathrm{\Lambda }]})^1$$
(4.5)
and recall that $`B`$ is the inverse of $`A`$.
###### Lemma 4.4
For any finite set $`\mathrm{\Lambda }E`$,
$$A_{[\mathrm{\Lambda }]}A_\mathrm{\Lambda }\text{and}B_{[\mathrm{\Lambda }]}B_\mathrm{\Lambda }.$$
(4.6)
Proof. We first prove the inequality $`A_{[\mathrm{\Lambda }]}A_\mathrm{\Lambda }`$. By Lemma 4.2,
$`A_{[\mathrm{\Lambda }]}`$ $`=`$ $`I_\mathrm{\Lambda }+((IK)_\mathrm{\Lambda })^1`$ (4.7)
$``$ $`I_\mathrm{\Lambda }+P_\mathrm{\Lambda }(IK)^1P_\mathrm{\Lambda }`$
$`=`$ $`A_\mathrm{\Lambda },`$
where $`I_\mathrm{\Lambda }:=P_\mathrm{\Lambda }IP_\mathrm{\Lambda }`$. The second inequality in (4.6) can be shown in two ways. We introduce both of them. First, as before, we define $`A(\epsilon )=A+\epsilon `$ and $`B(\epsilon )=A(\epsilon )^1`$ for $`\epsilon >0`$. By the same way used in (4.7) we can show
$$B(\epsilon )_{[\mathrm{\Lambda }]}:=(A(\epsilon )_{[\mathrm{\Lambda }]})^1B(\epsilon )_\mathrm{\Lambda },$$
(4.8)
where $`A(\epsilon )_{[\mathrm{\Lambda }]}:=K(\epsilon )_\mathrm{\Lambda }(IK(\epsilon )_\mathrm{\Lambda })^1`$ with $`K(\epsilon ):=A(\epsilon )(I+A(\epsilon ))^1`$. Since $`K(\epsilon )K`$ uniformly as $`\epsilon 0`$ we have
$$\underset{\epsilon 0}{lim}B(\epsilon )_{[\mathrm{\Lambda }]}=(A_{[\mathrm{\Lambda }]})^1=B_{[\mathrm{\Lambda }]}.$$
(4.9)
On the other hand, by (3.26)
$$B(\epsilon )_\mathrm{\Lambda }B_\mathrm{\Lambda }\text{uniformly as }\epsilon 0.$$
(4.10)
The inequality $`B_{[\mathrm{\Lambda }]}B_\mathrm{\Lambda }`$ follows from (4.8)-(4.10).
The second way is to use Lemma 4.1. $`B_{[\mathrm{\Lambda }]}`$ can be rewritten as $`B_{[\mathrm{\Lambda }]}=I_\mathrm{\Lambda }+(K_\mathrm{\Lambda })^1`$. Since $`K=A(I+A)^1`$, $`K`$ satisfies the conditions in the hypothesis (H) of Section 2. Applying Lemma 4.1(a) for the pair of operators $`K`$ and $`K^1`$, we have
$`B_{[\mathrm{\Lambda }]}`$ $``$ $`I_\mathrm{\Lambda }+P_\mathrm{\Lambda }K^1P_\mathrm{\Lambda }`$
$`=`$ $`P_\mathrm{\Lambda }(I+K^1)P_\mathrm{\Lambda }`$
$`=`$ $`P_\mathrm{\Lambda }A^1P_\mathrm{\Lambda }=B_\mathrm{\Lambda }.`$
The proof is completed. $`\mathrm{}`$
We are now ready to prove Theorem 2.4.
Proof of Theorem 2.4. Let $`x_0E`$ and $`\xi 𝒳`$ be any configuration with $`x_0\xi `$. For a convenience we define an auxiliary configuration $`\overline{\xi }𝒳`$ as
$$\overline{\xi }:=E(\xi \{x_0\}).$$
(4.11)
From the definition (2.26) and Proposition 3.1(b) we have the equality:
$$\alpha _{[\mathrm{\Lambda }]}=(A_{[\mathrm{\Lambda }]}(x_0\xi _\mathrm{\Lambda },x_0\xi _\mathrm{\Lambda })^1(x_0,x_0))^1.$$
(4.12)
By the first inequality in Lemma 4.4 and using Proposition 3.1 once more we have the bound
$$\alpha _{[\mathrm{\Lambda }]}(A_\mathrm{\Lambda }(x_0\xi _\mathrm{\Lambda },x_0\xi _\mathrm{\Lambda })^1(x_0,x_0))^1=\alpha _\mathrm{\Lambda },$$
(4.13)
where $`\alpha _\mathrm{\Lambda }`$ is defined in (2.22) with $`R_1:=\xi `$ (and $`\mathrm{\Lambda }_1=\mathrm{\Lambda }R_1=\mathrm{\Lambda }\xi \xi _\mathrm{\Lambda }`$). Now by Proposition 3.1(b) and (c) we have
$$\beta _{[\mathrm{\Lambda }]}:=(a_{[\mathrm{\Lambda }]})^1=(B_{[\mathrm{\Lambda }]}(x_0\overline{\xi }_\mathrm{\Lambda },x_0\overline{\xi }_\mathrm{\Lambda })^1(x_0,x_0))^1,$$
(4.14)
where $`B_{[\mathrm{\Lambda }]}=(A_{[\mathrm{\Lambda }]})^1`$. By the second inequality of Lemma 4.4 we also have the bound
$$\beta _{[\mathrm{\Lambda }]}\beta _\mathrm{\Lambda },$$
(4.15)
where, again, $`\beta _\mathrm{\Lambda }`$ is defined in (2.23) with $`R_2:=\overline{\xi }`$. Now we take the limit of $`\mathrm{\Lambda }`$ increasing to the whole space $`E`$. Since $`\alpha _\mathrm{\Lambda }\alpha `$ as $`\mathrm{\Lambda }`$ increases to $`E`$ we have from (4.13)
$$\underset{\mathrm{\Lambda }E}{lim\; sup}\alpha _{[\mathrm{\Lambda }]}\alpha .$$
(4.16)
On the other hand, since $`\beta _\mathrm{\Lambda }\beta `$ as $`\mathrm{\Lambda }E`$, we have also from (4.15)
$$\underset{\mathrm{\Lambda }E}{lim\; inf}\alpha _{[\mathrm{\Lambda }]}=(\underset{\mathrm{\Lambda }E}{lim\; sup}\beta _{[\mathrm{\Lambda }]})^1\beta ^1=\alpha .$$
(4.17)
The last equality comes from Theorem 2.3. From (4.16) and (4.17) we get $`lim_{\mathrm{\Lambda }E}\alpha _{[\mathrm{\Lambda }]}=\alpha `$, which was to be shown. $`\mathrm{}`$
Let us now turn to the proof of Theorem 2.5. For the proof of Gibbsianness we will follow the method developed in for continuum models. We will first define a Gibbsian specification by introducing an interaction. Then we will prove that the DPP of our concern is admitted to the specification. We refer also to \[16, Section 6\]. The proof of uniqueness will be shown by following the method of .
Let $`A`$ be an operator that satisfies the conditions in the hypothesis (H). For any finite configuration $`\xi 𝒳`$, we define an interaction potential of the particles in $`\xi `$ by
$$V(\xi ):=\mathrm{log}detA(\xi ,\xi ).$$
(4.18)
Notice that $`V(\xi )>0`$ for all finite configurations $`\xi 𝒳`$. For any $`\mathrm{\Lambda }_1,\mathrm{\Lambda }_2E`$ with $`\mathrm{\Lambda }_1\mathrm{\Lambda }_2=\mathrm{}`$, and for any configurations $`\xi _{\mathrm{\Lambda }_1}`$ and $`\xi _{\mathrm{\Lambda }_2}`$ on the sets $`\mathrm{\Lambda }_1`$ and $`\mathrm{\Lambda }_2`$, respectively, the mutual potential energy $`W(\xi _{\mathrm{\Lambda }_1};\xi _{\mathrm{\Lambda }_2})`$ is defined to satisfy
$$V(\xi _{\mathrm{\Lambda }_1}\xi _{\mathrm{\Lambda }_2})=V(\xi _{\mathrm{\Lambda }_1})+V(\xi _{\mathrm{\Lambda }_2})+W(\xi _{\mathrm{\Lambda }_1};\xi _{\mathrm{\Lambda }_2}).$$
(4.19)
Now for each $`\zeta _\mathrm{\Lambda }𝒳_\mathrm{\Lambda }`$ and $`\xi 𝒳`$, we define the energy of the particle configuration $`\zeta _\mathrm{\Lambda }`$ on $`\mathrm{\Lambda }`$ with boundary condition $`\xi `$ by
$$H_\mathrm{\Lambda }(\zeta _\mathrm{\Lambda };\xi ):=\underset{\mathrm{\Delta }E}{lim}(V(\zeta _\mathrm{\Lambda })+W(\zeta _\mathrm{\Lambda };\xi _{\mathrm{\Delta }\mathrm{\Lambda }})),$$
(4.20)
whenever the limit exists. As a matter of fact, $`H_\mathrm{\Lambda }(\zeta _\mathrm{\Lambda };\xi )`$ is well-defined for all $`\zeta _\mathrm{\Lambda }𝒳_\mathrm{\Lambda }`$ and $`\xi 𝒳`$ as shown in the following lemma:
###### Lemma 4.5
Suppose that the operator $`A`$ satisfies the conditions in the hypothesis (H). Then for any $`\zeta _\mathrm{\Lambda }𝒳_\mathrm{\Lambda }`$ and $`\xi 𝒳`$, the value $`H_\mathrm{\Lambda }(\zeta _\mathrm{\Lambda };\xi )`$ in (4.20) is well-defined as a finite number.
Proof. The proof is very similar to the one given for continuum model in \[21, Lemma 3.2\]. We define first for each bounded set $`\mathrm{\Delta }\mathrm{\Lambda }`$
$$H_{\mathrm{\Lambda };\mathrm{\Delta }}(\zeta _\mathrm{\Lambda };\xi ):=V(\zeta _\mathrm{\Lambda })+W(\zeta _\mathrm{\Lambda };\zeta _{\mathrm{\Delta }\mathrm{\Lambda }}).$$
(4.21)
From the definitions (4.18)-(4.19) we get
$$H_{\mathrm{\Lambda };\mathrm{\Delta }}(\zeta _\mathrm{\Lambda };\xi )=\mathrm{log}\frac{detA(\zeta _\mathrm{\Lambda }\xi _{\mathrm{\Delta }\mathrm{\Lambda }},\zeta _\mathrm{\Lambda }\xi _{\mathrm{\Delta }\mathrm{\Lambda }})}{detA(\xi _{\mathrm{\Delta }\mathrm{\Lambda }},\xi _{\mathrm{\Delta }\mathrm{\Lambda }})}.$$
(4.22)
Denoting $`Q_\mathrm{\Lambda }`$ for the projection on $`l^2(\zeta _\mathrm{\Lambda }\xi _{\mathrm{\Lambda }^c})`$ onto $`l^2(\zeta _\mathrm{\Lambda })`$, $`H_{\mathrm{\Lambda };\mathrm{\Delta }}(\zeta _\mathrm{\Lambda };\xi )`$ can be rewritten as (cf. Projection 3.1)
$$H_{\mathrm{\Lambda };\mathrm{\Delta }}(\zeta _\mathrm{\Lambda };\xi )=\mathrm{log}det(Q_\mathrm{\Lambda }A(\zeta _\mathrm{\Lambda }\xi _{\mathrm{\Delta }\mathrm{\Lambda }},\zeta _\mathrm{\Lambda }\xi _{\mathrm{\Delta }\mathrm{\Lambda }})^1Q_\mathrm{\Lambda })^1.$$
(4.23)
By using the projection-inversion lemma, Lemma 4.2, we see that $`H_{\mathrm{\Lambda };\mathrm{\Delta }}(\zeta _\mathrm{\Lambda };\xi )`$ decreases as $`\mathrm{\Delta }`$ increases. Hence the limit
$$H_\mathrm{\Lambda }(\zeta _\mathrm{\Lambda };\xi )=\underset{\mathrm{\Delta }E}{lim}H_{\mathrm{\Lambda };\mathrm{\Delta }}(\zeta _\mathrm{\Lambda };\xi )$$
(4.24)
exists. We now show the finiteness of the limit value. Let $`\zeta _\mathrm{\Lambda }=\{x_1,\mathrm{},x_n\}`$ be an enumeration of the sites in $`\zeta _\mathrm{\Lambda }`$. Then we can rewrite the quantity inside the logarithm in (4.22) as
$`{\displaystyle \frac{detA(\zeta _\mathrm{\Lambda }\xi _{\mathrm{\Delta }\mathrm{\Lambda }},\zeta _\mathrm{\Lambda }\xi _{\mathrm{\Delta }\mathrm{\Lambda }})}{detA(\xi _{\mathrm{\Delta }\mathrm{\Lambda }},\xi _{\mathrm{\Delta }\mathrm{\Lambda }})}}`$ (4.25)
$`=`$ $`{\displaystyle \frac{detA(x_1,\mathrm{},x_n\xi _{\mathrm{\Delta }\mathrm{\Lambda }},x_1,\mathrm{},x_n\xi _{\mathrm{\Delta }\mathrm{\Lambda }})}{detA(x_2,\mathrm{},x_n\xi _{\mathrm{\Delta }\mathrm{\Lambda }},x_2,\mathrm{},x_n\xi _{\mathrm{\Delta }\mathrm{\Lambda }})}}\mathrm{}{\displaystyle \frac{detA(x_n\xi _{\mathrm{\Delta }\mathrm{\Lambda }},x_n\xi _{\mathrm{\Delta }\mathrm{\Lambda }})}{detA(\xi _{\mathrm{\Delta }\mathrm{\Lambda }},\xi _{\mathrm{\Delta }\mathrm{\Lambda }})}}.`$
By Theorem 2.3, each term in the r.h.s. converges to a strictly positive number as $`\mathrm{\Delta }`$ increases to $`E`$. The proof is complete. $`\mathrm{}`$
The finiteness of the values $`H_\mathrm{\Lambda }(\zeta _\mathrm{\Lambda };\xi )`$ for all $`\zeta _\mathrm{\Lambda }𝒳_\mathrm{\Lambda }`$ and $`\xi 𝒳`$ says that any configuration $`\xi 𝒳`$ is “physically possible”as noted in \[12, p 16\].
Let us now define the Gibbsian specification. Define a partition function on the set $`\mathrm{\Lambda }`$ with a boundary condition $`\xi 𝒳`$ as
$$Z_\mathrm{\Lambda }(\xi ):=\underset{\zeta _\mathrm{\Lambda }\mathrm{\Lambda }}{}\mathrm{exp}[H_\mathrm{\Lambda }(\zeta _\mathrm{\Lambda };\xi )].$$
(4.26)
Then we define a probability distribution on the particle configurations as
$$\gamma _\mathrm{\Lambda }(\zeta _\mathrm{\Lambda };\xi ):=\frac{1}{Z_\mathrm{\Lambda }(\xi )}\mathrm{exp}[H_\mathrm{\Lambda }(\zeta _\mathrm{\Lambda };\xi )].$$
(4.27)
Let the set $`\{0,1\}`$ be equipped with a discrete topology and $`\mathrm{\Omega }:=\{0,1\}^E`$ with a product topology. Let $``$ be the Borel $`\sigma `$-algebra on $`\mathrm{\Omega }`$. For any subset $`\mathrm{\Delta }E`$ we let $`_\mathrm{\Delta }`$ be the $`\sigma `$-algebra on $`\mathrm{\Omega }`$ such that the map $`\xi _x=1`$ is measurable for all $`x\mathrm{\Delta }`$. We notice that $`_E=`$. By the natural mapping between $`\mathrm{\Omega }`$ and $`𝒳`$, we define $`\sigma `$-algebras $`_\mathrm{\Delta }`$, $`\mathrm{\Delta }E`$, and $``$ on $`𝒳`$. The Gibbsian specification is defined as follows : for any measurable set $`A`$ and $`\xi 𝒳`$, we define
$$\gamma _\mathrm{\Lambda }(A|\xi ):=\underset{\zeta _\mathrm{\Lambda }\mathrm{\Lambda }}{}\gamma _\mathrm{\Lambda }(\zeta _\mathrm{\Lambda };\xi )1_A(\zeta _\mathrm{\Lambda }\xi _{\mathrm{\Lambda }^c}),$$
(4.28)
where $`1_A`$ denotes the indicator function on the set $`A`$. It is not hard to check that the system $`(\gamma _\mathrm{\Lambda })_{\mathrm{\Lambda }E}`$ defines a specification, i.e., it satisfies the following properties:
1. $`\gamma _\mathrm{\Lambda }(|\xi )`$ is a probability measure for each $`\xi 𝒳`$;
2. $`\gamma _\mathrm{\Lambda }(A|)`$ is $`_{\mathrm{\Lambda }^c}`$-measurable for all $`A`$;
3. $`\gamma _\mathrm{\Lambda }(A|)=1_A()`$ if $`A_{\mathrm{\Lambda }^c}`$;
4. $`\gamma _\mathrm{\Delta }\gamma _\mathrm{\Lambda }(A|\xi ):=_{\zeta _\mathrm{\Delta }\mathrm{\Delta }}\gamma _\mathrm{\Delta }(\zeta _\mathrm{\Delta }|\xi )\gamma _\mathrm{\Lambda }(A|\zeta _\mathrm{\Delta }\xi _{\mathrm{\Delta }^c})=\gamma _\mathrm{\Delta }(A|\xi )`$ for all $`\mathrm{\Lambda }\mathrm{\Delta }E`$ and $`\xi 𝒳`$.
A probability measure $`\mu `$ on $`(𝒳,)`$ is said to be admitted to the specification $`(\gamma _\mathrm{\Lambda })_{\mathrm{\Lambda }E}`$, or a Gibbs measure, if it satisfies the DLR-equations:
$$\mu (A)=\gamma _\mathrm{\Lambda }(A|\xi )𝑑\mu (\xi ),\text{for any }A\text{ and }\mathrm{\Lambda }E.$$
(4.29)
The DLR condition says that for any $`\mathrm{\Lambda }E`$ and $`A`$, the conditional expectation $`E^\mu [1_A|_{\mathrm{\Lambda }^c}]`$ has a version $`\gamma _\mathrm{\Lambda }(A|)`$:
$$E^\mu [1_A|_{\mathrm{\Lambda }^c}](\xi )=\gamma _\mathrm{\Lambda }(A|\xi ),\mu \text{-a.a. }\xi .$$
(4.30)
From the equation (4.25) we easily see that for any $`\mathrm{\Lambda }E`$, $`\zeta _\mathrm{\Lambda }\{x_1,\mathrm{},x_n\}𝒳_\mathrm{\Lambda }`$, and $`\xi 𝒳`$,
$`H_\mathrm{\Lambda }(\zeta _\mathrm{\Lambda };\xi )`$
$`=`$ $`H_{\{x_1\}}(\{x_1\};\xi )+H_{\{x_2\}}(\{x_2\};\{x_1\}\xi )+H_{\{x_n\}}(\{x_n\};\{x_1,\mathrm{},x_{n1}\}\xi ).`$
This says that all the values $`H_\mathrm{\Lambda }(\zeta _\mathrm{\Lambda };\xi )`$ are determined by the values $`H_{\{x\}}(\{x\};\xi )`$. Now then the DLR condition (4.30) is equivalent to saying that (cf. and \[16, Section 6\])
$$\frac{E^\mu [\xi _x=\{x\}|_{\{x\}^c}](\xi )}{E^\mu [\xi _x=\mathrm{}|_{\{x\}^c}](\xi )}=\mathrm{exp}[H_{\{x\}}(\{x\};\xi )],xE.$$
(4.31)
Proof of Theorem 2.5. Gibbsianness. As noted above, it is enough to show the relation (4.31). Let $`x_0E`$ be a fixed point and let $`\xi 𝒳`$. Then by (4.22) and (4.24),
$$\mathrm{exp}[H_{\{x_0\}}(\{x_0\};\xi )]=\underset{\mathrm{\Delta }E}{lim}\frac{detA(x_0\xi _{\mathrm{\Delta }\{x_0\}},x_0\xi _{\mathrm{\Delta }\{x_0\}})}{detA(\xi _{\mathrm{\Delta }\{x_0\}},\xi _{\mathrm{\Delta }\{x_0\}})}.$$
(4.32)
On the other hand, by (2.25)-(2.26)
$`{\displaystyle \frac{E^\mu [\xi _{\{x_0\}}=\{x_0\}|_{\{x_0\}^c}](\xi )}{E^\mu [\xi _{\{x_0\}}=\mathrm{}|_{\{x_0\}^c}](\xi )}}`$ $`=`$ $`\underset{\mathrm{\Delta }E}{lim}{\displaystyle \frac{E^\mu [\xi _{\{x_0\}}=\{x_0\}|_{\mathrm{\Delta }\{x_0\}}](\xi _{\mathrm{\Delta }\{x_0\}})}{E^\mu [\xi _{\{x_0\}}=\mathrm{}|_{\mathrm{\Delta }\{x_0\}}](\xi _{\mathrm{\Delta }\{x_0\}})}}`$ (4.33)
$`=`$ $`\underset{\mathrm{\Delta }E}{lim}{\displaystyle \frac{detA_{[\mathrm{\Delta }]}(x_0\xi _{\mathrm{\Delta }\{x_0\}},x_0\xi _{\mathrm{\Delta }\{x_0\}})}{detA_{[\mathrm{\Delta }]}(\xi _{\mathrm{\Delta }\{x_0\}},\xi _{\mathrm{\Delta }\{x_0\}})}}.`$
By Theorem 2.4 the two limits in (4.32) and (4.33) are the same and this proves that the DPP $`\mu `$ corresponding to the operator $`A(I+A)^1`$ is a Gibbs measure admitted to the specification $`(\gamma _\mathrm{\Lambda })_{\mathrm{\Lambda }E}`$ in (4.27)-(4.28).
Uniqueness. Let us now address to the uniqueness problem of the Gibbs measure. The arguments in the sequel parallel those in \[16, Section 6\]. Suppose that $`\nu `$ is a probability measure admitted to the specification $`(\gamma _\mathrm{\Lambda })_{\mathrm{\Lambda }E}`$, i.e., $`\nu `$ satisfies the condition (4.30):
$$E^\nu [1_A|_{\mathrm{\Lambda }^c}](\xi )=\gamma _\mathrm{\Lambda }(A|\xi ),\nu \text{-a.a. }\xi 𝒳\text{ for all }\mathrm{\Lambda }E.$$
(4.34)
Let $`F:𝒳`$ be a function of the form
$$F(\xi )=1_{\{\xi _{\mathrm{\Lambda }_0}=X\}},\text{for some }\mathrm{\Lambda }_0E\text{ and }X\mathrm{\Lambda }_0.$$
(4.35)
We will show that for such functions $`F`$,
$$\nu (F)=\mu (F).$$
(4.36)
Since those functions $`F`$ generate the $`\sigma `$-algebra $``$, $`\nu `$ then should be $`\mu `$ and the uniqueness follows.
Let $`\mathrm{\Lambda }E`$ be any set with $`\mathrm{\Lambda }_0\mathrm{\Lambda }`$. Then by (4.34)
$$E^\nu [F|_{\mathrm{\Lambda }^c}](\xi )=\frac{1}{Z_\mathrm{\Lambda }(\xi )}\underset{Y\mathrm{\Lambda }\mathrm{\Lambda }_0}{}\mathrm{exp}[H_\mathrm{\Lambda }(XY;\xi )].$$
(4.37)
Notice that the partition function $`Z_\mathrm{\Lambda }(\xi )`$ can be rewritten as follows. Let $`\mathrm{\Phi }^{(\mathrm{\Lambda };\xi )}`$ be a matrix of size $`|\mathrm{\Lambda }|`$ whose components are given by
$$\mathrm{\Phi }^{(\mathrm{\Lambda };\xi )}(x,y):=A(x,y)(P_{\xi _{\mathrm{\Lambda }^c}}A(,x),P_{\xi _{\mathrm{\Lambda }^c}}A(,y))_{\xi _{\mathrm{\Lambda }^c};A},$$
(4.38)
where, as before, $`A(,x)`$ is a function on $`E`$: $`A(,x)(z)=A(z,x)`$, $`zE`$, which belongs to $`_+`$, and $`P_{\xi _{\mathrm{\Lambda }^c}}`$ is the restriction operator restricting the functions on $`E`$ to the set $`\xi _{\mathrm{\Lambda }^c}`$, and $`(,)_{\xi _{\mathrm{\Lambda }^c};A}`$ is the inner product of the RKHS $`_{\xi _{\mathrm{\Lambda }^c};A}`$ with RK $`A_{\xi _{\mathrm{\Lambda }^c}}`$, the restriction of $`A`$ to the set $`\xi _{\mathrm{\Lambda }^c}`$. By Theorem 3.2, the matrix $`\mathrm{\Phi }^{(\mathrm{\Lambda };\xi )}`$ is well-defined. In an informal level, we can write $`\mathrm{\Phi }^{(\mathrm{\Lambda };\xi )}(x,y)`$ as
$$\mathrm{\Phi }^{(\mathrm{\Lambda };\xi )}(x,y)=A(x,y)A(x,\xi _{\mathrm{\Lambda }^c})A(\xi _{\mathrm{\Lambda }^c},\xi _{\mathrm{\Lambda }^c})^1A(\xi _{\mathrm{\Lambda }^c},y).$$
(4.39)
We refer to \[16, p 1559\] for the same matrix, where, however, $`A`$ is strictly positive. For each finite $`\mathrm{\Delta }\mathrm{\Lambda }`$, we let
$`\mathrm{\Phi }^{(\mathrm{\Lambda },\mathrm{\Delta };\xi )}(x,y)`$ $`:=`$ $`A(x,y)(P_{\xi _{\mathrm{\Delta }\mathrm{\Lambda }}}A(,x),P_{\xi _{\mathrm{\Delta }\mathrm{\Lambda }}}A(,y))_{\xi _{\mathrm{\Delta }\mathrm{\Lambda }};A},`$
$`=`$ $`A(x,y)A(x,\xi _{\mathrm{\Delta }\mathrm{\Lambda }})A(\xi _{\mathrm{\Delta }\mathrm{\Lambda }},\xi _{\mathrm{\Delta }\mathrm{\Lambda }})^1A(\xi _{\mathrm{\Delta }\mathrm{\Lambda }},y),x,y\mathrm{\Lambda }.`$
By Theorem 3.3,
$$\underset{\mathrm{\Delta }E}{lim}\mathrm{\Phi }^{(\mathrm{\Lambda },\mathrm{\Delta };\xi )}(x,y)=\mathrm{\Phi }^{(\mathrm{\Lambda };\xi )}(x,y),x,y\mathrm{\Lambda }.$$
(4.40)
Moreover, it is obvious that for any $`X\mathrm{\Lambda }`$
$$\mathrm{exp}[H_{\mathrm{\Lambda },\mathrm{\Delta }}(X;\xi )]=det(\mathrm{\Phi }^{(\mathrm{\Lambda },\mathrm{\Delta };\xi )}(X,X))$$
(4.41)
and hence
$$\mathrm{exp}[H_\mathrm{\Lambda }(X;\xi )]=det(\mathrm{\Phi }^{(\mathrm{\Lambda };\xi )}(X,X)).$$
(4.42)
Therefore we get
$$Z_\mathrm{\Lambda }(\xi )=\underset{X\mathrm{\Lambda }}{}\mathrm{exp}[H_\mathrm{\Lambda }(X;\xi )]=\underset{X\mathrm{\Lambda }}{}det(\mathrm{\Phi }^{(\mathrm{\Lambda };\xi )}(X,X))=det(I+\mathrm{\Phi }^{(\mathrm{\Lambda };\xi )}).$$
(4.43)
By using the expression (4.42) we see that
$`{\displaystyle \underset{Y\mathrm{\Lambda }\mathrm{\Lambda }_0}{}}\mathrm{exp}[H_\mathrm{\Lambda }(XY;\xi )]`$ $`=`$ $`{\displaystyle \underset{Y\mathrm{\Lambda }\mathrm{\Lambda }_0}{}}det(\mathrm{\Phi }_{XY}^{(\mathrm{\Lambda };\xi )})`$ (4.44)
$`=`$ $`det(P_{\mathrm{\Lambda }\mathrm{\Lambda }_0}+\mathrm{\Phi }_{X(\mathrm{\Lambda }\mathrm{\Lambda }_0)}^{(\mathrm{\Lambda };\xi )}).`$
Here we have put $`\mathrm{\Phi }_{XY}^{(\mathrm{\Lambda };\xi )}\mathrm{\Phi }^{(\mathrm{\Lambda };\xi )}(XY,XY)`$, etc. We insert (4.43)-(4.44) into the r.h.s. of (4.37) and after a short computation we obtain the expression for $`E^\nu [F|_{\mathrm{\Lambda }^c}](\xi )`$ in (4.37) (see \[16, eq. (6.47)\] for the details):
$`E^\nu [F|_{\mathrm{\Lambda }^c}](\xi )`$ (4.45)
$`=`$ $`{\displaystyle \frac{det(P_X[\mathrm{\Phi }_{\mathrm{\Lambda }_0}^{(\mathrm{\Lambda };\xi )}\mathrm{\Phi }^{(\mathrm{\Lambda };\xi )}(\mathrm{\Lambda }_0,\mathrm{\Lambda }\mathrm{\Lambda }_0)(I+\mathrm{\Phi }_{\mathrm{\Lambda }\mathrm{\Lambda }_0}^{(\mathrm{\Lambda };\xi )})^1\mathrm{\Phi }^{(\mathrm{\Lambda };\xi )}(\mathrm{\Lambda }\mathrm{\Lambda }_0,X)]P_X)}{det(I+\mathrm{\Phi }_{\mathrm{\Lambda }_0}^{(\mathrm{\Lambda };\xi )}\mathrm{\Phi }^{(\mathrm{\Lambda };\xi )}(\mathrm{\Lambda }_0,\mathrm{\Lambda }\mathrm{\Lambda }_0)(I+\mathrm{\Phi }_{\mathrm{\Lambda }\mathrm{\Lambda }_0}^{(\mathrm{\Lambda };\xi )})^1\mathrm{\Phi }^{(\mathrm{\Lambda };\xi )}(\mathrm{\Lambda }\mathrm{\Lambda }_0,\mathrm{\Lambda }_0))}}.`$
We will show that
$`\underset{\mathrm{\Lambda }E}{lim}[(I+\mathrm{\Phi }^{(\mathrm{\Lambda };\xi )})_{\mathrm{\Lambda }_0}\mathrm{\Phi }^{(\mathrm{\Lambda };\xi )}(\mathrm{\Lambda }_0,\mathrm{\Lambda }\mathrm{\Lambda }_0)(I+\mathrm{\Phi }_{\mathrm{\Lambda }\mathrm{\Lambda }_0}^{(\mathrm{\Lambda };\xi )})^1\mathrm{\Phi }^{(\mathrm{\Lambda };\xi )}(\mathrm{\Lambda }\mathrm{\Lambda }_0,\mathrm{\Lambda }_0)]`$ (4.46)
$`=`$ $`(I+A)_{\mathrm{\Lambda }_0}A(\mathrm{\Lambda }_0,\mathrm{\Lambda }_0^c)(I+A)(\mathrm{\Lambda }_0^c,\mathrm{\Lambda }_0^c)^1A(\mathrm{\Lambda }_0^c,\mathrm{\Lambda }_0)`$
$`=`$ $`(P_{\mathrm{\Lambda }_0}(I+A)^1P_{\mathrm{\Lambda }_0})^1.`$
In fact, by using a similar computation as in Proposition 3.1(b) we have for any $`f_0l^2(\mathrm{\Lambda }_0)`$,
$`(f_0,[(I+\mathrm{\Phi }^{(\mathrm{\Lambda };\xi )})_{\mathrm{\Lambda }_0}\mathrm{\Phi }^{(\mathrm{\Lambda };\xi )}(\mathrm{\Lambda }_0,\mathrm{\Lambda }\mathrm{\Lambda }_0)(I+\mathrm{\Phi }_{\mathrm{\Lambda }\mathrm{\Lambda }_0}^{(\mathrm{\Lambda };\xi )})^1\mathrm{\Phi }^{(\mathrm{\Lambda };\xi )}(\mathrm{\Lambda }\mathrm{\Lambda }_0,\mathrm{\Lambda }_0)]f_0)_{l^2(\mathrm{\Lambda }_0)}`$ (4.47)
$`=`$ $`\underset{fl^2(\mathrm{\Lambda }\mathrm{\Lambda }_0)}{inf}(f_0f,(I+\mathrm{\Phi }^{(\mathrm{\Lambda };\xi )})(\mathrm{\Lambda },\mathrm{\Lambda })(f_0f))_{l^2(\mathrm{\Lambda })}`$
$`=`$ $`\underset{fl^2(\mathrm{\Lambda }\mathrm{\Lambda }_0)}{inf}(f_0f_{l^2(\mathrm{\Lambda })}^2+\underset{gl^2(\xi _\mathrm{\Lambda }^c)}{inf}(f_0fg,A(f_0fg))_{l^2(E)})`$
$`=`$ $`\underset{hl^2(\mathrm{\Lambda }_0^c)}{inf}(f_0h,(P_\mathrm{\Lambda }+A)(f_0h))_{l^2(E)}.`$
Since $`P_\mathrm{\Lambda }I`$ strongly as $`\mathrm{\Lambda }E`$, it is obvious that the last expression in (4.47) converges as $`\mathrm{\Lambda }E`$ to
$`\underset{hl^2(\mathrm{\Lambda }_0^c)}{inf}(f_0h,(I+A)(f_0h))_{l^2(E)}`$ (4.48)
$`=`$ $`(f_0,[(I+A)_{\mathrm{\Lambda }_0}A(\mathrm{\Lambda }_0,\mathrm{\Lambda }_0^c)(I+A)(\mathrm{\Lambda }_0^c,\mathrm{\Lambda }_0^c)^1A(\mathrm{\Lambda }_0^c,\mathrm{\Lambda }_0)]f_0)_{l^2(\mathrm{\Lambda }_0)}.`$
Eqs. (4.47)-(4.48) prove (4.46). Recall the operator $`K=A(I+A)^1`$ which gives the DPP $`\mu `$. We have
$`(IK)_{\mathrm{\Lambda }_0}`$ $`=`$ $`P_{\mathrm{\Lambda }_0}(I+A)^1P_{\mathrm{\Lambda }_0}`$ (4.49)
$`=`$ $`[(I+A)_{\mathrm{\Lambda }_0}A(\mathrm{\Lambda }_0,\mathrm{\Lambda }_0^c)(I+A)(\mathrm{\Lambda }_0^c,\mathrm{\Lambda }_0^c)^1A(\mathrm{\Lambda }_0^c,\mathrm{\Lambda }_0)]^1`$
and
$`A_{[\mathrm{\Lambda }_0]}={\displaystyle \frac{K_{\mathrm{\Lambda }_0}}{(IK)_{\mathrm{\Lambda }_0}}}`$ $`=`$ $`(IK)_{\mathrm{\Lambda }_0}^1`$ (4.50)
$`=`$ $`A(\mathrm{\Lambda }_0,\mathrm{\Lambda }_0)A(\mathrm{\Lambda }_0,\mathrm{\Lambda }_0^c)(I+A)(\mathrm{\Lambda }_0^c,\mathrm{\Lambda }_0^c)^1A(\mathrm{\Lambda }_0^c,\mathrm{\Lambda }_0).`$
We thus get, by using (4.45)-(4.46) and (4.49)-(4.50),
$`\nu (F)`$ $`=`$ $`\underset{\mathrm{\Lambda }E}{lim}E^\nu [F|_{\mathrm{\Lambda }^c}](\xi )`$
$`=`$ $`det(IK_{\mathrm{\Lambda }_0})det(P_XA_{[\mathrm{\Lambda }_0]}P_X)`$
$`=`$ $`det(P_XK_{\mathrm{\Lambda }_0}+P_{\mathrm{\Lambda }_0X}(IK_{\mathrm{\Lambda }_0}))`$
$`=`$ $`\mu (F).`$
Now then $`\nu `$ must be $`\mu `$ and we have proven the uniqueness of the Gibbs measure. $`\mathrm{}`$
## A Appendix
In this Appendix we discuss the hypothesis (H) in Section 2 by giving some examples. For simplicity we take $`E:=`$, the set of integers. We give three typical examples.
(i) The case that $`A`$ is bounded and has a bounded inverse. In this case all the norms $`_{}`$, $`_0`$, and $`_+`$ are equivalent and the spaces $`_{}`$, $`_0`$, and $`_+`$ are the same as sets. Obviously, $`_{}`$ is functionally completed. The Gibbsianness of the DPP for the operator $`A(I+A)^1`$ with $`A`$ being in this category has already been shown by Shirai and Takahashi .
(ii) The case of diagonal matrices. Suppose that $`A`$ is a diagonal matrix with diagonal elements $`\alpha _x>0`$ with $`\alpha _x`$ being bounded and decreasing to zero as $`x\mathrm{}`$. It is not hard to show that the hypothesis (H) is satisfied for those operators $`A`$. In fact, $`_{}`$ consists of those functions $`f:E`$ such that $`_{xE}\alpha _x|f(x)|^2<\mathrm{}`$. In other words, if $`g=(g(x))_{xE}_0`$ is any element of $`_0`$ then the vector $`f(\alpha _x^{1/2}g(x))_{xE}`$ belongs to $`_{}`$ and all the elements of $`_{}`$ are of this type.
(iii) Perturbation of diagonal matrices. Let $`D`$ be any diagonal matrix of the type in the case (ii) above. Let $`A:=C^{}DC`$, where $`C`$ is a matrix such that $`C`$ and its inverse $`C^1`$ have off-diagonal elements that decrease sufficiently fast as the distance from the diagonal become far. To say more concretely, let $`C(x,y)`$ and $`C^1(x,y)`$ be the matrix components of $`C`$ and $`C^1`$, respectively. We assume that there exist positive numbers $`m>0`$ and $`M>0`$ such that
$$mC(x,x)M\text{ and }mC^1(x,x)M\text{ for all }xE,$$
(A.1)
and $`C(x,y)`$ and $`C^1(x,y)`$ converge to zero sufficiently fast as $`|xy|\mathrm{}`$. Then $`A`$ satisfies the conditions in (H). Here we give an example. Let $`D`$ be a diagonal matrix with diagonal elements $`\alpha _x`$, $`xE`$. We assume that there is $`k`$ such that
$$\alpha _x^1(1+|x|)^k,xE.$$
(A.2)
Let $`C`$ be a bounded operator with bounded inverse $`C^1`$ such that there is $`m2(k+1)`$ and
$$|C(x,y)|\frac{1}{1+|xy|^m}\text{ and }|C^1(x,y)|\frac{1}{1+|xy|^m}.$$
(A.3)
Such an operator $`C`$ can, for example, be obtained by taking its convolution kernel function as the Fourier series of strictly positive and sufficiently smooth function on the circle. We prove that $`_{}`$ is functionally completed. It is enough to show that the pre-Hilbert space $`(_0,_{})`$ satisfies the conditions (i) and (ii) of Theorem 3.2. First we show that for any $`yE`$, $`f(y)`$ is continuous in $`(_0,)_{}`$. As noted in the Subsection 2.1, it is equivalent to show that $`e_y_+`$. But, we have
$`e_y_+^2=(e_y,A^1e_y)_0`$ $`=`$ $`((C^1)^{}e_y,D^1(C^1)^{}e_y)_0`$ (A.4)
$`=`$ $`{\displaystyle \underset{xE}{}}\alpha _x^1|(C^1)^{}e_y(x)|^2`$
$``$ $`{\displaystyle \underset{xE}{}}(1+|x|)^k{\displaystyle \frac{1}{(1+|xy|^m)^2}}<\mathrm{}.`$
Next, notice that for any $`f_0`$,
$$f_{}^2=(f,C^{}DCf)_0=(Cf,DCf)_0=(Cf_{}^{(D)})^2,$$
(A.5)
where $`_{}^{(D)}`$ is the “$``$”-norm for $`AD`$. Since we have observed in case (ii) that the space $`_{}^{(D)}`$, completion of $`_0`$ w.r.t. $`_{}^{(D)}`$-norm, is functionally completed, it is enough to show that given any sequence $`\{f_n\}_0`$ which is $`_{}`$-Cauchy and such that $`f_n(y)0`$ as $`n\mathrm{}`$ for all $`yE`$, $`Cf_n(y)0`$ as $`n\mathrm{}`$ for all $`yE`$, because $`\{Cf_n\}`$ is $`_{}^{(D)}`$-Cauchy. We observe that there is a constant $`b>0`$ such that
$$|f_n(y)|b(1+|y|)^k,yE.$$
(A.6)
In fact,
$$|f_n(y)|=|(e_y,f_n)_0|e_y_+f_n_{}.$$
(A.7)
Since $`\{f_n\}`$ is $`_{}`$-Cauchy, $`f_n_{}`$ is bounded uniformly for $`n`$. On the other hand from (A.4), it is not hard to see that there exists $`b_1>0`$ such that
$$e_y_+b_1(1+|y|)^k,yE.$$
(A.8)
This proves (A.6). Now we have
$`Cf_n(y)`$ $`=`$ $`(e_y,Cf_n)_0`$
$`=`$ $`(C^{}e_y,f_n)_0`$
$`=`$ $`{\displaystyle \underset{xE}{}}\overline{C^{}e_y(x)}f_n(x)`$
$`=`$ $`{\displaystyle \underset{x(y)}{}}\overline{C^{}e_y(x)}f_n(x)+{\displaystyle \underset{x(y)^c}{}}\overline{C^{}e_y(x)}f_n(x),`$
where $`(y)`$ is any sufficiently large but finite set containing $`y`$. Since $`f_n(x)0`$ for all $`xE`$, the first term in the last expression converges to $`0`$ as $`n\mathrm{}`$. By using (A.3) and (A.6) we have
$$|\underset{x(y)^c}{}\overline{C^{}e_y(x)}f_n(x)|b\underset{x(y)^c}{}\frac{1}{1+|xy|^m}(1+|x|)^k.$$
(A.9)
Once $`(y)`$ has been taken sufficiently large, the quantity in the r.h.s. of (A.9) becomes as much small as we wish. This completes the proof.
Acknowledgments. The author thanks Prof. Y. M. Park and Dr. C. Bahn for fruitful discussions.
|
warning/0506/quant-ph0506073.html
|
ar5iv
|
text
|
# Entanglement monotones and maximally entangled states in multipartite qubit systems
## I Introduction
Entanglement is one the most striking features of quantum mechanics, but it is also one of its most counterintuitive consequences of which we still have rather incomplete knowledge Bell87 . Although the concentrated effort during the past decade has produced an impressive progress, there is no general qualitative and quantitative theory of entanglement.
A pure quantum-mechanical state of distinguishable particles is called disentangled with respect to a given partition $`𝒫`$ of the system iff it can be written as a tensor product of the parts of this partition. In the opposite case, the state must contain some finite amount of entanglement. The question then is to characterize and quantify this entanglement.
As to measuring the amount of entanglement in a given pure multipartite state, the first major step was made by Bennett et al. BennettDiVincenzo96 who discovered that the partial entropy of a party in a bipartite quantum state is a measure of entanglement. It coincides (asymptotically) with the entanglement of formation. Subsequently, the entanglement of formation of a two-qubit state was related to the concurrence Hill97 ; Wootters98 . Interestingly, by exploiting the knowledge of the mixed-state concurrence, the so-called 3-tangle $`\tau _3`$ which is a measure for three-partite pure states could be derived Coffman00 . This was a remarkable step since, loosely speaking, it opened the path to studying multipartite entanglement on solid grounds. Further, it was noticed by Uhlmann that antilinearity is an important property of operators that measure entanglement Uhlmann . A particularly interesting consequence of the 3-tangle formula was presented by Dür et al. who found that there are two inequivalent classes of sharing entanglement among three partiesDuer00 .
Another important aspect of the research on entanglement measures was the question regarding the requirements for a function that represents an entanglement monotone MONOTONES . It turned out that the essential property to be satisfied is non-increasing behavior on average under stochastic local operations and classical communication (SLOCC) SLOCC ; Duer00 . Later, Verstraete et al. have demonstrated that all homogeneous positive functions of pure-state density matrices that remain invariant under determinant-one SLOCC operations are entanglement monotones VerstraeteDM03 .
Despite the enormous effort, the only truly operational entanglement measure for arbitrary mixed states at hand, up to now, is the concurrence. For pure states we have a slightly farther view up to systems of two qutrits Cereceda03 ; Briand03 , and for three qubits, due to the 3-tangle. Various multipartite entanglement measures for pure-states have been proposed; but most of these measures do not yield zero for all possible product states (e.g. Refs. Barnum03 ; HEYDARI04 ; Wallach ; Wong00 ). This motivated the quest for an operational entanglement measure emerging from one requirement only: that it be zero for product states (not only for completely separable pure states). In particular, the goal has been to explore the idea that entanglement monotones are related to antilinear operators as pointed out for the concurrence by Uhlmann Uhlmann . Here we show that it is possible to construct a filter, i.e., an operator that has zero expectation value for all product states. It will turn out that these filters are entanglement monotones by construction. Interestingly, the two-qubit concurrence and the 3-tangle have various equivalent filter representations (see below). In order to illustrate the application of the method to a nontrivial example, we will present filters for up to six-qubit states that are able to distinguish inequivalent types of genuine multipartite entanglement.
Before finding a measure for genuine multipartite entanglement, one first has to agree about a definition of maximal multipartite entanglement:
###### Definition I.1
A pure $`q`$-qubit state $`|\psi _q`$ has maximal genuine multipartite entanglement, i.e. q-tangle, if and only if
* All reduced density matrices of $`|\psi _q`$ with rank $`2`$ (this includes all $`(q1)`$-site and single-site ones) are maximally mixed
* all $`p`$-site reduced density matrices of $`|\psi _q`$, have zero $`p`$-tangle; $`1<p<q`$.
* there is a canonical form of any maximally $`q`$-tangled state, for which properties (i) and (ii) are unaffected by phase factors, i.e. they are phase invariant.
A stronger form of condition (i) appeared in Ref. Gisin98 , where it is demanded that all reduced density matrices be maximally mixed.
Notice that the first condition induces that all reduced density matrices of the state have rank larger than $`1`$. This excludes product states of whatsoever kind. We emphasize that we use the term genuine $`q`$-qubit entanglement in a more restricted sense than, e.g., in Ref. Duer00 ; in particular, the only class of three-qubit states with genuine three-partite entanglement is represented by the GHZ state.
Some remarks are in order: whereas the first two requirements are well motivated, since the first means a maximal gain of information when a bit of information is read out of a maximally entangled state, and the second excludes hybrids of many different types of entanglement (somewhat following the idea of entanglement as a resource whose amount can be distributed among possibly different types of entanglement only; see e.g. Ref. Coffman00 ), we have no good argument in favor of the third, except that maximally entangled states for two and three qubits have such a canonical form.
## II Combs and filters
The basic concept is that of the comb. We define a comb of first order as an antilinear operator $`A`$ with zero expectation value for all states of a certain Hilbert space $``$. That is,
$$\psi \left|A\right|\psi =\psi \left|LC\right|\psi =\psi \left|L\right|\psi ^{}0$$
(1)
for all $`|\psi `$, where $`L`$ is a linear operator and $`C`$ is the complex conjugation. Here $`A`$ necessarily has to be antilinear (a linear operator with this property is zero itself). For simplicity we abbreviate
$$\psi \left|LC\right|\psi =:L_C.$$
(2)
Note that the complex conjugation is included in the definition of the expectation value $`\mathrm{}_C`$ in Eq. (2).
We will use the comb operators <sup>1</sup><sup>1</sup>1 If the antilinear operator $`A=LC`$ is a comb (with the complex conjugation $`C`$), for the sake of brevity we will also call the linear operator $`L`$ a comb. in order to construct the desired filters which are defined as antilinear operators whose expectation values vanish for all product states. While a comb is a local, i.e., a single-qubit operator, a filter is a non-local operator that acts on the whole multi-qubit state. It is worth mentioning already at this point that such a filter is invariant under $`𝒫`$-local unitary transformations if the combs have this property. Even more, it is invariant under the complex extension of the corresponding unitary group which is isomorphic to the special linear group. Since the latter represents the SLOCC operations for qubits SLOCC ; Duer00 , the filters will be entanglement monotones by construction.
We focus on multipartite systems of qubits (i.e., spin 1/2). The local Hilbert space is $`_j=\text{ }\mathrm{C}^2=:𝔥`$ for all $`j`$. We need the Pauli matrices $`\sigma _0:=1\mathrm{l}`$, $`\sigma _1:=\sigma _x`$, $`\sigma _2:=\sigma _y`$, and $`\sigma _3:=\sigma _z`$. It is straight forward to verify that the only single-qubit comb is the operator $`\sigma _y`$:
$$\psi \left|\sigma _yC\right|\psi =\sigma _y_C0.$$
Since its expectation value is a bi(anti-)linear expression in the coefficients of the state we denote it a comb of order 1. In general we will call a comb to be of order n if its expectation value is $`2n`$-linear in the coefficients of the state. There is one independent single-qubit comb which is of 2nd order. One can verify that for an arbitrary single-qubit state
$$0=\sigma _\mu _C\sigma ^\mu _C:=\underset{\mu ,\nu =0}{\overset{3}{}}\sigma _\mu _Cg^{\mu ,\nu }\sigma _\nu _C,$$
(3)
with $`g^{\mu ,\nu }=\mathrm{diag}\{1,1,0,1\}`$ being very similar to the Minkowski metric.<sup>2</sup><sup>2</sup>2Note that the only $`SL(2,\text{ }\mathrm{C})`$ invariant linear operator of order two is very similar to its antilinear counterpart: $`0=\sigma _\mu \sigma ^\mu :=_{\mu ,\nu =0}^3\sigma _\mu g^{\mu ,\nu }\sigma _\nu `$ with $`g^{\mu ,\nu }=\mathrm{diag}\{1,1,1,1\}`$. Both combs are $`SL(2,\text{ }\mathrm{C})`$ invariant OScombs .
It will prove useful to introduce the embedding
$$_n:\begin{array}{ccc}& & _n=^n\\ |\psi & & _n|\psi =|\psi ^n.\end{array}$$
(4)
Further define the product $``$ for operators $`O`$, $`P`$: $``$ such that
$$OP:\begin{array}{ccc}_2& & _2\\ OP_2(|\psi )& =& O|\psi P|\psi .\end{array}$$
(5)
Then we have the single-site ($`=\text{ }\mathrm{C}^2`$) comb $`\sigma _y`$ for $`_1=`$ and $`\sigma _\mu \sigma ^\mu `$ for $`_2`$.
These two one-site combs are sufficient to construct filters for multipartite qubit systems, which are entanglement monotones by construction. For $`n`$-qubit filters we will use the symbol $`^{(n)}`$. Filters for two qubits are
$`_1^{(2)}`$ $`=`$ $`\sigma _y\sigma _y`$ (6)
$`_2^{(2)}`$ $`=`$ $`{\displaystyle \frac{1}{3}}(\sigma _\mu \sigma _\nu )(\sigma ^\mu \sigma ^\nu ).`$ (7)
Both forms are explicitly permutation invariant, and they are filters since, if the state were a product, the combs would annihilate its expectation value. From the filters we obtain the pure-state concurrence in two different equivalent forms:
$`C`$ $`=`$ $`\left|_1^{(2)}_C\right|`$ (8)
$`C^2`$ $`=`$ $`\left|_2^{(2)}_C\right|{\displaystyle \frac{1}{3}}\left|\sigma _\mu \sigma _\nu _C\sigma ^\mu \sigma ^\nu _C\right|.`$
While the first form in Eq. (8) has the well-know convex-roof extension of the pure-state concurrence via the matrix Hill97 ; Wootters98 ; Uhlmann
$$R=\sqrt{\rho }\sigma _y\sigma _y\rho ^{}\sigma _y\sigma _y\sqrt{\rho }$$
(9)
it can be shown that the convex roof extension of the second form in Eq. (8) is related to
$`Q`$ $`=`$ $`\sqrt{\rho }\sigma _\mu \sigma _\nu \rho ^{}\sigma _\kappa \sigma _\lambda `$
$`\rho \sigma ^\mu \sigma ^\nu \rho ^{}\sigma ^\kappa \sigma ^\lambda \sqrt{\rho }.`$
and we find that $`QR^2`$.
Now let us consider the 3-tangle Coffman00 . For states of three qubits we find, e.g.,
$`_1^{(3)}`$ $`=`$ $`(\sigma _\mu \sigma _y\sigma _y)(\sigma ^\mu \sigma _y\sigma _y)`$ (11)
$`_2^{(3)}`$ $`=`$ $`{\displaystyle \frac{1}{3}}(\sigma _\mu \sigma _\nu \sigma _\lambda )(\sigma ^\mu \sigma ^\nu \sigma ^\lambda ).`$ (12)
Both $`_1^{(3)}`$ and $`_2^{(3)}`$ are filters and the latter is explicitly permutation invariant. From these operators the pure-state 3-tangle is obtained in the following way:
$`\tau _3`$ $`=`$ $`\left|_1^{(3)}_C\right|=\left|_2^{(3)}_C\right|`$ (13)
Interestingly, all three-qubit filters are powers of the 3-tangle as entanglement measure. We mention, however, that there is no immediate extension to mixed states as in the case of the ’alternative’ two-qubit concurrence, Eq. (II).
## III Filters for four-qubit states
Classifications of four-qubit states with respect to their entanglement properties have been studied, e.g., in Refs. VerstraeteDMV02 ; BriandLT03 ; Miyake03 . Here we introduce three four-qubit filter operators and study the three classes of entangled states they are measuring.
A four-qubit filter has the property that its expectation value for a given state is zero if the state is separable, i.e., if there is a one-qubit or a two-qubit part which can be factored out (note that for a three-qubit filter it is enough to extract one-qubit parts only). An expression that obeys this requirement for any single qubit and any combination of qubit pairs is given by
$$_1^{(4)}=(\sigma _\mu \sigma _\nu \sigma _y\sigma _y)(\sigma ^\mu \sigma _y\sigma _\lambda \sigma _y)(\sigma _y\sigma ^\nu \sigma ^\lambda \sigma _y).$$
(14)
Recall that any combination of the type $`\sigma _\mu \sigma _y`$ ($`\mu 2`$) represents a two-qubit comb. Note that the expectation value of an $`n`$th-order four-qubit filter has to be taken with respect to the corresponding $`_n`$, see Ref. OScombs ). It is straightforward to check that for a four-qubit GHZ state
$$|\mathrm{\Phi }_1=\frac{1}{\sqrt{2}}(|0000+|1111)$$
(15)
we have $`\mathrm{\Phi }_1\left|_1^{(4)}\right|\mathrm{\Phi }_1^{}=1`$. However, there is another state for which $`_1^{(4)}_C`$ does not vanish. For
$$|\mathrm{\Phi }_5=\frac{1}{\sqrt{6}}(\sqrt{2}|1111+|1000+|0100+|0010+|0001)$$
(16)
we find $`\mathrm{\Phi }_5\left|_1^{(4)}\right|\mathrm{\Phi }_5^{}=8/9`$. Interestingly, (16) is the only maximally entangled state measured by the four qubit hyperdeterminantMiyake03 , which is a homogeneous function of degree $`24`$. Its value for this state is $`({\displaystyle \frac{8}{9}})^4`$, i.e. exactly the same as of the 24th order homogeneous invariant $`\left(_1^{(4)}\right)^4`$, which however also measures the GHZ state. The hyperdeterminant of the four qubit GHZ state is zero.
Besides the 3rd-order filter $`_1^{(4)}`$ there exist also filters of 4th order and of 6th order. Examples are
$`_2^{(4)}`$ $`=`$ $`(\sigma _\mu \sigma _\nu \sigma _y\sigma _y)(\sigma ^\mu \sigma _y\sigma _\lambda \sigma _y)`$ (17)
$`(\sigma _y\sigma ^\nu \sigma _y\sigma _\tau )(\sigma _y\sigma _y\sigma ^\lambda \sigma ^\tau )`$
$`_3^{(4)}`$ $`=`$ $`{\displaystyle \frac{1}{2}}(\sigma _\mu \sigma _\nu \sigma _y\sigma _y)(\sigma ^\mu \sigma ^\nu \sigma _y\sigma _y)(\sigma _\rho \sigma _y\sigma _\tau \sigma _y)`$
$`(\sigma ^\rho \sigma _y\sigma ^\tau \sigma _y)(\sigma _y\sigma _\rho \sigma _\tau \sigma _y)(\sigma _y\sigma ^\rho \sigma ^\tau \sigma _y).`$
While $`_2^{(4)}`$ measures only GHZ-type entanglement ($`\mathrm{\Phi }_2\left|_2^{(4)}\right|\mathrm{\Phi }_2^{}=1`$) the 6th-order filter $`_3^{(4)}`$ has the non-zero expectation values 1/2 for the GHZ state and 1 for yet another state,
$$|\mathrm{\Phi }_4=\frac{1}{2}(|1111+|1100+|0010+|0001).$$
(18)
$`_1^{(4)}`$ and $`_2^{(4)}`$ have zero expectation value for this state (as well as the hyperdeterminant). Finally, all four-qubit filters $`_j^{(4)}`$ ($`j=1,2,3`$) have zero expectation value for the four-qubit W state $`1/2(|0111+|1011+|1101+|1110)`$.
The states $`|\mathrm{\Phi }_j`$ are the maximally entangled states for four qubits; they satisfy all three requirements in Def. I.1, including the stronger condition (i) from Ref. Gisin98 . Note that they cannot be transformed into one another by SLOCC operations: A state with a finite expectation value for one filter cannot be transformed by means of SLOCC operations into a state with zero expectation value for the same filter. For example, $`_2^{(4)}`$ detects the GHZ state $`|\mathrm{\Phi }_1`$ but gives zero for the other two states. Therefore, the four-qubit entanglement in those states must be different from that of the GHZ state.
Hence, there are at least three inequivalent types of genuine entanglement for four qubits <sup>3</sup><sup>3</sup>3In fact, there are exactly three maximally entangled states for four qubits. This will be discussed in a forthcoming publication.. We mention that the three maximally entangled states $`|\mathrm{\Phi }_j`$ are not distinguished by the classification for pure four-qubit states of Ref. VerstraeteDMV02 . This can be seen by computing the expectation values of the four-qubit filters and the reduced one-qubit density matrices for each of the nine class representatives of Ref. VerstraeteDMV02 . Only the classes 1–4 and 6 have non-vanishing “4-tangle”. The corresponding local density matrices can be completely mixed only for class 1. Therefore, all three states $`|\mathrm{\Phi }_j`$ must belong to that class.
## IV Filters for more qubits
In this section we will continue the discussion from the previous section and demonstrate how general multipartite filters are constructed. It is not the scope of this work to discuss independence and completeness of a given set of filters, nor to “taylor” a filter for a given single class of entanglement. We only emphasize that every filter is an invariant and that linear homogeneous combinations and in fact any homogeneous function of them is an invariant, as well. Thus, when a sufficient set of independent filters is known together with their weights for the corresponding entanglement classes, such a taylored invariant can be constructed. This invariant, though, is not expected to be simply the modulus square of some filter.
For five qubits we find four independent filters, their independence becoming clear from their values on a set of maximally entangled states. In order to compactify the formulas, the tensor product symbol $``$ will be omitted.
$`_1^{(5)}`$ $`=`$ $`(\sigma _{\mu _1}\sigma _{\mu _2}\sigma _{\mu _3}\sigma _y\sigma _y)(\sigma ^{\mu _1}\sigma ^{\mu _2}\sigma _y\sigma _{\mu _4}\sigma _y)`$
$`(\sigma _{\mu _5}\sigma _y\sigma ^{\mu _3}\sigma ^{\mu _4}\sigma _y)(\sigma ^{\mu _5}\sigma _y\sigma _y\sigma _y\sigma _y)`$
$`_2^{(5)}`$ $`=`$ $`(\sigma _{\mu _1}\sigma _{\mu _2}\sigma _{\mu _3}\sigma _y\sigma _y)(\sigma ^{\mu _1}\sigma _y\sigma _y\sigma _{\mu _4}\sigma _{\mu _5})`$
$`(\sigma _y\sigma ^{\mu _2}\sigma _y\sigma _y\sigma _y)(\sigma _y\sigma _y\sigma ^{\mu _3}\sigma _y\sigma _y)`$
$`(\sigma _y\sigma _y\sigma _y\sigma ^{\mu _4}\sigma _y)(\sigma _y\sigma _y\sigma _y\sigma _y\sigma ^{\mu _5})`$
$`_3^{(5)}`$ $`=`$ $`(\sigma _{\mu _1}\sigma _{\mu _2}\sigma _{\mu _3}\sigma _y\sigma _y)(\sigma ^{\mu _1}\sigma ^{\mu _2}\sigma _{\mu _4}\sigma _y\sigma _y)`$
$`(\sigma _{\mu _5}\sigma _y\sigma ^{\mu _3}\sigma _{\mu _6}\sigma _y)(\sigma ^{\mu _5}\sigma _y\sigma ^{\mu _4}\sigma _{\mu _7}\sigma _y)`$
$`(\sigma _{\mu _8}\sigma _y\sigma _y\sigma ^{\mu _6}\sigma _{\mu _9})(\sigma ^{\mu _8}\sigma _y\sigma _y\sigma ^{\mu _7}\sigma ^{\mu _9})`$
$`_4^{(5)}`$ $`=`$ $`{\displaystyle \frac{1}{8}}(\sigma _{\mu _1}\sigma _{\mu _2}\sigma _{\mu _3}\sigma _y\sigma _y)(\sigma ^{\mu _1}\sigma ^{\mu _2}\sigma ^{\mu _3}\sigma _y\sigma _y)`$
$`(\sigma _{\mu _4}\sigma _y\sigma _{\mu _5}\sigma _{\mu _6}\sigma _y)(\sigma ^{\mu _4}\sigma _y\sigma ^{\mu _5}\sigma ^{\mu _6}\sigma _y)`$
$`(\sigma _{\mu _7}\sigma _y\sigma _y\sigma _{\mu _8}\sigma _{\mu _9})(\sigma ^{\mu _7}\sigma _y\sigma _y\sigma ^{\mu _8}\sigma ^{\mu _9})`$
The set of maximally entangled states distinguished by these filters is
$`|\mathrm{\Psi }_2`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2}}}(|11111+|00000)`$ (23)
$`|\mathrm{\Psi }_4`$ $`=`$ $`{\displaystyle \frac{1}{2}}(|11111+|11100+|00010+|00001)`$ (24)
$`|\mathrm{\Psi }_5`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{6}}}(\sqrt{2}|11111+|11000+|00100+|00010+|00001)`$ (25)
$`|\mathrm{\Psi }_6`$ $`=`$ $`{\displaystyle \frac{1}{2\sqrt{2}}}(\sqrt{3}|11111+|10000+|01000+|00100+|00010+|00001)`$ (26)
The index of the state indicates the number of Fock-states in the normal form of the state and will be termed its length; a deeper discussion of the maximally entangled states and their connection to the filters that measure them is beyond the scope of this article and will be reported elsewhere.
The states $`|\mathrm{\Psi }_2`$$`|\mathrm{\Psi }_5`$ satisfy all three requirements of definition I.1 for being a maximally entangled states. It is interesting that $`|\mathrm{\Psi }_6`$ instead satisfies only the 1st and the 3rd requirement but contains fourtangle as measured by the filter $`_3^{(4)}`$.
For six quibits we only exemplarily write two independent filters but indicate how to construct filters for a general number of qubits.
$`_1^{(6)}`$ $`=`$ $`(\sigma _{\mu _1}\sigma _{\mu _2}\sigma _y\sigma _y\sigma _y\sigma _y)(\sigma ^{\mu _1}\sigma _y\sigma _{\mu _3}\sigma _y\sigma _y\sigma _y)`$
$`(\sigma _{\mu _6}\sigma _y\sigma _y\sigma _{\mu _4}\sigma _y\sigma _y)(\sigma _y\sigma _y\sigma ^{\mu _3}\sigma _y\sigma _{\mu _5}\sigma _y)`$
$`(\sigma ^{\mu _6}\sigma ^{\mu _2}\sigma _y\sigma ^{\mu _4}\sigma ^{\mu _5}\sigma _y)`$
$`_2^{(6)}`$ $`=`$ $`(\sigma _{\mu _1}\sigma _{\mu _2}\sigma _y\sigma _y\sigma _y\sigma _y)(\sigma ^{\mu _1}\sigma _y\sigma _{\mu _3}\sigma _y\sigma _y\sigma _y)`$
$`(\sigma _{\mu _6}\sigma ^{\mu _2}\sigma ^{\mu _3}\sigma _{\mu _4}\sigma _y\sigma _y)(\sigma _y\sigma _y\sigma _y\sigma ^{\mu _4}\sigma _{\mu _5}\sigma _y)`$
$`(\sigma ^{\mu _6}\sigma _y\sigma _y\sigma _y\sigma ^{\mu _5}\sigma _y)`$
$`\mathrm{}`$
$`_i^{(6)}`$ $`=`$ $`(\sigma _\mu _{}\sigma _\mu _{}\sigma _y\sigma _y\sigma _y\sigma _y)(\sigma _\mu _{}\sigma _y\sigma _\mu _{}\sigma _y\sigma _y\sigma _y)`$
$`(\sigma _\mu _{}\sigma _{}\sigma _{}\sigma _\mu _{}\sigma _y\sigma _y)(\sigma _\mu _{}\sigma _{}\sigma _{}\sigma _{}\sigma _\mu _{}\sigma _y)(\sigma _{}\sigma _{}\sigma _{}\sigma _{}\sigma _{}\sigma _{})\mathrm{}`$
where in the latter formula all the $`\mu _{}`$ are to be contracted properly; in the $`\sigma _{}`$ the “$``$” either have to be substituted by indices which then have to be contracted properly or by $`\sigma _y`$. This also indicates how higher filters can be constructed and suggests that for a filter of an $`n`$-qubit system, at least $`_{n1}`$ be needed. It is worthwhile to mention that the above list is not meant to be exhaustive, nor did we explicitly check for permutation invariance, which eventually could help cristalizing the “proper” filters. The set of maximally entangled states to be distinguished by the six qubit filters is
$`|\mathrm{\Xi }_2`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{2}}}(|111111+|000000)`$ (30)
$`|\mathrm{\Xi }_4`$ $`=`$ $`{\displaystyle \frac{1}{2}}(|111111+|111100+|000010+|000001)`$ (31)
$`|\mathrm{\Xi }_5`$ $`=`$ $`{\displaystyle \frac{1}{\sqrt{6}}}(\sqrt{2}|111111+|111000+|000100+|000010+|000001)`$ (32)
$`|\mathrm{\Xi }_6`$ $`=`$ $`{\displaystyle \frac{1}{2\sqrt{2}}}(\sqrt{3}|1\mathrm{}1+|110000+|00|W_4)`$ (33)
$`|\mathrm{\Xi }_7`$ $`=`$ $`{\displaystyle \frac{1}{2\sqrt{2}}}(\sqrt{3}|111111+|W_6)`$ (34)
where $`|W_4:=|1000+|0100+|0010+|0001`$ and $`|W_6`$ analoguously are the W state for four and six qubits. We want to mention that only the states up to length $`5`$ are free of any subtangle and that only for states up to length $`4`$ all reduced density matrices are maximally mixed Gisin98 .
The filter values for the maximally entangled states are reported in the table. The states are classified by the length of their normal form. An “X” indicates that the corresponding state does not occur. Whereas the tangles for
four/five qubits discriminate all three/four maximally entangled states, the two six-tangles we explicitely wrote only attribute to those states with minimal length (the GHZ) and with maximal length. This table shows that the indicated states correspond to different entanglement SLOCC classes. In fact there is a relation between the length of the state and the degree of multilinearity of the filter, which will be reported on in another publication.
## V Conclusions
We have presented a new and efficient way of generating entanglement monotones. It is based on operators which we called filters. The expectation values of these operators are zero for all possible product states, not only for the completely factoring case. The building blocks of the filters (denoted combs) guarantee invariance under $`SL(2,\text{ }\mathrm{C})^N`$ for qubits. As a consequence, all filters are automatically entanglement monotones. They are measures of genuine multipartite entanglement. This circumvents the difficult task to construct entanglement monotones from the essentially known (linear) local unitary invariants.
As an immediate result of our method the concurrence for pure two-qubit states is reproduced. Moreover, we have found an alternative expression for the concurrence with the corresponding convex roof extension based on the corresponding filter operator. The application of the method to pure three-qubit states yields several operator-based expressions for the 3-tangle, including an explicitly permutation-invariant form.
Further advantages of this approach are the feasibility of constructing specific monotones that vanish for certain separable (pure) states and the applicability of this concept to partitions into subsystems other than qubits (i.e. qutrits…). The methods permits in a direct manner quantification and classification of multipartite entanglement. We demonstrate this with the explicit expressions for four- up to six-qubit entanglement measures that for the first time detect three different types of genuine four-qubit entanglement and four different types of five-qubit entanglement; the types of genuine four-qubit entanglement are not distinguished by the classification of four-qubit states in Ref. VerstraeteDMV02 .
As to $`N`$-qubit systems, there remain various interesting questions. Clearly, it would be desirable to have a recipe how to build invariant combs for more complicated systems (e.g. higher spin). It would also be interesting to know what characterizes a complete set of filters for any given $`N`$. While it is not obvious how the convex roof construction for two qubits can be generalized, we believe that the operator form of the $`N`$-tangles in terms of filters makes it easier to solve this problem. The question is whether there is a systematic way to obtain a convex-roof construction for a given filter with general multi-linearity.
Acknowledgments – We would like to thank L. Amico, R. Fazio, and especially A. Uhlmann for stimulating discussions. This work was supported by the EU RTN grant HPRN-CT-2000-00144, the Vigoni Program of the German Academic Exchange Service and the Sonderforschungsbereich 631 of the German Research Foundation. J.S. holds a Heisenberg fellowship from the German Research Foundation.
|
warning/0506/hep-th0506128.html
|
ar5iv
|
text
|
# 1 Introduction
## 1 Introduction
In the $`\beta `$-deformed $`𝒩=4`$ SYM theory, the 3 $`𝒩=1`$ chiral superfields $`\mathrm{\Phi }_i`$ have interaction superpotential
$$W=h\mathrm{Tr}(q\mathrm{\Phi }_1\mathrm{\Phi }_2\mathrm{\Phi }_3\frac{1}{q}\mathrm{\Phi }_1\mathrm{\Phi }_3\mathrm{\Phi }_2),$$
(1.1)
with complex parameters $`h`$ and $`q`$, the latter usually written as $`q=e^{i\pi \beta }`$. The undeformed $`𝒩=4`$ theory is obtained for $`q=\pm 1`$ and $`|h|=2g`$, where $`g`$ is the gauge coupling. The deformed theory has $`𝒩=1`$ SUSY with $`U(1)_R`$ and $`U(1)\times U(1)`$ flavor symmetries. It was pointed out in that renormalization group beta functions vanish if one condition on the constants $`q,h`$ is satisfied, giving an exactly marginal deformation of $`𝒩=4`$ SYM with $`𝒩=1`$ SUSY. New attention has been focused on this deformed theory, for gauge group $`SU(N)`$, after its gravity dual was found by Lunin and Maldacena<sup>3</sup><sup>3</sup>3 Earlier partial results were obtained in . It is a solution of Type IIB supergravity with product metric $`AdS_5\times \stackrel{~}{S}_5`$, where $`\stackrel{~}{S}_5`$ is a deformed 5-sphere. The solution has many interesting features. Further developments have appeared in the subsequent recent literature -.
In this note, we present several results most of which concern the properties of 2- and 3-point functions of $`SU(2,2|1)`$ chiral primary operators in the deformed theory in the weak coupling limit, thus exploring the perturbative dynamics of the exactly marginal deformation. For this purpose we combine the methods of and Appendix D of . The major conclusion is that the order $`\lambda =g^2N`$ radiative corrections to these correlators vanish, which is evidence that they are “protected”, that is given exactly by their free field values. In the undeformed $`𝒩=4`$ theory, this “protected” property was first observed in the strong coupling limit through the $`AdS_5\times S_5`$ gravity dual . It may be possible to carry out a similar analysis for the Lunin-Maldacena solution on the deformed $`\stackrel{~}{S}_5`$, although the problem of Kaluza-Klein decomposition of the coupled fluctuations is quite difficult. The spin chain method of calculation of anomalous dimensions, initiated in for the $`𝒩=4`$ theory, has also been applied to the $`\beta `$-deformation in , and information on 3-point functions may also be available through this method . Spin chain techniques are elegant and useful, but the simple weak coupling methods used herein lead to a clear and worthwhile picture of the correlators of chiral primaries.
## 2 The Central Charge from the Gravity Dual
If the Lunin-Maldacena solution is indeed the dual of the $`\beta `$-deformed $`𝒩=4`$ theory, then correlation functions computed from gravity by techniques developed early in the study of the AdS/CFT correspondence should agree with field theory correlators in the limit $`N\mathrm{}`$ and $`\lambda =g^2N>>1`$. Operators in the gauge theory are dual to fluctuations of IIB gravity theory fields about the L-M background solution. In most cases therefore, analysis of the fluctuations is required before correlation functions can be studied, and this is difficult as stated above.
One important case in which detailed fluctuation analysis is not required is the 2-point function $`T_{\mu \nu }T_{\rho \sigma }`$ which determines the central charge $`c`$ of the CFT. An exactly marginal deformation should not change the central charge from its value $`c=N^2/4`$ in the undeformed $`𝒩=4`$ theory. The gravity calculation of this correlator requires only the metric fluctuations in the $`AdS_5`$ part of the metric. We are thus interested in reducing the gravitational part of the IIB action on the $`\stackrel{~}{S}_5`$ internal space, so we write
$`S_{10}`$ $`=`$ $`{\displaystyle \frac{1}{2\kappa _{10}^2}}{\displaystyle d^{10}x\sqrt{G_{10}}\left[R_{10}+\mathrm{}\right]}`$ (2.2)
$`=`$ $`{\displaystyle \frac{A}{2\kappa _{10}^2}}{\displaystyle d^5x\sqrt{g_5}\left[R_5+12/L^2+\mathrm{}\right]}.`$ (2.3)
For a direct product metric, the factor $`A`$ would be $`Vol(\stackrel{~}{S}_5)`$, as first shown in and further applied in . Things are slightly more complicated in the L-M case because the Einstein frame metric is a warped product given by
$$ds_E^2=L^2G^{1/4}[ds_{AdS_5}^2+\underset{i}{}(d\mu _i^2+G\mu _i^2d\varphi _i^2)+|\stackrel{~}{\gamma }|^2L^4G\mu _1^2\mu _2^2\mu _3^2(\underset{i}{}d\varphi _i)^2],$$
(2.4)
$`G^1`$ $`=`$ $`1+|\stackrel{~}{\gamma }|^2L^4(\mu _1^2\mu _2^2+\mu _2^2\mu _3^2+\mu _1^2\mu _3^2),`$ (2.5)
$`L^4`$ $`=`$ $`4\pi N.`$ (2.6)
The $`\stackrel{~}{S}_5`$ metric is parameterized by 3 angles $`\varphi _i`$ and 3 positive coordinates $`\mu _i`$ which are constrained to satisfy $`_i\mu _i^2=1`$. The complex parameter $`\stackrel{~}{\gamma }`$ is the parameter in the L-M solution which describes a general $`\beta `$-deformation with $`\beta =\gamma +i\sigma `$. We are using (3.24) of . Denoting the $`\stackrel{~}{S}_5`$ metric in (2.4) by $`ds^2=h_{ij}dy^idy^j=_i(d\mu _i^2+\mathrm{})`$, we see that the factor $`A`$ in (2.2) is given by
$$A=𝑑y^1\mathrm{}𝑑y^6\sqrt{h}G^{5/4}G^{1/4}\delta (\underset{i}{}\mu _i^21).$$
(2.7)
It is now straightforward to calculate $`h=deth_{ij}=G^2\mu _1^2\mu _2^2\mu _3^2`$ and to observe that the integrand in (2.7) is independent of the deformation parameter $`\stackrel{~}{\gamma }`$. This is sufficient to show that the deformation, as described by the L-M solution, does not change the central charge $`c`$. The central charge in the L-M solution was also studied in .
Actually there is a general argument that $`n`$-point functions of $`T_{\mu \nu }`$ are unchanged by the deformation. Recall that the L-M solution is generated by an $`SL(2,R)`$ transformation which changes the modular parameter of a 2-torus in the 10 dimensional spacetime, but leaves the Einstein frame metric of the complementary 8 dimensions invariant. A $`U(1)\times U(1)`$ group, dual to the flavor group in the field theory, acts on the torus. Bulk fields which are invariant under $`U(1)\times U(1)`$, such as the 5-dimensional metric dual to the stress tensor, give boundary correlators which are unaffected by the $`SL(2,R)`$ transformation which produces the deformation. <sup>4</sup><sup>4</sup>4We are grateful to Juan Maldacena who supplied this argument.
## 3 Two-point correlation functions at weak coupling.
The main purpose of this section is to study the lowest order radiative corrections to 2-point functions of various scalar operators in the deformed theory. Among other things, we will confirm the absence of anomalous dimensions for the chiral primary operators identified in , and we will find that an operator not in this list has vanishing anomalous dimension in lowest order. The techniques we develop here will be applied to 3-point correlators in the next section. It is important that we use the gauge group SU(N), since the deformed U(N) theory is not conformal.
### 3.1 Preliminaries
In a general $`𝒩=1`$ SUSY gauge theory with chiral superfields $`\mathrm{\Phi }^A`$ and general cubic superpotential
$$W=\frac{1}{6}Y_{ABC}\mathrm{\Phi }^A\mathrm{\Phi }^B\mathrm{\Phi }^C,$$
(3.8)
the $`\beta `$-function for the couplings $`Y_{ABC}`$ is determined by the anomalous dimension matrix of the elementary $`\mathrm{\Phi }^A`$ via
$$\beta _{ABC}=Y_{ABD}\gamma _C^D+Y_{ADC}\gamma _B^D+Y_{DBC}\gamma _A^D.$$
(3.9)
To 1-loop order the anomalous dimension matrix is
$$16\pi ^2\gamma _B^A=\frac{1}{2}Y^{BCD}\overline{Y}_{ACD}2g^2C(R)_B^A,C(R)_B^A=(T^aT^a)_B^A,$$
(3.10)
It may also be seen from (6b) of that if $`\gamma ^{(1)}=0`$ and
$`Q=T(R)3C(G)=0`$, then $`\gamma ^{(2)}`$ also vanishes. Then the gauge coupling $`\beta `$-function $`\beta _{NSVZ}`$ also vanishes through 3-loop order.
In the $`\beta `$-deformed $`𝒩=4`$ theory, $`\mathrm{\Phi }^A\mathrm{\Phi }_i^a`$, while $`C(R)=N\delta ^{ab}\delta _{ij}`$,
$`C(G)=N`$, and $`T(R)=3N`$, so $`Q=0`$. Color symmetry and the $`Z_3Z_3`$ symmetry, see , of the superpotential (1.1) require that $`\gamma `$ is diagonal, i.e. $`\gamma _B^A\gamma \delta ^{ab}\delta _{ij}`$. One readily obtains from (3.10)
$$\gamma ^{(1)}=\{\frac{1}{4}|h|^2(N(|q|^2+\frac{1}{|q|^2})\frac{2}{N}|q\frac{1}{q}|^2))2g^2N\}.$$
(3.11)
The condition $`\gamma ^{(1)}=0`$ ensures that the deformed $`SU(N)`$ theory is conformal invariant through 2-loop order. This condition may be modified by higher loop corrections, but the diagonal property ensures that there is always a fixed point set of real codimension 1 in the space of couplings $`h,q,g^2.`$ The fixed line of $`𝒩=4`$ SYM theory is attained when $`q=\pm 1`$ and $`|h|^2=4g^2`$.
For gauge group $`U(N)`$, the situation is somewhat different. The color singlet fields $`\mathrm{\Phi }_i^0`$ have no D-type interactions, but they do couple through the deformed superpotential
$$W_{U(N)}=\frac{1}{\sqrt{N}}h\mathrm{\Phi }_1^0(q\frac{1}{q})\mathrm{tr}(\mathrm{\Phi }_2\mathrm{\Phi }_3)+cyclic+W_{SU(N)}.$$
(3.12)
Singlet and non-singlet fields have different anomalous dimensions, namely
$`16\pi ^2\gamma _{sing}`$ $`=`$ $`{\displaystyle \frac{N}{4}}|h|^2|q{\displaystyle \frac{1}{q}}|^2`$ (3.13)
$`16\pi ^2\gamma _{nonsing}`$ $`=`$ $`{\displaystyle \frac{N}{4}}|h|^2(|q|^2+{\displaystyle \frac{1}{|q|^2}})2g^2N.`$ (3.14)
The marginal deformation is lost, but singlet couplings flow to zero at long distance. The IR theory is conformal with the same Lagrangian description as the $`SU(N)`$ theory.
We are interested in the 2-point (and later 3-point) correlators of single trace operators of the general form tr$`(Z_1^jZ_2^kZ_3^l)`$ where $`Z_i`$ is the lowest component of the superfield $`\mathrm{\Phi }_i`$. We now make two observations which greatly simplify the required calculations. We refer to the component Lagrangian which is given in the Appendix, see (7.1). The first observation concerns the relation of self-energy corrections to internal scalar lines in the undeformed and deformed theories. These corrections include a spinor loop with F-term vertices from the Yukawa interaction in (7.1). The amplitude is proportional to the term involving $`|h|^2`$ in $`\gamma ^{(1)}`$ in (3.11) and thus appears to depend on the deformation parameter $`q`$. However we must work on the fixed point locus where $`\gamma ^{(1)}=0`$, so the spinor loop is really proportional to $`g^2`$ and independent of $`q`$.
This fact allows us to apply the arguments of and Appendix B of that order $`g^2N`$ radiative corrections due to D-term interactions cancel with self-energy insertions in 2- and 3-point functions. The reason is that D-term effects do not depend on the flavor quantum number of the fields $`Z_i`$. So their contribution to correlators of tr$`Z_1^jZ_2^kZ_3^l`$ is the same as for the the single flavor operators tr$`Z_1^{j+k+l}`$. The lowest order radiative corrections to these correlators, which have no F-term contributions (other than self-energy) then vanish by the combinatoric arguments of . These arguments are valid to lowest order in $`g^2N`$ and all orders in $`1/N`$. They imply that we need only consider contributions from the quartic F-term interaction in (7.1).
Since we will be primarily interested in chiral primary operators, we now describe the chiral ring of the $`\beta `$-deformed theory . To classify these operators we define three separate $`U(1)`$ flavor groups such that the field $`Z_i`$ carries unit charge under the $`i`$th $`U(1)`$ and the other fields are uncharged. For general $`\beta `$ the single trace operators of the chiral ring have the following charge assignments,
$$(J,0,0),(0,J,0),(0,0,J),(J,J,J).$$
(3.15)
Special chiral operators (dual to strings which wind contractible cycles in $`\stackrel{~}{S}_5`$ ) exist for the special values $`\beta =m/n`$ where $`m`$ and $`n`$ are mutually prime integers. They carry the $`U(1)`$ charges
$$(n_1,n_2,n_3),n_1=n_2=n_3(\mathrm{mod}n).$$
(3.16)
As we will see from explicit calculations in the case $`(J,J,J)`$, the actual operator with vanishing anomalous dimension is not tr$`(Z_1^JZ_2^JZ_3^J)`$ but rather a specific sum over permutations of the $`Z_i`$ fields involved. The same is true for the special cases $`(n_1,n_2,n_3)`$.
### 3.2 Correlators of $`\mathrm{tr}(Z_i^J)=𝒪_i^J.`$
Applying the arguments above, it is obvious that the 2-point functions $`𝒪_i^J\overline{𝒪_i^J}`$ (no sum on $`i`$) have no lowest order radiative corrections, since the quartic F-term interaction does not contribute. In fact, to lowest order, all correlators of $`𝒪_i^J`$ agree with those in the undeformed theory. Thus all 3-point functions and all extremal $`n`$-point functions , i.e.
$$𝒪_i^{J_1}\mathrm{}𝒪_i^{J_k}\overline{𝒪}_i^{J_1+\mathrm{}+J_k}$$
(3.17)
are protected at lowest order. However, non-extremal correctors such as
$$𝒪_i^{J_1}𝒪_i^{J_2}\overline{𝒪}_i^{J_3}\overline{𝒪}_i^{J_4}$$
(3.18)
are not protected, since they have non-vanishing $`D`$-term radiative corrections.
### 3.3 The operators tr$`(Z_i^JZ_j)`$ for $`ij`$ and $`J>1`$
In the $`𝒩=4`$ theory these operators have protected 2- and 3-point correlators because they are obtained by applying an SU(3) lowering operator to tr$`Z_i^{J+1}`$ which is clearly part of the symmetrized traceless tr$`X^k`$ in the common $`SO(6)`$ designation. Since the SU(3) symmetry is broken to $`U(1)U(1)`$ by the deformation we would not expect that these operators remain chiral primary.
We now develop the “effective operator” method, see Appendix D of , which we use in most of our calculations. This method allows us to obtain full results for two-point functions with only half the combinatoric labor.
To be definite we take $`i=1,j=2`$. Other cases can be trivially obtained from this one. Radiative corrections to the 2-point correlator $`\mathrm{tr}Z_1^JZ_2(x)\mathrm{tr}\overline{Z}_2\overline{Z}_1^J(y)`$ are obtained in two stages. In the first step we calculate the effective operator obtained from the Wick contractions of $`Z(x)`$ and $`\overline{Z}(z)`$ fields in one factor of the F-term vertex in (7.1),
$$\mathrm{tr}(Z_1^JZ_2)\mathrm{tr}\left(T^a(\overline{q}\overline{Z}_2\overline{Z}_1\frac{1}{\overline{q}}\overline{Z}_1\overline{Z}_2)\right).$$
(3.19)
We ignore propagator factors temporarily, but keep track of the $`q`$ dependence. To simplify the calculations we keep only leading terms in $`N`$ from the splitting/joining rules (7.41) and we drop Wick contractions corresponding to nonplanar diagrams. It is quite straightforward to obtain the effective operator
$$𝒪=(\overline{q}\frac{1}{\overline{q}})N\mathrm{tr}(Z_1^{J1}a)$$
(3.20)
In the second step we contract with the trace factor in the conjugate operator, i.e.
$$(Z_1^{J1}a)(a\overline{Z}_1^{J1})=(Z_1^{J1}\overline{Z}_1^{J1})=N^J$$
(3.21)
Finally we put things together, starting with the free-field term containing $`J+1`$ propagators and a factor $`N^{Z+1}`$ from processing the traces. We then add the above result for the interaction with propagators restored and the regulated integral (7.37) used. The correlator is then
$$\mathrm{tr}Z_1^JZ_2(x)\mathrm{tr}\overline{Z}_2\overline{Z}_1^J(y)=\frac{N^{J+1}}{(4\pi ^2)^{J+1}}\frac{1}{(xy)^{2(J+1)}}[1\gamma \mathrm{ln}(M^2(xy)^2)]$$
(3.22)
with anomalous dimension $`\gamma =|h|^2N|q\frac{1}{q}|^2/8\pi ^2`$. Note that $`\gamma `$ vanishes when $`q\pm 1`$ which is the limit of the undeformed $`𝒩=4`$ theory.
The leading $`N`$ approximation is a major simplification for $`J>1`$ and it is appropriate for comparison with gravity results using AdS/CFT. Of course it gives an incomplete answer in the field theory.
### 3.4 The special case tr$`(Z_iZ_j)`$ for $`ij`$
In this case it is easy to include terms of all orders in $`N`$. The effective operator is
$`𝒪`$ $`=`$ $`\mathrm{tr}(Z_iZ_j)\mathrm{tr}\left(T^a(\overline{q}\overline{Z}_j\overline{Z}_i{\displaystyle \frac{1}{\overline{q}}}\overline{Z}_i\overline{Z}_j)\right)`$ (3.23)
$`=`$ $`(\overline{q}1/\overline{q})[{\displaystyle \frac{N^21}{N}}(1{\displaystyle \frac{1}{N}})(T^a)]`$ (3.24)
which vanishes in $`SU(N)`$, indicating that to lowest order the operator behaves as a chiral primary.
Of course, this could be an accident of lowest order perturbation theory, and higher loop corrections should be studied. At 3-loop order there are many contributing Feynman diagrams including both quartic and Yukawa interactions from (7.1), so this appears to be a difficult problem. Supergraph methods, as used in , may be advantageous.<sup>5</sup><sup>5</sup>5In it has been shown that the anomalous dimension of tr$`(Z_iZ_j)`$ vanishes to 3-loop order.
It is curious to repeat the lowest order calculation for gauge group $`U(N)`$, the effective operator becomes
$$𝒪=(q\frac{1}{q})N^{\frac{3}{2}}\delta ^{a0}.$$
(3.25)
The 2-point function then takes the form (3.22) with $`\gamma =|h|^2N|q\frac{1}{q}|^2/8\pi ^2`$ and $`J=1`$. So the operator acquires anomalous dimension in this case.
### 3.5 $`(1,1,1)`$ operator
In this sector we expect that there is one linear combination $`\mathrm{tr}(Z_1Z_2Z_3)+\alpha \mathrm{tr}(Z_1Z_3Z_2)`$ which has vanishing anomalous dimension. We apply the splitting joining rules in (7.41) to work out the effective operators
$`𝒪_{123}=\mathrm{tr}(Z_1Z_2Z_3)\mathrm{tr}\left(T^a(\overline{q}\overline{Z}_2\overline{Z}_1{\displaystyle \frac{1}{\overline{q}}}\overline{Z}_1\overline{Z}_2)\right)`$
$`=[N\overline{q}{\displaystyle \frac{2}{N}}(\overline{q}{\displaystyle \frac{1}{\overline{q}}})]\mathrm{tr}(Z_3T^a)`$
$`𝒪_{132}=\mathrm{tr}(Z_1Z_3Z_2)\mathrm{tr}\left(T^a(\overline{q}\overline{Z}_2\overline{Z}_1{\displaystyle \frac{1}{\overline{q}}}\overline{Z}_1\overline{Z}_2)\right)`$
$`=[{\displaystyle \frac{N}{\overline{q}}}{\displaystyle \frac{2}{N}}(\overline{q}{\displaystyle \frac{1}{\overline{q}}})]\mathrm{tr}(Z_3T^a)`$
One must add the similar contributions from the other two cyclic permutations in the $`F`$-term Lagrangian. These give contributions which differ from (3.5) only by the replacements $`\mathrm{tr}(Z_3T^a)\mathrm{tr}(Z_1T^a)`$ and $`\mathrm{tr}(Z_3T^a)\mathrm{tr}(Z_2T^a)`$.
The 1-loop anomalous dimension is obtained by contracting the effective operators for the linear combinations $`\mathrm{tr}(Z_1Z_2Z_3)+\alpha \mathrm{tr}(Z_1Z_3Z_2)`$ with their conjugates. It is clear that the anomalous dimension vanishes if and only if the operator itself vanishes. This fixes the complex coefficient $`\alpha `$. For arbitrary $`q`$ and $`N`$ one obtains the following BPS operator:
$$𝒪^{(1,1,1)}=\mathrm{tr}(Z_1Z_2Z_3)+\frac{(N^22)\overline{q}^2+2}{N^22+2\overline{q}^2}\mathrm{tr}(Z_1Z_3Z_2).$$
(3.27)
To leading order in $`N`$, the coefficient $`\alpha \overline{q}^2`$.
### 3.6 $`(2,2,0)`$ operator
It is straightforward to use the effective operator method to investigate the linear combination $`\mathrm{tr}(Z_1Z_1Z_2Z_2)+c\mathrm{tr}(Z_1Z_2Z_1Z_2)`$. To leading order in $`N`$, the effective operator is
$$𝒪=N[(2c\overline{q}\frac{1}{\overline{q}})\mathrm{tr}(Z_1Z_2T^a)+(\overline{q}\frac{2c}{\overline{q}})\mathrm{tr}(Z_2Z_1T^a)].$$
(3.28)
The anomalous dimension vanishes to 1-loop order only if $`q^4=1`$, which is what we expect from the classification of chiral primary operators. There are two cases to consider:
i) $`q=\pm 1`$ with $`c=\frac{1}{2}`$. This means that we have the undeformed $`𝒩=4`$ SYM theory, and the operator is the second $`SU(3)`$ descendent of the primary $`\mathrm{tr}(Z_1^4)`$.
ii) $`q=\pm i`$ with $`c=\frac{1}{2}`$. This is the expected chiral primary chiral operator of type $`(2,2,0)`$ which occurs for the special value $`\beta =1/2`$. With $`1/N`$ corrections included, this operator becomes
$$𝒪^{(2,2,0)}=\mathrm{tr}(Z_1Z_1Z_2Z_2)\frac{1}{2}\left(\frac{N^28}{N^24}\right)\mathrm{tr}(Z_1Z_2Z_1Z_2).$$
(3.29)
### 3.7 BPS operators at large $`N`$
So far, we constructed some special members of the chiral primary operators. Our results above are valid for all values of $`N`$. It becomes cumbersome determine the chiral primaries that are composed of more canonical fields by the method of vanishing effective operator. However, one can easily work out the general form of these operators when $`N`$ is large, as below.
The $`(J_1,J_2,J_3)`$ operator is given as,
$$𝒪=\underset{\pi }{}c_\pi \mathrm{tr}(\pi Z_1^{J_1}Z_2^{J_2}Z_3^{J_3})$$
(3.30)
where $`\pi `$ is the sum over all distinct permutations modulo cyclicity of the trace<sup>6</sup><sup>6</sup>6It may be interesting to derive this definition of the chiral primaries from a $`qalgebra`$ method as in .. Determination of the the complex coefficients $`c_\pi `$, turns out to be easy in the large $`N`$ limit:
$$c_\pi =\frac{\overline{q}^{2k_\pi }}{s_\pi },$$
(3.31)
where $`k`$ is a positive integer and $`s`$ is a symmetry factor of the permutation.
To specify $`k_\pi `$ and $`s_\pi `$ we express the permutation $`\pi `$ in terms of the elementary exchanges (12), (23) and (31). We assign $`c_1=1`$ for the identity permutation. Then $`k_\pi `$ is obtained by introducing a factor $`\overline{q}^2`$ for each exchange. For example,
$$223311\overline{q}^4212313.$$
The symmetry factor is the number of repeated arrays in the permutation: e.g. for 213213 $`s=2`$, for 212121 $`s=3`$. Division by the symmetry factor is required for the cancellation among contributions to the effective operators.
Let us illustrate the basic mechanism for the cancelation of the one-loop radiative corrections to (3.30) with the special $`(n,n,0)`$ operators. They are chiral primary when $`\beta =m/n`$. These operators are conjectured to be dual to strings that rotate along the two contractible cycles in the deformed $`S^5`$ with momenta $`n_1=n_2=n`$ and winding numbers $`w_1=m`$, $`w_2=m`$ . Consider the F-term interactions of the following array of fields.
$$(Z_2\mathrm{}Z_2Z_1\mathrm{}Z_1)$$
(3.32)
The quartic interaction that involves the pair of fields in the middle of this array gives a contribution to the effective operator that is proportional to
$$\frac{1}{\overline{q}}(Z_2\mathrm{}T^a\mathrm{}Z_1)$$
in the limit $`N\mathrm{}`$. In case of the special BPS operator this contribution cancels out the contribution that comes from a term in (3.30) that is obtained from (3.32) by shifting the position of $`Z_2`$ one step right. This contribution is proportional to,
$$\overline{q}\left(\frac{1}{\overline{q}}\right)^2(Z_2\mathrm{}T^a\mathrm{}Z_1)$$
in the limit $`N\mathrm{}`$ so it cancels out the first contribution above. Similarly, one can see that all of the F-term contributions are paired up in a manner similarly and cancel out provided that $`q^{2n}=1`$.
This last requirement can be seen as follows. There are $`n`$ $`Z_2`$ fields in the $`(n,n,0)`$ operator that pair up with their neighbor $`Z_1`$ fields to form an F-quartic vertex. Let us label these fields by $`a`$, $`b`$, $`c,\mathrm{}`$ as in the fig. 1. As can easily be seen from the figure the interactions of the first $`n1`$ $`Z_2`$ fields cancel out by the above mechanism. However in order to cancel the interactions of the $`n`$th field by the same mechanism one needs to impose the cyclicity condition $`q^{2n}=1`$. Note that this requirement coincides with the definition of the chiral ring.
We list several examples of the prescription (3.32). The case of $`n=2`$ is already discussed at the end of the previous section where it was simple enough to determine the operator for all orders in $`N`$. In the next case $`n=3`$ one has the following two special operators (for large $`N`$),
$`𝒪^{(3,3,0)}`$ $`=`$ $`\mathrm{tr}(Z_2Z_2Z_2Z_1Z_1Z_1)+q^2\mathrm{tr}(Z_2Z_2Z_1Z_2Z_1Z_1)+`$ (3.33)
$`+q^4\mathrm{tr}(Z_2Z_2Z_1Z_1Z_2Z_1)+{\displaystyle \frac{1}{3}}\mathrm{tr}(Z_2Z_1Z_2Z_1Z_2Z_1)`$
where $`q=e^{2i\pi /3}`$ and $`q=e^{4i\pi /3}`$. These two operators are related to each other by the relabeling of the fields $`Z_1Z_2`$.
Finally, for $`n=4`$, q-variation method results in the following special BPS operators (for large $`N`$),
$`𝒪^{(4,4,0)}`$ $`=`$ $`\mathrm{tr}(Z_2Z_2Z_2Z_2Z_1Z_1Z_1Z_1)+q^2\mathrm{tr}(Z_2Z_2Z_2Z_1Z_2Z_1Z_1Z_1)`$
$`+q^4\mathrm{tr}(Z_2Z_2Z_2Z_1Z_1Z_2Z_1Z_1)+q^6\mathrm{tr}(Z_2Z_2Z_2Z_1Z_1Z_1Z_2Z_1)`$
$`+q^4\mathrm{tr}(Z_2Z_2Z_1Z_2Z_2Z_1Z_1Z_1)+q^6\mathrm{tr}(Z_2Z_2Z_1Z_2Z_1Z_2Z_1Z_1)`$
$`+\mathrm{tr}(Z_2Z_2Z_1Z_2Z_1Z_1Z_2Z_1)+{\displaystyle \frac{1}{2}}\mathrm{tr}(Z_2Z_2Z_1Z_1Z_2Z_2Z_1Z_1)`$
$`+q^6\mathrm{tr}(Z_2Z_2Z_1Z_1Z_2Z_1Z_2Z_1)+{\displaystyle \frac{q^4}{4}}\mathrm{tr}(Z_2Z_1Z_2Z_1Z_2Z_1Z_2Z_1)`$
where the independant choices for $`q`$ are $`e^{i\pi /4}`$ and $`e^{i\pi /2}`$ and $`e^{i3\pi /4}`$.
It is straightforward to check that the F-term radiative corrections to the two-point functions of these operators vanish as $`N\mathrm{}`$. However we expect to obtain modifications for finite $`N`$.
It is straightforward but somewhat more complicated to show that the prescription (3.30) determines an operator with vanishing anomalous dimension for all values of $`J_1,J_2,J_3`$ which correspond to chiral primaries.
## 4 Three point functions of chiral primary operators
In this section we consider three-point correlators of the chiral primary operators listed in (3.15, 3.16) and their complex conjugates. We note that all three-point functions are extremal and can be schematically written as $`\overline{𝒪}𝒪_1𝒪_2`$ where $`𝒪_1`$ and $`𝒪_2`$ contain only $`Z_{1,2,3}`$ and $`\overline{𝒪}`$ contains only conjugate fields.
As discussed in Sec 3.1 above, the combinatoric arguments of imply that the $`D`$-term interactions cancel provided that one works on the fixed point locus where $`\gamma ^{(1)}=0`$.
We now argue that lowest order $`F`$-term interactions also vanish because the effective operator method gives the same operator which occurred in the computation of two-point functions, and that operator thus vanishes. The one-loop radiative correction is given by Wick contractions of the three point function with the F-term quartic vertex:
$$\underset{i,a}{}\overline{𝒪}\frac{W}{Z_i^a}\frac{\overline{W}}{\overline{Z}_i^a}𝒪_1^{J_1}(y)𝒪_2^{J_2}.$$
(4.34)
Wick contractions of $`\overline{𝒪}\frac{W}{Z_i^a}`$ give the same effective operator $`\overline{𝒪}_i^a`$ which occurs in the computation of the F-term corrections to the 2-point function of $`\overline{𝒪}`$. These corrections vanish if $`\overline{𝒪}`$ is chiral primary. Therefore the one-loop radiative correction to the three point function is absent.
These arguments generalize immediately to extremal $`n`$-point functions of all chiral primaries.
## 5 Conclusions
We have shown that lowest order calculations by the effective operator method confirm the assignment of chiral primary operators of and suggest that their two- and three-point correlation functions (and extremal correlators) have vanishing radiative corrections. It would be interesting to confirm this suggestion in further study and to ascertain the properties of $`\mathrm{tr}(Z_iZ_j)`$, with $`ij`$, which we have shown to be protected to lowest order although it was not previously recognized as a chiral primary. There are far fewer chiral primary operators in the $`\beta `$-deformed theory than in its undeformed $`𝒩=4`$ parent. But it is striking that their correlators appear to enjoy the same properties that have been established in the latter case.
## 6 Acknowledgments
The authors thank Juan Maldacena for much encouraging and useful correspondence during the course of this work. We also thank Ken Intriligator and Carlos Nuñez for useful discussions.
## 7 Appendix
### 7.1 Conventions
We use the following conventions:
$$\mathrm{tr}(T^aT^b)=\delta ^{ab},[T^a,T^b]=i\sqrt{2}f^{abc}T^c,$$
(7.35)
With these conventions, the action of the deformed theory can be written in the $`𝒩=1`$ component notation as follows:
$``$ $`=`$ $`\mathrm{tr}[{\displaystyle \frac{1}{4}}F_{\mu \nu }^2+{\displaystyle \frac{1}{2}}\overline{\lambda }D/\lambda +\overline{D_\mu Z^i}D_\mu Z^i+{\displaystyle \frac{1}{2}}\overline{\psi }^iD/\psi ^i+g(\overline{\lambda }\overline{Z}^iL\psi ^i\overline{\psi }^iRZ^i\lambda )`$
$`+{\displaystyle \frac{1}{4}}g^2[\overline{Z}^i,Z^i]^2]`$
$`+{\displaystyle \frac{h}{2}}\mathrm{tr}\left(q\overline{\psi }_2LZ_3\psi _1{\displaystyle \frac{1}{q}}\overline{\psi }_1LZ_3\psi _2\right)+{\displaystyle \frac{\overline{h}}{2}}\mathrm{tr}\left(\overline{q}\overline{\psi }_1R\overline{Z}_3\psi _2{\displaystyle \frac{1}{\overline{q}}}\overline{\psi }_2R\overline{Z}_3\psi _1\right)`$
$`+|h|^2\mathrm{tr}\left(T^a(qZ_2Z_3{\displaystyle \frac{1}{q}}Z_3Z_2)\right)\mathrm{tr}\left(T^a(\overline{q}\overline{Z}_3\overline{Z}_2{\displaystyle \frac{1}{\overline{q}}}\overline{Z}_2\overline{Z}_3)\right)+\mathrm{cyclic}`$
where $`D_\mu Z=_\mu i\frac{g}{\sqrt{2}}[A_\mu ,Z]`$ and similarly for the fermions. The first two lines contain D-type interactions, and the last two lines give the F-type interactions. Note that cylic permutations of the labels 123 must be added in the F-terms. It is the quartic F-type interaction of the scalars that plays the major role in our computations. The splitting/joining rules could be used to simplify this term, but the factored form is the most convenient for the “effective operator” method used in most of our calculations.
The Lagrangian above is valid for Euclidean signature in which we consider Wick contractions with $`e^{S_{int}}`$. The scalar propagator is $`Z(x)\overline{Z(y)}=1/4\pi ^2(xy)^2`$. The “bubble graph” with 2-quartic vertices leads to the space-time integral
$$\frac{1}{(2\pi )^4}\frac{d^4z}{z^4(zx)^4}=\frac{\mathrm{ln}M^2x^2}{8\pi ^2x^4}.$$
(7.37)
This result is obtained by (partial) differential regularization, see .
Our methd of calculation is based on the splitting/joining rules for traces . We present these in a compact notation in which the generators of the fundamental of $`SU(N)`$ are replaced by their index values, i.e. $`T^aa,a=1,\mathrm{},N^21`$, and the trace of an arbitrary $`N\times N`$ matrix is denoted by $`\mathrm{Tr}(M)(M)`$. The following rules can then be used to evaluate traces and products thereof which involve sums over repeated indices:
$`\begin{array}{cccccc}\hfill (MaM^{}a)& =& (M)(M^{})\frac{1}{N}(MM^{})\hfill & \hfill (ab)& =& \delta ^{ab}\hfill \\ \hfill (Ma)(aM^{})& =& (MM^{})\frac{1}{N}(M)(M^{})\hfill & \hfill (a)& =& 0\hfill \\ \hfill aa& =& \frac{N^21}{N}I\hfill & \hfill (I)& =& N\hfill \end{array}`$ (7.41)
To treat $`U(N)`$ we add the generator $`T^0=I/\sqrt{(}N)`$. The relations in the first column of (7.41) are valid for $`U(N)`$ if the $`\frac{1}{N}`$ terms are dropped. The relations in the second column are valid except for the change $`(a)\sqrt{N}\delta ^{a0}.`$
|
warning/0506/astro-ph0506738.html
|
ar5iv
|
text
|
# Survey for Transiting Extrasolar Planets in Stellar Systems. II. Spectrophotometry and Metallicities of Open Clusters
## 1 Introduction
Galactic open clusters occupy a wide range of ages, sizes, locations in the Galaxy, initial mass functions, and metallicities, and are found throughout the Galactic disk. Thus, open clusters are excellent tracers of Galactic disk properties and give insight into the formation of the disk (Janes & Adler, 1982; Janes & Phelps, 1994; Friel, 1995; de La Fuente Marcos & de La Fuente Marcos, 2004). Unfortunately, the fundamental properties of open clusters are difficult to determine. Current estimates of these properties vary widely; there is only marginal agreement in the literature about the age, metallicity, reddening, or Galactocentric distance of any given open cluster. Part of the disparity between estimates is due to the variety of methods used to determine these quantities. The published metallicity estimates of an open cluster are especially prone to disagreement because each technique has different sensitivities to metallicity patterns, and it is likely that none of the methods, with the possible exception of high-resolution spectroscopy, commonly used to estimate stellar metallicity is measuring the “true” metallicity of the stars. Additionally, the relationship between the many different techniques is often unknown. The past few decades have yielded significant advances in the study of open clusters, but there is still much work to be done toward determining the fundamental properties of the more than 1500 currently known Galactic open clusters.
Nonetheless, open clusters are extremely useful objects to study because each cluster represents a homogeneous set of stars. All stars in an open cluster form at the same time and in the same circumstances, and thus are expected to have the same age, metallicity, and Galactocentric distance. For this reason, open clusters are good testbeds for many types of Galactic studies. For example, they may be ideal objects in which to search for extra-solar planets (e.g., Gonzalez, 1997). Furthermore, open clusters can be metal-rich objects. Metal-rich environments have recently received attention due to the suggestion (e.g., Santos et al., 2000; Gonzalez & Laws, 2000) that extra-solar planets form more readily around metal-rich stars. If this is the case, metal-rich open clusters may be excellent places to search for extra-solar planets. This is the primary motivation for this work; planet searches will be conducted in metal-rich open clusters identified in this study. These programs will be carried out by the Survey for Transiting Extrasolar Planets in Stellar Systems (STEPSS) program (e.g. Burke et al., 2003).
In this paper we present new, accurate metallicity estimates for seven open clusters that were presumed to be metal-rich based on literature searches. We have obtained moderate resolution spectroscopy of giants in the target-clusters and use spectrophotometric line indices to produce metallicity estimates for the target-clusters.
## 2 Observational Procedure
In this work we follow the method of Friel (1987) to determine metallicities using spectrophotometric line indices as proxies for high-resolution metallicity estimates. This procedure was adopted in several subsequent papers, including Thogerson et al. (1993), Friel & Janes (1993), and most recently Friel et al. (2002). Here we quantitatively assess the precision of this methodology, which is described below.
### 2.1 Sample Selection
The target-clusters were selected as a precursor to a search for extra-solar planets. Clusters to be observed were chosen on the basis of a high metallicity estimate cited in the literature, as well as their location in the Galaxy. We used the Lynga (1987) database to select target-clusters, with selection criteria of \[Fe/H\] $`>0.2`$ so as not to exclude metal-rich clusters with large errors in their metallicity determination. We required the target-clusters to have distances $`>`$ 300 pc, so that the stars would be bright enough to observe on our 2-m class telescopes, and to have declinations $`\delta `$ (1950) $`>15^o`$, so as to be observable from the northern hemisphere. We did not put any constraints on the ages of the clusters; indeed, the selected open clusters span a wide range of ages. These criteria yielded a set of 26 candidate open clusters.
These target-clusters were observed with the MDM “Echelle” CCD imager at the 1.3 m McGraw-Hill telescope of the MDM Observatory<sup>1</sup><sup>1</sup>1See http://www.astro.lsa.umich.edu/obs/mdm/ in a preliminary observing run. During this run we obtained images of the target-clusters as well as instrumental color-magnitude diagrams (CMDs) of each cluster. We use these images and instrumental CMDs to select target-cluster stars.
Using these preparatory images, we further required the selected open clusters to be optically distinguishable as a cluster in the 1.3 m images (the imager has a 17′field of view). This constraint ensures a large number of observable stars, both in the clump (for this work) and in the cluster in general (for use in the planet searches). The clusters were also explicitly required to have several giant stars in or near the clump, based on the instrumental CMD, in order to determine accurately the metallicity of each target-cluster. The seven target-clusters selected from these criteria are NGC 1245, NGC 2099, NGC 2324, NGC 2539, NGC 2682, NGC 6705, and NGC 6819. We select giants with $`(BV)_01`$ (typically red clump giants) in each of the target-clusters. For reference, the names of the stars in each target-cluster used here are those given in the WEBDA open cluster database<sup>2</sup><sup>2</sup>2See http://obswww.unige.ch/webda/. We also give the references for the identifications for each cluster in §4.2
We also observed stars of known metallicity to use as calibrators. We took calibration stars mainly from the survey of Friel (1987), who presents a large sample of bright field stars and cluster giants with accompanying measured indices and previously derived metallicities. We took bright metal-poor calibration stars from Cottrell & Sneden (1986), while additional giant stars from the old open cluster NGC 2682 were taken from Friel & Janes (1993). In order to minimize calibration errors between target-cluster stars and the calibrators, we selected calibration stars from these papers that have a color of $`BV=1\pm 0.1`$ mag, roughly the same (dereddened) color as the target-cluster stars. Additionally, we observed a few calibration stars with $`BV<0.9`$, in order to ensure some overlap with the color of the target-cluster stars.
### 2.2 Observations and Data Reduction
The spectral data were obtained at the MDM Observatory using the CCDS spectrograph<sup>3</sup><sup>3</sup>3See http://www.astronomy.ohio-state.edu/MDM/CCDS/. We obtained one night of data (2002 January 31) on the 2.4 m Hiltner telescope, while the other seven nights of data (during 2002 March 21 - April 01) were obtained on the 1.3 m McGraw-Hill telescope. The spectra cover the wavelength range 3800-5400 Å with the 350 l/mm (1.33 Å/pix) grating. The full-width at half-maximum (FWHM) of an unresolved emission line is $``$3 pixels, yielding a spectral resolution ($`\lambda `$/$`\mathrm{\Delta }\lambda )`$ of $``$1150. All spectra on both telescopes use an 87 $`\mathrm{\mu m}`$ slit for uniformity. This slit size translates to 1″ on the 2.4 m and 1$`\stackrel{}{\mathrm{.}}`$8 on the 1.3 m. Most spectra have formal signal-to-noise ratios (S/N) $`\stackrel{>}{}`$100 per pixel.
We observed the target-cluster stars interspersed with calibration stars of known metallicity. Thirty-five observations of 33 target-cluster stars and 56 observations of 38 calibration stars were made. Figures 17 show the reduced spectra of all of the stars observed in each target-cluster. Figure 8 shows a selected sample of the calibration stars observed. Multiple observations of some stars were included in order to ensure reproduction of Friel’s method, to check repeatability of the measurements between telescopes, and also to have many measurements on which to base the calibration.
We reduced the data using standard IRAF<sup>4</sup><sup>4</sup>4IRAF is distributed by the National Optical Astronomy Observatory, which is operated by the Association of Universities for Research in Astronomy, Inc., under cooperative agreement with the National Science Foundation. routines following the usual practice of overscan subtraction, division by a flat field, and extraction of the spectra. Flat fields and comparison lamp spectra were obtained using lamps within the instrument. We correct the spectra for atmospheric extinction using the standard KPNO (Kitt Peak) extinction curves provided by IRAF.
On each night we observed a spectrophotometric standard star from Strom (1977). The application of only one standard per night (regardless of the airmass of the observations) may introduce a small color effect into the spectra since the airmass has a color term and we do nothing to account for the airmass of the observations. Due to the nature of the measurements (indices with continuum bands on either side of the lines), this is not a significant concern. No attempt was made to do spectrophotometry, i.e., we did not align the slit with the parallactic angle at each telescope pointing. This seems to have had little effect on our results.
### 2.3 Measurement of Spectrophotometric Indices
Several recent papers (e.g. Strader & Brodie, 2004) use spectrophotometric indices as an estimator for the metallicity and other properties of individual stars. The indices are measured for both the program stars as well as calibration stars with well-known metallicity; the calibration star data are fit to obtain a calibration for the metallicities of the target-cluster stars. This practice is desirable because the indices may be measured with moderate resolution spectra to obtain a relatively accurate (i.e., generally good to $``$ 0.15 dex) metallicity estimate for the star, and obviates the need for difficult-to-obtain high-resolution spectroscopy. Lick indices (Burstein et al., 1984) are commonly used in this technique and are applied to extragalactic sources for which it is often impossible to obtain high S/N, high-resolution spectra. Friel (1987) slightly altered the Lick indices to account for differences between Galactic and extragalactic sources and applied them to giant stars in the Galaxy.
The indices are defined as a ratio between a bandpass that incorporates several strong metal lines and continuum measurements on either side of this bandpass that include little flux from metal lines. The final measured indices (several per star) may then be combined to yield a metallicity estimate for each star. For example, the Fe4680 index is defined as
$$\mathrm{Fe4680}=2.5log\frac{_{4636.00}^{4723.00}F_\lambda 𝑑\lambda /87\AA }{_{4606.00}^{4636.00}F_\lambda 𝑑\lambda /30\AA +_{4736.00}^{4773.00}F_\lambda 𝑑\lambda /37\AA }.$$
(1)
The line and continuum regions of the indices used in this work are given in Table 1, as well as the species measured by each index. This table is reproduced from Friel (1987).
Figure 9 shows a sample moderate resolution spectrum obtained on the MDM 2.4 m telescope. The central bandpasses of each measured index are shown across the bottom of the figure. Figure 10 shows an enlarged section of the same spectrum, showing the central bandpass and accompanying continuum bandpasses for the Fe4680 index.
## 3 Analysis
### 3.1 Metallicities of Calibration Stars
Although the calibration stars are selected from Friel (1987), Cottrell & Sneden (1986), and Friel & Janes (1993), many of these stars have more recent metallicity estimates than those cited in these original references. In order to minimize errors introduced into our analysis from the calibration stars, we selected only field giants from Friel (1987) that had metallicity estimates from a single source. Specifically, we chose metallicity estimates from the Cayrel de Strobel et al. (2001) (hereafter CdS) catalog. This catalog is a compilation of previously published atmospheric parameters determined from high-resolution, high S/N spectra from a variety of sources. The 2001 version of the catalog contains bright F, G, and K stars in the Galactic field as well as in certain associations. We selected thirty-eight metallicity standards from Friel (1987) that had at least one metallicity determination in the CdS catalog; we obtained 56 measurements of these 38 stars.
For stars with more than one entry in the CdS catalog we combine the available entries as follows. If there are two entries we average the two quoted metallicities. For stars with three or more entries, we discard metallicity measurements more than one standard deviation away from the median of all the measurements. We then average the remaining measurements. When only one measurement was present in the catalog, we adopt it.
Many of the entries in CdS do not have associated error estimates; those that do are of order 0.1 dex. We estimate errors by finding the standard deviation of the remaining metallicities once spurious measurements are discarded as described above. For stars with only one or two measurements, or those for which only two measurements remained, we assume an error of 0.1 dex. However, this may be an overestimate of these errors, as our method reproduces metallicities of most of these field stars to higher precision (see §4.1).
### 3.2 The Indices: Reproducibility and Comparison to Previous Index Measurements
Table 2 gives the measured indices of the calibration stars observed in this work along with their $`V`$ mag and $`BV`$ colors (from the Hipparcos Catalog, ESA, 1997) and metallicities (derived from the CdS database, as described above). Table 3 shows these data for the target-cluster stars.
We obtained all of the spectra with the CCDS spectrometer, and we made most of the observations with the MDM 1.3 m telescope. Since all measurements were obtained with the same instrument, we do not expect large systematic differences between the data obtained on each telescope. Several stars were observed on both of the MDM Observatory telescopes. We compared the indices for the stars observed on both telescopes; the results are shown in Table 4. The average difference between the target-cluster stars measured on the two telescopes is 0.00 and the standard deviation between the stars measured on each telescope is 0.03. This difference is well within the errors of the measurements and suggests no systematic difference in the indices measured on the two telescopes.
We also observed some stars twice on the 1.3 m run. We compare the indices for the stars observed twice on the 1.3 m run, as well as the indices of the stars observed multiple times over both runs. No stars were observed twice on the 2.4 m run. The results are again shown in Table 4. Note that in both cases the average difference between any two measurements is 0.00. The standard deviation for the stars measured twice during the 1.3 m run is 0.04 while that for all stars measured twice (over both runs) is 0.03.
Finally, we compare our measured indices for the calibration stars to those given in Friel (1987). For all of the stars that were measured both by Friel (1987) and in this work, we found the difference between Friel’s value of each index and ours and computed the average and standard deviation for all of the indices. Our indices reproduce Friel’s well; the average difference of the indices is $`0.012`$ and the standard deviation is 0.054.
### 3.3 Reddening
In order to determine quantitatively if the interstellar reddening to the target-clusters affects our results, we simulated the effects of reddening on a sample spectrum. We used IRAF utilities to redden artificially the spectra and measured the effect of reddening on each of the indices. We applied an E(B-V) of 1.0 (much more extreme than any of the target-clusters should actually have; see footnotes to Table 3), and re-calculated the index values. For some indices the difference in the index between the raw spectrum and the dereddened spectrum was higher than the others (Ca4226 and G-band were 0.05 and 0.11 respectively). For the remaining nine indices the difference was 0.03 or less (roughly the precision of the index measurements). However, none of the target-clusters have nearly this high of a reddening value (see the footnotes to Table 3). At the approximate reddening values applicable to our target-clusters, the effect of reddening was well below the other sources of error. We conclude that reddening does not significantly affect our results.
## 4 Results
### 4.1 Fit to Calibrators
We begin by writing the predicted metallicity as a linear function of the 11 observed spectrophotometric line indices $`f_i`$ ($`i=1\mathrm{}11`$), $`[\mathrm{Fe}/\mathrm{H}]_{\mathrm{pred}}=_{i=0}^{11}a_if_i`$, where $`f_0=1`$. We determine the coefficients $`a_i`$ by making a linear fit to 56 stars that have spectroscopically determined metallicities from CdS of \[Fe/H\]$`>1.5`$.
We determine which of these 11 spectrophotometric line indices are redundant as follows. In the fit we arbitrarily assign each index an error of unity. Since there are $`5612=44`$ degrees of freedom (dof), the resulting $`\chi ^2/`$dof of the fit gives the (square of the) scatter about the fit. We then try removing each of the 12 parameters in succession and find the parameter that causes $`\chi ^2`$ to increase by the least. Since the number of dof has also increased (by 1), the $`\chi ^2/`$dof may have increased or decreased. If it has decreased, we regard the parameter as redundant, remove it, and then try removing each of the remaining 11 parameters. We continue this process until $`\chi ^2/`$dof begins to increase.
We derive the metallicities of the target-clusters using the 56 calibrator stars. We find that the target-cluster stars all have \[Fe/H\]$`>0.7`$. Further, we find that thirteen calibrator stars have extremely high scatter in this fit ($`\sigma >`$0.15), possibly due to large errors in their CdS metallicity determinations. We therefore repeat the entire procedure but restricted to the sample of 43 calibrator star measurements that have \[Fe/H\]$`>0.7`$, discarding the high-scatter stars. In this case, we find 5 indices are removed (Fe4530, Fe5011, Mg, Fe5270, Fe5335). The equation used to combine the indices and produce the metallicities of the stars is
$`[m/H]=`$
$`3.180415+4.398434\times \text{Fe4065}+1.105911\times \text{CN4216}+`$
$`2.480848\times \text{Ca4226}+0.725162\times \text{G-band}+`$
$`2.835423\times \text{Fe4680}+1.620898\times \text{Fe4920}.`$
The coefficients of the indices are also presented in Table 5.
The scatter about the relation is $`\sqrt{\chi ^2/\mathrm{dof}}=0.0709`$. This number reflects the quadrature sum of the root-mean-square (rms) error in the spectroscopic metallicities and the intrinsic scatter of the method.
Even though the target-cluster giants are selected to have approximately the same $`BV`$ color as the calibration stars, we find that the derived metallicities of the target-cluster giants have a non-negligible dependence on color. This effect is due to the fact that some of the stars we observed were not in the red giant clump but rather were more evolved stars on the giant branch. Thus these stars have different surface temperatures and gravities than the rest of our clump stars. This effect is readily calibrated by removing the color dependence of the metallicity.
We compare metallicity residual (relative to CdS metallicity) to $`BV`$ color for the five calibrator stars and five target-cluster stars in NGC 2682 (the target-cluster with the most well-determined metallicity in the literature). The $`BV`$ colors of these stars are dereddened using E(B-V)=0.05 (Montgomery et al., 1993). For the five program giants (with no literature metallicity values) we assume the average literature value for the cluster metallicity of \[Fe/H\]=$`0.05`$. The equation of the fit is
$$[Fe/H]_{CdS}[m/H]_{thispaper}=1.4937\times (BV)1.5858.$$
(3)
The derived color correction is then applied to the 35 target-cluster stars individually to obtain the final metallicity of each star. We assume that the spread in ages of the target-clusters has a negligible effect on this fit. More explicitly, we assume the errors associated with this assumption are smaller than our other sources of error. The data and the derived fit are shown in Figure 11.
Figure 12 compares our derived metallicities (after applying the color correction) to those of CdS. The scatter in this relation is $``$0.08 dex, which is only 0.01 dex larger than the scatter in the derivation of the original relation. We therefore estimate the overall error in the method to be $``$0.08 dex. This implies that the technique of applying a color correction to the derived metallicities produces acceptable results.
Note that we use the term \[m/H\] to refer to our metallicity estimates. Since each of the spectrophotometric line indices is selected to incorporate several metal absorption lines, our method must measure some aspect of the star’s metallicity. But since the indices comprise several different metal species, this method does not necessarily measure only the iron abundance of the star.
### 4.2 Cluster Metallicities and Comparison to Previous Determinations
Table 6 shows the metallicities determined for each of the target-cluster stars, the mean metallicities of each target-cluster, and the scatter in the target-cluster star metallicities. We report the error for each target-cluster as the error in the mean of the target-cluster giant stars: the error in the method (0.08 dex) divided by the square root of the number of giant stars measured in each target-cluster.
Note that our method differs in a significant way from previous papers using spectrophotometric line indices to derive metallicities: we combine all of the measured indices at once using $`\chi ^2`$ analysis, whereas others (e.g., Friel & Janes, 1993; Thogerson et al., 1993; Friel et al., 2002) use each index individually to derive a metallicity for each star, then average the derived metallicities together. Our analysis indicates that several of the indices provide no additional sensitivity to metallicity for the stars in our sample, and we do not use them at all.
Below we discuss the application of our fit to each target-cluster individually and compare the derived metallicity to existing values in the literature. Most of the open clusters studied here do not have a large number of previous metallicity determinations.
A new catalog of open clusters (Dias et al., 2002) improves upon and replaces the Lynga (1987) database and includes metallicity estimates for all of our target-clusters. Although the on-line catalog does not include references for the metallicity estimates of the clusters, Dias (2005, private communication) has provided the references used in the catalog. We cite the references given to us by Dias for each of the clusters in the discussion below.
#### 4.2.1 NGC 1245
NGC 1245 has a metallicity of \[m/H\]=$`0.14\pm `$0.04 derived from seven measurements of six stars. The scatter in the derived metallicities is 0.09 dex, close to what we expect based on the error in the method. The identifiers used in Table 3 are taken from Chincarini (1964). The cluster has a well-defined red clump and all target-cluster stars were taken from the clump. We observed seven of the 27 clump giants from out instrumental CMD of NGC 1245. Unfortunately, WEBDA does not give membership probabilities for these stars.
NGC 1245 is studied extensively by Burke et al. (2004) (hereafter referred to as Paper I); new photometry is reported and fundamental properties of the cluster are derived, including a new metallicity value based on color-magnitude diagram profile fitting. It was selected as the first of several clusters for a targeted transiting planet search by the STEPSS program. Paper I reports a metallicity for NGC 1245 of \[Fe/H\]=$`0.05\pm `$ 0.03 (statistical) $`\pm `$ 0.08 (systematic). This value is derived from detailed isochrone fitting, using a $`\chi ^2`$ fit to accurate photometric observations to derive physical parameters of the cluster and quantifying the systematic errors in the method. This method yields perhaps the most reliable derivation of cluster parameters of any of those referred to below, with the possible exception of the high-resolution spectroscopy determinations.
It should be noted that in §3.1 of Paper I a best-fit value of the total-to-selective extinction $`R_V`$=2.3 is derived, yielding a metallicity of \[Fe/H\]=$`0.26`$. This value was in fact not used in the determination as it was assumed to be too low and the fiducial value of $`R_V`$=3.2 is adopted. Our derived metallicity agrees equally well with both determinations of Paper I. The assumed total-to-selective extinction may often be a large source of error in the determination of fundamental cluster parameters using isochrone fitting methods, since $`R_V`$ is not 3.2 uniformly throughout the Galaxy (see, e.g., Gould et al., 2001).
Wee & Lee (1996) also study the metallicity of NGC 1245 with Washington photometry. They derive a metallicity of \[Fe/H\]=$`0.04\pm `$0.05. Gratton (2000) cite a value of \[Fe/H\]=$`+0.1\pm `$0.15, derived by applying corrections based on high-resolution spectroscopy to low-resolution metallicity determinations.
The metallicity of NGC 1245 derived here, \[m/H\]=$`0.14\pm 0.04`$, is somewhat lower than previously published metallicities for this cluster. In particular, our metallicity for NGC 1245 reproduces the value cited in Paper I to only $``$2 sigma. We agree with the Gratton (2000) metallicity to 1.5 sigma and the Wee & Lee (1996) metallicity to 1.6 sigma.
#### 4.2.2 NGC 2099
The metallicity of NGC 2099 is \[m/H\]=$`+0.05\pm `$0.05 and is derived from measurements of eight stars. The metallicity determinations of the target-cluster stars in NGC 2099 have a scatter of 0.14 dex, about twice what we expect. Target-cluster names in Table 3 are taken from van Zeipel & Lindgren (1921). We observed eight of the 23 clump giants in this cluster’s instrumental CMD; all of the target-cluster giants in NGC 2099 are taken from the well-populated clump. The observed stars all had membership probabilities of greater than $`75\%`$, according to WEBDA.
NGC 2099 is a relatively unstudied cluster with few metallicity measurements. It is also the second cluster (after NGC 1245) to be observed by the STEPSS transiting planet search. Twarog et al. (1997) are some of the few to have studied NGC 2099. They use DDO photometry to derive \[Fe/H\]=+0.089$`\pm `$0.146. Our value of \[m/H\]=$`+0.05\pm `$0.05 is consistent with the metallicity cited by Twarog et al. (1997) to 0.2 sigma.
#### 4.2.3 NGC 2324
We derive a metallicity for NGC 2324 of \[m/H\]=$`0.06\pm `$0.04 from three measurements of four stars. The scatter in metallicity determinations among the stars observed in NGC 2324, 0.07 dex, is close to the expected level given the error in the method. Identifiers for these target-cluster stars are from Piatti et al. (2004). This cluster has a relatively well-defined clump; our target-cluster stars were selected from the seven giants in the clump. WEBDA reports very low membership probabilities for two of these stars; 0850 is given as $`35\%`$ and 1006 is $`11\%`$. The third star, 1552, does not have a membership probability reported in WEBDA. These numbers are surprising, since our results have low scatter and the stars fall within the region of the clump in the CMD.
Friel et al. (2002) studied NGC 2324 with the spectrophotometric line index method used in this paper. They report a metallicity for NGC 2324 of \[Fe/H\]=$`0.15\pm `$0.16. Others have also studied this cluster: Piatti et al. (2004) use their own reddening data with the Washington photometric data of Geisler et al. (1992) to derive a metallicity of \[Fe/H\]=$`0.31\pm `$0.04. Geisler et al. (1992) themselves derive \[Fe/H\]$`1`$ for this cluster. Cameron (1985) uses UV excess methods to derive \[Fe/H\]=-0.163.
We report \[m/H\]=$`0.06\pm `$0.04 for NGC 2324. Our derived metallicity is consistent with the Friel et al. (2002) value to 0.5 sigma and with the Piatti et al. (2004) value to 4.4 sigma.
#### 4.2.4 NGC 2539
The metallicity of NGC 2539 is \[m/H\]=$`0.04\pm `$0.03, derived from four stars. The standard deviation of the metallicity estimates of the cluster stars observed in NGC 2539 is 0.05 dex. We use the Lapasset et al. (2000) identifiers for the target-cluster stars in this cluster. Our instrumental CMD of NGC 2539 showed a small clump with only a few stars; we observed all four of these clump stars. WEBDA does not report membership probabilities for this cluster, although due to the low scatter in metallicities we believe all of our target-cluster stars to in fact be members of the cluster.
Clariá & Lapasset (1986) derive a metallicity of \[Fe/H\]=$`+0.2`$ from DDO data; they also derive a value of \[Fe/H\]=$`0.2\pm `$0.1 from iron lines. Twarog et al. (1997) publish an \[Fe/H\] of +0.137$`\pm `$0.062 determined from DDO photometry.
Our derived metallicity differs from that of Twarog et al. (1997) by 2.6 sigma. We agree with the Clariá & Lapasset (1986) metallicity derived from iron lines to 1.5 sigma.
#### 4.2.5 NGC 2682
We report a metallicity of \[m/H\]=$`0.05\pm `$0.02 for NGC 2682 from five measurements of four cluster stars. The scatter is about half what we expect given the error in the method, 0.04 dex. Fagerholm (1906) provides the identifications for the stars in this target-cluster. All of our target-cluster stars have greater than $`93\%`$ probabilities of being cluster members, according to WEBDA. NGC 2682 has the most evolved giant branch of any of the target-clusters; in this cluster we selected four stars from the instrumental CMD that were located along the giant branch.
NGC 2682 (M67) is one of the oldest known open clusters in the Galaxy; it is also by far the most-studied cluster in our sample, with many more metallicity determinations than discussed here. Previous studies agree that NGC 2682 has an approximately solar metallicity. Cayrel de Strobel (1990) uses high-resolution spectroscopy to obtain a metallicity for NGC 2682 of \[Fe/H\]=+0.039$`\pm `$0.09. Hobbs & Thorburn (1991) use high-resolution spectroscopy of weak iron lines to find \[Fe/H\]=$`0.04\pm 0.12`$. Both Friel & Janes (1993) and Friel et al. (2002) derive the metallicity of this cluster from indices used in this paper; they obtain metallicities of \[Fe/H\]=$`0.09\pm `$0.07 and \[Fe/H\]=$`0.15\pm `$0.05, respectively. Burstein et al. (1986) use a moderate resolution iron index to derive a metallicity of \[Fe/H\]=$`0.1_{0.04}^{+0.12}`$. Cameron (1985) uses the cluster’s UV excess to derive \[Fe/H\]=$`0.029`$. Twarog et al. (1997) use DDO photometry to find \[Fe/H\]=+0.000$`\pm `$0.092.
Our derived metallicity for NGC 2682 agrees well with most metallicities given in the literature. This metallicity agrees with Cayrel de Strobel (1990) to 0.96 sigma and with Hobbs & Thorburn (1991) to 0.1 sigma. Our determination agrees with other metallicities derived from spectrophotometric line index methods as well: we reproduce the Friel & Janes (1993) value to 0.5 sigma and the Friel et al. (2002) value to 1.8 sigma. We agree with the lower error of Burstein et al. (1986) to 1.1 sigma; we agree with Twarog et al. (1997) to 0.5 sigma.
#### 4.2.6 NGC 6705
The metallicity of NGC 6705 is \[m/H\]=+0.14$`\pm `$0.08, derived from four stars. The stars have a scatter of 0.16 dex. We use the identifiers of McNamara et al. (1977) for NGC 6705. These stars all had membership probabilities in WEBDA of $`98\%`$ or greater. This cluster has a very sparse instrumental CMD; we observed four of the six giants in the well-defined clump.
Cameron (1985) also studied NGC 6705. For this cluster they derive \[Fe/H\]=+0.070 with UV excess methods. Twarog et al. (1997) give \[Fe/H\]=+0.136$`\pm `$0.086, derived by shifting spectroscopic measurements made by Thogerson et al. (1993) onto a scale in common with other literature values. Thogerson et al. (1993) derive \[Fe/H\]=$`+0.21\pm `$0.09 using spectrophotometric line indices.
We derive a metallicity for NGC 6705 of \[m/H\]=+0.14$`\pm `$0.08. This agrees very well with previous estimates; it is within 0.03 sigma of the metallicity derived by Twarog et al. (1997), and agrees with Thogerson et al. (1993) to 0.6 sigma.
#### 4.2.7 NGC 6819
NGC 6819 has a metallicity of \[m/H\]=$`+0.07\pm `$0.12. This cluster has the largest scatter in metallicity determinations among cluster stars in our sample (0.24 dex). It is possible that some of the stars in our sample are not in fact cluster members. Unfortunately, WEBDA does not report membership probabilities for any of our stars in NGC 6819. Furthermore, the four stars selected in this cluster lie slightly below the giant branch of this evolved cluster. This may also account for some of the scatter in the derived metallicities of these stars. The identifiers for this target-cluster are taken from Auner (1974).
NGC 6819 is another old open cluster that has been studied more than the younger clusters. The cluster’s metallicity is derived in two papers using the index method: both Friel & Janes (1993) and Thogerson et al. (1993) derive \[Fe/H\]=$`+0.05\pm `$0.11. Bragaglia et al. (2001) use high-resolution spectroscopy to measure metal lines in three red clump stars in NGC 6819; they derive a metallicity of \[Fe/H\]=+0.09$`\pm `$0.03. Twarog et al. (1997) give \[Fe/H\]=+0.074$`\pm `$0.123, again obtained by shifting spectroscopic measurements made by Friel & Janes (1993) and Thogerson et al. (1993) onto a scale in common with other literature values.
Our metallicity estimate for NGC 6819 is \[m/H\]=$`+0.07\pm `$0.12, in excellent agreement with existing literature values. We agree with Friel & Janes (1993) and Thogerson et al. (1993) to 0.12 sigma, with Twarog et al. (1997) to 0.02 sigma, and with Bragaglia et al. (2001) to 0.16 sigma.
## 5 Conclusions
We have derived metallicity estimates for seven Galactic open clusters. We use moderate-resolution spectroscopy to estimate metallicities of giants in these clusters, calibrating our results with high-resolution metallicity determinations of field giants. We reproduce the high-resolution determinations to 0.08 dex. Our derived metallicities generally agree with those found in the literature; more importantly, we provide new metallicity estimates for several clusters that are not well studied.
This method is of interest because it requires little observing time, relatively small telescopes, and the results are arrived at quickly compared to more complicated methods (such as those used in Paper I). This method is also more robust than many photometric techniques, as it relies on quantitative spectroscopy and relative index measurements, not on the generally unpredictable photometricity of the night sky. We have demonstrated that it reproduces the metallicities of calibration stars of known metallicity well. The measured indices are easily reproducible on different telescopes and even, as evidenced by comparison of our indices to the published indices of Friel (1987), by different instruments and observers.
The metallicity estimates given here will allow for selection of open clusters in which to conduct future transiting extra-solar planet searches.
The authors wish to thank W. Dias for making an updated version of part of his open cluster catalog available to us prior to publication. We have made extensive use of the WEBDA database as well as the VizieR Service. This work was supported by NASA grant NAGS-10678. A. Gould was supported by grant AST 02-01266 from the NSF.
|
warning/0506/math-ph0506018.html
|
ar5iv
|
text
|
# 1 Skew-symmetric differential forms
## 1 Skew-symmetric differential forms
### 1.1 Some properties of manifolds.
A distinction of evolutionary skew-symmetric differential forms from exterior forms is connected with the properties of manifolds on which skew-symmetric forms are defined.
It is known that manifolds with structures of any type can serve as a basis of exterior differential forms. They have one common property, namely, locally they admit one-to-one mapping into the Euclidean subspaces and into other manifolds or submanifolds of the same dimension .
While describing the evolutionary processes in material systems (material media) one is forced to deal with manifolds which do not allow one-to-one mapping described above. These can be manifolds constructed of trajectories of material system elements (particles). Such manifolds, which can be called accompanying manifolds, are deforming variable manifolds. In real processes such varying deforming manifolds cannot be manifolds of the types for which the theory of exterior differential forms has been developed.
The differential forms defined on these manifolds are evolutionary ones. The coefficients of these differential forms and the characteristics of accompanying manifolds are interconnected and are varied as functions of evolutionary variables.
A difference of manifolds on which exterior and evolutionary forms are defined relates to the properties of metric forms of these manifolds.
Assume that on the manifold one can set the coordinate system with base vectors $`𝐞_\mu `$ and define the metric forms of manifold : $`(𝐞_\mu 𝐞_\nu )`$, $`(𝐞_\mu dx^\mu )`$, $`(d𝐞_\mu )`$. The metric forms and their commutators define the metric and differential characteristics of the manifold.
If metric forms are closed (the commutators are equal to zero), the metric is defined $`g_{\mu \nu }=(𝐞_\mu 𝐞_\nu )`$, and the results of translation over manifold of the point $`d𝐌=(𝐞_\mu dx^\mu )`$ and of the unit frame $`d𝐀=(d𝐞_\mu )`$ prove to be independent of the curve shape (the path of integration).
To describe the manifold differential characteristics and, correspondingly, the metric form commutators, one can use connectednesses \[2-4\]. If the components of metric form can be expressed in terms of connectedness $`\mathrm{\Gamma }_{\mu \nu }^\rho `$ , the expressions $`\mathrm{\Gamma }_{\mu \nu }^\rho `$, $`(\mathrm{\Gamma }_{\mu \nu }^\rho \mathrm{\Gamma }_{\nu \mu }^\rho )`$ and $`R_{\nu \rho \sigma }^\mu `$ are components of the commutators of the metric forms with zeroth- first- and third degrees. (The commutator of the second degree metric form is written down in a more complex manner , and therefore it is not presented here).
The closed metric forms define the manifold structure. And the commutators of metric forms define the manifold differential characteristics that specify the manifold deformation: bending, torsion, rotation, and twist. (For example, the commutator of the zeroth degree metric form $`\mathrm{\Gamma }_{\mu \nu }^\rho `$ characterizes the bend, that of the first degree form $`(\mathrm{\Gamma }_{\mu \nu }^\rho \mathrm{\Gamma }_{\nu \mu }^\rho )`$ characterizes the torsion, the commutator of the third -degree metric form $`R_{\nu \rho \sigma }^\mu `$ determines the curvature. (For manifolds with closed metric form of first degree the coefficients of connectedness are symmetric ones.)
Is is evident that the manifolds that are metric ones or possess the structure have closed metric forms. It is with such manifolds that the exterior differential forms are connected.
If the manifolds are deforming manifolds, this means that their metric form commutators are nonzero. That is, the metric forms of such manifolds turn out to be unclosed. The accompanying manifolds appearing to be deforming ones are the examples of such manifolds.
The skew-symmetric evolutionary differential forms whose basis are accompanying deforming manifolds are defined on manifolds with unclosed metric forms.
Below it will be shown that the specific properties of evolutionary skew-symmetric differential forms, which are defined on manifolds with unclosed metric forms, are connected with the properties of metric form commutators.
### 1.2 Specific features of skew-symmetric differential forms
The skew-symmetric differential form of degree $`p`$ ($`p`$-form) can be written down as
$$\omega ^p=\underset{\alpha _1\mathrm{}\alpha _p}{}a_{\alpha _1\mathrm{}\alpha _p}dx^{\alpha _1}dx^{\alpha _2}\mathrm{}dx^{\alpha _p}0pn$$
$`(1.1)`$
where the local basis obeys the condition
$$\begin{array}{c}dx^\alpha dx^\alpha =0\hfill \\ dx^\alpha dx^\beta =dx^\beta dx^\alpha \alpha \beta \hfill \end{array}$$
The differential of skew-symmetric differential form can be written in the form
$$d\omega ^p=\underset{\alpha _1\mathrm{}\alpha _p}{}da_{\alpha _1\mathrm{}\alpha _p}dx^{\alpha _1}dx^{\alpha _2}\mathrm{}dx^{\alpha _p}+\underset{\alpha _1\mathrm{}\alpha _p}{}a_{\alpha _1\mathrm{}\alpha _p}d(dx^{\alpha _1}dx^{\alpha _2}\mathrm{}dx^{\alpha _p})$$
$`(1.2)`$
where the second term is connected with the differential of the basis.
\[In further presentation a symbol of summing $``$ and a symbol of exterior multiplication $``$ will be omitted. Summation over repeated indices is implied.\]
The second term connected with the differential of the basis is expressed in terms of the metric form commutator. For differential forms defined on manifold with unclosed metric form one has $`d(dx^{\alpha _1}dx^{\alpha _2}\mathrm{}dx^{\alpha _p})0`$. And for manifold with closed metric form it is valid the following $`d(dx^{\alpha _1}dx^{\alpha _2}\mathrm{}dx^{\alpha _p})=0`$.
That is, for differential forms defined on the manifold with unclosed metric form the second term is nonzero, whereas for differential forms defined on the manifold with closed metric form the second term vanishes.
For example, let us consider the first-degree form $`\omega =a_\alpha dx^\alpha `$. The differential of this form can be written as
$$d\omega =K_{\alpha \beta }dx^\alpha dx^\beta $$
$`(1.2^{})`$
where $`K_{\alpha \beta }=a_{\beta ;\alpha }a_{\alpha ;\beta }`$ are the components of the commutator of the form $`\omega `$, and $`a_{\beta ;\alpha }`$, $`a_{\alpha ;\beta }`$ are the covariant derivatives. If we express the covariant derivatives in terms of the connectedness (if it is possible), they can be written as $`a_{\beta ;\alpha }=a_\beta /x^\alpha +\mathrm{\Gamma }_{\beta \alpha }^\sigma a_\sigma `$, where the first term results from differentiating the form coefficients, and the second term results from differentiating the basis. (In the Euclidean space the covariant derivatives coincide with ordinary ones since in this case the derivatives of the basis vanish). If we substitute the expressions for covariant derivatives into the formula for commutator components, we obtain the following expression for the commutator components of the form $`\omega `$
$$K_{\alpha \beta }=\left(\frac{a_\beta }{x^\alpha }\frac{a_\alpha }{x^\beta }\right)+(\mathrm{\Gamma }_{\beta \alpha }^\sigma \mathrm{\Gamma }_{\alpha \beta }^\sigma )a_\sigma $$
$`(1.3)`$
Here the expressions $`(\mathrm{\Gamma }_{\beta \alpha }^\sigma \mathrm{\Gamma }_{\alpha \beta }^\sigma )`$ entered into the second term are just the components of commutator of the first-degree metric form.
That is, the corresponding metric form commutator will enter into the differential form commutator.
If we substitute the expressions (1.3) for the skew-symmetric differential form commutator into formula (1.2’), we obtain the following expression for the differential of the first degree skew-symmetric form
$$d\omega =\left(\frac{a_\beta }{x^\alpha }\frac{a_\alpha }{x^\beta }\right)dx^\alpha dx^\beta +\left((\mathrm{\Gamma }_{\beta \alpha }^\sigma \mathrm{\Gamma }_{\alpha \beta }^\sigma )a_\sigma \right)dx^\alpha dx^\beta $$
The second term in the expression for the differential of skew-symmetric form is connected with the differential of the manifold metric form, which is expressed in terms of the metric form commutator.
While deriving formula (1.3) for the differential form commutator the connectednesses of special type were used. However, a similar result can be obtained by applying the connectednesses of arbitrary type or by using another way for finding the differential of the base coordinates. For differential forms of any degree the metric form commutator of corresponding degree will be included into the commutator of the skew-symmetric differential form.
As it is known , the differential of exterior differential form involves only a single term. There is no second term. This indicates that the metric form commutator vanishes. In other words, the manifold, on which the exterior differential form is defined, has a closed metric form.
The differential of the evolutionary differential form, which is defined on the manifold with unclosed metric forms, will contain two terms: the first term depends on the differential form coefficients and the other depends on the differential characteristics of the manifold.
Thus, the differentials and, correspondingly, the commutators of exterior and evolutionary forms are of different types.
What does this lead to?
The commutator of exterior differential form contains only one term obtained from the derivatives of the differential form coefficients. Such a commutator may be equal to zero or nonzero depending on the form coefficients. This means that the exterior differential form may be closed or unclosed.
In contrast to this, the evolutionary differential form cannot be closed. The terms of the commutator of evolutionary differential form have a different nature. Such terms cannot compensate one another. For this reason, the evolutionary form commutator, and hence, the differential of evolutionary form, prove to be nonzero. And this means that the evolutionary form cannot be closed.
### 1.3 Closed exterior differential forms
### Functional and utilitarian properties of closed exterior differential forms
Exterior differential form is called a closed one if its differential is equal to zero:
$$d\theta ^p=0$$
$`(1.4)`$
From condition (1.4) one can see that the closed form is a conservative quantity. This means that it can correspond to the conservation law, namely, to some conservative physical quantity.
The differential of exterior form is a closed form. The exterior form which is the differential of some other form
$$\theta ^p=d\theta ^{p1}$$
$`(1.5)`$
is called an exact form. (The exact forms prove to be closed automatically).
The closed inexact exterior forms possess utilitarian properties. These are exterior forms that are closed only on a certain structure (with the dimension being less than that of the initial manifold). In its metric form such a structure turns out to be a pseudostructure.
If the exterior form is closed only on pseudostructure, the closure condition is written as
$$d_\pi \theta ^p=0$$
$`(1.6)`$
And the pseudostructure $`\pi `$ obeys the condition
$$d_\pi {}_{}{}^{}\theta _{}^{p}=0$$
$`(1.7)`$
where $`{}_{}{}^{}\theta _{}^{p}`$ is the dual form.
From conditions (1.6) and (1.7) one can see that the pseudostructure and the closed exterior form constitute a conservative object, namely, a quantity that is conservative on the pseudostructure. Hence, such an object can correspond to some conservation law. It is from such object that the physical fields are formed. (More detail description will be given below).
The exact form is, by definition, a differential. In this case the differential is a total one. The closed inexact form is a differential too. And in this case the differential is an interior (on pseudostructure) differential, that is
$$\theta _\pi ^p=d_\pi \theta ^{p1}$$
$`(1.8)`$
Thus any closed form is a differential of the exterior form of a lower degree. From this it follows that the exterior form of lower degree may correspond to the potential, and the closed form by itself may correspond to the potential force. This is an additional example showing that the closed exterior form may have a physical sense. Here the two-fold nature of the closed form is revealed, on the one hand, as a conservative locally constant object, and, on the other hand, as a potential force.
Similarly to the differential connection between the exterior forms of sequential degrees, there is an integral connection. In particular, the integral theorems by Stokes and Gauss follow from the integral relation for $`p=1,2`$ in three-dimensional space.
Invariant properties of closed exterior differential forms. Since the closed form is a differential (a total one if the form is exact, or an interior one on the pseudostructure if the form is inexact), it is obvious that the closed form turns out to be invariant under all transformations that conserve the differential. The unitary transformations (0-form), the tangent and canonical transformations (1-form), the gradient and gauge transformations (2-form) and so on are examples of such transformations. These are gauge transformations for spinor, scalar, vector, tensor (3-form) fields. It should be pointed out that just such transformations are used in field theory.
Conjugacy and duality of exterior differential forms. Symmetries. The closure of exterior differential form, which leads to invariance, is a result of the conjugacy of elements of exterior or dual forms. The closure property of the exterior form means that any objects, namely, elements of the exterior form, components of elements, elements of the form differential, exterior and dual forms and others, turn out to be conjugated. It is a conjugacy that leads to realization of invariant and covariant properties of the exterior and dual forms.
Conjugacy is possible if there is one or other type of symmetry. Gauge symmetries, which are interior symmetries of field theory and with which gauge transformations are connected, are the symmetries of closed exterior differential forms. They are obtained as the result of conjugacy of any exterior form elements. The conservation laws for physical fields are connected with such interior symmetries.
(Below it will be shown that the exterior symmetries are symmetries of evolutionary differential forms, which are conditioned by degrees of freedom of material systems).
With the conjugacy another characteristic property of the exterior differential forms, namely, their duality, is connected. {The conjugacy is some identical connection between two operators or mathematical objects. The duality is a concept that means that one object carries a double meaning or that two objects with different meanings (of different physical nature) are identically connected. If one knows any dual object, one can obtain the other object}.
The connection of the exterior and dual forms is an example of the duality. The exterior form and the dual form correspond to the objects of different nature: the exterior form corresponds to the physical (i.e. algebraic) quantity, and the dual form corresponds to some spatial (or pseudospatial) structure. At the same time, under conjugacy the duality of these objects manifests itself, that is, if one form is known, it is possible to find the other form. In the case of closed inexact form the relation between the closed exterior form and the dual one leads to forming a differential-geometrical structure. (Physical structures that form physical fields are examples of such structures).
The duality is also revealed in the fact that, if the degree of the exterior form equals $`p`$, the dimension of the structure equals $`Np`$, where $`N`$ is the space dimension.
As the closed exterior form possesses invariant properties and the dual form corresponding to that possesses covariant properties, the invariance and covariance is one of examples of duality of the exterior differential forms.
The further example of the duality of the closed form is connected with the fact that, on the one hand, the closed exterior form is conservative quantity, and, on the other hand, the closed form can correspond to potential force. (Below the physical meaning of this duality will be elucidated, and it will be shown in respect to what the closed form manifests itself as potential force and with what the conservative physical quantity is connected).
As it will be shown below, the duality of closed differential forms has a fundamental physical meaning. The duality is a tool, which untangles the mutual connection, the mutual changeability and the transitions between different physical objects in evolutionary processes.
### Specific features of the mathematical apparatus of exterior differential forms.
Operators of the theory of exterior differential forms. In differential calculus the derivatives are basic elements of the mathematical apparatus. By contrast, the differential is an element of mathematical apparatus of the theory of exterior differential forms. It enables one to analyze the conjugacy of derivatives in various directions, which extends the potentialities of differential calculus.
The operator of exterior differential $`d`$ is an abstract generalization of ordinary mathematical operations of the gradient, curl, and divergence in the vector calculus . \[Here it should be emphasized that usual concepts of gradient, curl and divergence are operators applied to vector, whereas a gradient, curl and divergence in the theory of exterior forms, which are obtained as a result of exterior differentiation, are operators applied to pseudovector (an axial vector).\]
If, in addition to the exterior differential, we introduce the following operators: (1) $`\delta `$ for transformations that convert the form of $`p+1`$ degree into the form of $`p`$ degree, (2) $`\delta ^{}`$ for cotangent transformations, (3) $`\mathrm{\Delta }`$ for the $`d\delta \delta d`$ transformation, (4)$`\mathrm{\Delta }^{}`$ for the $`d\delta ^{}\delta ^{}d`$ transformation, one can write down the operators in the field theory equations in terms of these operators that act on the exterior differential forms. The operator $`\delta `$ corresponds to Green’s operator, $`\delta ^{}`$ does to the canonical transformation operator, $`\mathrm{\Delta }`$ does to the d’Alembert operator in 4-dimensional space, and $`\mathrm{\Delta }^{}`$ corresponds to the Laplace operator . It can be seen that the operators of the exterior differential form theory are connected with many operators of mathematical physics.
The mathematical apparatus of exterior differential forms extends the potentialities of the integral calculus. As it has been pointed out, the exterior differential forms were introduced as integrand expressions for definition of the integral invariants. The closure condition for exterior differential forms makes it possible to find the integrability conditions. This fact is of great importance in applications.
Identical relations of exterior differential forms. The basis of the mathematical apparatus of closed exterior differential forms is formed by identical relations. (Below it will be shown that at the basis of the mathematical apparatus of evolutionary differential forms there lie nonidentical relations, and it will be shown that identical relations for exterior differential forms are obtained from nonidentical relations for evolutionary forms.)
The identical relations of exterior differential forms are an mathematical expression of various kinds of the conjugacy or the duality of closed differential forms. (Since the conjugacy is a certain connection between two operators or mathematical objects, it is evident that the relations can be used to express the conjugacy mathematically.) They describe the conjugacy of any objects: the form elements, components of each element, exterior and dual forms, exterior forms of various degrees, and others. The identical relations that are connected with different kinds of conjugacy elucidate invariant, structural and group properties of exterior forms, which is of great importance in applications.
The identical relations for exterior differential forms follow from the closure conditions of differential forms, namely, vanishing the form differential (see formulae (1.4), (1.6), (1.7)) and the conditions connecting the forms of consequent degrees (see formulae (1.5), (1.8)).
The identical relation, which connects exterior forms of consequent degrees, express the fact that each closed exterior form is a differential of some exterior form. In general form such an identical relation can be written as
$$d_\pi \varphi =\theta _\pi ^p$$
$`(1.9)`$
In this relation the form in the right-hand side has to be a closed one. (As it will be shown below, the identical relations are satisfied only on pseudostructures).
In identical relation (1.9) in one side it stands a closed form and in other side does the differential of some differential form (the differential is a closed form as well). The examples of such identical relations are a) the Poincare invariant $`ds=Hdt+p_jdq_j`$, b) the second principle of thermodynamics $`dS=(dE+pdV)/T`$, d) the conditions on characteristics in the theory of differential equations and so on. The requirement that the function is an antiderivative (the integrand is a differential of a certain function) can be written in terms of such an identical relation. Existence of harmonic function is written by means of the identical relation: the harmonic function is a closed form, that is, a differential.
In addition to relations in differential forms, from the closure conditions of differential forms and the conditions connecting the forms of consequent degrees the identical relations of other types are obtained. This is also connected with the fact that the exterior forms can have the tensor, differential or integral representation. As the example of such relations it can serve
integral identical relations (formulae by Newton, Leibnitz and Green and integral relations by Stokes, Gauss-Ostrogradskii, etc);
tensor identical relations (vector and tensor identical relations that connect the operators of gradient, curl, divergence and so on\];
identical relations between derivatives (the Cauchi-Riemann conditions in the theory of complex variables, the transversality condition in the calculus of variations, the canonical relations in the Hamilton formalism, the thermodynamic relations between derivatives of thermodynamic functions , the condition that the derivative of implicit function is subject to, the eikonal relations and so on.
As the examples of identical relations it can serve such relations as the gauge relations in electromagnetic field theory, the tensor relations between connectednesses and their derivatives in gravitation (the symmetry of connectednesses with respect to subscripts, the Bianchi identity, the conditions imposed on the Christoffel symbols) and so on.
The identical relations of exterior differential forms appear practically in all branches of physics, mechanics and thermodynamics.
The physical meaning of identical relations for exterior differential forms will be disclosed below using evolutionary differential forms.
Nondegenerate transformations. One of the fundamental methods in the theory of exterior differential forms is an application of nondegenerate transformations. The nondegenerate transformations, if applied to identical relations, enable one to obtain new identical relations and new closed exterior differential forms.
Nondegenerate transformations in the theory of exterior differential forms are those that conserve the differential. As it has been pointed, the unitary, tangent, canonical, gradient, and gauge transformations are examples of such transformations.
From description of operators of exterior differential forms one can see that they are operators which execute some transformations. All these transformations are connected with the nondegenerate transformations of exterior differential forms.
The significance of the nondegenerate transformations consists in the fact that they allow to find closed differential forms.
As one can see, the properties of closed exterior differential forms reveal in many branches of physics.
In the second part of the paper it will be demonstrated a role of exterior differential forms in field theory. This role is connected with the fact that closed exterior forms reflect the properties of conservation laws for physical fields.
### 1.4 Evolutionary skew-symmetric differential forms
The properties of exterior differential forms and specific features of their mathematical apparatus have a great utilitarian and functional importance. However, the question of how closed exterior differential forms are obtained and how the process of conjugating different objects is realized remains unsolved.
The mathematical apparatus of evolutionary differential forms enables us to answer this question.
### Nonclosure of evolutionary differential forms.
It has been shown above that the exterior skew-symmetric differential forms can be closed. Such skew-symmetric differential forms possess invariant properties, and the basis of their mathematical apparatus includes identical relations and degenerate transformations. (Just on these properties the field theory is based.)
The evolutionary forms, as opposed to exterior forms, cannot be closed. This relates to the fact that the manifolds, on which the evolutionary forms are defined, have unclosed metric forms. The commutator of unclosed metric form, which is nonzero, enters into the evolutionary form commutator. Such commutator of differential form cannot vanish. This means that the evolutionary form differential is nonzero. Hence, the evolutionary differential form, in contrast to the case of the exterior form, cannot be closed and does not possess invariant properties.
As it will be shown below, the basis of mathematical apparatus of evolutionary differential forms, which are unclosed forms, includes nonidentical relations and degenerate transformations. (Nonidentical relations under degenerate transformations are just the relations that generate identical relations for exterior forms.)
### Nonidentical relations of evolutionary differential forms
The identical relations of closed exterior differential forms reflect a conjugacy or duality of any objects. The evolutionary forms, being unclosed, cannot directly describe the conjugacy of any objects. But they allow a description of the process in which the conjugacy may appear (that is, to describe how closed exterior differential forms are generated). Such a process is described by nonidentical relations.
The identical relations establish exact correspondence between the quantities (or objects) involved into this relation. It is possible in the case when the quantities involved into the relation are measurable (invariant) ones. In the nonidentical relations one of the quantities is unmeasurable.
\[The concept of “nonidentical relation” may appear to be inconsistent. However, as it will be shown below, this has a deep meaning.\]
The relation of evolutionary forms is a relation between the differential, which is a closed differential form and is a measurable (invariant) quantity, and the evolutionary form that is unclosed form and hence is not a measurable quantity. Such a relation cannot be identical.
Nonidentical relations of such type appear in descriptions of physical processes.
Nonidentical relations may be written as
$$d\psi =\omega ^p$$
$`(1.10)`$
Here $`\omega ^p`$ is the $`p`$-degree evolutionary form, which is unclosed form, $`\psi `$ is a certain form of degree $`(p1)`$, and the differential $`d\psi `$ is a closed form of degree $`p`$.
In the left-hand side of this relation it stands the form differential, i.e. a closed form that is an invariant object. In the right-hand side it stands the unclosed form that is not an invariant object. Such a relation cannot be identical. {Nonidentical relation was analyzed in paper J.L.Synge ”Tensorial Metods in Dynamics” (1936). And yet it was allowed a possibility to use the sign of equality in nonidentical relation.}
One can see a difference of relations for exterior forms and evolutionary ones. In the right-hand side of identical relation (1.9) it stands a closed form, whereas the form in the right-hand side of nonidentical relation (1.10) is an unclosed one.
Relation (1.10) is an evolutionary relation as it involves the evolutionary form.
Such nonidentical relations appear, for example, while investigating the integrability of any differential equations that describe various processes.
As the example one can inspect the partial differential equation of the first order. Let
$$F(x^i,u,p_i)=0,p_i=u/x^i$$
$`(1.11)`$
be a partial differential equation of the first order.
Let us consider the functional relation
$$du=\theta $$
$`(1.12)`$
where $`\theta =p_idx^i`$ Here $`\theta =p_idx^i`$ is the differential form of the first degree (summation over repeated indices is implied).
In the general case, for example, when differential equation (1.11) describes any physical processes, functional relation (1.12) turns out to be nonidentical.
For this relation be identical, the differential form $`\theta =p_idx^i`$ must also be a differential (like the left-hand side of relation (1.12)), that is, it has to be a closed exterior differential form. To do this, it requires the commutator $`K_{ij}=p_j/x^ip_i/x^j`$ of the differential form $`\theta `$ has to vanish.
In the general case, from equation (1.11) it does not follow (explicitly) that the derivatives $`p_i=u/x^i`$, which obey to the equation (and given boundary or initial conditions of the problem), make up a differential. Without any supplementary conditions the commutator $`K_{ij}`$ of the differential form $`\theta `$ is not equal to zero. The form $`\theta =p_idx^i`$ turns out to be unclosed and is not a differential like the left-hand side of relation (1.12). Functional relation (1.12) appears to be nonidentical.
If to write functional relation (1.12) in the form
$$dup_idx^i=0$$
$`(1.12^{})`$
we get the well-known Pfaff equation for partial differential equation. However, the nonidentical relation cannot be treated as an equation in differentials. To solve equation (1.12’) means to find the derivatives $`p_i`$ of original equation (1.11) which make up a differential. In this case the derivatives $`p_i`$ of equation (1.11) that do not obey these conditions are ignored, although they satisfy original equation (1.11) and boundary and initial conditions. But in the analysis of relation (1.12) all derivatives that satisfy original equation (1.11) and boundary or initial conditions are accounted for, and their role in the physical process under consideration is investigated.
The nonidentity of functional relation means that the original equation is nonintegrable one, that is, the derivatives of this equation are not reduced (without additional condition) to the identical relation of the type $`d\psi =dU`$.
It can be shown that, for every differential equation describing any physical processes, the relevant functional relation constructed of derivatives will be nonidentical ones.
Below a role of such relations in mathematical physics will be disclosed.
While investigating real physical processes one often meets relations that are nonidentical. Usually one treats nonidentical relations like equations, namely, one finds only the values at which nonidentical relation becomes identic. In doing so, it is not accounted for the fact that the nonidentical relation is often obtained from the description of some physical process and it has physical meaning at every stage of the physical process rather than at the stage when the additional conditions are satisfied.
The approach to nonidentical relation as to a relation rather then to the equation in differentials gives fundamentally new physical results.
Selfvariation of nonidentical relation. Nonidentical relation possesses an unique evolutionary property, namely, this relation appears to be aselfvarying relation.
The nonidentical relation is selfvarying one since, firstly, it is nonidentical, namely, it contains two objects one of which appears to be unmeasurable (noninvariant), and, secondly, it is an evolutionary relation, namely, the variation of any object of the relation in some process leads to the variation of another object, and, in turn, the variation of the latter leads to variation of the former. Since one of the objects is an unmeasurable quantity, the other cannot be compared with the first one, and hence, the process of mutual variation cannot terminate.
Varying the evolutionary form coefficients leads to varying the first term of the commutator (see Eq.(1.3) for the case of the first-order evolutionary form). In accordance with this variation it varies the second term, that is, the metric form of the manifold varies. Since the metric form commutators specifies the manifold differential characteristics that are connected with the manifold deformation, this points to the manifold deformation. This means that the evolutionary form basis varies. In turn, this leads to variation of the evolutionary form, and the process of intervariation of the evolutionary form and the basis is repeated. The processes of variation of the evolutionary form and the basis are governed by the evolutionary form commutator and it is realized according to the evolutionary relation.
Selfvariation of the evolutionary relation goes on by exchange between the evolutionary form coefficients and the manifold characteristics. This is an exchange between physical quantities and space-time characteristics. (This is an exchange between quantities of different nature).
The process of the evolutionary relation selfvariation cannot come to the end. This is indicated by the fact that both the evolutionary form commutator and the evolutionary relation involve unmeasurable quantities. From this it follows that the evolutionary nonidentical relation cannot be converted into identical relation. However, from this nonidentical relation it can be obtained identical relation.
It appears that this is only possible under degenerate transformation.
### Degenerate transformations. Obtaining identical relation from nonidentical one
To obtain the identical relation from the evolutionary nonidentical relation, it is necessary that the closed exterior differential form should be derived from the evolutionary differential form included into evolutionary relation.
However, as it has been shown above, the evolutionary form cannot be a closed form. For this reason the transition from evolutionary form is possible only to the inexact closed exterior form that is defined only on pseudostructure.
To the pseudostructure it is assigned a closed dual form (whose differential vanishes). For this reason the transition from the evolutionary form to the closed inexact exterior form proceeds only when the conditions of vanishing the dual form differential are realized, in other words, when the metric form differential or commutator becomes equal to zero.
The metric form of original manifold, on which the evolutionary form (and correspondingly, nonidentical relation) is defined, is not closed. The transition from unclosed metric form with nonzero differential to closed metric form with the differential being equal to zero is possible only as a degenerate transformation, that is, as a transformation which does not conserve the differential.
The conditions of vanishing the dual form differential (vanishing the metric form commutator) are the conditions of degenerate transformation.
Vanishing the metric form commutator under degenerate transformation points to emergence of closed metric form, and this corresponds to realization of pseudostructure (the dual form).
Vanishing one term of the evolutionary form commutator, namely, the metric form commutator, leads to that the second term of the commutator also vanishes. This is due to the fact that the terms of the evolutionary form commutator correlate between one another. The evolutionary form commutator and, correspondingly, the differential, vanish on pseudostructure, and this means that it appears the unclosed inexact exterior form.
If the conditions of degenerate transformation are realized, from the unclosed evolutionary form, which differential is nonzero $`d\omega ^p0`$, one can obtain a differential form closed on pseudostructure. The differential of this form equals zero. That is, it is realized the transition
$`d\omega ^p0`$ (degenerate transformation) $`d_\pi \omega ^p=0`$, $`d_\pi {}_{}{}^{}\omega _{}^{p}=0`$
On the pseudostructure $`\pi `$ evolutionary relation (1.10 ) transforms into the relation
$$d_\pi \psi =\omega _\pi ^p$$
$`(1.13)`$
which proves to be an identical relation. Indeed, since the form $`\omega _\pi ^p`$ is a closed one, on the pseudostructure this form turns out to be a differential of some differential form. In other words, this form can be written as $`\omega _\pi ^p=d_\pi \theta `$. Relation (1.13 ) is now written as
$$d_\pi \psi =d_\pi \theta $$
There are differentials in the left-hand and right-hand sides of this relation. This means that the relation is an identical one.
Thus, it is evident that under degenerate transformation the identical on pseudostructure relation can be obtained from the evolutionary nonidentical relation. This is due to the realization of closed metric form, and correspondingly to the realization of closed exterior form, and this points to the emergency of pseudostructure and a conservative quantity on pseudostructure. The pseudostructure with conservative quantity makes up a differential-geometrical structure (the properties of such structure will be defined below).
Under degenerate transformation the evolutionary form differential vanishes only on pseudostructure. This is an interior (being equal to zero) differential. The total differential of the evolutionary form is nonzero. The evolutionary form remains to be unclosed, and for this reason the original relation, which contains the evolutionary form, remains to be nonidentical.
Under realization of new additional conditions a new identical relation can be obtained. As a result, the nonidentical evolutionary relation can generate identical relations.
(It can be shown that all identical relations of the exterior differential form theory are obtained from nonidentical relations by applying the degenerate transformation.)
The conditions of degenerate transformation, that is, additional conditions, can be realized, for example, if it will appear any symmetries of the metric form coefficients or their derivatives. This can happen under selfvariation of the nonidentical relation.
(While describing material system the conditions of degenerate transformation are connected with degrees of freedom of a material system).
To the conditions of degenerate transformation there corresponds a requirement that some functional expressions become equal to zero. Such functional expressions are Jacobians, determinants, the Poisson brackets, residues, and others.
Mathematically the degenerate transformation is realized as a transition from one frame of reference to another (nonequivalent) frame of reference. This is a transition from the frame of reference connected with the manifold whose metric forms are unclosed to the frame of reference being connected with pseudostructure. The first frame of reference cannot be inertial or locally- inertial frame. The evolutionary form and nonidentical evolutionary relation are defined in the noninertial frame of reference. But the closed exterior form obtained and the identical relation are obtained with respect to the locally-inertial frame of reference.
Relation between degenerate transformations of exterior forms and nondegenerate transformations of evolutionary forms. In the theory of closed exterior forms only nondegenerate transformations, which conserve the differential, are used. Nondegenerate transformations of evolutionary forms are transformations that do not conserve the differential. Nevertheless, these transformations are mutually connected. The degenerate transformations execute a transition from original deforming manifold to pseudostructures. And the nondegenerate transformations execute a transition from one pseudostructure to another. As the result of degenerate transformation the closed inexact exterior form arises, and this points to origination of the differential-geometrical structure. But under nondegenerate transformation the transition from one closed form to another takes place, and this points to the transition from one differential-geometrical structure to another. When describing the evolutionary processes in material media using evolutionary forms it will be shown that the degenerate transformation describes an origination of physical structures (which are differential-geometrical structures) in material media and that nondegenerate transformation describes a transition from one physical structure to another.
### Mechanism of realization of conjugated objects and operators.
The mathematical apparatus of evolutionary differential forms reveals the process of mutual topological variation of the evolutionary form and the manifold. Under such variation the conditions of degenerate transformation can be realized, and the exterior differential form closed on the pseudostructure can arise (spontaneously).
To the closed exterior forms it can be assigned conjugated operators, whereas to the evolutionary forms there correspond nonconjugated operators. The transition from the evolutionary form to the closed exterior form is that from nonconjugated operators to conjugated ones. This is expressed as the transition from the nonzero differential (unclosed evolutionary form) to the differential (closed exterior form) that equals zero.
Here it should be emphasized that the properties of deforming manifolds and skew-symmetric differential forms on such manifolds, namely, evolutionary forms, play a principal role in the process of conjugating.
The process of conjugating includes the following points:
1) selfvariation of nonidentical relation, namely, mutual variations of the evolutionary form coefficients (which have the algebraic nature) and of manifold characteristics (which have the geometric nature) described by nonidentical evolutionary equation, and 2) realization of the degenerate transformation.
Hence one can see that the process of conjugating is a mutual exchange between the quantities of different nature and the degenerate transformation under additional conditions. Here it should be pointed that the condition of degenerate transformation (vanishing some functional expressions like, for example, Jacobians, determinants and so on) may be realized spontaneously while selfvarying the nonidentical relation if any symmetries appear. It is possible if the system (that is described by this relation) possesses any degrees of freedom.
One can see that the process of realization of conjugated operators or objects is described by the nontraditional mathematical apparatus, namely, by nonidentical relations and degenerate transformations.
As it has been shown above, closed exterior forms appear in many mathematical formalisms. Practically all conjugated objects are explicitly or implicitly connected with closed exterior forms. And yet it can be shown that closed exterior forms are generated by evolutionary differential forms, which are skew-symmetric differential forms defined on the deforming varying manifolds.
By comparing the exterior and evolutionary forms one can see that they possess the opposite properties. Yet it was shown that the evolutionary differential forms generate the closed exterior differential forms. This elucidates that the exterior and evolutionary differential forms are unified whole.
### 1.5 Evolutionary forms: Realization of differential-geometrical structures and forming pseudometric and metric manifolds
### Realization of differential-geometrical structures.
The process of realization of closed exterior form relates to the realization of closed metric form, which is a form being dual with respect to the closed exterior form. The closed exterior form and corresponding dual form make up a new conjugated object. Since the closed exterior form is a conservative quantity (because the differential of closed exterior form equals zero) and the dual form describes the pseudostructure, such conjugated object (closed exterior form and dual form) describes the differential-geometrical structure, namely, the pseudostructure with conservative quantity. As it has been already noted, such structures are the example of the differential-geometrical G-structures. (The physical structures, which forms physical fields, and corresponding conservation laws, are just such structures).
The mathematical apparatus of evolutionary differential forms describing the process of generation of closed inexact exterior differential forms, describes thereby the process of origination of the differential-geometrical structure.
Obtaining differential-geometrical structures is a process of conjugating the objects. Such process is, firstly, a mutual exchange between the quantities of different nature (for example, between the algebraic and geometric quantities or between the physical and spatial quantities), and, secondly, the establishment of exact correspondence (conjugacy) of these objects. This process is described by selfvariation of nonidentical relation and degenerate transformation.
While describing the process of realization of the differential-geometrical structure the following questions arise. What does generate such structures, and what is the reason of such processes? These problems will be discussed in the next sections while studying physical processes in material media (which are controlled by conservation laws).
Below it is shown by what are defined the characteristics of the structures originated, how are such structures classified, how are formed the pseudometric and metric manifolds from these structures.
Characteristics of the differential-geometrical structures realized. For the sake of convenience in the subsequent presentation the differential- geometrical structures, to which inexact exterior and dual forms obtained from the evolutionary forms correspond, will be called the binary structures or Bi-Structures.
Since the closed exterior and dual differential forms, which correspond to Bi-Structure arisen, were obtained from the nonidentical relation that involves the evolutionary form, it is evident that the characteristics of such structure have to be connected:
a) with those of the evolutionary form and of the deforming manifold on which this form is defined,
b) with the values of commutators of the evolutionary form and the manifold metric form, and
c) with the conditions of degenerate transformation as well.
The condition of degenerate transformation corresponds to a realization of the closed metric (dual) form and defines the pseudostructure.
Vanishing the interior commutator of the evolutionary form (on pseudostructure) corresponds to a realization of the closed (inexact) exterior form and points to emergence of conservative (invariant) quantity.
When Bi-Structure originates, the value of the total commutator of the evolutionary form containing two terms is nonzero. These terms define the following characteristics of Bi-Structures:
a) the first term of evolutionary form commutator (which is composed of the derivatives of the evolutionary form coefficients) defines the value of the discrete change of conservative quantity, that is, the quantum, which the quantity conserved on the pseudostructure undergoes at the transition from one pseudostructure to another;
b) the second term (which is composed of the derivatives of coefficients of the metric form connected with the manifold) specifies the characteristics of Bi-Structures, which fixes the character of the initial manifold deformation taking place before Bi-Structures arose. (Spin is such an example). (This characteristics fixes the deformation of original manifold that proceeded in the process of originating the differential-geometrical structure and was described by selfvariation of nonidentical relation).
Thus, the conditions of degenerate transformation determine the pseudostructures; the first term of the evolutionary form commutator determines the value of the discrete change (the quantum) of conservative quantity; the second term of the evolutionary form commutator specifies the characteristics that fixes the character of the initial manifold deformation.
The discrete (quantum) change of conservative quantity proceeds in the direction that is normal to the pseudostructure. (Jumps of the derivatives normal to the potential surfaces are examples of such changes.)
Above it has been noted that the evolutionary form and the nonidentical relation are obtained while describing the physical processes that proceed in material systems. For this reason it is evident that the characteristics of Bi-Structure must also be connected with the characteristics of the material system being described.
Classification of differential-geometrical structures realized. The closed forms that correspond to Bi-Structures are generated by the evolutionary relation which includes the evolutionary form of $`p`$ degree. Therefore, the structures originated can be classified by the parameter $`p`$.
The other parameter is the degree of closed forms generated by the nonidentical evolutionary relation.
The nonidentical evolutionary relation with the forms of degree $`p`$ will be called the evolutionary relation of degree $`p`$. Under degenerate transformation from nonidentical relation of degree $`p`$ an identical relation is obtained. This identical relation contains the closed form of degree $`k=p`$.
To find how the parameter $`k=p`$ changes, one has to consider the problem of integration of the nonidentical evolutionary relation.
The identical relation with closed form of degree $`p`$ obtained can be integrated, since the right-hand side of such a relation can be expressed in terms of differential (as well as the left-hand side).
By integrating the identical relation obtained one can get the nonidentical relation of degree $`(p1)`$.
The nonidentical relation of degree $`(p1)`$ obtained can be integrated once again if the corresponding degenerate transformation is realized, and the identical relation is formed. This identical relation will already include the closed form of degree $`k=p1`$.
In this manner the evolutionary nonidentical relation of degree $`p`$ may generate closed (on the pseudostructure) exterior forms of sequential degrees $`k=p,\mathrm{},k=0`$ and corresponding identical relations.
Thus, one can see that Bi-structures, to which there are assigned the closed (on the pseudostructure) exterior forms, can depend on two parameters. These parameters are the degree of evolutionary form $`p`$ (in the evolutionary relation) and the degree of created closed forms $`k`$.
In addition to these parameters, another parameter appears, namely, the dimension of space. If the evolutionary relation generates the closed forms of degrees $`k=p`$, $`k=p1`$, …, $`k=0`$, to them there are assigned the pseudostructures of dimensions $`(Nk)`$, where $`N`$ is the space dimension.
### Forming pseudometric and metric manifolds.
At this point it should be noted that at every stage of the evolutionary process it is realized only one element of pseudostructure, namely, a certain minipseudostructure. (The example of minipseudoctructure is the wave front. The wave front is an eikonal surface (the level surface), i.e. a surface with conservative quantity.)
While varying the evolutionary variable the minipseudostructures form the pseudostructure.
Manifolds with closed metric forms are formed by pseudostructures. They are obtained from the deforming manifolds with unclosed metric forms. In this case the initial deforming manifold (on which the evolutionary form is defined) and the manifold with closed metric forms originated (on which the closed exterior form is defined) are different spatial objects.
It takes place the transition from the initial (deforming) manifold with unclosed metric form to the pseudostructure, namely, to the manifold with closed metric forms created. Mathematically this transition (the degenerate transformation) proceeds as a transition from one frame of reference to another, nonequivalent, frame of reference.
The pseudostructures, on which the closed inexact forms are defined, form the pseudomanifolds.
To the transition from pseudomanifolds to metric space it is assigned the transition from closed inexact differential forms to exact exterior differential forms.
It was shown above that the evolutionary relation of degree $`p`$ can generate (with using the degenerate transformations) closed forms of degrees $`0,\mathrm{},p`$. While generating closed forms of sequential degrees $`k=p,k=p1,\mathrm{},k=0`$ the pseudostructures of dimensions $`(n+1k)`$ are obtained. As a result of transition to the exact closed form of zero degree the metric structure of the dimension $`n+1`$ is obtained.
Sections of the cotangent bundles (Yang-Mills fields), cohomologies by de Rham, singular cohomologies, pseudo-Riemannian and pseudo-Euclidean spaces, and others are examples of the psedustructures and spaces that are formed by pseudostructures. Euclidean and Riemannian spaces are examples of metric manifolds that are obtained when changing to the exact forms. Here it should be noted that the examples of pseudometric spaces are potential surfaces (surfaces of a simple layer, a double layer and so on). In these cases the type of potential surfaces is connected with the above listed parameters.
Conservative quantities (closed exterior inexact forms) defined on pseudomanifolds (closed dual forms) constitute some fields. (The physical fields are the examples of such fields.) The fields of conservative quantities are formed from closed exterior forms at the same time when the manifolds are created from the pseudoctructures.
Since the closed metric form is dual with respect to some closed exterior differential form, the metric forms cannot become closed by themselves, independently of the exterior differential form. This proves that the manifolds with closed metric forms are connected with the closed exterior differential forms. This indicates that the fields of conservative quantities are formed from closed exterior forms at the same time when the manifolds are created from the pseudoctructures. The specific feature of manifolds with closed metric forms that have been formed is that they can carry some information. That is, the closed exterior differential forms and manifolds, on which they are defined, are mutually connected objects.
One can see that the evolutionary forms possess the properties, which enable one to describe the evolutionary processes, namely, the processes of generating the differential-geometrical structures and forming manifolds. In other mathematical formalisms there are no such possibilities that the mathematical apparatus of evolutionary and exterior skew-symmetrical forms possesses.
The evolutionary differential forms, which generate the closed exterior forms, disclose the mechanism of origination of physical structures forming physical fields and the determinacy of these processes.
Before describing a role of the evolutionary differential forms in mathematical physics and field theory, it is necessary to show the relation between the closed exterior forms and existing invariant field theories.
## 2 Role of exterior forms in field theory
The role of exterior forms in field theory is due to the fact that they reflect the properties of conservation laws.
### 2.1 Conservation laws for physical fields (exact conservation laws)
Owing to the development of science the concept of “conservation laws” in thermodynamics, physics and mechanics contains different meanings.
In areas of physics related to the field theory and in the theoretical mechanics “the conservation laws” are those according to which there exist conservative physical quantities or objects. These are the conservation laws that above were called “exact”.
In mechanics and physics of continuous media the concept of “conservation laws” is related to the conservation laws for energy, linear momentum, angular momentum, and mass that establish the balance between the change of physical quantities and external action. These are balance conservation laws.
In thermodynamics the conservation laws are associated with the principles of thermodynamics.
Below it will be shown the relation between exact and balance conservation laws and the connection of conservation laws with the principles of thermodynamics.
The exact conservation laws are those that state an existence of conservative physical quantities or objects. The exact conservation laws are related to physical fields {The physical fields are a special form of the substance, they are carriers of various interactions such as electromagnetic, gravitational, wave, nuclear and other kinds of interactions.}
The closed exterior differential forms correspond to the exact conservation laws. Indeed, from the closure conditions of the exterior differential form (see formulas (1.4), (1.6), (1.7)) it is evident that the closed exterior differential form is a conservative quantity. In this case the closed inexact exterior differential form and the corresponding dual form describe a conservative object, namely, there is a conservative quantity only on some pseudostructure $`\pi `$. From this one can see that the closed exterior differential form can correspond to the exact conservation law.
The closure conditions for the exterior differential form ($`d_\pi \theta ^p=\mathrm{\hspace{0.17em}0}`$) and the dual form ($`d_\pi ^{}\theta ^p=\mathrm{\hspace{0.17em}0}`$) are mathematical expressions of the exact conservation law.
It has been pointed above that the pseudostructure (dual form) and the conservative quantity (the closed exterior form) define the differential-geometrical (binary) structure (Bi-structure), which is the example of G-Structure. It is evident that such structure corresponds to the exact conservation law.
It is such structures (pseudostructures with a conservative physical quantity) that correspond to exact conservation law, that are, the physical structures from which physical fields are formed.
Equations for the physical structures ($`d_\pi \theta ^p=\mathrm{\hspace{0.17em}0}`$, $`d_\pi ^{}\theta ^p=\mathrm{\hspace{0.17em}0}`$) turn out to coincide with the mathematical expression for the exact conservation law.
The mathematical expression for the exact conservation law and its connection with physical fields can be schematically written in the following manner:
$$\{\begin{array}{cc}d_\pi \theta ^p=0\hfill & \\ d_\pi {}_{}{}^{}\theta _{}^{p}=0\hfill & \end{array}\{\begin{array}{cc}& \\ \theta ^p\hfill & \\ & \\ {}_{}{}^{}\theta _{}^{p}\hfill & \end{array}\text{}\text{physical structures}\text{physical fields}$$
It should be emphasized that the closed inexact exterior forms correspond to the physical structures that form physical fields. The exact exterior forms correspond to the material system elements. (About this it will be said below).
It can be shown that the field theories, i.e. the theories that describe physical fields, are based on the invariant and metric properties of the closed exterior differential and dual forms that correspond to exact conservation laws.
### 2.2 Closed exterior differential forms in invariant field theories. (Exact conservation laws and specific features of existing field theories)
The properties of closed exterior differential forms correspond to the conservation laws for physical fields. Therefore, the mathematical principles of the theory of closed exterior differential forms lie at the basis of existing field theories.
The equations that are equations of the existing field theories are those obtained on the basis of the properties of the exterior differential form theory.
The Hamilton formalism is based on the properties of closed exterior and dual forms of the first degree. The corresponding equation of the field has the form:
$$\frac{s}{t}+H(t,q_j,p_j)=\mathrm{\hspace{0.17em}0},\frac{s}{q_j}=p_j$$
$`(2.1)`$
where $`s`$ is the field function (the state function) for the action functional $`S=L𝑑t`$. Here $`L(t,q_j,\dot{q}_j)`$ is the Lagrange function, $`H`$ is the Hamilton function: $`H(t,q_j,p_j)=p_j\dot{q}_jL`$, $`p_j=L/\dot{q}_j`$. To this equation it is assigned the closed exterior form of the first degree, which is the Poincare invariant $`ds=Hdt+p_jdq_j`$. The closure conditions for this form and corresponding dual form constitute the Hamiltonian systems:
$$\frac{dq_j}{dt}=\frac{H}{p_j},\frac{dp_j}{dt}=\frac{H}{q_j}$$
$`(2.2)`$
The Schrődinger equation in quantum mechanics is an analog to equation (2.1) (where the conjugated coordinates are changed by operators), and the Heisenberg equation is an analog to the appropriate integral of the equation (2.2) that have the form of canonical relations. The conjugacy of Dirac’s bra- and cket- vectors in quantum mechanics corresponds to the closure condition of the zero degree exterior form . {The duality of closed forms manifests itself in the approaches by Schrődinger and Heisenberg. Whereas the zero degree closed exterior form corresponds to the Schrődinger equation, the closed dual form corresponds to the Heisenberg equation. It can be pointed out that, whereas the equations by Shrődinger and Heisenberg describe the behavior of the potential obtained from the zero degree closed form, Dirac’s bra- and cket\- vectors constitute the zero degree closed exterior form itself as the result of conjugacy (vanishing the scalar product)}.
It is evident that the closed exterior and dual forms of zero degree correspond to quantum mechanics.
The properties of closed exterior and dual forms of the second degree lie at the basis of the electromagnetic field equations. The Maxwell equations may be written as $`d\theta ^2=0`$, $`d^{}\theta ^2=0`$, where $`\theta ^2=\frac{1}{2}F_{\mu \nu }dx^\mu dx^\nu `$ (here $`F_{\mu \nu }`$ is the strength tensor).
Closed exterior and dual forms of the third degree correspond to the gravitational field.
From the above said one can see that to each type of physical fields there corresponds a closed exterior form of appropriate degree. (However, to the physical field of given type it can be assigned closed forms of less degree. In particular, to the Einstein equation for gravitational field it corresponds the first degree closed form, although it was pointed out that the type of a field with the third degree closed form corresponds to the gravitational field.)
The connection between field theory and closed exterior differential forms supports the invariance of field theory.
And here it should underline that field theories are based on the properties of closed inexact forms. This is explained by the fact that only inexact exterior forms can correspond to the physical structures that form physical fields. The condition that the closed exterior forms, which constitute the basis of field theory equations, are inexact ones reveals in the fact that essentially all existing field theories include a certain elements of noninvariance, i.e. they are based either on functionals that are not identical invariants (such as Lagrangian, action functional, entropy) or on equations (differential, integral, tensor, spinor, matrix and so on) that have no identical invariance (integrability or covariance). Such elements of noninvariance are, for example, the nonzero value of the curvature tensor in Einstein’s theory , the indeterminacy principle in Heisenberg’s theory, the torsion in the theory by Weyl , the Lorentz force in electromagnetic theory , an absence of general integrability of the Schrődinger equations, the Lagrange function in the variational methods, an absence of the identical integrability of the mathematical physics equations, and that of identical covariance of the tensor equations, and so on. Only if we assume elements of noncovariance, we can obtain closed inexact forms that correspond to physical structures.
And yet, the existing field theories are invariant ones because they are provided with additional conditions under which the invariance or covariance requirements have to be satisfied. It is possible to show that these conditions are the closure conditions of exterior or dual forms. Examples of such conditions may be the identity relations: canonical relations in the Schrődinger equations, gauge invariance in electromagnetic theory, commutator relations in the Heisenberg theory, symmetric connectednesses, identity relations by Bianchi in the Einstein theory, cotangent bundles in the Yang-Mills theory, the Hamilton function in the variational methods, the covariance conditions in the tensor methods, etc. The field theory postulates are the expression of such conditions.
The connection between the field theory equations and closed exterior forms shows that to every physical field it is assigned the appropriate degree of closed exterior form. The type of gauge transformations used in field theory depends on the degree of closed exterior differential form.
This shows that it is possible to introduce a classification of physical fields according to the degree of closed exterior form. But within the framework of only exterior differential forms one cannot understand how this classification is explained. (This can be elucidated only by application of evolutionary differential forms.)
The exterior differential forms, whose properties correspond to the conservation laws, constitute the basis of the invariant field theories.
The existing field theories allow to describe the physical fields. However, because these theories are invariant ones they cannot answer the question about the mechanism of originating physical structures that form physical fields. The origination of physical structures and forming physical fields are evolutionary processes, and hence they cannot be described by the invariant field theories. Only evolutionary theory can do this. The theory of exterior and evolutionary forms can be such a theory.
## 3 Role of evolutionary forms in mathematical physics and field theory
The role of evolutionary forms in mathematical physics and field theory (as well as the role of exterior forms) is due to the fact that they reflect the conservation laws. However, these conservation laws are those not for physical fields but for material media. These are balance conservation laws.
### 3.1 Balance conservation laws
The balance conservation laws are those that establish the balance between the variation of a physical quantity and the corresponding external action. These are the conservation laws for the material systems (material media).
The balance conservation laws are the conservation laws for energy, linear momentum, angular momentum, and mass.
In the integral form the balance conservation laws express the following : a change of a physical quantity in an elementary volume over a time interval is counterbalanced by the flux of a certain quantity through the boundary surface and by the action of sources. Under transition to the differential expression the fluxes are changed by divergences.
The equations of the balance conservation laws are differential (or integral) equations that describe a variation of functions corresponding to physical quantities \[8, 12 - 14\]. If the material system is not a dynamical one (as in the case of a thermodynamic system), the equations of the balance conservation laws can be written in terms of increments of physical quantities and governing variables.
(The specific forms of these equations for thermodynamical and gas dynamical material systems and the systems of charged particles will be presented in the Appendix).
But it appears that, even without knowledge of the concrete form of these equations, with the help of the differential forms one can see specific features of these equations that elucidate the properties of the balance conservation laws. To do so it is necessary to study the conjugacy (consistency) of these equations.
{The necessity of studying the conjugacy of the equations describing any process has a physical meaning. If these equations (or derivatives with respect to different variables) be not conjugated, the solutions to corresponding equations prove to be noninvariant: they are functionals rather then functions. The realization of the conditions (while varying variables), under which the equations become conjugated ones, leads to that the relevant solution becomes invariant. It will be shown below that the transition to the invariant solution, which can be obtained only using evolutionary forms, describes the mechanism of evolutionary transition from one quality to another, which leads to emergence of physical structures}.
Equations are conjugate if they can be contracted into identical relations for the differential, i.e. for a closed form.
Let us analyze the equations that describe the balance conservation laws for energy and linear momentum.
We introduce two frames of reference: the first is an inertial one (this frame of reference is not connected with the material system), and the second is an accompanying one (this system is connected with the manifold built by the trajectories of the material system elements). The energy equation in the inertial frame of reference can be reduced to the form:
$$\frac{D\psi }{Dt}=A$$
$`(3.1)`$
where $`D/Dt`$ is the total derivative with respect to time, $`\psi `$ is the functional of the state that specifies the material system, $`A`$ is the quantity that depends on specific features of the system and on external energy actions onto the system. {The action functional, entropy, wave function can be regarded as examples of the functional $`\psi `$. Thus, the equation for energy presented in terms of the action functional $`S`$ has a similar form: $`DS/Dt=L`$, where $`\psi =S`$, $`A=L`$ is the Lagrange function. In mechanics of continuous media the equation for energy of an ideal gas can be presented in the form : $`Ds/Dt=\mathrm{\hspace{0.17em}0}`$, where $`s`$ is entropy. In this case $`\psi =s`$, $`A=\mathrm{\hspace{0.17em}0}`$. It is worth noting that the examples presented show that the action functional and entropy play the same role.}
In the accompanying frame of reference the total derivative with respect to time is transformed into the derivative along the trajectory. Equation (3.1) is now written in the form
$$\frac{\psi }{\xi ^1}=A_1$$
$`(3.2)`$
here $`\xi ^1`$ is the coordinate along the trajectory.
In a similar manner, in the accompanying frame of reference the equation for linear momentum appears to be reduced to the equation of the form (see, for example, Appendix)
$$\frac{\psi }{\xi ^\nu }=A_\nu ,\nu =\mathrm{\hspace{0.17em}2},\mathrm{}$$
$`(3.3)`$
where $`\xi ^\nu `$ are the coordinates in the direction normal to the trajectory, $`A_\nu `$ are the quantities that depend on the specific features of the system and external (with respect to local domain) force actions.
Eqs. (3.2), (3.3) can be convoluted into the relation
$$d\psi =A_\mu d\xi ^\mu ,(\mu =\mathrm{\hspace{0.17em}1},\nu )$$
$`(3.4)`$
where $`d\psi `$ is the differential expression $`d\psi =(\psi /\xi ^\mu )d\xi ^\mu `$.
Relation (3.4) can be written as
$$d\psi =\omega $$
$`(3.5)`$
here $`\omega =A_\mu d\xi ^\mu `$ is the skew-symmetrical differential form of the first degree.
Since the balance conservation laws are evolutionary ones, the relation obtained is also an evolutionary relation.
Relation (3.5) was obtained from the equation of the balance conservation laws for energy and linear momentum. In this relation the form $`\omega `$ is that of the first degree. If the equations of the balance conservation laws for angular momentum be added to the equations for energy and linear momentum, this form in the evolutionary relation will be the form of the second degree. And in combination with the equation of the balance conservation law of mass this form will be the form of degree 3.
Thus, in the general case the evolutionary relation can be written as
$$d\psi =\omega ^p$$
$`(3.6)`$
where the form degree $`p`$ takes the values $`p=\mathrm{\hspace{0.17em}0},1,2,3`$.. (The evolutionary relation for $`p=\mathrm{\hspace{0.17em}0}`$ is similar to that in the differential forms, and it was obtained from the interaction of energy and time.)
In relation (3.5) the form $`\psi `$ is the form of zero degree. And in relation (3.6) the form $`\psi `$ is the form of $`(p1)`$ degree.
Let us show that the evolutionary relation obtained from the equation of the balance conservation laws proves to be nonidentical.
To do so we shall analyze relation (3.5).
In the left-hand side of evolutionary relation (3.5) there is a differential that is a closed form. This form is an invariant object. The right-hand side of relation (3.5) involves the differential form $`\omega `$, that is not an invariant object because in real processes, as it is shown below, this form proves to be unclosed.
For the form to be closed the differential of the form or its commutator must be equal to zero.
Let us consider the commutator of the form $`\omega =A_\mu d\xi ^\mu `$. The components of the commutator of such a form can be written as follows:
$$K_{\alpha \beta }=\left(\frac{A_\beta }{\xi ^\alpha }\frac{A_\alpha }{\xi ^\beta }\right)$$
$`(3.7)`$
(here the term connected with the manifold metric form has not yet been taken into account).
The coefficients $`A_\mu `$ of the form $`\omega `$ have been obtained either from the equation of the balance conservation law for energy or from that for linear momentum. This means that in the first case the coefficients depend on the energetic action and in the second case they depend on the force action. In actual processes energetic and force actions have different nature and appear to be inconsistent. The commutator of the form $`\omega `$ constructed from the derivatives of such coefficients is nonzero. This means that the differential of the form $`\omega `$ is nonzero as well. Thus, the form $`\omega `$ proves to be unclosed and is not a measurable quantity.
This means that the relation (3.5) involves an unmeasurable term. Such a relation cannot be an identical one.
Hence, without the knowledge of particular expression for the form $`\omega `$, one can argue that for actual processes the relation obtained from the equations corresponding to the balance conservation laws proves to be nonidentical.
In similar manner it can be shown that general relation (3.6) is also nonidentical.
As it was noted the differential forms $`\omega ^p`$ and relations (3.6) are defined on accompanying manifold, that is, on the manifold made up by the trajectories of elements of material system. This manifold is deforming one. The metric forms of such manifold cannot be closed. The differential forms defined on such manifold are evolutionary differential forms, which have been described in the fourth section.
The differential forms $`\omega ^p`$ are evolutionary forms, and relations (3.5) and (3.6) are examples of nonidentical relations for evolutionary forms.
In the Appendix the derivation of nonidentical relations for thermodynamic and gas dynamic systems as well as for the system of charged particles is presented and their brief analysis is given.
Thus, one can see that to the conservation laws for physical fields (exact conservation laws) there correspond the properties of closed inexact forms, whereas to the conservation laws for material media there correspond the properties of evolutionary differential forms.
The nonidentity of the relation obtained from the equations of balance conservation laws means that the equations of balance conservation laws turn out to be nonconjugated (thus, if from the energy equation we obtain the derivative of $`\psi `$ in the direction along the trajectory and from the momentum equation we find the derivative of $`\psi `$ in the direction normal to the trajectory and then we calculate their mixed derivatives, from the condition that the commutator of the form $`\omega `$ is nonzero it follows that the mixed derivatives prove to be noncommutative).
The nonconjugacy of the equations of balance conservation laws reflects the properties of balance conservation laws, namely, their noncommutativity. This property have a governing importance for the evolutionary processes.
### 3.2 Evolutionary process in material medium and origination of physical structures
In this subsection the mechanism of evolutionary processes in material media, which are accompanied by the emergence of physical structure, is described on the basis of the mathematical apparatus of evolutionary differential forms.
The conservation laws are shown to play a governing role in evolutionary processes.
\[To emphasize a connection between the mathematical and physical principles, some of the principles listed above will be included into the titles of some subsections. It will be used double titles, namely, those having the physical meaning and those having the mathematical one. The corresponding mathematical principles are presented in brackets.\]
### Nonequilibrium of the material system. (Nonidentity of the evolutionary relation)
It was shown above that the evolutionary relation that was obtained from the balance conservation law equations proves to be nonidentical. This points to the noncommutativity of the balance conservation laws. By analyzing the behavior of the nonidentical evolutionary relation one can understand to what result the noncommutativity of the balance conservation laws leads.
Before we start analyzing the evolutionary relation some relevant concepts such as a “local domain” of a material system, “accompanying manifolds”, “nonequilibrium” and “locally equilibrium” states of a material system are to be explained.
The local domain of material system. The local domain of material system is the element and its vicinity. In deriving the evolutionary relation with the first degree form (for the balance conservation laws of energy and linear momentum) it was considered the local domain of material system that involves the material system element and its vicinity, this is, the material system element was an element of the local domain. For the evolutionary relation with the second degree form the local domain for the evolutionary relation with the first degree form will serve as the element of the local domain, and so on. For the evolutionary relation with zero degree form the element of the material system will serve as the local domain (rather than the element).
Accompanying manifold. The accompanying manifold is the manifold constructed by the trajectories of the elements of corresponding local domains. For the evolutionary relation with $`p=1`$ the accompanying manifold is the manifold formed from the trajectories of the material system elements because in this case the elements of the material system in itself serve as the elements of local domain. It is to be noted that the accompanying manifold is constructed of trajectories of the elements of the local domain, rather than of material system (see the concept of a “local domain”).
Nonequilibrium and locally equilibrium states of material system. It is evident that the nonequilibrium is connected with actions of some forces. One has to distinguish between external forces and internal ones (in the subsequent analysis one will also deal with potential forces). Here external forces (external actions) mean the forces that act onto the local domain. Internal forces are those that act inside the local domain of the material system. The nonequilibrium is connected with action of internal forces. The nonequilibrium state is the state of a material system when internal forces act in each local domain. If there are no internal forces, the state of material system is in equilibrium. If there are no internal forces only in a particular domain of the material system but there are internal forces in the neighboring local domains, such a state of the material system will be referred to as the state “in local equilibrium”. In this case the total state of the material system is not equilibrium.
We are coming now to the analysis of the evolutionary relation.
It was mentioned above that the noncommutativity of the balance conservation laws is connected with a state of the material system. This is reflected by the evolutionary relation.
It should be emphasized that the evolutionary relation treats a state of the local domain of the material system.
{The evolutionary relation takes into account the interaction of an element of the local domain with its vicinity. The noncommutativity of the balance conservation laws points to the fact that the element and its vicinity prove to be nonconjugated.}
Let us consider evolutionary relation (3.5).
If the evolutionary relation proves to be identical, one can obtain the differential $`d\psi `$ and find the state function $`\psi `$, this will indicate that the material system state is in equilibrium. But if the evolutionary relation be nonidentical, this indicates an absence of the differential $`d\psi `$ and nonequilibrium of the material system state. (Hereafter the differential $`d\psi `$ will be called the state differential as it specifies the material system state. This is a closed form. If the state differential be an exact closed form, this corresponds to the equilibrium system state, whereas, if the state differential be an inexact closed form, this will correspond to the locally equilibrium state).
The evolutionary relation gives a possibility to determine either presence or absence of the differential (the closed form). And this allows us, firstly, to recognize whether the material system state is in equilibrium, in local equilibrium or not in equilibrium, and secondly, to determine the conditions of transition from one state into another (this explains the mechanism of such a transition). If it is possible to determine the differential $`d\psi `$ from the evolutionary relation, this indicates that the system is in equilibrium or locally equilibrium state. And if the differential cannot be determined, then this means that the system is in a nonequilibrium state.
It is evident that if the balance conservation laws be commutative, the evolutionary relation would be identical and from that it would be possible to get the differential $`d\psi `$, this would indicate that the material system is in the equilibrium state.
However, as it has been shown, in real processes the balance conservation laws are noncommutative. The evolutionary relation is not identical and from this relation one cannot get the differential $`d\psi `$. This means that the system state is nonequilibrium.
The nonequilibrium state means that there is an internal force in the material system. It is evident that the internal force originates at the expense of some quantity described by the evolutionary form commutator. (If the evolutionary form commutator be zero, the evolutionary relation would be identical, and this would point to the equilibrium state, i.e. the absence of internal forces.) Everything that gives a contribution into the evolutionary form commutator leads to emergence of the internal force.
Thus, the nonidentity of the evolutionary relation obtained from the equations of the balance conservation laws points to the noncommutativity of the balance conservation laws and the nonequilibrium material system state produced as a result. A quantity described by the evolutionary differential form commutator serves as the internal force.
Each external action, as the result of which a change of physical quantities has been produced, has the nature different from that of the material system itself. For this reason the changed physical quantities cannot directly become the physical quantities of the material system itself. (The noncommutativity of the balance conservation laws does not allow a direct transition of the external actions into the physical quantities of the material system). The changed physical quantities prove to be inconsistent. As a result it arises an unmeasurable quantity that is described by the commutator of the evolutionary form $`\omega ^p`$ and acts as an internal force.
To become the consistent physical quantities of the material system itself, the modified physical quantities have to come to agreement with the properties of the material system. Such transitions, as it will be shown below, are also governed by the balance conservation laws.
### Selfvariation of nonequilibrium state of material system. (Selfvariation of the evolutionary relation)
What does the material system nonequilibrium indicated by the nonidentity of the evolutionary relation results in?
While describing the properties of evolutionary forms it has been shown that, the evolutionary and nonidentical relation is a selfvarying one.
Such a specific feature of the evolutionary relation explains the particulars of the material system, namely, the selfvariation of its nonequilibrium state.
The selfvariation mechanism of the nonequilibrium state of the material system can be understood if analyze the selfvariation of the evolutionary relation go on. For this purpose we have to analyze the topological properties of the evolutionary form commutator.
The evolutionary form in the evolutionary relation is defined on the accompanying manifold that for real processes appears to be the deformable manifold because it is formed simultaneously with a change of the material system state and depends on the physical processes. Such a manifold cannot be a manifold with closed metric forms. Hence, the term containing the characteristics of the manifold will be included into the evolutionary form commutator in addition to the term connected with derivatives of the form coefficients. The interaction between these terms of different nature describes a mutual change of the state of the material system.
Let us examine this with an example of the commutator of the form $`\omega =A_\mu d\xi ^\mu `$ that is included into evolutionary relation (3.5).
We assume that at the beginning the accompanying manifold was that with the first degree closed metric form. In this case the commutator of the form $`\omega `$ can be written as (3.7). If at the next instant any action affects the material system, this commutator turns out to be nonzero. The state of the material system becomes nonequilibrium and it will arise an internal force whose action will lead to a deformation of the accompanying manifold. The metric form commutator of the accompanying manifold, which specifies the deformation, will become nonzero (that is, the metric form of the accompanying manifold will be unclosed). In the commutator of the form $`\omega `$ it will appear an additional term, that specifies a deformation of the manifold and is a commutator of the manifold metric form. If it is possible to define the coefficients of connectedness $`\mathrm{\Gamma }_{\alpha \beta }^\sigma `$ (for a nondifferentiable manifold they are skew-symmetric ones), the form commutator may be written as (1.3), where $`A_\alpha =a_\alpha `$.
The emergence of the second term can only change the commutator and cannot make it zero (because the terms of the commutator have different nature). The further deformation (torsion) of the manifold will go on. This leads to a change of the metric form commutator, produces a change of the evolutionary form and its commutator and so on. Such a process is governed by the nonidentical evolutionary relation and, in turn, produces a change of the evolutionary relation.
The process of selfvariation of the evolutionary relation points to a change of the material system state. But the material system state remains nonequilibrium in this process because the internal forces do not vanish due to the evolutionary form commutator remaining nonzero.
At this point it should be emphasized that such selfvariation of the material system state proceeds under the action of internal (rather than external) forces. That will go on even in the absence of external forces. That is, the selfvariation of the nonequilibrium state of the material system takes place.
Here it should be noted that in a real physical process the internal forces can be increased (due to the selfvariation of the nonequilibrium state of the material system). This can lead to the development of instability in the material system . {For example, this was pointed out in the works by Prigogine . “The excess entropy” in his works is analogous to the commutator of a nonintegrable form for the thermodynamic system. “Production of excess entropy” leads to the development of instability}.
### Transition of the material system into a locally equilibrium state. Origination of the physical structures. (Degenerate transformation. Emergence of closed exterior forms)
Thus, it was shown that in real processes the material system is in a nonequilibrium state (with an internal force). This follows from the analysis of the nonidentical evolutionary relation obtained from the balance conservation law equations.
Now the question arises whether the material system can got rid of the internal force and transfer into the equilibrium state?
The internal force is described by the evolutionary form commutator. But the evolutionary form commutator cannot vanish. This means that the internal force, which is described by the evolutionary form commutator, cannot disappear. That is, the material system cannot be transformed into the equilibrium (without internal forces) state.
However, the material system can change from the nonequilibrium state into the locally equilibrium state. This follows from the evolutionary differential form properties. Under degenerate transformation the identical relation can be obtained from the nonidentical evolutionary relation. That is, from nonidentical relation (3.6) it is obtained the identical on pseudostructure relation
$$d_\pi \psi =\omega _\pi ^p$$
$`(3.8)`$
where the form $`\omega _\pi ^p`$ is one closed on pseudostructure.
The identical relation obtained from the nonidentical evolutionary relation under degenerate transformation integrates the state differential and the closed inexact exterior form. The availability of the state differential $`d_\pi \psi `$ indicates that the material system state becomes a locally equilibrium state (that is, the local domain of the system under consideration changes into the equilibrium state). The availability of the exterior closed inexact form $`\omega _\pi ^p`$ means that the physical structure is present. This shows that the transition of material system into the locally equilibrium state is accompanied by the origination of physical structures.
The conditions of degenerate transformation are connected with symmetries that can be obtained from the coefficients of commutators of evolutionary and metric forms. Such symmetries can be due to the degrees of freedom of material system and its elements. The translational degrees of freedom, internal degrees of freedom of the system elements, and so on can be examples of such degrees of freedom.
As it was noted above, to the degenerate transformation it must correspond vanishing of some functional expressions, such as Jacobians, determinants, the Poisson brackets, residues and others. Vanishing these functional expressions is the closure condition for a dual form. And it should be emphasize once more that the degenerate transformation is realized as the transition from the accompanying noninertial coordinate system to the locally inertial system.
The availability of the degrees of freedom in the material system indicates that it is allowed the degenerate transformation, which, in turns, allows the state of the material systems to be transformed from a nonequilibrium state to a locally equilibrium state. But, for this to take place in reality it is necessary that the additional conditions connected with the degrees of freedom of the material system be realized. It is selfvariation of the nonequilibrium state of the material system describe by the selfvarying evolutionary relation that could give rise to realization of the additional conditions. This can appear only spontaneously because it is caused by internal (rather than external) reasons (the degrees of freedom are the characteristics of the system rather than of external actions).
Under degenerate transformation the evolutionary form differential vanishes only along a certain direction. In other words, the interior differential equal to zero is realized. But in this case the total differential of the evolutionary form is nonzero. The evolutionary form commutator does not vanish. Vanishing the interior differential of the evolutionary form means that there is a closed inexact form, and this points to the locally equilibrium state of the material system. At the same time a nonzero value of the total differential of evolutionary form means that the form remains unclosed. This shows that the total state of the material system remains nonequilibrium.
Thus, from the properties of the nonidentical evolutionary relation and those of the evolutionary form one can see that under realization of the additional condition (which is a condition of degenerate transformation) the transition of the material system state from nonequilibrium to locally equilibrium state can be realized. Such transition is accompanied by emergency of physical structures. It should be emphasized once again that such a transition can occur only spontaneously.
Here we should recall once again that the closed inexact form is a quantity with double meaning, namely, it is both the conservative quantity and the measurable quantity that acts as a potential force. The transition of the material system from nonequilibrium into a locally equilibrium state (which is indicated by the formation of a closed form) means that the unmeasurable quantity described by the nonzero commutator of the nonintegrable differential form $`\omega ^p`$, that acts as an internal force, transforms into the measurable quantity. It is evident that it is just the measurable quantity that acts as a potential force. In other words, the internal force transforms into a potential force.
Thus, the mathematical apparatus of the evolutionary differential forms elucidates a mechanism of the evolutionary process in material media and of the emergency of physical structures. This mechanism involves the following steps.
1) The external actions onto the material system are transformed into the unmeasurable quantity that acts as an internal force and brings the material system into the nonequilibrium state. (The nonzero value of the evolutionary form commutator. Nonidentity of the evolutionary relation obtained from the balance conservation laws).
2) Selfvariation of the nonequilibrium state of the material system. The deformation of accompanying manifold. (Selfvariation of the nonidentical evolutionary relation. The topological properties of the evolutionary form commutator).
3) Realization of the degrees of freedom of the material system in the process of selfvariation of the nonequilibrium state of the system itself. (Degenerate transformations).
4) Transition of the material system from the nonequilibrium state into the locally equilibrium one: the transition of an internal force into a potential force. The emergence of physical structures. (Formation of closed inexact forms and obtaining the differential $`d_\pi \psi `$ that specifies the state of the material system).
{Here it should be emphasized the following.
The evolutionary relation nonidentity that follows from the conservation law noncommutativity just reflects the overdeterminacy of the set of the balance conservation law equations. Actually, a number of the balance conservation law equations is equal to a number of desired physical quantities that specify the material system. But since the physical quantities relate to the same material system, it has to be some connection between them. (This connection is executed by the function that specifies the system state.) And as the physical quantities are related to each other, then the set of the balance conservation law equations proves to be overdetermined one. A realization of the additional conditions, when from the nonidentical evolutionary relation it follows the identical relation, there corresponds to that from the overdetermined set of equations it results the set of consistent equations from which one can find the desired physical quantities. As the additional conditions may be realized only in the discrete manner, then the solutions to this set may be only quantized.
In the book by A.Pais “The Science and the Life of Albert Einstein” the author wrote: “He (Einstein) hoped that the idea of the overdeterminacy will lead to getting the discrete solutions. He also believed that from the future theory it will be possible to derive the partly localized solutions that would correspond to particles that carries the quantized electric charge”.}
## 4 Evolutionary forms: Properties of physical structures. Formation of physical fields and manifolds
In Section 3 the mechanism of evolutionary processes in material media was described, and it was shown that the evolutionary processes lead to origination of the physical structures. (These are just such structures that form physical fields.)
In the present Section the properties of the physical structures and their connection with material media are described. The mechanism of forming physical fields and corresponding manifolds is described.
### 4.1 Characteristics of physical structures. (Characteristics of differential forms)
The exterior form closed on the pseudostructure in combination with the dual form determining the pseudostructure constitute the differential-geometrical structure named the Bi-Structure.
The physical structure is such differential-geometrical structure.
The physical structure is an object obtained by conjugating the conservative physical quantity, which is described by inexact closed exterior form, and the pseudostructure, which is described by relevant dual form.
What characteristics do the physical structures possess?
The closed exterior forms corresponding to physical structures are conservative quantities. These conservative quantities describe certain charges.
Under transition from one structure to another the conservative quantity corresponding to the closed exterior form discretely changes, and the pseudostructure also changes discretely.
Discrete changes of the conservative quantity and pseudostructure are determined by the value of the evolutionary form commutator, which the commutator has at the instant when the physical structure originates. The first term of the evolutionary form commutator obtained from the derivatives of the evolutionary form coefficients controls the discrete change of the conservative quantity. The second one obtained from the derivatives of the metric form coefficients of the initial manifold controls the pseudostructure change.
Spin is the example of the second characteristic. Spin is a characteristic that determines a character of the manifold deformation before origination of the quantum. (The spin value depends on the form degree.)
Breaks of the derivatives of the potential along the direction normal to the potential surface, breaks of the derivative in transition throughout the characteristic surfaces and in transition throughout the wave front, and others are the examples of discrete change of the conservative quantity.
A discrete change of the conservative quantity and that of the pseudostructure produce the quantum that is obtained while going from one structure to another. The evolutionary form commutator formed at the instant of the structure origination determine characteristics of this quantum.
### 4.2 Connection between physical structures originated and material systems. (Identical relation: connection between the state differential and the closed inexact exterior form)
Since closed inexact exterior forms corresponding to physical structure are obtained from the evolutionary relation for the material system, it follows that physical structures are generated by the material systems. (This is controlled by the conservation laws.) The closed exterior forms obtained correspond to the state differential for material system. The differentials of entropy, action, potential and others are the examples of such differentials.
In this manner the physical structures are connected with the material system, its elements, its local domains. The characteristics of physical structure are determined by the characteristics of the material system that generates these physical structures.
The equation of the pseudostructure (dual form) is obtained, as it has been shown, from the condition of degenerate transformation. And this relates to the degrees of freedom of the material system.
The characteristics of inexact closed exterior forms are defined by the characteristics of evolutionary forms following from the balance conservation laws for material media and, hence, depend on the characteristics of material media.
Here it should call attention to one more fact. The material system can generate physical structures only if the system is in nonequilibrium state, that is, if it experience the influence of any actions. As it has been shown, the internal forces arisen are described by the evolutionary form commutator. Therefore, the characteristics of the physical structures arisen will also depend on the characteristics of the evolutionary form commutator describing any actions onto the material system.
In material system the origination of physical structure reveals as a new measurable and observable formation that spontaneously arises in material system. $`\{`$As the examples it can be fluctuations, pulsations, waves, vortices, and creating massless particles.$`\}`$.
In the physical process this formation is spontaneously extracted from the local domain of material system and so it allows the local domain of material system to get rid of an internal force and come into the locally equilibrium state.
The formation created in a local domain of material system (at the cost of unmeasurable quantity that acts in the local domain as an internal force) and liberated from that, begins acting onto the neighboring local domain as a force. This is a potential force, this fact is indicated by the double meaning of the closed exterior form (on the one hand, a conservative quantity, and, on other hand, a potential force). (This action was produced by the material system in itself, and therefore this is a potential action rather than an arbitrary one).
The neighboring domain of the material system works over this action that appears to be external with respect to that. If in the process the conditions of conjugacy of the balance conservation laws turn out to be satisfied again, the neighboring domain will create a formation by its own, and this formation will be extracted from this domain. (If the conjugacy conditions are not realized, the process is finished.) In such a way the formation can move relative to the material system. (Waves are the example of such motions).
The extraction of a formation from the local domain of the system is accompanied by emergence of the break surfaces in the material system. The contact breaks are the examples of such surfaces. These breaks do not propagate relative to the material system as shocks or shock waves do. The black holes may be such break surfaces.
The observed formation and the physical structure are not identical objects. If the wave be such a formation, then the wave front is the physical structure. In this case the wave element is a minipseudostructure.
How is the created formation connected with a change of physical quantities of the material system?
Assume that at some instant the local domain of the material system was in equilibrium. That is, its physical quantities, for example, energy and momentum were consistent and simultaneously measurable physical quantities. Then under the effect of external (with respect to the local domain) actions the physical quantities were changed and ceased to be consistent measurable quantities. When the degrees of freedom of the material system are realized, this allows the physical quantities, which were changed at the expense of external actions, to redistribute in such a way as to become measurable physical quantities, namely, the inherent (corresponding to the nature of material system) quantities of material system.
It is evident that the transition from the initial measurable physical quantities to new physical quantities realized is discrete one.
That is, the proper measurable physical quantities are changed discretely. This discrete change of physical quantities is revealed as a formation created.
### Potential forces. (Duality of closed exterior forms as conservative quantities and as potential forces)
As it was shown above, the unmeasurable quantity, that acts as an internal force and has been stored at the cost of all external actions giving contribution into the commutator, is converted into a measurable quantity that acts as a potential force. This is indicated by the availability of a closed inexact form that can correspond to the potential force.
Where, from what, and on what does the potential force act?
The potential force is an action of the created (quantum) formation onto the local domains of the material system over which it is translated. And if the internal force acts in the interior of the local domain of the material system (and it caused it to deform), the potential force acts onto the neighboring domain. The local domain gets rid of its internal force and modifies it into a potential force that acts onto the neighboring domains. An unmeasurable quantity, that acts in a local domain as an internal force, is transformed into a measurable quantity of the observable formation (and the physical structure as well) that is emitted from the local domain and acts onto the neighboring domain as a force equal to this quantity.
The potential forces, as well as the internal forces, originate at the cost of the external actions, but the potential forces (unlike interior ones) are connected with the measurable quantities.
Unlike arbitrary external forces, the potential forces are those that originate at the expense of external actions processed by the material system.
If the external actions equal zero (the evolutionary form commutator be equal to zero), then internal and potential forces equal zero.
Thus, one has to recognize the forces of three types: 1) external forces (external actions), 2) internal forces that originate in local domains of material system due to the fact that the physical quantities of the material system changed by external actions turn out to be inconsistent, and 3) the potential forces are forces of the action of the formations (corresponding to physical structures) onto a material system.
The potential forces are the source of originating internal forces in the neighboring domains of the material system, on which the potential forces act (such as the forces that are external with respect to that domain). For this reason the total state of a material system can remain nonequilibrium just without additional external actions (nonpotential forces).
The potential force, whose value is conditioned by the quantity of the commutator of the evolutionary form $`\omega ^p`$ at the instant of the formation production, acts normally to the pseudostructure, i.e. with respect to the integrating direction, along which the interior differential (the closed form) is formed. The potential forces are described, for example, by jumps of the derivatives in the direction normal to the characteristics, to the potential surfaces and so on (as well as the physical structure characteristics). This corresponds to the fact that the evolutionary form commutators along these directions are nonzero.
The duality of the closed inexact form as a conservative quantity and as a potential force shows that the potential forces are the action of formations corresponding to the physical structures onto the material system.
Here the following should be pointed out. The physical structures are generated by local domains of the material system. They are the elementary physical structures. By combining with one another they can form the large-scale structures and physical fields.
### Characteristics of the formation created: intensity, vorticity, absolute and relative speeds of propagation of the formation. (Value of the evolutionary form commutator, the properties of material system)
As it was already mentioned, in the material system a created physical structure is revealed as an observable formation. It is evident that the characteristics of the formation, as well as those of the created physical structure, are determined by the evolutionary form and its commutator and by the material system characteristics.
Analysis of the formation and its characteristics allows a better understanding of the specific features of physical structures.
Since the formation is a result of converting an unmeasurable quantity described by the evolutionary form commutator into a measurable physical quantity, it is evident that the intensity of the formation created (as well as the discrete change of the physical structure characteristics) is controlled by the quantity that was stored by the evolutionary form commutator at the instant when the formation appeared.
The first term of the commutator constructed of the derivatives of the form coefficients controls the intensity of the formation, whereas the second term that specifies the deformation of the accompanying manifold (bending, torsion, curvature) is fixed as any internal characteristics of the formation originated (which corresponds to vorticity, for example).
In the preceding subsection it was shown that the formation emerged in the local domain of material system acts onto the neighboring local domain of material system. Such action is determined by potential force. It is evident that the formation intensity is revealed as a potential force.
The other characteristics of the formation obserrved are absolute and relative speeds of the formation propagation.
As it was pointed out above, the observed formation that appears in material system moves along material system. The physical structure is a front of such moving formation.
Here it should be emphasized once again that the moving formation at each instant appears as a newly appeared formation.
The absolute propagation speed of the formation originated (speed in the inertial frame of reference) is obtained from the condition of degenerate transformation that corresponds to conjugacy of the balance conservation laws. As it was already pointed out, these conditions are determined by the material system characteristics and are connected with the degrees of freedom of material system.
These conditions are the differential equations of pseudostructure. They specify the rate of formatting pseudostructure, namely, the speed of the front of the formation emerged. This is just an absolute speed of the formation emerged.
Here it should be emphasized that the speed of the formation originated, though it is determined by the material system characteristics, is not a parameter of the system itself. This is a quantity that at every instant while the formation moves with respect to material system is realized anew as the condition of conjugacy of the balance conservation laws.
If the material system is homogeneous, the speed of translation will have the same values. However, it is not constant because it is formed anew at every instant of the evolutionary processes.
The relative speed (speed in the accompanying frame of reference) is a speed of the formation translation relative to material system.
The relative speed is equal to the absolute speed minus the velocity of the local domain elements of material system (or of the elements of material system if $`p=1`$). That is, the relative speed of the observed formation is determined by the degrees of freedom of material system and by the velocities of the elements of local domain.
In such a way the following correspondence between the characteristics of the formations emerged and characteristics of the evolutionary forms, of the evolutionary form commutators and of the material system is established:
1) an intensity of the formation (a potential force) $``$ the value of the first term in the commutator of nonintegrable form at the instant when the formation is created;
2) vorticity (an analog of spin) $``$ the second term in the commutator that is connected with the metric form commutator;
3) an absolute speed of propagation of the created formation (the speed in the inertial frame of reference) $``$ additional conditions connected with degrees of freedom of material system;
4) a speed of the formation propagation relative to material system $``$ additional conditions connected with degrees of freedom of material system and the velocity of elements of local domain.
Analysis of formations originated in material system that correspond to physical structures enables us to clarify some properties of physical structures.
The rate of varying the pseudostructures that are described by dual forms defines the absolute speed of formations. That is, the pseudostructures are connected with the front of the formation translation. Since a certain physical quantity is conserved on the pseudostructure (the closed form), the pseudostructure is a level surface. The equation of pseudostructure is the equation of eikonal surface. (The eikonal is an example of physical structure.)
It can be shown that the equations of the characteristic surfaces, the surfaces of potential (of simple layer, double layer), the residue equations and so on, obtained from the equations of mathematical physics serve as the equations for pseudostructures. {In the papers the connection of equations for one, two, … eikonals with the equations for characteristics, with the Hamilton equation, and others was shown}.
The mechanism of creating the pseudostructures lies at the basis of forming the pseudometric surfaces and their transition into metric spaces (see the next subsection). It should be pointed out that the eigenvalues and the coupling constants appear as the conjugacy conditions for exterior or dual forms, the numerical constants are the conjugacy conditions for exact forms.
### 4.3 Formation of pseudometric and metric spaces. (Integration of the nonidentical evolutionary relation)
The mechanism of forming pseudometric and metric spaces is connected with the creation of pseudostructures.
In Subsection (1.5) it has been shown that the evolutionary nonidentical relation containing the evolutionary form of degree $`p`$ may generate the pseudostructure and the closed (on the pseudostructure) exterior forms of sequential degrees $`k=p,\mathrm{},k=0`$. And yet the pseudostructure dimensionality depends on the dimension of space on which the exterior forms are defined.
What is implied by the concept “space”?
While deriving the evolutionary relation two frames of reference were used and, correspondingly, two spatial objects. The first frame of reference is an inertial one, which is connected with the space where material system is situated and is not directly connected with material system. This is an inertial space, it is a metric space. (As it will be shown below, this space is also formed by the material system itself.) The second frame of reference is a proper one, it is connected with the accompanying manifold, which is not a metric manifold.
As it is known, the form degree cannot be greater than the space dimension. Therefore, if the dimension of the inertial space is $`n`$, the maximal degree of the form will be $`n`$. For material system in such a space there work the balance conservation laws that are convoluted into the evolutionary relation of the degree $`p`$. In the real processes practically all conservation laws allowed in the space of a given dimension interact with one another. For given $`n`$ the value of $`p`$ will be practically a maximal value, that is, $`p=n`$.
Assume that $`n=2`$ and $`p=2`$. If the form $`\omega ^2`$ is an unclosed form, the commutator of this form will act in the space of $`n+i=2+i`$ dimension (here the additional dimension is denoted by an imaginary unit as it does not commutate with other dimensions). This means that the deformed accompanying manifold will not be imbedded into the original inertial space of dimension $`n`$. This leads to that the dimension of the space formed increases by one.
It was shown above that the evolutionary relation of degree $`p`$ can generate (in the presence of the degenerate transformations) closed forms of the degree $`0kp`$ on the pseudostructures. While generating closed forms of sequential degrees $`k=p`$, $`k=p1`$, …, $`k=0`$ the pseudostructures of dimensions $`(n+1k)`$: 1, …, $`n+1`$ are obtained. As a result of transition to the exact closed form of zero degree the metric structure of the dimension $`n+1`$ is obtained. Under influence of the external action (and in the presence of degrees of freedom) the material system can transfer the initial inertial space into the space of the dimension $`n+1`$. $`\{`$It is known that the skew-symmetric tensors of the rank $`k`$ correspond to closed exterior differential forms, and the pseudotensors of the rank $`(Nk)`$, where $`N`$ is the space dimension, correspond to the relevant dual forms. The pseudostructures correspond to such tensors, but on the space formed with the dimension $`n+1`$. That is, $`N=n+1`$}.
Under the effect of external actions (and in the presence of degrees of freedom) the material system can convert the initial inertial space of the dimension $`n`$ into the space of the dimension $`n+1`$. Thus, a certain stage of forming the metric space is completed. Every material system has the cycle that includes four stages $`(n=0,1,2,3)`$. The cycle ends and a new cycle can begin. (This corresponds to one system being embedded into another one). The mechanism of formation of the pseudostructures and the metric structures can explain, in particular, how the internal structure of the elements of material system is formed. (See Subsection 6.1).
So it can be seen that the inertial spaces are not absolute spaces where actions are developed, they are spaces generated by material systems.
The pseudo-Riemann and pseudo-Euclidean spaces and others can be regarded as examples of pseudostructures and spaces that are formed in a similar manner. The Riemann and Euclidean spaces are the example of metric manifolds obtained in changing to exact forms. Below we present a brief analysis of the space corresponding to gravitational field.
### Space of gravitational field
Material system (medium), which generates gravitational field, is a cosmological system. What can be said about the pseudo-Riemann manifold and Riemann space? The distinctive property of the Riemann manifold is an availability of the curvature. This means that the metric form commutator of the third degree is nonzero. Hence, it does not equal zero the evolutionary form commutator of the third degree $`p=3`$, which involves into itself the metric form commutator. That is, the evolutionary form that enters into the evolutionary relation is unclosed, and the relation is nonidentical.
When realizing pseudostructures of the dimensions $`1`$, $`2`$, $`3`$, and $`4`$ and obtaining the closed inexact forms of the degrees $`k=3`$, $`k=2`$, $`k=1`$, and $`k=0`$, the pseudo-Riemann space is formed, and the transition to the exact form of zero degree corresponds to the transition to Riemann space.
It is well known that while obtaining the Einstein equations it was suggested that there are fulfilled the conditions : the Bianchi identity is satisfied, the coefficients of connectedness are symmetric, the condition that the coefficients of connectedness are the Christoffel symbols, and an existence of the transformation, under which the coefficients of connectedness vanish. These conditions are the conditions of realization of the degenerate transformations for nonidentical relations obtained from the evolutionary relation of the degree $`p=3`$ and after changing to the exact relations. In this case to the Einstein equation there corresponds the identical relations of the first degree.
The above described mechanism of forming the manifolds elucidates the connection between the space and material objects. In his paper S.Weinberg gives more than one historical concepts of the space. He wrote that the connection of the space with material objects was pointed out by Leibnitz who believed that there is no philosophical necessity in any concept of space apart from that following from the connections with material objects. In addition, S.Weinberg cites another similar concept, namely, the Mach principle, which claims that in the definition of inertial system the masses of Earth and celestial bodies play a role. The idea of physical space as a continuum whose properties are governed by the matter was realized by A.Einstein. The above described mechanism of formatting manifolds is one more substantiation of the connection of the space with material objects.
### 4.4 Forming physical fields. Classification of physical structures. (Parameters of closed and dual forms)
Since the physical structures are generated by numerous local domains of material system and at numerous instants of realizing various degrees of freedom of material system, it is evident that they can generate fields. In this manner physical fields are formed. To obtain the physical structures that form a given physical field, one has to examine the material system corresponding to this field and the appropriate evolutionary relation. In particular, in the Appendix it is shown that, for to obtain the thermodynamical structures (fluctuations, phase transitions, etc), one has to analyze the evolutionary relation for thermodynamical systems, to obtain the gas dynamic ones (waves, jumps, vortices, pulsations) one has to employ the evolutionary relation for gas dynamic systems, for the electromagnetic field one must employ a relation obtained from equations for charged particles.
Closed forms that correspond to physical structures are generated by the evolutionary relation having the parameter $`p`$ that defines a number of interacting balance conservation laws. Therefore, the physical structures can be classified by the parameter $`p`$. The other parameter is a degree of closed forms generated by the evolutionary relation. As it was shown above, the evolutionary relation of degree $`p`$ can generate closed forms of degree $`0kp`$. Therefore, physical structures can be classified by the parameter $`k`$ as well. Closed exterior forms of the same degree realized in spaces of different dimensions prove to be distinguishable because the dimension of the pseudostructures, on which the closed forms are defined, depends on the space dimension. As a result, the space dimension also specifies the physical structures. This parameter determines the properties of physical structures rather than their type.
Hence, from the analysis of the evolutionary relation one can see that the type and the properties of the differential-geometrical structures and, consequently, of the physical structures (and, accordingly, of physical fields) for a given material system depend on a number of interacting balance conservation laws $`p`$, on the degree of closed forms realized $`k`$, and on the space dimension. By introducing a classification with respect to $`p`$, $`k`$, and space dimension we can understand an internal connection of various physical fields and interactions. Such a connection will be considered in Subsection (6.1).
## 5 Conservation laws. Symmetries. Causality
### 5.1 Conservation laws
Here it should be emphasized once again the role of the conservation laws in evolutionary processes and the connection between the conservation laws for material systems and those for physical fields (connection between balance conservation laws and exact conservation laws).
From the description of evolutionary process it was seen that in the evolutionary processes proceeded in material medium the balance conservation laws for energy, linear momentum, angular momentum, and mass play a controlling and governing role. Such a role of the conservation laws for material systems is due to its peculiarities, that is, they turn out to be noncommutative ones.
This noncommutativity of balance conservation laws reflects the fact that the external actions cannot directly be converted into quantities (measurable) of material system, and hence, they produce a certain nonmeasurable quantity, which acts as internal force and converts the material system into the nonequilibrium state.
The interaction of the noncommutative balance conservation laws controls the process of selfvarying the nonequilibrium state of material system. Such a process can cause the realization of the degrees of freedom of material system (if they are available). The degrees of freedom allow the nonmeasurable quantity, which acts as external force, to be converted into the measurable quantities of the material system itself. The physical structures are the result of realizing the measurable quantities of material system. They appear in material system as the production of new formations. And yet the material system is changed into the locally equilibrium state.
Since the exact conservation laws correspond to physical structures, the origination of physical structure points to a realization of the exact conservation law. From here one can see the connection between the balance and exact conservation laws.
The physical structures that correspond to the exact conservation laws are produced by material system in the evolutionary processes, which are based on the interaction of the noncommutative balance conservation laws.
Noncommutativity of the balance conservation laws and their controlling role in the evolutionary processes accompanied by emerging physical structures practically have not been taken into account in the explicit form anywhere. The mathematical apparatus of evolutionary differential forms enables one to take into account and to describe these points.
### 5.2 Symmetries
The exterior and evolutionary skew-symmetric differential forms, which describe the conservation laws, disclose thereby the properties and specific features of symmetries.
The gauge symmetries, which are interior symmetries of the field theory equations, are connected with the conservation laws for physical fields. These symmetries are those of (inexact) closed forms.
The closure property of the exterior form means that any objects, namely, elements of the exterior form, components of elements, elements of the form differential, exterior and dual forms, and others, turn out to be conjugated. And the conjugacy is possible only if there are symmetries of one or other type.
The gauge transformations of field theory, which are nondegenerate transformations of the closed exterior differential forms, are connected with the gauge symmetries. Since the closed exterior differential form is a differential (a total one if the form is exact, or an interior one on pseudostructure if the form is inexact), it is obvious that the closed form proves to be invariant under all transformations that conserve the differential. The unitary transformations (0-form), the tangent and canonical transformations (1-form), the gradient and gauge transformations (2-form) and so on are examples of such transformations. These are gauge transformations for spinor, scalar, vector, and tensor (3-form) fields.
The internal symmetries in field theory are those of closed exterior differential forms, whereas the external symmetries are symmetries of relevant dual forms.
It has been shown that the closed exterior forms and relevant dual forms, which correspond to the conservation laws for physical fields, are obtained from the evolutionary forms, which describe the balance conservation laws for material media. This proceeds under degenerate transformation, which is connected with the degrees of freedom of material system. The conditions of degenerate transformation defines the closed dual form. From this it follows that the external symmetries, namely, the symmetries of dual forms, are due to the degrees of freedom of material system. It is for this reason the exterior symmetries are spatial-temporal symmetries.
The realization of the closed dual form, which proceeds due to the degrees of freedom of material system, leads to realization of the closed exterior form, that is, to the conjugacy of the differential form elements, and emergency of internal symmetries. From this one can see the connection between internal and external symmetries.
Whereas the internal symmetries are connected with the conservation laws for physical fields, the external symmetries caused by the degrees of freedom of material media are connected with the balance conservation laws for material media.
The nondegenerate transformations are connected with internal symmetries, and the degenerate transformations of evolutionary forms are connected with external symmetries.
### 5.3 Causality
The Encyclopedia gives the following definition of the concept of “Causality”. “The causality is a genetic connection between separate states of types and forms of matter in the process of its motion and development. The emergency of any objects and systems and their time-developments have their own grounds in the preceding states of matter; these grounds are known as causes and the changes produced by them are referred to as consequences. The essence of the causality is a production of the consequence by the cause; the consequence, which is governed by the cause, produces the reverse action to the cause”.
A determinacy of the above described evolutionary processes in material systems matches with this definition of the causality.
It can be understood that all external actions onto material system lead to the evolutionary processes. This is a cause of the evolutionary processes. The emergency of physical structures and formation of physical fields are the consequences of the evolutionary processes in material systems.
From the description of evolutionary processes one can see a connection between physical fields and material systems. Physical fields are generated by material systems. And the noncommutative balance conservation laws for material systems control these processes.
The causality of evolutionary processes in material systems that lead to emergency of physical structures is justified by the mathematical apparatus of evolutionary and exterior differential forms describing the balance conservation laws in material systems.
The evolutionary process in material system and the emergency of physical structures can take place, if
1) the material system is subject to an external action (the evolutionary form commutator in the evolutionary relation obtained from the balance conservation laws is nonzero),
2) the material system possesses the degrees of freedom (there are the conditions of degenerate transformation, under which from the nonidentical evolutionary relation an identical relation is obtained),
3) the degrees of freedom of material system have to be realized, that is possible only under selfvariations of the nonequilibrium state of material system (the conditions of degenerate transformation have to be satisfied, this is possible under selfvariation of the nonidentical evolutionary relation).
If these conditions (causes) are fulfilled, in the material system physical structures arise (this is indicated by the presence of closed inexact form obtained from the identical relation).
It is these structures that form physical fields.
Note that to each physical field it is assigned its own material system. The question of what physical system does correspond to a particular physical field is still an open question. Examples of such material systems are the thermodynamic, gas dynamical, cosmological systems, the system of charged particles, and so on. Maybe, for elementary particles the physical vacuum is such a system.
The evolutionary process may lead to creation of elements of its own material system (while obtaining the exact form of zero degree).
The emergency of physical structures in the evolutionary process proceeds spontaneously and is manifested as an emergency of certain observable formations. In this manner the causality of emerging various observable formations in material media is explained. Such formations and their manifestations are fluctuations, turbulent pulsations, waves, vortices, creating massless particles, and others.
The existing field theories that are invariant ones are based on some postulates. The investigation performed enables us to make the following conclusion. The postulates, which lie at the basis of the existing field theories, correspond to the conditions of conjugacy of the balance conservation laws for material systems that generate physical structures forming physical fields.
In physics there exists “the causality principle” . This principle establishes the permitted limits of mutual influence of physical events. “The causality principle” is a statement that is significantly narrower than the general philosophic understanding of causality formulated above. It does not clarify the causal and consequential connection of physical phenomena.
## 6 A certain aspects of quantum field theory and approaches to general field theory
The importance of evolutionary forms in mathematical physics and field theory consists in the fact that they disclose the mechanism of forming physical fields and substantiates the determinacy of corresponding processes. From the description of evolutionary processes in material media using the evolutionary skew-symmetric differential forms it follows that material media (material systems) generate physical fields. And the conservation laws, which turn out to be noncommutative, control these processes.
Connection of physical fields with material media and the classification of physical structures based on parameters that specify the conservation laws for material media enable one to see internal relations of different physical fields and their common foundations.
### 6.1 On interactions and classification of physical structures and physical fields
As it was shown above, the type of physical structures (and, accordingly, of physical fields) generated by the evolutionary relation depends on the degree of differential forms $`p`$ and $`k`$ and on the dimension of original inertial space $`n`$ (here $`p`$ is the degree of the evolutionary form in the nonidentical relation that is connected with a number of interacting balance conservation laws, and $`k`$ is the degree of closed form generated by the nonidentical relation). By introducing the classification by numbers $`p`$, $`k`$, $`n`$ one can understand the internal connection between various physical fields. Since physical fields are the carriers of interactions, such classification enables one to see the connection between interactions.
On the basis of the properties of evolutionary forms that correspond to the conservation laws one can suppose that such a classification may be presented in the form of the table given below. This table corresponds to elementary particles.
{It should be emphasized the following. Here the concept of “interaction” is used in a twofold meaning: an interaction of the balance conservation laws that relates to material systems, and the physical concept of “interaction” that relates to physical fields and reflects the interactions of physical structures, namely, it is connected with the exact conservation laws}.
Recall that the interaction of balance conservation laws for energy and linear momentum corresponds to the value $`p=1`$, with the balance conservation law for angular momentum in addition this corresponds to the value $`p=2`$, and with the balance conservation law for mass in addition it corresponds to the value $`p=3`$. The value $`p=0`$ corresponds to interaction between time and the balance conservation law for energy.
TABLE
In the Table names of the particles created are given. Numbers placed near particle names correspond to the space dimension. In braces {} the sources of interactions are presented. In the next to the last row we present massive particles (the elements of material system) formed by interactions (the exact forms of zero degree obtained by sequential integrating the evolutionary relations with the evolutionary forms of degree $`p`$ correspond to these particles). In the bottom row the dimension of the metric structure created is presented.
From the Table one can see the correspondence between the degree $`k`$ of the closed forms realized and the type of interactions. Thus, $`k=0`$ corresponds to the strong interaction, $`k=1`$ corresponds to the weak interaction, $`k=2`$ corresponds to the electromagnetic interaction, and $`k=3`$ corresponds to the gravitational interaction. The degree $`k`$ of the closed forms realized and the number of interacting balance conservation laws determine the type of interactions and the type of particles created. The properties of particles are governed by the space dimension. The last property is connected with the fact that closed forms of equal degrees $`k`$, but obtained from the evolutionary relations acting in spaces of different dimensions $`n`$, are distinctive because they are defined on the pseudostructures of different dimensions (the dimension of the pseudostructure $`(n+1k)`$ depends on the dimension of initial space $`n`$). For this reason the realized physical structures with closed forms of equal degrees $`k`$ are distinctive in their properties.
In the Table one cycle of forming physical structures is presented. This cycle involves four levels, to each of which there correspond their own values of $`p`$ ($`p=0,1,2,3`$) and space dimension $`n`$. The structures formed in the previous cycle serve as a source of interactions for the first level of new cycle. The consequent cycles reflect the properties of material systems subsequently imbedded. And yet a given level has specific properties that are inherent characteristics of the same level in another cycles. This can be seen, for example, from comparison of the cycle described and the cycle in which to the exact forms there correspond conductors, semiconductors, dielectrics, and neutral elements. The properties of the elements of the third level, namely, of neutrons in one cycle and of dielectrics in another are identical to the properties of so called ”magnetic monopole” .
The Table presented provides the idea about the dimension of pseudostructures and metric structures.
In the bottom row of the Table the dimension $`N`$ of the metric structure formed is presented. From original space of the dimension $`0`$ the metric space of the dimension $`1`$ (it can occurs to be time) can be realized. From space of the dimension $`1`$ the metric space of the dimension $`2`$ (time and coordinate) can appear and so on. From original space of the dimension $`3`$ it can be formed the metric space of the dimension $`4`$ (time and $`3`$ coordinates). Such space is convoluted and a new dimension cannot already be realized. This corresponds to ending the cycle. (Such metric space with corresponding physical quantity defined by the exact exterior form is the element of new material system.)
The method of studying evolutionary processes developed on the basis of exterior and evolutionary differential forms with using the properties of the conservation laws may serve as an approach to general field theory.
### 6.2 Mathematical apparatus of skew-symmetric differential forms as the basis of general field theory
At Section 2 it was shown the role of closed exterior forms in field theory. The properties of closed exterior differential forms explicitly or implicitly manifest themselves essentially in all formalisms of field theory, such as the Hamilton formalism, tensor approaches, group methods, quantum mechanics equations, the Yang-Mills theory and others.
Such a role of closed exterior forms in field theory is related to the fact that these forms reflect the properties of the conservation laws for physical fields. The differential-geometrical structures described by closed (inexact) exterior forms and relevant dual forms are physical structures that make up physical fields. The degrees of closed exterior forms set the classification of physical fields and interactions. To the strong, weak, electromagnetic, and gravitational interactions there correspond the closed exterior forms of zeroth, first, second, and third degrees. Gauge transformations for spinor, scalar, vector, and tensor fields are transformations of closed exterior forms of zeroth, first, second, and third degrees.
Such properties of closed exterior forms can be useful in establishing the unified field theory.
And the theory of skew-symmetric differential forms, which unites the theory of closed exterior forms and the theory of evolutionary forms (generating the closed inexact exterior forms), can serve as an approach to the general field theory. Such a theory enables one not only to describe physical fields, but also shows how are the physical fields produced, what does generate them, what is a cause of these processes.
The field theories that are based on exact conservation laws allow to describe physical fields. However, because these theories are invariant ones they cannot answer the question about the mechanism of originating physical structures that form physical fields. The originating physical structures and forming physical fields are evolutionary processes, and hence they cannot be described by the invariant field theories. Only evolutionary theory can do this. The mathematical apparatus of evolutionary differential forms can serve as the mathematical apparatus of such theory.
The basic mathematical foundations that describe the evolutionary process in material systems, and the mechanism of originating physical structures evidently must be included into the evolutionary field theory.
It should be noted that it is rather difficult to realize all these mathematical foundations and in many cases this turns out to be impossible. The difficulties may be caused by a derivation of the evolutionary relation that describes the mechanism of originating physical structures and forming physical fields. To do this one has to know the equations of the balance conservation laws for material systems that generate given physical fields. The problems can be also caused by the fact that these equations have to be written in the frame of reference that is related to the deforming accompanying manifold. Moreover, this can lead to difficulties in the process of obtaining closed exterior forms from evolutionary relation, because for to do this it is necessary to obtain the additional conditions that correspond to the degrees of freedom of material system and have to be realized by themselves when changing the evolutionary relation.
These problems may appear to be practically unsolvable.
However, a knowledge of the basic mathematical principles of evolutionary theory may be helpful while studying the mechanism of originating physical fields.
The results of qualitative investigations of evolutionary processes using the mathematical apparatus of evolutionary differential forms enables one to see the common properties that unify all physical fields. The physical fields are generated by material media, and at the basis of this it lies the interaction of the noncommutative conservation laws of energy, linear momentum, angular momentum, and mass for material media. This explains the causality of physical phenomena and clarifies the essence of postulates that lie at the basis of existing field theories.
These results enable one to understand what do the properties of physical structures depend on and with which parameters is the classification of physical structures connected, and hence to see the internal connections between various physical fields.
The properties of physical structures depend primarily on which material systems (media) generate physical structures (but the physical structures generated by different material media possess common properties as well).
One of the parameters, according to which it is possible to classify physical structures and physical fields, is the number of interacting balance conservation laws. This is the parameter $`p`$ that ranges from 0 to 3. (Recall that the case $`p=1`$ corresponds to interaction of the balance conservation laws of energy and linear momentum, the case $`p=2`$ does to that of energy, linear momentum, and angular momenta, the case $`p=3`$ corresponds to interaction of the balance conservation laws of energy, linear and angular momenta, and mass, and to $`p=0`$ it corresponds an interaction between time and the balance conservation law of energy.) This parameter, which is the evolutionary form degree that enters into the evolutionary relation, specifies a type of physical fields. (So, the electromagnetic field is obtained from interaction between the balance conservation laws of energy and linear and angular momenta. The gravitational field is obtained as the result of interactions between the balance conservation laws of energy, linear momentum, angular momentum, and mass.)
The other parameter is the degree of closed differential forms realized from given evolutionary relation. The values of these parameters denoted by $`\kappa `$ range from $`p`$ to $`0`$. This parameter, which corresponds to physical structures realized, characterizes the connection between physical structures and exact conservation laws. It specifies a type of interaction of physical fields.
Since the realization of physical structures proceeds discretely, this emphasizes a quantum character of physical fields.
One more parameter is the dimension $`n`$ of the space in which the physical structures are generated. This parameter points to the fact that the physical structures, which belong to common type of the exact conservation laws, can be distinguished by their space structure. (The classification with respect to these parameters not only elucidates the connections between the physical fields generated by material media, but explains the mechanism of creating the elements of material media themselves and demonstrates the connections between material media as well).
By comparison of the invariant and evolutionary approaches to field theory one can state the following. Physical fields are described by invariant field theory that is based on exact conservation laws. The properties of closed exterior differential forms lie at the basis of mathematical apparatus of the invariant theory. The mechanism of forming physical fields can be described only by evolutionary theory. The evolutionary theory that is based on the balance conservation laws for material systems is just such a theory. It is evident that as the general field theory it must serve a theory that involves the basic mathematical foundations of the evolutionary and invariant field theories. As an approach to such general field theory it can serve the theory of skew-symmetric differential forms, which unites the theories of closed exterior forms and evolutionary forms.
The papers on the theory of evolutionary skew-symmetric differential forms written in 2002-2005 years are in ArXive (http://arXiv.org/find)
Appendix
The analysis of balance conservation laws for
thermodynamic and gas dynamic systems and for
the system of charged particles
### Thermodynamic systems
The thermodynamics is based on the first and second principles of thermodynamics that were introduced as postulates . The first principle of thermodynamics, which can be written in the form
$$dE+dw=\delta Q$$
$`(A.1)`$
follows from the balance conservation laws for energy and linear momentum (but not only from the conservation law for energy). This is analogous to the evolutionary relation for the thermodynamic system. Since $`\delta Q`$ is not a differential, relation (A.1) which corresponds to the first principle of thermodynamics, as well as the evolutionary relation, appears to be a nonidentical relation. This points to a noncommutativity of the balance conservation laws (for energy and linear momentum) and to a nonequilibrium state of the thermodynamic system.
If condition of the integrability be satisfied, from the nonidentical evolutionary relation, which corresponds to the first principle of thermodynamics, it follows an identical relation. It is an identical relation that corresponds to the second principle of thermodynamics.
If $`dw=pdV`$, there is the integrating factor $`\theta `$ (a quantity which depends only on the characteristics of the system), where $`1/\theta =pV/R`$ is called the temperature $`T`$ . In this case the form $`(dE+pdV)/T`$ turns out to be a differential (interior) of some quantity that referred to as entropy $`S`$:
$$(dE+pdV)/T=dS$$
$`(A.2)`$
If the integrating factor $`\theta =1/T`$ has been realized, that is, relation (A.2) proves to be satisfied, from relation (A.1), which corresponds to the first principle of thermodynamics, it follows
$$dS=\delta Q/T$$
$`(A.3)`$
This is just the second principle of thermodynamics for reversible processes. This takes place when the heat input is the only action onto the system.
If in addition to the heat input the system experiences a certain mechanical action (for example, an influence of boundaries), we obtain
$$dS>\delta Q/T$$
$`(A.4)`$
that corresponds to the second principle of thermodynamics for irreversible processes.
In the case examined above the differential of entropy (rather than entropy itself) becomes a closed form. $`\{`$In this case entropy manifests itself as the thermodynamic potential, namely, the function of state. To the pseudostructure there corresponds the state equation that determines the temperature dependence on the thermodynamic variables$`\}`$.
### Gas dynamical systems
We take the simplest gas dynamical system, namely, a flow of ideal (inviscid, heat nonconductive) gas .
Assume that the gas (the element of gas dynamic system) is a thermodynamic system in the state of local equilibrium (whenever the gas dynamic system itself may be in nonequilibrium state), that is, it is satisfied the relation
$$Tds=de+pdV$$
$`(A.5)`$
where $`T`$, $`p`$ and $`V`$ are the temperature, the pressure and the gas volume, $`s`$ and $`e`$ are entropy and internal energy per unit volume.
Let us introduce two frames of reference: an inertial one that is not connected with material system and an accompanying frame of reference that is connected with the manifold formed by the trajectories of the material system elements.
The equation of the balance conservation law of energy for ideal gas can be written as
$$\frac{Dh}{Dt}\frac{1}{\rho }\frac{Dp}{Dt}=\mathrm{\hspace{0.17em}0}$$
$`(A.6)`$
where $`D/Dt`$ is the total derivative with respect to time (if to denote the spatial coordinates by $`x_i`$ and the velocity components by $`u_i`$, $`D/Dt=/t+u_i/x_i`$). Here $`\rho =1/V`$ and $`h`$ are respectively the mass and the entalpy densities of the gas.
Expressing entalpy in terms of internal energy $`e`$ using the formula $`h=e+p/\rho `$ and using relation (A.5), the balance conservation law equation (A.6) can be put to the form
$$\frac{Ds}{Dt}=\mathrm{\hspace{0.17em}0}$$
$`(A.7)`$
And respectively, the equation of the balance conservation law for linear momentum can be presented as
$$\text{grad}s=(\text{grad}h_0+𝐔\times \text{rot}𝐔𝐅+𝐔/t)/T$$
$`(A.8)`$
where $`𝐔`$ is the velocity of the gas particle, $`h_0=(𝐔𝐔)/2+h`$, $`𝐅`$ is the mass force. The operator $`grad`$ in this equation is defined only in the plane normal to the trajectory.
Since the total derivative with respect to time is that along the trajectory, in the accompanying frame of reference equations (A.7) and (A.8) take the form:
$$\frac{s}{\xi ^1}=\mathrm{\hspace{0.17em}0}$$
$`(A.9)`$
$$\frac{s}{\xi ^\nu }=A_\nu ,\nu =2,\mathrm{}$$
$`(A.10)`$
where $`\xi ^1`$ is the coordinate along the trajectory, $`s/\xi ^\nu `$ is the left-hand side of equation (A.8), and $`A_\nu `$ is obtained from the right-hand side of relation (A.8).
Equations (A.9) and (A.10) can be convoluted into the equation
$$ds=A_\mu d\xi ^\mu $$
$`(A.11)`$
where $`A_\mu d\xi ^\mu =\omega `$ is the first degree differential form (here $`A_1=0`$,$`\mu =1,\nu `$).
Relation (A.11) is the evolutionary relation for gas dynamic system (in the case of local thermodynamic equilibrium). Here $`\psi =s`$. $`\{`$It worth notice that in the evolutionary relation for thermodynamic system the dependence of entropy on thermodynamic variables is investigated (see relation (A.5)), whereas in the evolutionary relation for gas dynamic system the entropy dependence on the space-time variables is considered$`\}`$.
Relation (A.11) appears to be nonidentical. To make it sure that this is true one must inspect the commutator of the form $`\omega `$.
Nonidentity of the evolutionary relation points to the nonequilibrium state and the development of the gas dynamic instability. Since the nonequilibrium state is produced by internal forces that are described by the commutator of the form $`\omega `$, it becomes evident that the cause of the gas dynamic instability is something that contributes into the commutator of the form $`\omega `$.
One can see (see (A.8)) that the development of instability is caused by not a simply connectedness of the flow domain, nonpotential external (for each local domain of the gas dynamic system) forces, a nonstationarity of the flow.
{In the case when gas is nonideal equation (A.9) can be written in the form
$$\frac{s}{\xi ^1}=A_1$$
where $`A_1`$ is the expression that depends on the energetic actions (transport phenomena: viscous, heat-conductive). In the case of reacting gas the extra terms connected with the chemical nonequilibrium state are added. These factors contributes to the commutator of the form $`\omega `$.}
All these factors lead to emergency of internal forces, that is, to nonequilibrium state and to development of various types of instability.
And yet for every type of instability one can find the appropriate term giving contribution to the evolutionary form commutator, which is responsible for this type of instability. Thus, there is an unambiguous connection between the type of instability and the terms that contribute to the evolutionary form commutator in the evolutionary relation. {In the general case one has to consider the evolutionary relations that correspond to the balance conservation laws for angular momentum and mass as well}.
As it was shown above, under realization of additional degrees of freedom it can take place the transition from the nonequilibrium state to the locally equilibrium one, and this process is accompanied by emergency of physical structures. The gas dynamic formations that correspond to these physical structures are shocks, shock waves, turbulent pulsations and so on. Additional degrees of freedom are realized as the condition of the degenerate transformation, namely, vanishing of determinants, Jacobians of transformations, etc. These conditions specify the integral surfaces (pseudostructures): the characteristics (the determinant of coefficients at the normal derivatives vanishes), the singular points (Jacobian is equal to zero), the envelopes of characteristics of the Euler equations and so on. Under crossing throughout the integral surfaces the gas dynamic functions or their derivatives undergo the breaks.
### Electromagnetic field
The system of charged particles is a material medium, which generates electromagnetic field.
If to use the Lorentz force $`𝐅=\rho (𝐄+[𝐔\times 𝐇]/c)`$, the local variation of energy and linear momentum of the charged matter (material system) can be written respectively as : $`\rho (𝐔𝐄)`$, $`\rho (𝐄+[𝐔\times 𝐇]/c)`$. Here $`\rho `$ is the charge density, $`𝐔`$ is the velocity of charged matter. These variations of energy and linear momentum are caused by energetic and force actions and are equal to values of these actions. If to denote these actions by $`Q^e`$, $`𝐐^i`$, the balance conservation laws can be written as follows:
$$\rho (𝐔𝐄)=Q^e$$
$`(A.12)`$
$$\rho (𝐄+[𝐔\times 𝐇]/c)=𝐐^i$$
$`(A.13)`$
After eliminating the characteristics of material system (the charged matter) $`\rho `$ and $`𝐔`$ by using the Maxwell-Lorentz equations , the left-hand sides of equations (A.12), (A.13) can be expressed only in terms of the strengths of electromagnetic field, and then one can write equations (A.12), (A.13) as
$$c\text{div}𝐒=\frac{}{t}I+Q^e$$
$`(A.14)`$
$$\frac{1}{c}\frac{}{t}𝐒=𝐆+𝐐^𝐢$$
$`(A.15)`$
where $`𝐒=[𝐄\times 𝐇]`$ is the Pointing vector, $`I=(E^2+H^2)/c`$, $`𝐆=𝐄\text{div}𝐄+\text{grad}(𝐄𝐄)(𝐄\text{grad})𝐄+\text{grad}(𝐇𝐇)(𝐇\text{grad})𝐇`$.
Equation (A.14) is widely used while describing electromagnetic field and calculating energy and the Pointing vector. But equation (A.15) does not commonly be taken into account. Actually, the Pointing vector $`𝐒`$ must obey two equations that can be convoluted into the relation
$$d𝐒=\omega ^\mathrm{𝟐}$$
$`(𝐀\mathbf{.16})`$
Here $`d𝐒`$ is the state differential being 2-form and the coefficients of the form $`\omega ^2`$ (the second degree form) are the right-hand sides of equations (A.14) and (A.15). It is just the evolutionary relation for the system of charged particles that generate electromagnetic field.
By analyzing the coefficients of the form $`\omega ^2`$ (obtained from equations (A.14) and (A.15), one can assure oneself that the form commutator is nonzero. This means that from relation (A.16) the Pointing vector cannot be found. This points to the fact that there is no such a measurable quantity (a potential).
Under what conditions can the Pointing vector be formed as a measurable quantity?
Let us choose the local coordinates $`l_k`$ in such a way that one direction $`l_1`$ coincides with the direction of the vector $`𝐒`$. Because this chosen direction coincides with the direction of the vector $`𝐒=[𝐄\times 𝐇]`$ and hence is normal to the vectors $`𝐄`$ and $`𝐇`$, one obtains that $`\text{div}𝐒=s/l_1`$, where $`S`$ is a module of $`𝐒`$. In addition, the projection of the vector $`𝐆`$ on the chosen direction turns out to be equal to $`I/l_1`$. As a result, after separating from vector equation (A.15) its projection on the chosen direction equations (A.14) and (A.15) can be written as
$$\frac{S}{l_1}=\frac{1}{c}\frac{I}{t}+\frac{1}{c}Q^e$$
$`(A.17)`$
$$\frac{S}{t}=c\frac{I}{l_1}+c𝐐^i$$
$`(A.18)`$
$$0=𝐆^{\prime \prime }c𝐐^{\prime \prime i}$$
Here the prime relates to the direction $`l_1`$, double primes relate to the other directions. Under the condition $`dl_1/dt=c`$ from equations (A.17) and (A.18) it is possible to obtain the relation in differential forms
$$\frac{S}{l_1}dl_1+\frac{S}{t}dt=\left(\frac{I}{l_1}dl_1+\frac{I}{t}dt\right)+(Q^edt+𝐐^idl_1)$$
$`(A.19)`$
Because the expression within the second braces in the right-hand side is not a differential (the energetic and force actions have different nature and cannot be conjugated), one can obtain a closed form only if this term vanishes:
$$(Q^edt+𝐐^idl_1)=\mathrm{\hspace{0.17em}0}$$
$`(A.20)`$
that is possible only discretely (rather than identically).
In this case $`dS=\mathrm{\hspace{0.17em}0}`$, $`dI=\mathrm{\hspace{0.17em}0}`$ and the modulus of the Pointing vector $`S`$ proves to be a closed form, i.e. a measurable quantity. The integrating direction (the pseudostructure) will be
$$\frac{S/t}{S/l_1}=\frac{dl_1}{dt}=c$$
$`(A.21)`$
The quantity $`I`$ is the second dual invariant.
Thus, the constant $`c`$ entered into the Maxwell equations is defined as the integrating direction.
1. Cartan E., Les Systemes Differentials Exterieus ef Leurs Application Geometriques. -Paris, Hermann, 1945.
2. Schutz B. F., Geometrical Methods of Mathematical Physics. Cambrige University Press, Cambrige, 1982.
3. Tonnelat M.-A., Les principles de la theorie electromagnetique et la relativite. Masson, Paris, 1959.
4. Encyclopedia of Mathematics. -Moscow, Sov. Encyc., 1979 (in Russian).
5. Bott R., Tu L. W., Differential Forms in Algebraic Topology. Springer, NY, 1982.
6. Wheeler J. A., Neutrino, Gravitation and Geometry. Bologna, 1960.
7. Haywood R. W., Equilibrium Thermodynamics. Wiley Inc. 1980.
8. Fock V. A., Theory of space, time, and gravitation. -Moscow, Tech. Theor. Lit., 1955 (in Russian).
9. Encyclopedic dictionary of the physical sciences. -Moscow, Sov. Encyc., 1984 (in Russian).
10. Dirac P. A. M., The Principles of Quantum Mechanics. Clarendon Press, Oxford, UK, 1958.
11. Pauli W. Theory of Relativity. Pergamon Press, 1958.
12. Dafermos C. M. In ”Nonlinear waves”. Cornell University Press, Ithaca-London, 1974.
13. Clark J. F., Machesney M., The Dynamics of Real Gases. Butterworths, London, 1964.
14. Tolman R. C., Relativity, Thermodynamics, and Cosmology. Clarendon Press, Oxford, UK, 1969.
15. Petrova L. I., On the problem of development of the flow instability. //Basic Problems of Physics of Shock Waves. Chernogolovka, 1987, V. 1, P. 1, 304-306 (in Russian).
16. Prigogine I., Introduction to Thermodynamics of Irreversible Processes. –C.Thomas, Springfild, 1955.
17. Rumer Yu. B. Investigations on 5-Optics. GITTL, Moscow, 1956 (in Russian).
18. Einstein A. The Meaning of Relativity. Princeton, 1953.
19. Weinberg S., Gravitation and Cosmology. Principles and applications of the general theory of relativity. Wiley & Sons, Inc., N-Y, 1972.
20. Dirac P. A. M., Proc. Roy. Soc., A133, 60 (1931).
21. Dirac P. A. M., Phys. Rev., 74, 817 (1948).
22. Liepman H. W., Roshko A., Elements of Gas Dynamics. Jonn Wiley, New York, 1957
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.